{"id":"9558eaeb-bb8c-4b13-94e0-05f137209218","arxiv_id":"2411.12263","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"high","formal_verification":"none","parameter_count":1,"one_line_summary":"PowerMove claims order-of-magnitude fidelity improvements on neutral atom quantum computers by using continuous qubit routing and a storage zone.","lead":"PowerMove is a compiler for neutral atom quantum computers that routes qubits directly between layouts and parks idle qubits in a storage zone during Rydberg gates. The paper claims large fidelity gains and much faster compilation than the leading compiler, but the evaluation rests on a simplified hardware model with no released code.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Sec. 5.3's Coll-Move grouping checks only pairwise endpoint order; it never verifies that atoms sharing an AOD row/column can be moved independently, so generated schedules may be physically unrealizable.","rationale":"The central claim is that PowerMove produces physically executable compiled circuits with large fidelity and execution-time improvements. That claim depends on the schedules being realizable on AOD hardware. The paper's only feasibility check for grouped 1Q moves is the pairwise order-preservation test in Sec. 5.3. That test is incomplete: it treats each 1Q move as an independent endpoint pair, ignoring that an AOD row or column is a single degree of freedom shared by all atoms in that row or column, and it ignores persistence of row/column identity across consecutive Coll-Moves. A concrete failure mode is two atoms co-located after a CZ gate: if both must move to different sites in the next stage, any attempt to move one through the AOD also moves the other unless the second is first transferred to SLM and reloaded later; the paper does not model or schedule around this. The evaluation is circular with respect to this assumption because the same pairwise criterion is used as both the optimizer's feasibility predicate and the measurement model. Secondary issues, such as the T_exe improvement column being computed from the non-storage variant rather than the full with-storage method, are real and should be corrected, but they only weaken one numeric claim. The physical-feasibility gap, if it lands, invalidates the schedules themselves and therefore the central claim. The proposed AOD-level checker would settle the question directly, so the reader's REJECT verdict is retained.","tokens_in":20356,"tokens_out":23756,"duration_ms":274213,"concrete_test":"Implement an AOD-level feasibility checker for the schedules emitted by PowerMove. For each Coll-Move, record the row/column index of every atom in the AOD and verify that (i) atoms in the same row/column share start and end coordinates, (ii) no two atoms in different rows/columns cross at intermediate times under linear or planner-controlled interpolation, and (iii) row/column index assignments are consistent across consecutive Coll-Moves unless an SLM transfer and reload occurs, with that transfer counted in N_trans and T_exe. Run this checker on the BV-70, QFT-29, and QSIM-rand-40 rows of Table 3. If any pairwise-conflict-free group fails (i)-(iii), the reported gains are invalid; if all pass, the concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"An AOD array moves all atoms in a row with one y-coordinate and all atoms in a column with one x-coordinate; individual atoms cannot be addressed independently within the same array. The conflict test in Sec. 5.3 only declares two 1Q moves conflicting if the relative x/y order of their endpoints flips. This is necessary for non-crossing row/column motion, but it is not sufficient for a multi-stage schedule. Pairwise order preservation permits two atoms that start in different rows/columns to end at the same y/x, and it does not constrain atoms that already share a row/column at the start of a move, such as the two atoms of a CZ pair brought to one site. If such atoms are assigned different destinations in the next stage, any Coll-Move that moves one will move the other, since they share row/column coordinates; separating them requires an extra SLM transfer and reload that the grouping algorithm neither checks nor costs. The paper provides no proof, cycle-accurate AOD simulation, or released code showing that every conflict-free group can be embedded in one AOD array with fixed row/column identities across the whole schedule. Because both the optimizer and the evaluation use the same pairwise criterion, the reported execution-time and fidelity gains inherit this unverified physical-feasibility assumption.