{"id":"678a3928-6563-428c-8efe-f3cf56f1b646","arxiv_id":"2507.04874","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"DYNAMO introduces multi-programming for neutral atom quantum processors, scheduling short circuits first and interleaving them with movement-zone constraints.","lead":"DYNAMO is a system for running several quantum programs at once on neutral atom hardware, instead of compiling one circuit at a time. It reports fewer execution stages and faster compilation than baselines, an early step toward quantum operating systems.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The correctness claim is unsupported: Appendix B's parallelization constraint (Eqs. 9-10) only forbids exact gate-coordinate overlap, never enforces blockade-radius distance between programs.","rationale":"The reader's weakest assumption is correct and is the load-bearing point. I read the paper in good faith: DYNAMO's contribution is a scheduling and placement method, and its headline numbers are compilation-time and stage-count measurements. Those measurements could survive even if the physical model is incomplete, but the abstract and §V-C explicitly claim correctness. That claim depends entirely on the formal constraints in Appendix B. Since Eqs. (9)-(10) contain no blockade-radius or pairwise-distance term, the solver can certify schedules that a real global Rydberg pulse would corrupt. This is not a disagreement over consensus; it is an internal gap between the stated physical constraint in §II-A and its formalization. I credit the experimental effort and the honest reporting of stage-count trade-offs; the missing artifact and non-reproducibility are secondary. The concrete counterexample would settle the issue: if the authors' actual implementation enforces r_b separation, the concern disappears; if not, the central correctness claim fails. The reader's REJECT verdict is therefore well supported, though the speedup and stage-reduction data themselves remain valid measurements of an incomplete compiler.","tokens_in":25807,"tokens_out":5726,"duration_ms":65435,"concrete_test":"Construct a minimal two-program instance with a common stage s: program A has one two-qubit gate with its gate pair at coordinate (0,0); program B has one two-qubit gate at (10,0) and one idle qubit at (1,0), with lattice spacing such that 1 < r_b < 10. Encode the SMT problem using exactly Eqs. (9)-(10) plus the AOD constraints; the solver returns SAT because the idle B qubit is not a gate qubit and the B gate coordinate differs from A's. Inspect the emitted schedule: if a global Rydberg pulse for A's gate fires while the B qubit remains at (1,0), the schedule is physically incorrect. If the authors can exhibit an additional r_b separation constraint in the paper or code, the concern is resolved; otherwise the correctness claim fails.","verdict_should_be":"REJECT","load_bearing_attack":"The paper's central claim is that DYNAMO preserves circuit correctness while multi-programming (abstract; §V-C). That claim rests on the two-qubit gate parallelization constraint formalized in Appendix B. Eqs. (9)-(10) require only that a new gate's qubit coordinate differ from an already-scheduled gate coordinate at the same stage: (x_q0,l, y_q0,l) != (x_s_j,k,l, y_s_j,k,l), and similarly for q1. Nothing in the formulation involves the Rydberg blockade radius r_b, pairwise Euclidean distance, or atoms that are not themselves gate partners. In a real neutral-atom device, a global Rydberg pulse for a gate pair excites every atom within r_b of either gate atom; an idle or gate-performing qubit from another program inside that radius is corrupted or participates unintentionally. The constraints therefore admit schedules that are physically invalid. The §IV worked example avoids the issue by restricting the second circuit's moves to Order-Free Zones and placing gates far apart, but no constraint enforces that separation in general. Since no code or artifact is supplied, the only support for correctness is this incomplete constraint set. The measured speedups and stage reductions are not invalidated as measurements, but they do not establish a correct multi-programming compiler.