A quantum processor based on coherent transport of entangled atom arrays – Nature

Dynamic reconfiguration in 2D tweezer arrays

Our experiments utilize the same apparatus described previously in ref. ²⁵. Inside our vacuum cell, ⁸⁷Rb atoms are loaded from a magneto-optical trap into a backbone array of programmable optical tweezers generated by an SLM⁵⁷. Atoms are rearranged in parallel into defect-free target positions in this SLM backbone⁵⁷ by additional optical tweezers generated from a crossed 2D AOD. Following the rearrangement procedure, we transfer selected atoms from the static SLM traps back into the mobile AOD traps, and then move these mobile atoms to their starting positions in the quantum circuit. During this entire process, the atoms are cooled with polarization-gradient cooling. Before running the quantum circuit, we take a camera image of the atoms in their initial starting positions, and following the circuit we take a final camera image to detect qubit states |0⟩ (atom presence) and |1⟩ (atom loss, following resonant pushout). We postselect all data on finding perfect rearrangement of the AOD and SLM atoms before running the circuit. In all experiments here, each atom remains in a single static or single mobile trap throughout the duration of the quantum circuit^20,58,59.

The crossed AOD system is composed of two independently controlled AODs (AA Opto Electronic DTSX-400) for x and y control of the beam positions. Both AODs are driven by independent arbitrary waveforms, which are generated by a dual-channel arbitrary waveform generator (M4i.6631-x8 by Spectrum Instrumentation) and then amplified through independent MW amplifiers (Minicircuits ZHL-5W-1). The time-domain arbitrary waveforms are composed of multiple frequency tones corresponding to the x and y positions of columns and rows, which are independently changed as a function of time for steering around the AOD-trapped atoms dynamically; the full x and y waveforms are calculated by adding together the time-domain profile of all frequency components with a given amplitude and phase for each component. For running quantum circuits, we program the positions of the AOD atoms at each gate location and then smoothly interpolate (with a cubic profile) the AOD frequencies as a function of time between gate positions. The cubic profile enacts a constant jerk onto the atoms, which allows us to move roughly five-to-ten-times faster (without heating and loss) than if we move at a constant velocity (linear profile). In our movement protocol, we only do stretches, compressions and translations of the AOD trap array: that is, the AOD rows and columns never cross each other to avoid atom loss and heating associated with two frequency components crossing each other.

We homogenize the AOD tweezer intensity throughout the whole atom trajectory to minimize dephasing induced by a time-varying magnitude of differential light shifts. To this end, we use a reference camera in the image plane to gauge the intensity of each AOD tweezer at each gate location and homogenize by varying the amplitude of each frequency component; during motion between two locations, we interpolate the amplitude of each individual frequency component.

The SLM tweezer light (830 nm) and the AOD tweezer light (828 nm) are generated by two separate, free-running titanium:sapphire lasers (M Squared, 18-W pump). Projected through a 0.5 numerical aperture objective, the SLM tweezers have a waist of roughly 900 nm (roughly 1,000 nm for AODs). When loading the atoms, the trap depths are about 2π × 16 MHz, with radial trap frequencies of about 2π × 80 kHz, and when running quantum circuits the trap depths are about 2π × 4 MHz, with radial trap frequencies of about 2π × 40 kHz.

Raman laser system

Fast, high-fidelity single-qubit manipulations are critical ingredients of the quantum circuits demonstrated in this work. To this end, we use a high-power 795-nm Raman laser system for driving global single-qubit rotations between magnetic sublevel m_F = 0 clock states. This Raman laser system is based on dispersive optics, developed and described in ref. ²⁶. The 795-nm light (Toptica TA pro, 1.8 W) is phase-modulated by an electro-optic modulator (Qubig), which is driven by microwaves at 3.4 GHz (Stanford Research Systems SRS SG384) that are doubled to 6.8 GHz and amplified. The laser phase modulation is converted to amplitude modulation for driving Raman transitions through use of a chirped Bragg grating (Optigrate)²⁶. IQ (in-phase and quadrature) control of the SG384 is used for frequency and phase control of the microwaves, which are imprinted onto the laser amplitude modulation and thus give us direct frequency and phase control over the hyperfine qubit drive.