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes PowerMove, a compiler for neutral-atom quantum computers with a zoned architecture. PowerMove jointly handles stage scheduling, qubit allocation, qubit movement, and storage-zone placement: a Stage Scheduler orders CZ stages to reduce inter-zone movement, a Continuous Router determines single-qubit moves for direct layout transitions, a Coll-Move grouping step packs these moves into collective AOD moves, and a Coll-Move Scheduler orders and parallelizes the resulting groups across multiple AOD arrays. The evaluation compares PowerMove with Enola and reports fidelity improvements of several orders of magnitude (for example, BV-70 rising from 6.92e-4 to 0.75), execution-time reductions of 1.71x to 3.46x, and compilation-time reductions up to 213.5x.","tokens_in":20567,"tokens_out":9955,"duration_ms":118350,"significance":"If the main claims are correct, this would be a valuable compiler contribution: it addresses an important and timely hardware feature (the zoned architecture), is evaluated against the strongest available baseline (Enola), and the improvements come from heuristics rather than from fitting parameters to the reported fidelity numbers, which is a strength. The ablation study and the multi-AOD analysis are also helpful. However, the central movement-grouping step is not shown to produce physically executable AOD schedules, and no code or simulator is provided to independently verify the schedules. The reported gains are therefore not yet supported as stated.","major_comments":[{"comment":"The conflict test used for Coll-Move grouping checks only whether the relative x/y order of the two endpoints flips. This condition is necessary for non-crossing row/column motion in an AOD, but it is not sufficient: it does not verify that the moves in a group can be assigned to rows and columns of a single AOD array with fixed identities, and it does not model the SLM transfers required to load or unload atoms whose row or column is shared. As a result, a collision-free group may contain moves that are physically unrealizable (for example, two atoms that must separate from a common CZ site while still sharing a row/column, or moves that require two rows/columns to coincide). Because the Table 3 evaluation uses this same grouping routine with no independent physical check, the reported fidelity and execution-time improvements rest on an unverified feasibility assumption. The paper needs either a proof that every conflict-free group is realizable by one AOD array, a cycle-accurate AOD simulation that validates the schedules, or a corrected grouping algorithm that incorporates row/column identity and transfer constraints.","section":"Section 5.3"},{"comment":"The multi-AOD scheduler partitions the already-grouped Coll-Moves into parallel batches, but it never assigns qubits to specific AOD arrays or verifies that a qubit's movement can be partitioned across arrays without extra SLM-to-AOD transfers. The claim that the transfer error is unchanged because 'the number of transfers does not change' is not justified: distributing groups across multiple AODs may require additional transfers to move atoms between arrays, which would change N_trans in Eq. (1). Without a resource-aware allocation, the multi-AOD speedups shown in Figure 7 are not demonstrated.","section":"Section 6.2"}],"minor_comments":[{"comment":"The stage-scheduling objective contains a free parameter alpha < 1, but the paper does not report a sensitivity analysis or justify the chosen value beyond the qualitative preference for move-in operations.","section":"Section 4.2"},{"comment":"In several rows the with-storage fidelity is lower than the non-storage fidelity (for example, QAOA-regular4-40: 0.42 vs 0.45, and VQE-30: 0.79 vs 0.81), which is not discussed in the text claiming that the storage zone 'significantly enhances fidelity.'","section":"Table 3"},{"comment":"The claimed average improvement of 313.86x for the storage zone is not defined; because the per-benchmark ratios span several orders of magnitude, please state whether this is a geometric or arithmetic mean and report the per-benchmark ratios.","section":"Section 7.2"},{"comment":"The benchmarks are randomly generated, but the evaluation does not report variance across seeds or error bars; adding this would strengthen the comparison.","section":"Section 7.3 / Figure 7"},{"comment":"The paper states that code will be open-sourced later; for a compiler evaluation, making the artifact available at submission time would materially improve reproducibility.","section":"Abstract / Section 9"},{"comment":"The description 'search around its current location to find the nearest empty site' is underspecified; please give the search order, distance metric, and tie-breaking rule.","section":"Section 5.2, Step 3"}],"recommendation":"major_revision","confidential_remarks":"This is a borderline case. The Coll-Move feasibility gap is serious enough that a rejection could also be defended, because the headline results may be built on physically unrealizable schedules. I would only accept a revision that adds a rigorous physical model or AOD simulation, re-runs the evaluation with corrected schedules, and ideally releases the artifact."