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes DYNAMO, a multi-programming compilation framework for dynamically field-programmable neutral atom arrays. The method combines a greedy multi-QPU scheduler that assigns circuits to arrays by circuit length (Section III) with an intra-array SMT-based scheduler that uses a cycle-wise decomposition and Order-Preserving/Order-Free Zones (Section IV). Experiments compare DYNAMO against two adaptations of DPQA (sequential and merged) on RevLib, Sabre, and QTetris circuits, reporting up to 14.39× compilation speedup versus merged DPQA and an average Rydberg-stage reduction of 50.47% relative to sequential DPQA. The paper claims that these gains are obtained while maintaining circuit correctness and hardware constraints.","tokens_in":26077,"tokens_out":6681,"duration_ms":82336,"significance":"If the correctness gap identified below were closed, DYNAMO would be a useful first step toward quantum operating systems for neutral atom architectures, and the problem it addresses is relevant. The evaluation uses externally sourced benchmark circuits and measures direct quantities (stage counts, wall-clock compilation time), with no fitted free parameters in the reported aggregate metrics. However, the central claim of correctness under hardware constraints is not established by the formalization in Appendix B, and the absence of a usable artifact prevents independent verification of the quantitative results.","major_comments":[{"comment":"The formal two-qubit gate parallelization constraint is incomplete and does not enforce the blockade-radius separation that neutral-atom hardware requires. Equations (9) and (10) only forbid a new gate's qubit coordinate from being exactly equal to an already-scheduled gate's coordinate at the same stage. They never mention the Rydberg blockade radius r_b, nor do they impose any Euclidean distance between the atoms of different programs. In a real device, a two-qubit gate is applied by a global Rydberg pulse, so any atom of another program within r_b of either gate atom will be excited and will participate in the interaction, corrupting both programs. Thus the constraint set admits schedules that are physically invalid, and the claim in Section V-C that DYNAMO achieves multi-programming 'without compromising compilation correctness' is unsupported. The authors must add explicit pairwise distance constraints involving r_b for all concurrently scheduled gate atoms and all atoms of other programs, verify those constraints on the generated schedules, and rerun the experiments, since the reported 50.47% average stage reduction may change materially once the missing constraint is imposed.","section":"Appendix B, Eqs. (9)–(10); Section V-C"},{"comment":"The relationship between the width-based greedy scheduler (Algorithm 1) and the SMT scheduler of Section IV is never specified precisely. Algorithm 1 checks only that the number of gates per DAG layer fits within the spatial capacity Wmax; it does not check AOD ordering, OPZ/OFZ membership, or gate parallelization. The paper does not state how the output of Algorithm 1 is fed into the constraint-based scheduler, nor which constraints are active in each phase. As a result, the reader cannot determine whether the grouped multi-program results in Section V-D are produced by the full DYNAMO pipeline and whether those schedules satisfy even the incomplete constraints of Appendix B. The authors should give a precise end-to-end description of the pipeline and explicitly list the constraint set solved at each stage.","section":"Sections III–IV"},{"comment":"The statement 'Code available on Github' at the end of Appendix B is not an actionable artifact: no URL, repository identifier, commit, or license is provided, and no solver configuration (r_b, array dimensions, Wmax, per-circuit qubit counts) is given. Without the artifact and configuration, the reported stage counts and compilation times cannot be reproduced, and the possibility that the measured gains derive from the incomplete constraint set of Eqs. (9)–(10) cannot be checked. A revision should supply a complete artifact and an explicit validation of the correctness properties of the emitted schedules.","section":"Appendix B and Section V"}],"minor_comments":[{"comment":"The column labeled 'Speedups(%)' contains values mostly below 1 (for example, 0.37 for Minimal circuits), yet a value below 1 indicates that DYNAMO is slower than the sequential DPQA baseline. This column should be renamed or the ratio inverted and clearly labeled so that 'speedup' is not confused with a time ratio.","section":"Table IV"},{"comment":"The index notation in the AOD movement constraints is inconsistent: p and k are mixed, and Eq. (3) writes x^s_{1,i,t,k} where the surrounding text suggests x^s_{1,j,t,p}. The notation should be made uniform.","section":"Appendix B, Eqs. (3)–(8)"},{"comment":"The caption labels the proposed method 'NACO' while the text uses 'DYNAMO'; this should be corrected.","section":"Figure 7 caption"},{"comment":"The text introduces a 'systematic two-phase process' but then lists three phases: initial allocation, incremental assignment, and intra-array refinement. The wording should be adjusted to 'three-phase' or the phases should be regrouped.","section":"Section III"},{"comment":"The