The Raman laser illuminates the atom plane from the side in a circularly polarized elliptical beam with waists of 40 μm and 560 μm on the thin axis and the tall axis, respectively, with a total average optical power of 150 mW on the atoms. The large vertical extent ensures <1% inhomogeneity across the atoms, and shot-to-shot fluctuations in the laser intensity are also <1%. For Figs. 1–3, we operate our Raman laser at a blue-detuned intermediate-state detuning of 180 GHz, resulting in two-photon Rabi frequencies of 1 MHz and an estimated scattering error per π pulse of 7 × 10⁻⁵ (that is, 1 scattering event per 15,000 π pulses)²⁶. For Fig. 4, to shorten the duration of the coherent mapping pulse sequence, we increase the Raman laser power and operate at a smaller blue-detuned intermediate-state detuning of 63 GHz, with a corresponding two-photon Rabi frequency of 3.2 MHz and an estimated scattering error per π pulse of 2 × 10⁻⁴.

Robust single-qubit rotations

For almost all single-qubit rotations in this work (other than XY8 and XY16 self-correcting sequences), we implement robust single-qubit rotations in the form of composite pulse sequences. These composite pulse sequences are well known in the NMR community^27,60 and can be highly insensitive to pulse errors such as amplitude or detuning miscalibrations. Our dominant source of coherent single-qubit errors arise from ≲1% amplitude drifts and inhomogeneity across the array; as such, we primarily use the ‘BB1’ (broadband 1) pulse sequence, which is a sequence of four pulses that implements an arbitrary rotation on the Bloch sphere while being insensitive to amplitude errors to sixth order^27,60. We benchmark the performance of these robust pulses in Extended Data Fig. 3a. Furthermore, by applying a train of BB1 pulses, we find an accumulated error consistent with the estimated scattering limit (not plotted here), suggesting that the scattering limit roughly represents our single-qubit rotation infidelities (about 3 × 10⁻⁴ error per BB1 pulse owing to the increased length of the composite pulse sequence). Randomized benchmarking⁶¹ can be applied in future studies to further study single-qubit rotation fidelity.

Qubit coherence and dynamical decoupling

In our 830-nm traps, hyperfine qubit coherence is characterized by inhomogeneous dephasing time ({T}_{2}^{ast }) = 4 ms (not plotted here), T₂ = 1.5 s (XY16 with 128 total π pulses) and relaxation time T₁ = 4 s (including atom loss) (Extended Data Fig. 3b, c). All of our experiments in this work are performed in a d.c. magnetic field of 8.5 G. Coherence can be further improved by using further-detuned optical tweezers (with trap depth held constant, the tweezer differential lightshifts decrease as 1/Δ and 1/T₁ decreases as 1/Δ³ (ref. ⁶²), where Δ is the detuning of the trap wavelength) and shielding against magnetic field fluctuations. For practical QEC operation, atom loss can be detected in a hardware-efficient manner⁴⁶ and the atom then replaced from a reservoir, which could in principle be continuously reloaded by a magneto-optical trap for reaching arbitrarily deep circuits.

All of our transport sequences^20,58,59 are accompanied by dynamical decoupling sequences. The number of pulses we use is a trade-off between preserving qubit coherence while minimizing pulse errors. We interchange between two types of dynamical decoupling sequence: XY8 and XY16 sequences, composed of phase-alternated individual π pulses that are self-correcting for amplitude and detuning errors^28,63, and Carr–Purcell–Meiboom–Gill (CPMG)-type dynamical decoupling sequences composed of robust BB1 pulses. The CPMG-BB1 sequence is more robust to amplitude errors but incurs more scattering error. We empirically optimize for any given experiment by choosing between these different sequences and a variable number of decoupling π pulses, optimizing on either single-qubit coherence (including the movement) or the final signal. Typically, our decoupling sequences are composed of a total 12–18 π pulses.

Movement effects on atom heating and loss

We study here the effects of movement on atom loss and heating in the harmonic oscillator potential given by the tweezer trap. Motion of the trap potential is equivalent to the non-inertial frame of reference where the harmonic oscillator potential is stationary but the atom experiences a fictitious force given by F(t) = −ma(t), where m is the mass of the particle and a(t) is the acceleration of the trap as a function of time^64,65. By following ref. ⁶⁶ (equation 5.4), we find the average vibrational quantum number increase ΔN is given by

where (tilde{a}({omega }_{0})) is the Fourier transform of a(t) evaluated at the trap frequency ω₀, and the zero-point size of the particle ({x}_{{rm{zpf}}}equiv sqrt{hbar /(2m{omega }_{0})}), where (hbar ) is the reduced Planck constant. ΔN is the same for all initial levels of the oscillator⁶⁶. Experimentally, we apply an acceleration profile a(t) = jt to the atom, from time −T/2 to +T/2 to move a distance D with constant jerk j. We calculate (|tilde{a}(omega ){|}^{2}), simplify using ω₀T ≫ 1, and assume a small range of trap frequencies to average the oscillatory terms, resulting in