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a real contribution to neutral-atom compilation—the continuous router and first-time storage-zone handling are new relative to Enola and company—but the paper's load-bearing assumption about physically realizable collective moves is unproven, and the evaluation does not independently validate it. I'd send it to a serious referee, expecting major revision.\n\nWhat's good: the problem is real and current. Enola's revert-to-initial-layout movement is indeed wasteful, and the storage zone is an obvious next feature. The continuous router's idea of letting the current allocation guide the next stage's movements, rather than routing back through a fixed layout, is sensible and, as far as I know, new. The stage scheduler and Coll-Move scheduler are simple but reasonable greedy heuristics, and the multi-AOD parallelization is a nice practical addition. The experimental sweep across QAOA, QFT, BV, VQE, and QSim is broad, and the compilation-time speedups over Enola are striking.\n\nThe soft spots are serious. Section 5.3 defines a conflict between two 1Q moves solely as an x- or y-order flip. That is necessary for AOD non-crossing but not sufficient. The paper never proves that every conflict-free group is realizable by one AOD array whose rows and columns move in tandem. In particular, two atoms that already share a row or column—e.g., a CZ pair at one site—cannot be independently assigned different y or x destinations in a single collective move; separating them needs an SLM transfer and reload, which the grouping neither checks nor costs. The stress-test note is right. Without a feasibility proof or a cycle-accurate AOD simulation, the generated schedules may be physically unrealizable, which would invalidate the reported fidelity and timing numbers.\n\nThe evaluation also has issues. Both the optimizer and the reported results use the same simplified fidelity model, with no error bars and no code released. The headline execution-time improvement (up to 3.46x) comes from the non-storage variant; the with-storage variant is often slower than non-storage and sometimes worse in fidelity (e.g., QAOA-regular4-40, VQE-30). The paper's own Table 3 shows this, which the text glosses over.\n\nNet: the core idea deserves attention, but the key correctness claim is unresolved. A serious referee should ask for a proof or a detailed AOD motion simulation, plus code or a reproducibility artifact. If the feasibility issue is fixable, this could become a useful compiler; right now it's a promising system with an unverified physical foundation.","headline":"PowerMove is a genuine new compiler idea for neutral-atom machines, but its collective-move feasibility criterion is unproven and the evaluation leans on the same unverified model.","tokens_in":21123,"tokens_out":4770,"would_cite":false,"duration_ms":47782,"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":"PowerMove claims that jointly optimizing gate scheduling, qubit allocation, qubit movement, and the zoned architecture raises neutral-atom circuit fidelity by orders of magnitude while cutting execution and compilation time.","keywords":["neutral atom quantum computing","zoned architecture","qubit movement","acousto-optic deflector","compilation","gate scheduling","fidelity optimization","circuit routing"],"falsifier":"Construct a small example on a grid where each pair of single-qubit moves preserves relative x/y order, but no single AOD frame can realize all moves because its rows and columns must move together and cannot cross. Simulating or implementing PowerMove's grouping on such a case would settle whether its collision-free criterion is physically sufficient; if such a case exists, the movement schedules and reported fidelity and execution-time gains would need to be recomputed.","tokens_in":20109,"feed_emoji":"⚛️","tokens_out":6251,"duration_ms":61723,"temperature":0.7,"pith_summary":"PowerMove is a compiler for neutral-atom quantum computers that treats the zoned architecture as a first-class optimization