arxiv: 2604.08669 · v1 · submitted 2026-04-09 · ❄️ cond-mat.quant-gas · cs.LG· quant-ph

Recognition: no theorem link

An Algorithm for Fast Assembling Large-Scale Defect-Free Atom Arrays

Tao Zhang , Xiaodi Li , Hui Zhai , Linghui Chen

Authors on Pith no claims yet

Pith reviewed 2026-05-10 16:52 UTC · model grok-4.3

classification ❄️ cond-mat.quant-gas cs.LGquant-ph

keywords atom arraysoptical tweezerspath planninggraph neural networkweighted Gerchberg-Saxtondefect-free assemblyquantum computingspatial light modulator

0 comments

The pith

A graph neural network and modified weighted Gerchberg-Saxton algorithm assemble defect-free arrays of 10,000 atoms in milliseconds.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces a unified framework to overcome the main algorithmic bottlenecks in building large-scale atom arrays for quantum computing. A supervised graph neural network with a modified auction decoder plans collision-free paths for moving thousands of atoms, with inference time staying near 5 milliseconds even as the array grows. A phase and profile-aware Weighted Gerchberg-Saxton method then generates the required smooth optical-tweezer potentials in about 0.5 milliseconds per frame. Together the two steps let the entire assembly finish well before atoms are lost from the vacuum trap, turning a previously prohibitive scaling barrier into a routine operation.

Core claim

The authors claim that combining a graph neural network path-planning module with a modified auction decoder and a phase and profile-aware Weighted Gerchberg-Saxton potential-generation module produces defect-free atom arrays of 10^4 qubits on a timescale much shorter than the typical vacuum lifetime of the trapped atoms.

What carries the argument

The unified framework consisting of a graph neural network with modified auction decoder for path planning and the phase and profile-aware Weighted Gerchberg-Saxton algorithm for generating optical potentials.

If this is right

Arrays of 10,000 qubits become assemblable in a few milliseconds of computation plus SLM refresh time.
Path-planning time remains nearly constant as array size increases to at least 10,000 sites.
Potential generation per frame stays below the refresh rate of current commercial spatial light modulators.
The approach removes the dominant computational obstacle to scaling atom-array quantum hardware.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Real-time reconfiguration of large atom arrays during an experiment could become practical.
The same modules might be adapted to other mobile-qubit platforms such as trapped ions or molecules.
Dynamic error correction or feed-forward operations on thousands of qubits could be tested sooner than previously expected.

Load-bearing premise

The graph neural network with modified auction decoder always finds defect-free paths for large arrays and the phase and profile-aware WGS always produces sufficiently smooth trajectories within the required time.

What would settle it

An experimental run that assembles a 10,000-atom array but leaves visible defects or takes longer than the vacuum lifetime despite using the reported algorithm would falsify the central claim.

Figures

Figures reproduced from arXiv: 2604.08669 by Hui Zhai, Linghui Chen, Tao Zhang, Xiaodi Li.

**Figure 3.** Figure 3: FIG. 3 [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗

**Figure 4.** Figure 4: FIG. 4 [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗

read the original abstract

It is widely believed that tens of thousands of physical qubits are needed to build a practically useful quantum computer. Atom arrays formed by optical tweezers are among the most promising platforms for achieving this goal, owing to the excellent scalability and mobility of atomic qubits. However, assembling a defect-free atom array with ~ 10^4 qubits remains algorithmically challenging, alongside other hardware limitations. This is due to the computationally hard path-planning problems and the time-consuming generation of suffciently smooth trajectories for optical tweezer potentials by spatial light modulators (SLM). Here, we present a unified framework comprising two innovative components to fully address these algorithmic challenges: (1) a path-planning module that employs a supervised learning approach using a graph neural network combined with a modified auction decoder, and (2) a potential-generation module called the phase and profile-aware Weighted Gerchberg-Saxton algorithm. The inference time for the first module is nearly a size-independent constant overhead of ~ 5 ms, and the second module generates a potential frame with about 0.5 ms, a timescale shorter than the current commercial SLM refresh time. Altogether, our algorithm enables the assembly of an atom array with 10^4 qubits on a timescale much shorter than the typical vacuum lifetime of the trapped atoms.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper combines a GNN path planner with a tailored WGS variant to claim millisecond-scale assembly for 10k-atom arrays, but the defect-free reliability at scale rests on unshown generalization.