Several relevant insights can be gleaned from this formula. First, this expression indicates our ability to move large distances D with comparably small increases in time T. Furthermore, to maintain a constant ΔN, the movement time (Tpropto {omega }_{0}^{-3/4}). Moreover, to perform a large number of moves k for a deep circuit, we can estimate ΔN ∝ k/T⁴, suggesting that we can increase our number of moves from, for example, 5 to 80 by slowing each move from 200 μs to 400 μs. Move speed could be further improved with different a(t) profiles, but inevitably with finite resources such as trap depth, quantum speed limits will eventually prevent arbitrarily fast motion of qubits across the array³⁰.

We now compare equation (2) with our experimental observations. In Fig. 1d we start to observe atom loss when we move 55 μm in 200 μs under a constant negative jerk. This speed limit is consistent with our above estimates: using ω₀ = 2π × 40 kHz and x_zpf = 38 nm, we predict ΔN ≈ 6 for this move, corresponding to the onset of tangible heating at this move speed. More quantitatively, we assume a Poisson distribution with mean N and variance N and integrate the population above some critical N_max upon which the atom will leave the trap. From this analysis we find atom retention is given by (frac{1}{2}(1+{rm{erf}}[({N}_{{rm{max }}}-N)/sqrt{2N}])).

Extended Data Fig. 2a, b measures the atom retention as a function of move time T and trap frequency ω₀/2π. Using the functional form above, for both sets of measurements, we extract an N_max of about 30, corresponding to adding about 30 excitations before exciting the atom out of the trap. Such a limit is physically reasonable as the absolute trap depth of 4 MHz implies only about 100 levels, the atom starts at finite temperature, and moreover the effective trap frequency reduces once the anharmonicity of the trap starts to play a role. We note that these estimates are only approximate (we roughly estimate ω₀ for the trap depths used during the motion), but nonetheless suggests our motion limit is consistent with physical limits for our chosen a(t). Our analysis here also neglects the acoustic lensing effects associated with ramping the AOD frequency, which causes astigmatism by focusing one axis to a different plane and thus deforms the trap and reduces the peak trap intensity (and ω₀) as given by the Strehl ratio.

Additional heating and loss during the circuit can also be caused by repeated short drops for performing two-qubit gates, where the tweezers are briefly turned off to avoid anti-trapping of the Rydberg state and light shifts of the ground–Rydberg transition. However, drop–recapture measurements in Extended Data Fig. 2c suggest that the 500-ns drops we use experimentally have a negligible effect until hundreds of drops per atom (corresponding to hundreds of CZ gates). We find that atom loss and heating as a function of number of drops are well described by a diffusion model, which would then predict that reducing atom temperature by a factor of 2× (reducing thermal velocity by (sqrt{2}times )) and reducing the drop time t_drop by 2×, together would increase the number of possible CZ gates per atom to thousands.

Two-qubit CZ gates implementation

We implement our two-qubit gates and calibrations following ref. ⁵. Specifically, the two-qubit CZ gate is implemented by two global Rydberg pulses, with each pulse at detuning Δ and length τ, and with a phase jump ξ between the two pulses. The pulse parameters are chosen such that qubit pairs, adjacent and under the Rydberg blockade constraint, will return from the Rydberg state back to the hyperfine qubit manifold with a phase depending on the state of the other qubit. The numerical values for these pulse parameters are:

For our experiments in Figs. 1–3, we operate with a two-photon Rydberg Rabi frequency of Ω/2π = 3.6 MHz, giving a theoretical τ = 190 ns and a theoretical Δ/(2π) = −1.36 MHz. We choose the negative detuning sign to help minimize excitation into the m_j = +1/2 Rydberg state (m_j denotes magnetic sublevel of the 70S_1/2 Rydberg state), which is detuned by about 24 MHz under the field of 8.5 G (and experiences a three-times lower coupling to the Rydberg laser than the desired m_j = −1/2 state owing to reduced Clebsch–Gordan coefficients). In this work, we operate with strong blockade between adjacent qubits, with Rydberg–Rydberg interactions V₀/2π ranging from 200 MHz to 1 GHz. In Fig. 4, we operate with Ω/2π = 4.45 MHz for the two-qubit gates.