resource. Its central thesis is that the four decisions a neutral-atom compiler faces—when to fire parallel two-qubit gates, where to place qubits, how to move them, and when to shelter them in a storage zone—constrain one another and must be made jointly rather than in sequence. On the paper's benchmarks, the jointly optimized compilation raises circuit fidelity by several orders of magnitude over the previous leading compiler, with a 70-qubit Bernstein-Vazirani circuit improving from about $7\\times 10^{-4}$ to $0.75$, and reduces execution time by 1.71x to 3.46x and compilation time by up to 213.5x. The reason to care is that fidelity, not qubit count, is the limiting resource in near-term neutral-atom computation; if these numbers hold, much larger algorithms become usable and the storage zone stops being a hardware curiosity.","feed_headline":"Neutral-atom compiler lifts 70-qubit fidelity from 0.0007 to 0.75","feed_subtitle":"PowerMove schedules stages, routing, and storage together, cutting execution time up to 3.5x and compile time up to 213x.","key_machinery":"The load-bearing machinery has three pieces. First, a stage scheduler that partitions CZ gates by greedy edge coloring and reorders the stages so successive stages share as many interacting qubits as possible, with an asymmetry that favors sending qubits into storage over pulling them out. Second, a continuous router that formulates each qubit's next-stage destination as a target site coordinate, classifies qubits as static, mobile, or undecided by zone choice and clustering risk, then greedily groups the single-qubit moves into collective moves using a pairwise conflict test: two moves conflict if they reverse their relative x-order or y-order. Third, a Coll-Move scheduler that executes groups in descending order of move-ins minus move-outs and, with multiple AOD arrays, partitions groups across arrays to parallelize movement. Together these convert the compiler's problem from separate layout-to-layout transitions into one continuous layout-transition problem with storage-zone awareness.","core_discovery":"The paper claims to be the first neutral-atom compiler to integrate the storage zone while allowing direct transitions between computation layouts instead of returning to an initial layout after every Rydberg stage. Its continuous router assigns every qubit a target site for the next stage, including sites inside or outside the storage zone, and then groups the resulting single-qubit movements into collective moves that respect the constraint that rows and columns of a mobile trap array move in tandem and cannot cross. The stage scheduler orders CZ stages to minimize qubit interchange between the storage and computation zones, and the Coll-Move scheduler orders and parallelizes collective moves to maximize dwell time in storage. In evaluation, the with-storage configuration removes essentially all excitation error, improves fidelity versus the no-storage variant by 313.86x on average, and the full pipeline beats the previous best compiler in every reported benchmark.","pith_inferences":["A testable extension: the pairwise x/y-order conflict check is a necessary condition, but the paper does not show it is sufficient for an AOD frame whose rows and columns must move in tandem; a brute-force check on small grids would validate or falsify the grouping method.","The same continuous-routing idea could apply to other reconfigurable quantum platforms, such as trapped-ion shuttling, if the movement constraint is expressed as an order-preservation invariant.","The fidelity model treats storage-zone decoherence as negligible; re-running the evaluation with a finite storage-zone $T_2$ would show how much of the reported 313.86x with-storage gain depends on that hardware assumption.","The reported compilation-time scaling suggests the heuristics may reach thousands of qubits, but deeper circuits with more CZ layers would test whether greedy stage ordering degrades relative to global scheduling."],"forward_implications":["Near-term neutral-atom circuits previously limited by decoherence and excitation errors can run at useful fidelity; in the paper, a 70-qubit Bernstein-Vazirani circuit goes from $6.92\\times10^{-4}$ to $0.75$ with storage.","Eliminating the return-to-home layout means fewer transfers and shorter movement paths, directly cutting execution time by 1.71x to 3.46x across benchmarks.","Storage-zone placement removes the Rydberg excitation error for non-interacting qubits, which grows with circuit size and becomes the dominant error for large programs.","Near-linear heuristics replace expensive maximum-independent-set