Rydberg Controlled-NOT Gates

• Rydberg CNOT gates are two-qubit controlled operations that harness strong, long-range Rydberg excitations to implement quantum logic.
• Various schemes, including blockade, anti-blockade, and geometric approaches, offer flexible routes to realize both CNOT and controlled-phase gates.
• Recent advances in pulse engineering and system optimization have achieved gate fidelities exceeding 99%, though they demand precise calibration of laser parameters.

Rydberg controlled-NOT gates are two-qubit controlled-$X$ operations realized by exploiting the strong, long-range interaction between Rydberg excitations, or by using Rydberg-mediated effective interactions in related photonic and solid-state settings. In the standard computational basis $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$, the gate maps $|00\rangle\to|00\rangle$, $|01\rangle\to|01\rangle$, $|10\rangle\to|11\rangle$, and $|11\rangle\to|10\rangle$. Across the literature, this action is obtained either directly through blockade, transition slow-down, anti-blockade, or optimized pulse shaping, or indirectly by implementing a controlled-phase gate such as $CZ=\mathrm{diag}(1,1,1,-1)$ and surrounding it with local basis changes (Shi, 2017, Han et al., 2014, Yu et al., 18 Mar 2026).

1. Logical structure and computational role

The logical status of a Rydberg CNOT depends on the chosen primitive. In several neutral-atom schemes the gate is implemented directly as the standard matrix

$$\mathrm{CNOT}= \begin{pmatrix} 1&0&0&0\ 0&1&0&0\ 0&0&0&1\ 0&0&1&0 \end{pmatrix},$$

with the first qubit as control and the second as target (Shi, 2017). In other schemes the primitive is a controlled-phase or a more general entangling unitary, and CNOT is obtained from

$$\mathrm{CNOT}_{A\to B}=(I\otimes H)\,CZ\,(I\otimes H),$$

or from local equivalence between a nontrivial entangling gate and a canonical controlled operation (Han et al., 2014, Zhao et al., 2017).

The qubit encoding is not unique. Some proposals use hyperfine ground states $|0\rangle,|1\rangle$ as the logical manifold and a Rydberg level $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$0 as an auxiliary state; others encode directly in $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$1 and $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$2, so that $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$3 and $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$4 (Yu et al., 18 Mar 2026, Zhao et al., 2017). This distinction is operationally important because direct $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$5-$\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$6 encodings simplify some geometric constructions, whereas hyperfine encodings reduce logical exposure to Rydberg decay outside the entangling interval.

A recurring structural point is that a Rydberg CNOT need not be the native entangling primitive even when the hardware is optimized for CNOT-class logic. The nonadiabatic geometric two-qubit gate proposed in one blockade-based scheme is explicitly presented as a generic entangling gate,

$\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$7

and the route to CNOT or $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$8 is through arbitrary one-qubit geometric gates and local conjugations rather than a direct closed-form CNOT pulse sequence (Zhao et al., 2017).

2. Blockade-based neutral-atom implementations

The canonical neutral-atom mechanism is the Rydberg blockade: excitation of one atom to $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$9 shifts the Rydberg level of a nearby atom and suppresses its excitation. A compact direct realization was given by Shi in a three-pulse protocol. Pulse 1 maps $|00\rangle\to|00\rangle$0 with a $|00\rangle\to|00\rangle$1 pulse on the control; pulse 2 is a target $|00\rangle\to|00\rangle$2-type $|00\rangle\to|00\rangle$3 pulse with equal-magnitude couplings $|00\rangle\to|00\rangle$4 and $|00\rangle\to|00\rangle$5 and duration $|00\rangle\to|00\rangle$6; pulse 3 returns $|00\rangle\to|00\rangle$7. When the control is $|00\rangle\to|00\rangle$8, blockade freezes the target. When the control is $|00\rangle\to|00\rangle$9, the target pulse swaps $|01\rangle\to|01\rangle$0, and the net mapping is exactly CNOT (Shi, 2017).

A conceptually different blockade construction uses a single global off-resonant Rydberg pulse applied symmetrically to both atoms. In that scheme, one computational state is coupled to $|01\rangle\to|01\rangle$1, $|01\rangle\to|01\rangle$2 is dark, $|01\rangle\to|01\rangle$3 and $|01\rangle\to|01\rangle$4 behave as single-atom off-resonant excitations with $|01\rangle\to|01\rangle$5, and $|01\rangle\to|01\rangle$6 evolves with the collective frequency $|01\rangle\to|01\rangle$7. Choosing $|01\rangle\to|01\rangle$8 and appropriate approximate solutions for $|01\rangle\to|01\rangle$9, where $|10\rangle\to|11\rangle$0, yields a diagonal controlled-phase gate that is locally equivalent to $|10\rangle\to|11\rangle$1, and hence to CNOT through Hadamards on the target (Han et al., 2014).