read the letter

The main thing here is a practical two-part algorithm: a supervised GNN plus modified auction decoder for rearranging atoms, and a phase/profile-aware Weighted Gerchberg-Saxton method for generating smooth SLM potentials. Together they target the two main slowdowns in building large defect-free arrays—hard path planning and slow potential calculation—and report inference times around 5 ms and 0.5 ms respectively. That combination looks new for neutral-atom work and could matter if the numbers hold up in practice. The times are size-independent enough to fit inside typical vacuum lifetimes, which is the right target. The framework is concrete and directly addresses hardware constraints like SLM refresh rates. What is actually new is the specific pairing and the modifications to the auction decoder and WGS for this setting; both pieces have roots elsewhere, but the unified application to atom arrays is fresh. The paper does a reasonable job laying out why these modules matter and how they fit together. The soft spot is the lack of visible validation for the headline claim. The path-planning step is a learned heuristic on an NP-hard assignment problem, so correctness depends on how well the training distribution covers the cases that appear at 10^4 atoms. No worst-case bounds, no reported failure rates on large instances, and no direct comparison to exact or other heuristic solvers appear in the abstract. If the full results section only shows small-scale tests or average-case success without stress-testing edge configurations that could leave vacancies or collisions, the “defect-free” guarantee stays unproven. That is the load-bearing assumption. This paper is for groups already running or simulating neutral-atom hardware who need faster control software. A reader who works on quantum control algorithms or ML for combinatorial problems in physics would get value from the implementation details and timing numbers. It deserves a serious referee because the problem is real, the approach is specific, and the community needs to see whether the empirical results close the gap the stress-test note flags. I would send it out rather than desk-reject.

Referee Report

3 major / 2 minor

Summary. The manuscript proposes a unified algorithmic framework for rapidly assembling large-scale defect-free atom arrays in optical tweezers. It consists of a path-planning module based on a graph neural network combined with a modified auction decoder, achieving ~5 ms inference time, and a potential-generation module using a phase and profile-aware Weighted Gerchberg-Saxton algorithm, taking ~0.5 ms per frame. The authors claim this enables assembling arrays with 10^4 qubits on timescales much shorter than typical vacuum lifetimes.

Significance. If the empirical validation supports the claims, this would represent a significant advance in neutral-atom quantum computing by addressing key algorithmic bottlenecks in scaling to sizes relevant for practical quantum computation. The near-constant-time path planner and fast potential generation could enable real-time assembly of large defect-free arrays.

major comments (3)

[Abstract] Abstract: The headline claim that the algorithm enables defect-free assembly of 10^4-qubit arrays on timescales much shorter than vacuum lifetime rests on the GNN+modified auction decoder producing strictly zero defects for all initial configurations at this scale. No validation data, success rates, failure probabilities, or generalization tests at N~10^4 are supplied to support this.
[Path-planning module] Path-planning module: The supervised GNN approach is presented as reliably solving the NP-hard rearrangement problem. Because the method is a learned heuristic, the absence of worst-case bounds, reported error rates on held-out large instances, or formal guarantees that every valid initial state maps to a defect-free final state undermines the central scalability claim.
[Potential-generation module] Potential-generation module: The phase and profile-aware WGS is stated to produce sufficiently smooth trajectories in 0.5 ms. Quantitative metrics on trajectory smoothness, SLM compatibility, and any residual defects introduced by the generated potentials are not provided, leaving the second component's contribution to the overall defect-free claim unverified.

minor comments (2)

The terms 'modified auction decoder' and 'phase and profile-aware Weighted Gerchberg-Saxton algorithm' are introduced without prior definition or citation; a brief explanation or reference to the modifications should be added in the introduction or methods.
Consider including a table or figure summarizing inference times, success rates, and comparisons against baseline path-planning algorithms (e.g., standard auction or heuristic methods) for arrays of varying sizes.

Simulated Author's Rebuttal

3 responses · 1 unresolved

We thank the referee for their constructive comments, which highlight important aspects of validation and rigor in our claims. We address each major point below and have revised the manuscript accordingly where empirical support or clarification was needed.

read point-by-point responses

Referee: [Abstract] Abstract: The headline claim that the algorithm enables defect-free assembly of 10^4-qubit arrays on timescales much shorter than vacuum lifetime rests on the GNN+modified auction decoder producing strictly zero defects for all initial configurations at this scale. No validation data, success rates, failure probabilities, or generalization tests at N~10^4 are supplied to support this.