solvers, cutting compilation time by up to 213.5x and making larger circuits compilable in practice.","Multiple independently controlled AOD arrays give a direct hardware tradeoff: more arrays shorten movement latency and reduce decoherence without changing transfer error."],"supporting_citations":[{"why":"Supplies the movement model and the leading-baseline fidelity and timing numbers that all comparisons must beat.","marker":"[8]"},{"why":"The zoned-architecture hardware reference; provides zone-separation distance, Rydberg excitation error, and gate fidelity values used in the fidelity model.","marker":"[6]"},{"why":"Establishes the AOD row/column tandem and non-crossing constraints and the acceleration limit that bound collective moves.","marker":"[7]"},{"why":"Solver-based qubit mapping baseline; motivates the need for scalable heuristics and supports the NP-hardness framing.","marker":"[42]"},{"why":"Solver-based dynamic-field compiler baseline whose stage-synthesis input format and scalability limits PowerMove builds on.","marker":"[44]"},{"why":"SWAP-based compiler baseline whose extra two-qubit gates PowerMove avoids by moving atoms instead.","marker":"[46]"},{"why":"Flying-ancilla routing baseline representing an alternative movement strategy compared via the leading baseline.","marker":"[47]"}],"fun_headline_variants":["Neutral-atom compiler: 0.0007 to 0.75 fidelity, 213x faster compilation","PowerMove: first to integrate storage zone for neutral-atom compilers","Fidelity leap: neutral-atom compiler reaches 0.75 from 0.0007","Compiler speeds neutral-atom execution 3.46x, compile 213x","PowerMove integrates storage zone, lifting fidelity by orders of magnitude"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that checking each pair of moves preserves left-right and top-bottom order is enough to guarantee that one mobile trap array, whose rows and columns must move together and cannot cross, can actually execute all the grouped movements in that layout transition.","fun_headline_variants_meta":{"raw":{"variants":["Neutral-atom compiler: 0.0007 to 0.75 fidelity, 213x faster compilation","PowerMove: first to integrate storage zone for neutral-atom compilers","Fidelity leap: neutral-atom compiler reaches 0.75 from 0.0007","Compiler speeds neutral-atom execution 3.46x, compile 213x","PowerMove integrates storage zone, lifting fidelity by orders of magnitude"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000958,"raw_usage":{"total_tokens":4062,"prompt_tokens":905,"completion_tokens":3157,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":521,"completion_tokens_details":{"reasoning_tokens":3047}},"tokens_in":521,"tokens_out":3157,"duration_ms":24059,"temperature":1.0,"reasoning_tokens":3047,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T17:44:52.103285+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Construct a small example on a grid where each pair of single-qubit moves preserves relative x/y order, but no single AOD frame can realize all moves because its rows and columns must move together and cannot cross. Simulating or implementing PowerMove's grouping on such a case would settle whether its collision-free criterion is physically sufficient; if such a case exists, the movement schedules and reported fidelity and execution-time gains would need to be recomputed.","supporting_citations":[{"cited_title":"Compilation for dynamically field-programmable qubit arrays with efficient and provably near-optimal scheduling","cited_arxiv_id":null,"evidence_quote":"Supplies the movement model and the leading-baseline fidelity and timing numbers that all comparisons must beat."},{"cited_title":"A quantum processor based on coherent transport of entangled atom arrays","cited_arxiv_id":null,"evidence_quote":"Establishes the AOD row/column tandem and non-crossing constraints and the acceleration limit that bound collective moves."},{"cited_title":"Qubit mapping for reconfigurable atom arrays","cited_arxiv_id":null,"evidence_quote":"Solver-based qubit mapping baseline; motivates the need for scalable heuristics and supports the NP-hardness framing."},{"cited_title":"Compiling quantum circuits for dynamically field-programmable neu- tral atoms array processors","cited_arxiv_id":null,"evidence_quote":"Solver-based dynamic-field compiler baseline whose stage-synthesis input format and scalability limits PowerMove builds on."},{"cited_title":"Atomique: A quantum compiler for reconfigurable neutral atom arrays","cited_arxiv_id":null,"evidence_quote":"SWAP-based compiler baseline whose extra two-qubit gates PowerMove avoids by moving atoms instead."}],"review_version":1}