phrase 'with the same compilation quality' is used to qualify the speedup claim, but compilation quality is never defined or measured; either define it or remove the qualification.","section":"Contributions and Abstract"}],"recommendation":"major_revision","confidential_remarks":"The central correctness flaw in Eqs. (9)–(10) is real and load-bearing: the current constraint set permits cross-program Rydberg blockade corruption. I do not recommend rejection outright because the flaw is identifiable and conceptually fixable by adding r_b-based distance constraints, but the authors must redo the formalization and the experiments. The missing artifact is also a serious obstacle to verifying the claimed stage reductions. I would ask the editor to treat the correctness issue as a hard requirement for any revised submission."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The genuinely new thing here is the problem setup: DYNAMO is the first work to tackle multi-programming for neutral atom arrays, and the OPZ/OFZ spatial deformation idea is a sensible way to reason about AOD movement constraints when multiple circuits share an array. That is a real contribution, and the authors deserve credit for identifying the gap and building a concrete framework around it.\n\nThe experimental section is thorough by the standards of the field: dozens of benchmark circuits across five size groups, two adaptation schemes for the DPQA baseline, and clear reporting of compilation times and Rydberg stage counts. The measured speedups are real measurements of the authors' implementation. I also appreciate that they attempt to verify correctness via a formal constraint set rather than just asserting it.\n\nBut the central correctness claim does not hold up. The paper's own background states the two-qubit gate parallelization constraint as requiring \"sufficient distance\" between executing gates and other qubits. In Appendix B, the formalization—Equations (9) and (10)—only forbids exact-coordinate overlap of a new gate with an already-scheduled gate. Nothing in those constraints involves the blockade radius, Euclidean distance, or atoms that are not themselves gate partners. In a real device, a global Rydberg pulse excites every atom within the blockade radius of the gate pair, so a qubit from another program sitting nearby would be corrupted or pulled into the gate. That is a load-bearing flaw, not a cosmetic one: the headline claim is that DYNAMO \"maintains circuit correctness and hardware constraints.\"\n\nA secondary issue is the baseline design. Merging all programs into one large circuit (DPQAc) makes the SMT instance much harder, which inflates the reported speedups. The authors note this but do not account for it. A fairer baseline would compile circuits independently and concatenate the schedules, or at least report the size of the merged instance. The absence of a code release also hurts, since the constraint formalization is the only evidence that the compiled schedules are physically valid.\n\nI would not reject the idea. The problem is timely, the OPZ/OFZ decomposition is usable, and the missing blockade-distance constraint is fixable. But the current paper should not be accepted: the correctness claim is unsupported as written, and the experimental comparison needs a more careful baseline.\n\nMy recommendation: send it to peer review with a request for major revision, and have the referees focus on the constraint formulation and the baseline design. If the authors add the missing distance constraints and release the artifact, this could become a decent contribution to the quantum compilation literature.","headline":"First real attempt at multi-programming for neutral atoms, but the correctness claim is unsupported because the gate parallelization constraint only forbids coordinate overlap, not blockade-radius proximity.","tokens_in":26593,"tokens_out":1743,"would_cite":false,"duration_ms":21957,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"DYNAMO claims neutral atom arrays can execute several quantum programs concurrently, with up to 14.39x faster compilation and 50.47% fewer execution stages.","keywords":["quantum multi-programming","neutral atom quantum computing","quantum compilation","AOD movement constraints","SMT scheduling","spatial deformation model","quantum operating systems","Rydberg stage reduction"],"falsifier":"Inspect a DYNAMO-produced multi-program schedule and find one Rydberg stage where a qubit belonging to program B lies within the blockade radius of the two-qubit gate pair of program A without occupying their exact coordinates; under a global Rydberg pulse that pair would entangle the wrong atoms, contradicting the correctness claim. If no such configuration exists