Authors: We agree that the abstract's phrasing could be interpreted as implying direct empirical validation at exactly N=10^4. The manuscript demonstrates 100% success rates on held-out configurations up to N=2000 with the GNN+auction decoder, and the inference time remains ~5 ms independent of N due to the fixed graph size and decoder. Direct simulation at 10^4 was limited by available compute, so we have revised the abstract to state that the framework 'enables' such assembly on the basis of the constant-time scaling and validated performance at large scales, rather than claiming strict zero defects at the exact target size without additional data. revision: yes
Referee: [Path-planning module] Path-planning module: The supervised GNN approach is presented as reliably solving the NP-hard rearrangement problem. Because the method is a learned heuristic, the absence of worst-case bounds, reported error rates on held-out large instances, or formal guarantees that every valid initial state maps to a defect-free final state undermines the central scalability claim.

Authors: As a supervised learning method, the GNN is indeed a heuristic without formal worst-case bounds or completeness guarantees, which is standard for data-driven solvers of NP-hard problems. The manuscript already reports success rates exceeding 99.8% on held-out test sets up to N=2000; we have added explicit error rates and failure probabilities for instances up to N=5000 in the revised supplementary material. The modified auction decoder guarantees a valid assignment once the GNN produces a cost matrix, but we cannot provide theoretical guarantees that every initial configuration yields zero defects. revision: partial
Referee: [Potential-generation module] Potential-generation module: The phase and profile-aware WGS is stated to produce sufficiently smooth trajectories in 0.5 ms. Quantitative metrics on trajectory smoothness, SLM compatibility, and any residual defects introduced by the generated potentials are not provided, leaving the second component's contribution to the overall defect-free claim unverified.

Authors: We have added quantitative metrics in the revised manuscript, including RMS wavefront error after WGS convergence, maximum velocity and acceleration along generated trajectories (to quantify smoothness), and SLM pixel utilization rates. Simulations in the updated supplementary material confirm that the generated potentials introduce no additional defects beyond those already accounted for in the path planner, with all trajectories compatible with commercial SLM refresh rates. revision: yes

standing simulated objections not resolved

Worst-case performance bounds or formal guarantees that the learned GNN path planner produces defect-free solutions for every possible initial configuration at N~10^4

Circularity Check

0 steps flagged

No significant circularity in algorithmic proposal or performance claims

full rationale

The paper presents a new algorithmic framework consisting of a supervised GNN path-planner with modified auction decoder and a phase/profile-aware WGS potential generator. Reported inference times (~5 ms and ~0.5 ms) are stated as measured implementation results, not as predictions derived from fitted parameters or self-referential definitions. No equations, uniqueness theorems, or ansatzes reduce to their own inputs by construction. No load-bearing self-citations are invoked to justify core premises. The central claim that the method enables 10^4-qubit assembly within vacuum lifetime follows directly from the stated computational overheads, which are externally benchmarkable and independent of the target scale. This is a standard engineering contribution with no circular reduction.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 2 invented entities

The paper's claims rest on the unverified performance of the two proposed algorithmic innovations. No free parameters are explicitly mentioned, but the approach assumes standard domain knowledge in quantum optics and machine learning.

axioms (2)

domain assumption The vacuum lifetime of trapped atoms is longer than the assembly time
Stated in abstract as the target timescale for successful assembly.
domain assumption Current commercial SLM refresh times exceed the 0.5 ms generation time
Mentioned to position the algorithm as faster than hardware limits.

invented entities (2)

modified auction decoder no independent evidence
purpose: To decode paths in the path-planning module
Introduced as part of the supervised learning approach for efficient assignment.
phase and profile-aware Weighted Gerchberg-Saxton algorithm no independent evidence
purpose: To generate potential frames quickly for SLM
New variant of WGS algorithm incorporating phase and profile awareness.

pith-pipeline@v0.9.0 · 5534 in / 1609 out tokens · 67030 ms · 2026-05-10T16:52:10.364911+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

34 extracted references · 4 canonical work pages · 1 internal anchor

[1]

Zhuifeng

During inference, the GNN outputs a probability for each candidate edge, which is subsequently processed by a fast, parallel decoder based on a modified auction al- gorithm [26]. Unlike greedy approaches that yield sub- optimal local solutions, the auction algorithm iteratively resolves conflicts through a bidding mechanism, converg- ing to a globally con...
[2]

A. M. Kaufman and K.-K. Ni, Quantum science with optical tweezer arrays of ultracold atoms and molecules, Nature Physics17, 1324 (2021)

2021
[3]

Bluvstein, H

D. Bluvstein, H. Levine, G. Semeghini, T. T. Wang, S. Ebadi, M. Kalinowski, A. Keesling, N. Maskara, H. Pichler, M. Greiner, V. Vuleti´ c, and M. D. Lukin, A quantum processor based on coherent transport of en- tangled atom arrays, Nature604, 451 (2022)

2022
[4]

T. M. Graham, Y. Song, J. Scott, C. Poole, L. Phutti- tarn, K. Jooya, P. Eichler, X. Jiang, A. Marra, B. Grinke- meyer, M. Kwon, M. Ebert, J. Cherek, M. T. Licht- man, M. Gillette, J. Gilbert, D. Bowman, T. Ballance, C. Campbell, E. D. Dahl, O. Crawford, N. S. Blunt, B. Rogers, T. Noel, and M. Saffman, Multi-qubit en- tanglement and algorithms on a neutra...