Managing spurious phases during CZ gates

The two-qubit gate from ref. ⁵ induces both an intrinsic single-qubit phase, as well as spurious phases that are primarily induced by the differential light shift from the 420-nm laser. Under certain configurations, the 420-nm-induced differential light shift on the hyperfine qubit can be exceedingly large (>8 MHz), yielding phase accumulations on the hyperfine qubit of about 6π. Small, percent-level variations of the 420-nm intensity can thus lead to significant qubit dephasing.

Reference ⁵ addresses this 420-induced-phase issue by performing an echo sequence: after the CZ gate, the 1,013-nm Rydberg laser is turned off, a Raman π pulse is applied and then the 420-nm laser is pulsed again to cancel the phase induced by the 420 light during the CZ gate. This method echoes out the 420-induced phase, but comes at a cost of a factor of two increase in the 420-induced scattering error, which is the dominant source of error in our two-qubit CZ gates.

Echo between CZ gates

To address these various issues, here we perform a Raman π pulse between each CZ gate to echo out spurious gate-induced phases on the hyperfine qubit (Extended Data Fig. 1). This approach has several advantages. The 420-induced phase is now cancelled by pairs of CZ gates, without explicitly applying additional 420-nm pulses to echo each individual CZ gate, thereby reducing the scattering error of the CZ gate in this work by a factor of approximately two. We estimate that this echo technique, having reduced the scattering error incurred during each gate, roughly compensates the increased scattering rate incurred by spreading our optical power over more space in 2D, thereby giving us comparable gate fidelites to the two-qubit CZ gate fidelities of ≥97.4(2)% reported in ref. ⁵. Furthermore, the echo between CZ gates also cancels the intrinsic single-qubit phase of the CZ gate, removing errors in the calibration of this parameter, as well as cancelling any other gate-induced spurious single-qubit phases such as a roughly 0.01-rad phase induced by pulsing the traps off for 500 ns for the two-qubit gate (Extended Data Fig. 1). In instances where the number of CZ gates we apply is odd, we perform the echo for the final CZ gate.

Sign of intermediate-state detuning

To further suppress the effect of the spurious, 420-induced phase, we operate our 420-nm laser to be red-detuned (by 2 GHz) from the 6P_3/2 transition. For red detunings, the light shift on the |0⟩ state and the |1⟩ state are of the same sign, minimizing the differential light shift, whereas for blue detunings <6.8 GHz, the light shift on the |0⟩ state and the |1⟩ state have opposite signs and amplify the differential light shift.

Sensitivity to axial trap oscillations

In typical Rydberg excitation timescales with optical tweezers, the axial trap oscillation frequencies of several kilohertz are inconsequential. Here with our circuits running as long as 1.2 ms, with Rydberg pulses throughout, we find that the axial trap oscillations can have important effects. In particular, the axial oscillations cause the atoms to make oscillations in and out of the Rydberg beams: at our estimated axial temperature of about 25 μK and axial oscillation frequency of 6 kHz, we estimate an axial spread (sqrt{langle {z}^{2}rangle }approx 1{rm{.3}},{rm{mu }}{rm{m}}). For our 20-μm-waist beams, the effect of this positional spread is relatively small on the pulse parameters of the CZ gate, but can be significant on the sensitive 420-induced phase we seek to cancel by echoing out the phase induced by CZ gates separated by about 200 μs (see previous section). When using 20-μm-waist beams, and a 2.5-GHz blue detuning of our 420-nm laser, we find that the dephasing due to the axial trap oscillations is significant (Extended Data Fig. 4). To remedy this deleterious effect, we increase the beam waist of our 420-nm laser to 35 μm (while maintaining constant intensity) and change the laser frequency to be 2-GHz red-detuned, together resulting in a significant reduction in the dephasing associated with improper echoing of the 420-nm pulse.