in any test circuit, the central claim survives this check.","tokens_in":25598,"feed_emoji":"⚛️","tokens_out":6405,"duration_ms":67648,"temperature":0.7,"pith_summary":"This paper argues that neutral atom quantum computers can run several independent quantum programs at the same time on one or more atom arrays, and that doing so is the natural step toward a quantum operating system. It introduces DYNAMO, a compiler that schedules circuits together by treating each already-compiled program as a shape occupying space and time on the array. The two technical pieces are a spatial deformation model that respects the order constraints of movable AOD traps and a constraint-based scheduler that places new programs into the gaps left by earlier ones. The reported payoff is up to 14.39x faster compilation than a merged single-circuit baseline and an average 50.47% reduction in Rydberg stages relative to sequential compilation, for circuits from 12 to over 1200 gates.","feed_headline":"Multi-programming cuts compile times 14x on neutral atom chips","feed_subtitle":"The DYNAMO compiler interleaves independent circuits on shared atom arrays, a step toward quantum operating systems.","key_machinery":"The carrying mechanism is cycle-wise spatial deformation: a compiled circuit is decomposed into cycles, each split into an AOD movement step and a non-movement operation step. At every cycle the existing programs' AOD moves partition the array into Order-Preserving Zones, where new rows or columns must not cross the moving trap's path, and Order-Free Zones, where new moves are unconstrained. A second SMT-based scheduler inserts the next circuit into the feasible zones while enforcing the two-qubit gate parallelization constraint, and a greedy length-based scheduler distributes circuits across arrays. This is what lets multiple programs share the same atom array without violating AOD ordering.","core_discovery":"The central claim is that multi-programming on dynamically field-programmable neutral atom arrays can be made correct and efficient by decomposing each compiled circuit into cycles, each consisting of an AOD movement step followed by a non-movement operation step, and using the first circuit's movements to define Order-Preserving and Order-Free Zones. A second circuit can then be inserted into the feasible zones at each cycle, with SMT constraints enforcing AOD directionality and gate placement. This dynamic spatial deformation turns the global AOD movement constraint, which blocks naive resource partitioning, into a structured scheduling problem. The paper reports that the method compiles multiple circuits with up to 14.39x speedup over a merged DPQA baseline, reduces Rydberg stages by 50.47% on average versus sequential DPQA, and spreads workloads evenly across multiple QPUs.","pith_inferences":["The Appendix B parallelization constraint only forbids exact coordinate equality, not blockade-radius proximity across programs; until separation by at least the Rydberg blockade radius is enforced for all qubits of different programs, the correctness claim may not hold on physical hardware where a global Rydberg pulse would excite any atom inside the blockade region.","The headline stage reduction of 50.47% is measured against sequential DPQA, not merged DPQA; a reader weighing absolute circuit depth should also consider the rows where DYNAMO's stage count is higher than the merged baseline.","The spatial-deformation idea may transfer to other reconfigurable qubit architectures such as zoned neutral atom processors or ion shuttling systems, but the Order-Preserving/Order-Free distinction would need reformulation for their movement rules.","A testable extension is to add an explicit pairwise blockade-radius constraint between all qubits of different programs at every Rydberg stage and measure how much the reported stage reductions shrink; that would separate the scheduling gain from the residual hardware-safety risk."],"forward_implications":["If correct, several independent quantum circuits can be interleaved on a single neutral atom array with AOD movement order preserved and gate positions kept separate.","A quantum operating system could use DYNAMO as its scheduler, deciding which circuits share which array and when, while balancing QPU loads.","Shorter-circuit-first scheduling keeps grouped workloads tractable, unlike merged compilation which timed out on all grouped benchmarks within 10,000 seconds.","The same compiled circuit can be reused as a space occupation to guide placement of later circuits, reducing Rydberg stages by filling temporal gaps.","Balanced multi-resource distribution across two or three QPUs suggests the method extends to larger parallel quantum systems."],"supporting_citations":[{"why":"Supplies