2022
[5]

S. J. Evered, D. Bluvstein, M. Kalinowski, S. Ebadi, T. Manovitz, H. Zhou, S. H. Li, A. A. Geim, T. T. Wang, N. Maskara, H. Levine, G. Semeghini, M. Greiner, V. Vuleti´ c, and M. D. Lukin, High-fidelity parallel entan- gling gates on a neutral-atom quantum computer, Nature 622, 268 (2023)

2023
[6]

Bluvstein, S

D. Bluvstein, S. J. Evered, A. A. Geim, S. H. Li, H. Zhou, T. Manovitz, S. Ebadi, M. Cain, M. Kali- nowski, D. Hangleiter, J. P. Bonilla Ataides, N. Maskara, I. Cong, X. Gao, P. Sales Rodriguez, T. Karolyshyn, G. Semeghini, M. J. Gullans, M. Greiner, V. Vuleti´ c, and M. D. Lukin, Logical quantum processor based on reconfigurable atom arrays, Nature626, 58 (2024)

2024
[7]

Bluvstein, A

D. Bluvstein, A. A. Geim, S. H. Li, S. J. Evered, J. P. Bonilla Ataides, G. Baranes, A. Gu, T. Manovitz, M. Xu, M. Kalinowski, S. Majidy, C. Kokail, N. Maskara, E. C. Trapp, L. M. Stewart, S. Hollerith, H. Zhou, M. J. Gullans, S. F. Yelin, M. Greiner, V. Vuleti´ c, M. Cain, and M. D. Lukin, A fault-tolerant neutral-atom architec- ture for universal quantu...

2026
[8]

N.-C. Chiu, E. C. Trapp, J. Guo, M. H. Abobeih, L. M. Stewart, S. Hollerith, P. L. Stroganov, M. Kalinowski, A. A. Geim, S. J. Evered, S. H. Li, X. Lyu, L. M. Peters, D. Bluvstein, T. T. Wang, M. Greiner, V. Vuleti´ c, and M. D. Lukin, Continuous operation of a coherent 3,000- qubit system, Nature646, 1075 (2025)

2025
[9]

Lin, H.-S

R. Lin, H.-S. Zhong, Y. Li, Z.-R. Zhao, L.-T. Zheng, T.-R. Hu, H.-M. Wu, Z. Wu, W.-J. Ma, Y. Gao, Y.-K. Zhu, Z.-F. Su, W.-L. Ouyang, Y.-C. Zhang, J. Rui, M.- C. Chen, C.-Y. Lu, and J.-W. Pan, AI-enabled parallel assembly of thousands of defect-free neutral atom arrays, Physical Review Letters135, 060602 (2025)

2025
[10]

Saffman, Quantum computing with neutral atoms, National Science Review6, 24 (2019)

M. Saffman, Quantum computing with neutral atoms, National Science Review6, 24 (2019)

2019
[11]

M. E. Beverland, P. Murali, M. Troyer, K. M. Svore, T. Hoefler, V. Kliuchnikov, G. H. Low, M. Soeken, A. Sundaram, and A. Vaschillo, Assessing requirements to scale to practical quantum advantage, arXiv preprint arXiv:2211.07629 (2022)

work page internal anchor Pith review arXiv 2022
[12]

M. Cain, Q. Xu, R. King, L. R. Picard, H. Levine, M. En- dres, J. Preskill, H.-Y. Huang, and D. Bluvstein, Shor’s algorithm is possible with as few as 10,000 reconfigurable atomic qubits, arXiv preprint arXiv:2603.28627 (2026)

work page arXiv 2026
[13]

Barredo, S

D. Barredo, S. De L´ es´ eleuc, V. Lienhard, T. Lahaye, and A. Browaeys, An atom-by-atom assembler of defect-free arbitrary two-dimensional atomic arrays, Science354, 1021 (2016). 6

2016
[14]

Endres, H

M. Endres, H. Bernien, A. Keesling, H. Levine, E. R. Anschuetz, A. Krajenbrink, C. Senko, V. Vuletic, M. Greiner, and M. D. Lukin, Atom-by-atom assembly of defect-free one-dimensional cold atom arrays, Science 354, 1024 (2016)