Bell-state preparation and fidelity

In Fig. 1, we prepare the |Φ⁺⟩ Bell state in the same way that is done in ref. ⁵. After initializing a pair of qubits in |00⟩, we apply X(π/2) pulse–CZ gate–X(π/4) pulse. We calculate and plot the raw resulting fidelity of this |Φ⁺⟩ Bell state as the sum of populations in |00⟩ and |11⟩, averaged with the fitted amplitude of parity oscillations (example in Fig. 1c), which measures the off-diagonal coherences. In Fig. 1d, upon significant loss from movement, this fidelity estimate becomes skewed because we begin measuring an artificially large population in |11⟩ (as state |1⟩ is detected as loss); accordingly, we estimate the |Φ⁺⟩ population as two times the population of |00⟩ once the population difference between |11⟩ and |00⟩ becomes greater than 0.1 (an arbitrary cut-off where the effects of loss start to become significant). In Fig. 1d, for moves slower than 300 μs, we achieve an average raw Bell-state fidelity after the moving of 94.8(2)%. If we do not move or attempt to preserve coherence for 500 μs (that is, if we measure immediately after preparing the Bell state), then we measure a raw Bell-state fidelity of 95.2(1)% (not plotted here), consistent with the results in ref. ⁵.

Analysis of error sources

We detail here some of our measured and estimated sources of error for an entire sequence (toric code preparation in particular, our deepest circuit). We find the total single-qubit fidelity after performing the entire sequence is roughly 96.5% for the toric code circuit, which we measure by embedding the entire experiment in a Ramsey sequence: that is, we perform a Raman π/2 pulse, do all motion and decoupling, and then do a final π/2 pulse with variable phase to measure total contrast. We are able to account for our single-qubit fidelity quantitatively as being composed of our known single-qubit errors in Extended Data Fig. 6c.

Estimated contributions to two-qubit gate error are summarized in Extended Data Fig. 6c. These estimates come from numerical simulations in QuTiP (version 4.5.0) with experimental parameters. The effects of intermediate state scattering and Rydberg decay are included via collapse operators in the Lindblad master equation solver. Other error contributions include finite-temperature random Doppler shifts and position fluctuations, as well as laser pulse-to-pulse fluctuations, all of which are simulated using classical Monte Carlo sampling of experiment parameters. Experimental parameters used for the simulations are as follows: blue and red Rabi frequencies (Ω_b, Ω_r) = 2π × (160, 90) MHz, 6P_3/2 intermediate state detuning of 2 GHz, intermediate state lifetime of 110 ns, 70S_1/2 Rydberg state lifetime of 150 μs, Rydberg blockade energy of 500 MHz, splitting to second Rydberg state of 24 MHz, radial and axial trap frequencies (ω_r, ω_z) = 2π × (40, 6) kHz, and temperature T = 20 μK. We can also use this modelling to project for future performance; by assuming a 10 times increase in available 1,013-nm intensity and that atoms are cooled to a temperature of 2 uK, we project a possible CZ gate fidelity of 99.7%, beyond the surface code threshold^38,67. Alkaline-earth atoms could also offer other routes to high-fidelity operations for QEC^68,69,70.

To understand how our various single-qubit and two-qubit errors contribute to our graph-state fidelities, we perform a stochastic simulation of the quantum circuit used for graph-state preparation (Extended Data Fig. 6a, b). We utilize the Clifford properties of our circuit, allowing for efficient numerical evaluation and random sampling of many possible error realizations. The simulation is performed under a realistic error model, where the rates of ambient depolarizing noise and atom loss are measured in the experiment (Extended Data Fig. 6c). The resulting stabilizer and logical qubit expectation values agree well with those measured experimentally.

Rydberg beam shaping and homogeneity

We shape our Rydberg beams into tophats of variable size through wavefront control using the phase profile on an SLM²⁵. This ability allows us to match the height of our beam profile to the experiment zone size of any given experiment, thereby maximizing our 1,013-nm light intensity and CZ gate fidelities. We optimize our Rydberg beam homogeneity until peak-to-peak inhomogenities are below <1%. To this end, we correct all aberrations up to the window of our vacuum chamber, as done in ref. ²⁵, which yields an inhomogeneity on the atoms of several per cent that we attribute to imperfections of the final window. To further optimize the homogeneity, we empirically tune aberration corrections on the tophat through Zernike polynomial corrections to the phase profile in the SLM plane (Fourier plane). With this procedure, we reduce peak-to-peak inhomogeneities to <1% over a range of 40–50 μm in the atom plane.