A further blockade-regime variant replaces excitation annihilation by transition slow-down. In this approach a ground-Rydberg cycling in one atom slows down a Rydberg-involved state transition of a nearby atom, and a two-pulse sequence with a target phase flip implements the exact computational mapping

$|10\rangle\to|11\rangle$2

The gate duration is approximately $|10\rangle\to|11\rangle$3, where $|10\rangle\to|11\rangle$4 is a negligible transient time to implement a phase change in the pulse, and the required ratio is $|10\rangle\to|11\rangle$5 (Shi, 2021).

3. Geometric and holonomic Rydberg CNOT constructions

Nonadiabatic geometric quantum computation introduces a different design principle: the entangling action is generated by cyclic evolution and parallel-transport conditions rather than by purely dynamical phase accumulation. In the blockade-based geometric proposal, two identical two-level Rydberg atoms are driven with

$|10\rangle\to|11\rangle$6

and, under $|10\rangle\to|11\rangle$7, the effective Hamiltonian reduces to

$|10\rangle\to|11\rangle$8

with a bright state $|10\rangle\to|11\rangle$9 and a dark state $|11\rangle\to|10\rangle$0 in $|11\rangle\to|10\rangle$1. For pulse area $|11\rangle\to|10\rangle$2, the resulting unitary is the entangling $|11\rangle\to|10\rangle$3 quoted above. The paper does not explicitly call it CNOT or $|11\rangle\to|10\rangle$4; instead, it relies on the standard theorem that any entangling two-qubit gate plus arbitrary local unitaries is universal, and it supplies arbitrary one-qubit geometric gates of the form $|11\rangle\to|10\rangle$5 (Zhao et al., 2017).

An anti-blockade holonomic route realizes CNOT directly as a parameter point of an arbitrary controlled-unitary family. In that setting each qubit is encoded in two hyperfine ground states, $|11\rangle\to|10\rangle$6 and $|11\rangle\to|10\rangle$7, with an auxiliary Rydberg state $|11\rangle\to|10\rangle$8. Large single-atom detunings and the anti-blockade condition $|11\rangle\to|10\rangle$9 reduce the dynamics to the effective three-level manifold $CZ=\mathrm{diag}(1,1,1,-1)$0, with

$CZ=\mathrm{diag}(1,1,1,-1)$1

Writing $CZ=\mathrm{diag}(1,1,1,-1)$2 and $CZ=\mathrm{diag}(1,1,1,-1)$3 produces bright and dark states and enables NHQC$CZ=\mathrm{diag}(1,1,1,-1)$4 pulse design through inverse engineering of $CZ=\mathrm{diag}(1,1,1,-1)$5 and $CZ=\mathrm{diag}(1,1,1,-1)$6. The resulting unitary is

$CZ=\mathrm{diag}(1,1,1,-1)$7

and the choice $CZ=\mathrm{diag}(1,1,1,-1)$8 gives

$CZ=\mathrm{diag}(1,1,1,-1)$9

which is the logical CNOT with atom 1 as control and atom 2 as target (Yu et al., 18 Mar 2026).

These two geometric strands differ in their relation to CNOT. The blockade-holonomy construction furnishes a universal entangling gate set from which CNOT is compiled. The anti-blockade NHQC$$\mathrm{CNOT}= \begin{pmatrix} 1&0&0&0\ 0&1&0&0\ 0&0&0&1\ 0&0&1&0 \end{pmatrix},$$0 construction yields CNOT as an explicit point in a continuous controlled-$$\mathrm{CNOT}= \begin{pmatrix} 1&0&0&0\ 0&1&0&0\ 0&0&0&1\ 0&0&1&0 \end{pmatrix},$$1 family that also contains $$\mathrm{CNOT}= \begin{pmatrix} 1&0&0&0\ 0&1&0&0\ 0&0&0&1\ 0&0&1&0 \end{pmatrix},$$2 and controlled-Hadamard as parameter choices (Zhao et al., 2017, Yu et al., 18 Mar 2026).