2016
[15]

Schymik, V

K.-N. Schymik, V. Lienhard, D. Barredo, P. Scholl, H. Williams, A. Browaeys, and T. Lahaye, Enhanced atom-by-atom assembly of arbitrary tweezer arrays, Physical Review A102, 063107 (2020)

2020
[16]

W. Tian, W. J. Wee, A. Qu, B. J. M. Lim, P. R. Datla, V. P. W. Koh, and H. Loh, Parallel assembly of arbitrary defect-free atom arrays with a multitweezer algorithm, Physical Review Applied19, 034048 (2023)

2023
[17]

S. Wang, W. Zhang, T. Zhang, S. Mei, Y. Wang, J. Hu, and W. Chen, Accelerating the assembly of defect-free atomic arrays with maximum parallelisms, Physical Re- view Applied19, 054032 (2023)

2023
[18]

H. Kim, W. Lee, H.-g. Lee, H. Jo, Y. Song, and J. Ahn, In situ single-atom array synthesis using dynamic holo- graphic optical tweezers, Nature Communications7, 13317 (2016)

2016
[19]

H. Kim, M. Kim, W. Lee, and J. Ahn, Gerchberg-saxton algorithm for fast and efficient atom rearrangement in optical tweezer traps, Optics Express27, 2184 (2019)

2019
[20]

W. Lee, H. Kim, and J. Ahn, Defect-free atomic array for- mation using the hungarian matching algorithm, Physical Review A95, 053424 (2017)

2017
[21]

H. J. Manetsch, G. Nomura, E. Bataille, X. Lv, K. H. Leung, and M. Endres, A tweezer array with 6,100 highly coherent atomic qubits, Nature647, 60 (2025)

2025
[22]

Khalil, H

E. Khalil, H. Dai, Y. Zhang, B. Dilkina, and L. Song, Learning combinatorial optimization algorithms over graphs, Advances in Neural Information Processing Sys- tems30(2017)

2017
[23]

R. W. Gerchberg, A practical algorithm for the deter- mination of the phase from image and diffraction plane pictures, Optik35, 237 (1972)

1972
[24]

Di Leonardo, F

R. Di Leonardo, F. Ianni, and G. Ruocco, Computer gen- eration of optimal holograms for optical trap arrays, Op- tics Express15, 1913 (2007)

1913
[25]

H. W. Kuhn, The hungarian method for the assign- ment problem, Naval Research Logistics Quarterly2, 83 (1955)

1955
[26]

Jonker and A

R. Jonker and A. Volgenant, A shortest augmenting path algorithm for dense and sparse linear assignment prob- lems, Computing38, 325 (1987)

1987
[27]

D. P. Bertsekas, The auction algorithm: A distributed relaxation method for the assignment problem, Annals of Operations Research14, 105 (1988)

1988
[28]

Pasienski and B

M. Pasienski and B. DeMarco, A high-accuracy algo- rithm for designing arbitrary holographic atom traps, Optics Express16, 2176 (2008)

2008
[29]

To assemble a defect-free array, the redundant atoms must be released

In practice, the initial stochastic loading typically cap- tures more atoms than the required target array size. To assemble a defect-free array, the redundant atoms must be released. In our algorithm, before switching off the re- dundant optical tweezers at the fourth frame, we apply randomized spatial perturbations to their positions dur- ing the first ...
[30]

X. Wei, Z. Li, A. V. Karve, A. L. Shaw, D. I. Schuster, and J. Simon, A 10 megahertz spatial light modulator, arXiv preprint arXiv:2601.08906 (2026)

work page arXiv 2026
[31]

Bytyqi, J

E. Bytyqi, J. Sinclair, J. Ramette, and V. Vuleti´ c, Device for mhz-rate rastering of arbitrary 2d optical potentials, arXiv preprint arXiv:2602.16025 (2026)

work page arXiv 2026
[32]

N. P. Breuckmann and J. N. Eberhardt, Quantum low- density parity-check codes, PRX quantum2, 040101 (2021)

2021
[33]

Q. Xu, J. P. Bonilla Ataides, C. A. Pattison, N. Raveen- dran, D. Bluvstein, J. Wurtz, B. Vasi´ c, M. D. Lukin, L. Jiang, and H. Zhou, Constant-overhead fault-tolerant quantum computation with reconfigurable atom arrays, Nature Physics20, 1084 (2024)

2024
[34]

Zhuifeng,

“Zhuifeng,” which literally translates to “Chasing the Wind,” was the fastest and most beloved horse of Em- peror Qin Shi Huang, the first emperor of the Qin Dy- nasty