Creation and optimization of graph layouts

We outline here a description of how we optimize our graph layouts for the cluster state, Steane code, surface code and toric code preparation. Our optimization in this work is heuristic, and future work can develop appropriate algorithms for designing optimal circuits through atom spatial arrangement and AOD trajectories. Extended Data Fig. 5 shows all of the graphs we create and the process for creating them. There are several parameters we optimize for. (1) Minimize the number of parallel two-qubit gate layers. (2) Minimize the total move distance for the moving atoms. (3) Have all moving atoms in one sublattice (all graphs realized here are bipartite) to facilitate the final local rotation of one sublattice. (4) Minimize the vertical extent of the array and the number of distinct rows (to maximize 1,013-nm light intensity and minimize sensitivity to beam inhomogeneity between the rows). (5) When ordering gates, apply two-qubit gates as early as possible in the circuit. If a gate layer induces a bit-flip (X error) then that error can propagate during subsequent gates (becoming a Z error on the other qubit), so gates should be in the earliest layer possible.

Local (sublattice) hyperfine rotations

We perform local rotations in the hyperfine basis by use of our horizontally propagating 420-nm beam, which imposes a differential light of several megahertz on the hyperfine qubit and can thus be used for realizing a fast Z rotation. To realize the local Y(π/2) rotation used throughout this work, we move one sublattice of atoms out of the 420-nm beam, then apply [global Y(π/4)]–[local Z(π)]–[global Y(π/4)]. This realizes a Y(π/2) rotation on one sublattice and a Z(π) rotation on the other sublattice (which is inconsequential as it then commutes with the immediately following measurement in the Z basis). To apply a Y(π/2) on the other sublattice of atoms, we add an additional global Z(π) (implemented by jumping the Raman laser phase) between the two Y(π/4) pulses. Future experiments will benefit from an additional set of locally focused beams for performing local Raman control of hyperfine qubit states, but we find that moving atoms works so efficiently (even for moving >50 μm to move out of the 420-nm beam) that this approach is well suited for producing a high-fidelity, homogeneous rotation on roughly half the qubits.

Local Rydberg initialization

We perform local Rydberg control to initialize the |ℤ₂⟩ = |rgrg…⟩ ≡ |r1r1…⟩ state for studying the dynamics of many-body scars. We achieve this local initialization by applying approximately 50-MHz light shifts between |1⟩ and |r⟩ using 810-nm tweezers generated by an SLM onto a desired subset of sites, and then apply a global Rydberg π pulse, which excites the non-light-shifted atoms. We use this approach here to prepare every other atom in each chain into |r⟩, but emphasize that as the locations of the SLM tweezers are fully programmable, this technique can be used to prepare any initial blockade-satisfying configuration of atoms in |1⟩ and |r⟩.

The 50-MHz biasing light shift is significantly larger than the Rydberg Rabi frequency Ω/2π = 4.45 MHz, leading to a Rydberg population on undesired sites of <1%. The t = 0 time point of Extended Data Fig. 10b shows the high-fidelity preparation of the |ℤ₂⟩ state using this approach. We note that with 810-nm light, even though the achieved biasing light shift is significant, the Raman-scattering-induced T₁ (of the hyperfine qubit) is still about 50 ms and thus leads to a scattering error ≲4 × 10⁻⁶ during the 200-ns pulse of the light-shifting tweezers. There can also be a motional effect from the biasing tweezers, with an estimated radial trapping frequency of 150 kHz, which we also deem to be negligible during the 200-ns pulse.

Rydberg Hamiltonian

In Fig. 4, we study dynamics under the many-body Rydberg Hamiltonian

where (hbar ) is the reduced Planck constant, Ω is the Rabi frequency, Δ is the laser frequency detuning, n_i = |r_i⟩⟨r_i| is the projector onto the Rydberg state at site i and ({sigma }_{i}^{x}=|{1}_{i}rangle langle {r}_{i}|+|{r}_{i}rangle langle {1}_{i}|)flips the atomic state. For the entanglement entropy measurements in this work, we choose lattice spacings where the nearest-neighbour interaction V₀ > Ω results in the Rydberg blockade, preventing adjacent atoms from simultaneously occupying |r⟩. In particular, the many-body experiments are performed on eight-atom chains, quenching to a time-independent H_Ryd with V₀/2π = 20 MHz, Ω/2π = 3.1 MHz and Δ/2π = 0.3 MHz. Quenching to small, positive Δ = 0.0173V₀ partially suppresses the always-positive long-range interactions and thereby is optimal for scar lifetime, as derived and shown experimentally in ref. ⁷¹.