4. Pulse engineering, fidelities, and error channels

A major recent trend is to treat the Rydberg CNOT as a waveform-design problem rather than a fixed pulse-sequence problem. In a native two-photon scheme for $$\mathrm{CNOT}= \begin{pmatrix} 1&0&0&0\ 0&1&0&0\ 0&0&0&1\ 0&0&1&0 \end{pmatrix},$$3, qubits are encoded in $$\mathrm{CNOT}= \begin{pmatrix} 1&0&0&0\ 0&1&0&0\ 0&0&0&1\ 0&0&1&0 \end{pmatrix},$$4 and $$\mathrm{CNOT}= \begin{pmatrix} 1&0&0&0\ 0&1&0&0\ 0&0&0&1\ 0&0&1&0 \end{pmatrix},$$5, the Rydberg state is $$\mathrm{CNOT}= \begin{pmatrix} 1&0&0&0\ 0&1&0&0\ 0&0&0&1\ 0&0&1&0 \end{pmatrix},$$6, and the interaction is modeled by a Förster-resonant exchange with $$\mathrm{CNOT}= \begin{pmatrix} 1&0&0&0\ 0&1&0&0\ 0&0&0&1\ 0&0&1&0 \end{pmatrix},$$7 at $$\mathrm{CNOT}= \begin{pmatrix} 1&0&0&0\ 0&1&0&0\ 0&0&0&1\ 0&0&1&0 \end{pmatrix},$$8. A single Gaussian red pulse with constant blue gives $$\mathrm{CNOT}= \begin{pmatrix} 1&0&0&0\ 0&1&0&0\ 0&0&0&1\ 0&0&1&0 \end{pmatrix},$$9; optimizing the red width raises this to $$\mathrm{CNOT}_{A\to B}=(I\otimes H)\,CZ\,(I\otimes H),$$0; dual-Gaussian red and blue pulses reach $$\mathrm{CNOT}_{A\to B}=(I\otimes H)\,CZ\,(I\otimes H),$$1; and adding a shaped two-photon detuning

$$\mathrm{CNOT}_{A\to B}=(I\otimes H)\,CZ\,(I\otimes H),$$2

yields raw fidelities $$\mathrm{CNOT}_{A\to B}=(I\otimes H)\,CZ\,(I\otimes H),$$3, $$\mathrm{CNOT}_{A\to B}=(I\otimes H)\,CZ\,(I\otimes H),$$4, and $$\mathrm{CNOT}_{A\to B}=(I\otimes H)\,CZ\,(I\otimes H),$$5 for $$\mathrm{CNOT}_{A\to B}=(I\otimes H)\,CZ\,(I\otimes H),$$6, together with conservative lower bounds $$\mathrm{CNOT}_{A\to B}=(I\otimes H)\,CZ\,(I\otimes H),$$7 and $$\mathrm{CNOT}_{A\to B}=(I\otimes H)\,CZ\,(I\otimes H),$$8 after technical imperfections are included (Li et al., 2023).

Single-pulse global optimization with a low-parameter Fourier ansatz has also been used directly in the weak van der Waals regime. For two qubits separated by $$\mathrm{CNOT}_{A\to B}=(I\otimes H)\,CZ\,(I\otimes H),$$9, and excluding fluctuation of Rydberg interactions, the reported single-pulse two-qubit CNOT fidelity is $|0\rangle,|1\rangle$0; in the same framework the three-qubit CNOT-class gate reaches $|0\rangle,|1\rangle$1 (Li et al., 2021). This suggests that reducing pulse count alone does not remove the need to optimize the full many-body trajectory in the computational and leakage subspaces.

Off-resonant single-pulse controlled-phase gates provide another performance benchmark for CNOT-class operations. With $|0\rangle,|1\rangle$2, $|0\rangle,|1\rangle$3, $|0\rangle,|1\rangle$4, and $|0\rangle,|1\rangle$5, the simulated average fidelity is $|0\rangle,|1\rangle$6 and the minimum fidelity is $|0\rangle,|1\rangle$7; with Doppler shift $|0\rangle,|1\rangle$8, the minimum fidelity remains $|0\rangle,|1\rangle$9 (Han et al., 2014). Although this primitive is a controlled-phase gate, the error budget is directly relevant to CNOT after the standard local basis changes.

The transition-slow-down protocol makes the speed–error trade-off especially explicit. For realistic parameters it gives sub-$\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$00s duration about $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$01, with predicted CNOT or Bell-state fidelities $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$02 at $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$03 and $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$04 at $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$05. The same analysis identifies Rydberg decay, Doppler dephasing, and laser amplitude mismatch as the dominant nonidealities rather than blockade leakage (Shi, 2021).