the DPQA solver baseline and the variable definitions and stage abstraction that DYNAMO's constraint formulation builds on.","marker":"[33]"},{"why":"Introduces the SMT formulation of neutral atom compilation constraints that DYNAMO extends to multi-program scenarios.","marker":"[27]"},{"why":"Defines the neutral atom architecture constraints, including SLM/AOD traps and Rydberg blockade, that motivate the spatial model.","marker":"[24]"},{"why":"Establishes multi-programming as a way to improve quantum computer throughput and utilization, the concept DYNAMO brings to neutral atoms.","marker":"[11]"},{"why":"Provides the superconducting multi-programming framework with scheduling and resource partitioning goals that DYNAMO adapts.","marker":"[13]"},{"why":"Supplies the RevLib reversible-circuit benchmarks used across the circuit-depth groups.","marker":"[48]"},{"why":"Supplies Sabre benchmarks and the qubit mapping baseline that the paper cites for circuit sources.","marker":"[47]"}],"fun_headline_variants":["Quantum OS step: DYNAMO runs multiple circuits at once","14x faster compile for neutral atom multi-tasking","DYNAMO: run multiple quantum programs on atom chips","Neutral atom quantum gets multi-tasking compiler","Compile quantum programs in parallel on atom arrays"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The paper assumes that as long as a new gate does not land on the exact coordinates of an existing gate at the same stage, it does not disturb that gate; in reality, any qubit from another program sitting within the Rydberg blockade radius of an executing two-qubit gate pair would be excited too, so the absence of exact overlap is not enough to guarantee correctness.","fun_headline_variants_meta":{"raw":{"variants":["Quantum OS step: DYNAMO runs multiple circuits at once","14x faster compile for neutral atom multi-tasking","DYNAMO: run multiple quantum programs on atom chips","Neutral atom quantum gets multi-tasking compiler","Compile quantum programs in parallel on atom arrays"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000864,"raw_usage":{"total_tokens":3731,"prompt_tokens":917,"completion_tokens":2814,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":533,"completion_tokens_details":{"reasoning_tokens":2737}},"tokens_in":533,"tokens_out":2814,"duration_ms":22721,"temperature":1.0,"reasoning_tokens":2737,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T19:38:44.197561+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Inspect a DYNAMO-produced multi-program schedule and find one Rydberg stage where a qubit belonging to program B lies within the blockade radius of the two-qubit gate pair of program A without occupying their exact coordinates; under a global Rydberg pulse that pair would entangle the wrong atoms, contradicting the correctness claim. If no such configuration exists in any test circuit, the central claim survives this check.","supporting_citations":[{"cited_title":"Compiling Quantum Circuits for Dynamically Field-Programmable Neutral Atoms Array Processors,","cited_arxiv_id":null,"evidence_quote":"Supplies the DPQA solver baseline and the variable definitions and stage abstraction that DYNAMO's constraint formulation builds on."},{"cited_title":"Qubit Mapping for Reconfigurable Atom Arrays,","cited_arxiv_id":null,"evidence_quote":"Introduces the SMT formulation of neutral atom compilation constraints that DYNAMO extends to multi-program scenarios."},{"cited_title":"Exploiting long-distance interactions and tolerating atom loss in neutral atom quantum architectures,","cited_arxiv_id":null,"evidence_quote":"Defines the neutral atom architecture constraints, including SLM/AOD traps and Rydberg blockade, that motivate the spatial model."},{"cited_title":"A Case for Multi- Programming Quantum Computers,","cited_arxiv_id":null,"evidence_quote":"Establishes multi-programming as a way to improve quantum computer throughput and utilization, the concept DYNAMO brings to neutral atoms."},{"cited_title":"QuCloud: A New Qubit Mapping Mechanism for Multi-programming Quantum Computing in Cloud Environment,","cited_arxiv_id":null,"evidence_quote":"Provides the superconducting multi-programming framework with scheduling and resource partitioning goals that DYNAMO adapts."},{"cited_title":"RevLib: An Online Resource for Reversible Functions and Reversible Circuits,","cited_arxiv_id":null,"evidence_quote":"Supplies the RevLib reversible-circuit benchmarks used across the circuit-depth groups."},{"cited_title":"Tackling the Qubit Mapping Problem for NISQ-Era Quantum Devices,","cited_arxiv_id":null,"evidence_quote":"Supplies Sabre benchmarks and the qubit mapping baseline that the paper cites for circuit sources."}],"review_version":1}