Coherent mapping protocol

As described in the text, we implement a coherent mapping protocol to transfer a generic many-body state in the {|1⟩, |r⟩} basis to the long-lived and non-interacting {|0⟩, |1⟩} basis. To achieve this mapping, immediately following the Rydberg dynamics we apply a Raman π pulse to map |1⟩ → |0⟩, and then a subsequent Rydberg π pulse to map |r⟩ → |1⟩ (ref. ⁷²).

Even for perfect Raman and Rydberg π pulses (on isolated atoms), there are three key sources of infidelity associated with this mapping process. (1) Any population in blockade-violating states (that is, two adjacent atoms both in |r⟩) will be strongly shifted off-resonance for the final Rydberg π pulse. As such, this atomic population will be left in the Rydberg state and lost. (2) Long-range interactions, for example, from next-nearest neighbours, will detune the final Rydberg π pulse from resonance and thus reduce pulse fidelity. As the long-range interactions are not the same for all many-body microstates, this effect cannot be mitigated by a simple shift of the detuning. (3) Dephasing of the state occurs throughout the duration of the Raman π pulse, predominantly from Doppler shifts between the ground states |0⟩ and |1⟩ and the Rydberg state |r⟩. Although these random on-site detunings are also present during the many-body dynamics, turning the Rydberg drive Ω off allows the system to freely accumulate phase and makes us particularly sensitive to dephasing errors.

We now detail our mitigation of the above error mechanisms. To minimize errors from (1), we perform our many-body dynamics with ({Omega }^{2}/(2{V}_{0}^{2})approx 0.01). This minimizes the probability of an atom to violate blockade to be of order 1%. To help minimize errors from (2), we increase the amplitude of the 420-nm laser for the final π pulse by a factor of 2×, such that (V_NNN/Ω)² = 0.005 (where V_NNN is the interactions with next-nearest neighbours), reducing pulse errors from long-range interactions to order 1%. Finally, to reduce errors from (3), we perform a fast Raman π pulse and leave only 150 ns between ending the many-body Rydberg dynamics and beginning the Rydberg π pulse. The 150-ns gap is comparably short relative to the ({T}_{2}^{ast }) ≈ 3–4 μs of the {|g⟩, |r⟩} basis, leading to a random phase accumulation of the order of about 0.02 × 2π rad per particle, but is further compounded by having entangled states of N particles in one copy accumulating a random phase relative to entangled states of N particles in the second copy. We study these various effects numerically in Extended Data Fig. 9c.

Finally, we note that the global Raman beam induces a light-shift-induced phase shift of about π on |0⟩ and |1⟩ relative to |r⟩ during the Raman π pulse. Similarly, the global 420-nm laser also induces a light-shift-induced phase shift of about π between |0⟩ and |1⟩ during the Rydberg π pulse. Although the measurements we perform here are interferometric (in other words, the singlet state we measure is invariant under global rotations) and thus not affected by these global phase shifts, in future work these phase shifts can be measured and accounted for where relevant.

Measuring entanglement entropy

The second-order Renyi entanglement entropy is given by ({S}_{2}({rm{A}})=) (-{log }_{2}{rm{Tr}},[{rho }_{{rm{A}}}^{2}]), where ({rm{Tr}}[{rho }_{{rm{A}}}^{2}]) is the state purity of reduced density matrix ρ_A on subsystem A. The purity can be measured with two copies by noticing that ({rm{Tr}},[{rho }_{{rm{A}}}^{2}]={rm{Tr}},[hat{{bf{S}}}{rho }_{{rm{A}}}otimes {rho }_{{rm{A}}}]) is the expectation value of the many-body SWAP operator (hat{{bf{S}}}) (ref. ^10,11). The many-body SWAP operator is composed of individual SWAP operators ({hat{{bf{s}}}}_{i}) on each twin pair, that is, (hat{{bf{S}}}={Pi }_{i}{hat{{bf{s}}}}_{i}) (with i ∈ A). Measuring this expectation value amounts to probing occurences of the singlet state (frac{|01rangle -|10rangle }{sqrt{2}}) (with eigenvalue −1 under ({hat{{bf{s}}}}_{i})), as all other ({hat{{bf{s}}}}_{i}) eigenstates have eigenvalue +1. Occurences of the singlet state in each twin pair, that is, the Bell state |Ψ⁻⟩, is extracted by a Bell measurement circuit (with an additional local Z(π), see next paragraph), which maps |Ψ⁻⟩ → |00⟩ and can thereafter be measured in the computational basis. As such, after performing the Bell measurement circuit, we analyse the resulting bit-string outputs and calculate the purity of any subsystem A by calculating (langle {Pi }_{mathrm{iin A}}{hat{{bf{s}}}}_{i}rangle ): that is, we measure purity as the average parity = ⟨(−1)^{observed |00⟩ pairs}⟩ within A. In the absence of experimental imperfections, the purity will equal 1 for the whole system, and be less than 1 for subsystems that are entangled with the rest of the system.