Across these schemes the main error channels recur with different weights: spontaneous emission from the intermediate and Rydberg states, blockade or anti-blockade mismatch, Doppler broadening, laser phase noise, laser intensity fluctuations, thermal motion, and inhomogeneous coupling. The geometric literature adds a further qualification: geometric robustness is tied to maintaining the designed cyclic path and pulse-area conditions rather than eliminating sensitivity to calibration altogether (Zhao et al., 2017, Yu et al., 18 Mar 2026).

5. Multiqubit, non-local, photonic, and related extensions

Rydberg CNOT physics extends naturally to multi-target and distributed operations. Building on the Rydberg–EIT proposal of Müller et al., Farouk and coauthors showed that heteronuclear Förster engineering can separate strong control–target interactions from weaker target–target interactions. In their optimized schemes the fidelity of a $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$06 gate, defined as one control and four targets, can be up to $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$07, while the fidelity of a $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$08 gate can be up to $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$09 under experimentally feasible conditions (2206.12176).

Non-local CNOTs can also be assembled from local Rydberg controlled-unitary primitives plus entanglement distribution. In the anti-blockade holonomic framework, a local CNOT is extended to distant qubits by entanglement transfer in a chain, Bell-basis manipulations, recovery operations $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$10, and a final local controlled-$\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$11 gate on the remote pair. The same construction is used for four-qubit entanglement transformations, including GHZ-to-cluster and GHZ-to-$\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$12 conversions (Yu et al., 18 Mar 2026).

Photonic realizations replace stationary qubits by stored or propagating optical excitations. A deterministic photonic $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$13 gate can be implemented by storing both control and target photons within an atomic ensemble using non-Rydberg EIT followed by a fast, single-step Rydberg excitation with global lasers; the paper concludes that with realistic experimental parameters $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$14 fidelity is achievable, and local Hadamards convert this $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$15 into CNOT (Khazali, 2022). An experimental photon-photon gate based on Rydberg interactions reported a controlled-NOT truth table with a fidelity of $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$16 and an entangling-gate fidelity of $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$17, both post-selected upon detection of a control and a target photon (Tiarks et al., 2018).

The broader notion of a Rydberg CNOT also appears outside neutral-atom arrays. A scheme based on the four-level Rydberg structure of a surface electron above liquid helium encodes two logical qubits in the $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$18 levels, uses an EIT-based three-level mechanism involving $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$19, and reports a CNOT gate fidelity of $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$20 with experimentally achievable parameters (Wang et al., 2023).

6. Conceptual distinctions, trade-offs, and recurring misconceptions

One persistent misconception is that “Rydberg CNOT gate” always denotes a direct blockade-based controlled-$\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$21 pulse sequence. In fact, several influential schemes implement $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$22, a controlled phase, or a generic entangling unitary and only then infer CNOT by local rotations. This is the case for the single-global-pulse neutral-atom controlled-phase gate, the nonadiabatic geometric blockade gate, and the single-step photonic $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$23 gate (Han et al., 2014, Zhao et al., 2017, Khazali, 2022).

A second misconception is that strong blockade is the only useful interaction regime. The literature includes anti-blockade NHQC$\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$24 controlled-$\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$25 gates with $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$26, transition slow-down gates in the blockade regime that do not operate by excitation annihilation, and globally optimized weak-van der Waals gates that explicitly avoid the requirement $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$27 as the defining design principle (Yu et al., 18 Mar 2026, Shi, 2021, Li et al., 2021).

A third distinction concerns pulse-count reduction. Single-pulse or single-cycle constructions remove switching overhead and can suppress some intrinsic errors, but they generally replace that simplicity with stricter requirements on waveform calibration, relative phase control, detuning management, and spatial homogeneity. The optimized two-photon neutral-atom CNOT and the single-step photonic $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$28 both exemplify this trade-off: fewer optical segments, but stronger dependence on accurately shaped $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$29, $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$30, $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$31, or on the relative intensity modulation of the excitation lasers (Li et al., 2023, Khazali, 2022).

Finally, the role of CNOT within larger Rydberg gate sets is not fixed. Some architectures continue to treat two-qubit CNOT or $\{|00\rangle,|01\rangle,|10\rangle,|11\rangle\}$32 as the universal primitive; others introduce alternative few-qubit gates intended to reduce CNOT depth. A three-qubit Rydberg parity gate based on dark-state resonance has been proposed as a substitute in parity-controlled circuits, and its authors explicitly argue that it can be a better substitute of the CNOT gate because it decreases circuit noise by reducing circuit depth (Rej et al., 10 Jan 2026). This does not displace the centrality of Rydberg CNOTs, but it clarifies that the most efficient neutral-atom logic layer may combine direct CNOT implementations with native multiqubit primitives rather than decomposing all control logic into repeated two-qubit gates.