A Bell measurement circuit can be decomposed into applying an X(π/2) rotation on one atom of the twin pair, then applying a CZ gate and then a global X(π/2) rotation. In other measurements, we realize a local X(π/2) by doing a global X(π/4) rotation, then local Z(π) rotation and then global X(π/4). However, we note that for this singlet measurement circuit, the first X(π/4) is redundant as the singlet state is invariant under global rotations, and so for the local X(π/2) we only apply the local Z(π) and then the second global X(π/4). This effectively realizes the X(π/2) on one qubit, up to a Z(π) on the other qubit (not shown in circuit diagram in Fig. 4). Under this simplification, the Bell measurement circuit to map |Ψ⁻⟩ → |00⟩ can be roughly understood as the reverse of the Bell-state preparation circuit in ref. ⁵, which is precisely how we calibrate the parameters of the Bell measurement.

Calibrating and benchmarking the interferometry

To validate the interferometry measurement (and check for proper calibration), we benchmark it separately from the many-body dynamics and coherent mapping protocol. We perform this benchmarking by preparing independent qubits in identical, variable single-qubit superpositions (through a global Raman pulse of variable time) and ensuring that the interferometry rarely results in |00⟩ for all the variable initial product states (Extended Data Fig. 8a). We find this is an important benchmarking step, because we find that small miscalibrations of the Bell measurement can lead to lower fidelity (that is, higher entropy) for different initial product states and thereby result in additional spurious signals in an entanglement entropy measurement. We note that this measurement is particularly sensitive to the single-qubit phase immediately before the final X(π/2) pulse (induced by the CZ gate and cancelled by a global Z(θ) pulse).

Additional many-body data and details

To benchmark our method of measuring entanglement entropy in a many-body system, in Extended Data Fig. 8b we study the entanglement dynamics after initializing two proximal atoms in |1⟩ and resonantly exciting to the Rydberg state for a variable time t. Under conditions of Rydberg blockade, this excitation results in two-particle Rabi oscillations between |11⟩ and the entangled state (|Wrangle =frac{1}{sqrt{2}}(|1rrangle +|r1rangle )) (top panel of Extended Data Fig. 8b)^3,13,72. The state purity of this two-particle system is measured by performing Bell measurements on atom pairs from two identical copies. Locally, the measured purity of the one-particle subsystem reduces to a value of about 0.5 when the system enters the maximally entangled |W⟩ state, at which point the reduced density matrix of each individual atom is maximally mixed. In contrast, the purity of the global, two-particle state remains high. The observation that the global-state purity is higher than the local-subsystem purity is a distinct signature of quantum entanglement^11,12.

For the data shown in Fig. 4c, e, we subtract the data by an extensive classical entropy as is done in ref. ¹². This fixed, time-independent offset is given by the entropy per particle, that is, (global entropy at quench time t = 0) × (subsystem size)/(global system size). In Extended Data Fig. 9a, we show the raw entanglement entropy measurements alongside numerics, to indicate the size of the extensive classical entropy contribution. In plotting, we also delay the theory curves by 10 ns to account for the fact that the Raman π pulse cuts off the final 10 ns of the Rydberg evolution, which is done to keep the coherent mapping gap as short as possible and minimize Doppler dephasing. Furthermore, in Extended Data Fig. 9b we plot the measured global purity and compare it with numerical simulations incorporating experimental errors (Extended Data Fig. 9c).

In Extended Data Fig. 10, we show additional many-body data on the eight-atom chain system, with the same parameters as those used in the main text. We show the measured single-site entropy of each site in the eight-atom chain for the |ℤ₂⟩ quench in Extended Data Fig. 10a. Furthermore, in Extended Data Fig. 10b, we plot the global Rydberg population, measured in both the {|1⟩, |r⟩} basis and the {|0⟩, |1⟩} basis.