{"id":"c8d7d8af-3d21-4c39-804f-150c938e1b45","arxiv_id":"2412.07969","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A numerical optimization framework automatically synthesizes dynamic quantum circuits for state and unitary preparation, achieving shallower circuits at the cost of extra ancillas and mid-circuit measurement errors.","lead":"This paper presents a software framework that automatically builds dynamic quantum circuits, which measure qubits mid-computation and use the results to choose later operations. The authors show these auto-generated circuits can be much shallower than standard circuits for preparing states and gates, a useful trade-off as mid-circuit measurement hardware improves.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 'any state or unitary' claim overreaches: the implemented generator lacks a completeness/convergence proof and has no random-target validation.","rationale":"The reader's conditional verdict is reasonable, but I only partially agree with the stated weakest assumption. The U3-only branch ansatz is not obviously the bottleneck: for any target V, one can choose an isometry Ub satisfying P_j = (I⊗<j|)Ub(I⊗|0^a>) = α_j U_j† V with local U_j, so local branch unitaries are expressively sufficient. The real gap is algorithmic: the paper gives no proof or systematic evidence that DC-QSearch/DC-Inst converges to a zero-cost solution for arbitrary targets, and Section VIII acknowledges the branch restriction is a simplification. The cost-function derivations are sound and the small-demonstration/hardware/noise analysis is transparent and valuable. The 'first practical procedure' claim also conflicts with the paper's own citation of [1] as a prior variational dynamic-circuit state-preparation method, which further supports softening the abstract's universality/novelty wording. I recommend keeping the conditional verdict rather than accepting or rejecting: the framework is a solid contribution, but the headline 'any state or unitary' needs either a completeness proof, random-target benchmarks, or explicit qualification.","tokens_in":28023,"tokens_out":20546,"duration_ms":234319,"concrete_test":"Run DC-QSearch on a suite of 20 Haar-random 4-qubit pure states and 20 Haar-random 4-qubit unitaries, each with one ancilla on a linear topology, under a fixed budget (e.g., 10^6 instantiations or 30 minutes per target). Record the fraction of targets whose dynamic cost (Eq. (6) for states, Eq. (10) for unitaries) reaches below 1e-6. A success rate materially below 100% would show the implementation does not substantiate the 'any' claim; a 100% success rate would support the practical claim and motivate supplying a formal completeness proof.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The abstract's central claim, 'automatically prepare any state or unitary operator,' requires a completeness argument for the implemented generator described in Section VI: Ub is produced by QSearch or the DC-Inst pattern, and every branch unitary Ui is restricted to one layer of single-qubit U3 gates. The paper proves the cost functions are faithful (Appendices A-C), but it does not prove that a zero-cost solution exists within this restricted ansatz for an arbitrary target, nor that the numerical instantiation/search loop converges to one. Section VIII explicitly says that allowing entangled branch unitaries 'might further reduce the circuit depth,' marking the U3-only restriction as a heuristic simplification. All demonstrations use structured, symmetric targets (GHZ, W, Dicke, Toffoli, long-range gates); no random dense state or unitary is tested. This is load-bearing because the headline assertion is exactly the universality claim. To be fair, a simple isometry-extension argument suggests local U3 branch unitaries are expressively sufficient in principle; the missing piece is the algorithmic guarantee that the search finds the required Ub. The evidence supports a qualified claim: 'for the demonstrated targets, and in principle for arbitrary targets given sufficient search depth,' rather than the unqualified 'any.'","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents AC/DC, a framework for automatically synthesizing dynamic quantum circuits that include mid-circuit measurements and classical feed-forward. The authors derive cost functions for state preparation (Eq. 6) and unitary preparation (Cdyn1, Eq. 9; Cdyn2, Eq. 10), prove their faithfulness in Appendices A-C, and integrate them into the open-source BQSKit synthesis toolkit via two algorithms, DC-QSearch and DC-Inst. They demonstrate the framework on GHZ, W, and Dicke state preparation; long-range entangling gates; multi-qubit gates; large-circuit optimization through partitioning; and a lattice-simulation application, with validation on simulators and quantum hardware. They also present a noise model analyzing when dynamic circuits can outperform unitary circuits.","tokens_in":28272,"tokens_out":4489,"duration_ms":45004,"significance":"If the results hold, this is a valuable step toward practical automated compilation for dynamic circuits. The cost functions are derived from first principles, and the faithfulness proofs in Appendices A-C are rigorous and reusable beyond the specific BQSKit implementation. The integration into an open-source framework, the hardware validation, and the noise trade-off analysis are concrete strengths that will benefit the community. The main caveat is that the headline claim of preparing 'any state or unitary operator' is not supported by the implemented generator as described; the implemented branch unitaries are restricted to single-qubit layers, and the large-circuit scaling claims rest on pattern extrapolation rather than direct synthesis. These issues are fixable with qualified claims and additional evidence, but they affect the central assertion of the paper.","major_comments":[{"comment":"The abstract's claim that the framework can 'automatically prepare any state or unitary operator' is not supported by the implemented generator. Section VI restricts every branch unitary Ui to a single layer of single-qubit U3 gates, and Section VIII explicitly acknowledges that allowing entangled branch unitaries 'might further reduce the circuit depth' and that simultaneous synthesis of Ub and all Ui is left to future work. The paper proves faithfulness of the cost functions but does not prove that a zero-cost solution exists within this restricted ansatz for an arbitrary target, nor does it provide a convergence guarantee for the numerical search. All demonstrations use structured, symmetric targets; no random dense state or unitary is tested. The claim should be qualified to 'for the demonstrated targets, and in principle for arbitrary targets given sufficient search depth and more general branch unitaries,' or supplemented with a completeness argument for the implemented ansatz.","section":"Abstract and Section VI"},{"comment":"The uniform-measurement restriction W = D H^{⊗n} is introduced as an 'empirical observation' and then used as an assumption in the type-1 cost function. This restriction may change the optimization landscape: a zero-cost solution may exist for the full W but not within the D H^{⊗n} family, so Cdyn1 with this restricted W is not proven faithful to the original dynamic-circuit problem. The paper does not show that uniform measurement probabilities are without loss of optimality for general targets; Figure 3 illustrates only a single next-nearest-neighbor CNOT. This should be stated as a heuristic with a discussion of its possible impact on completeness, or removed by optimizing W analytically as in Cdyn2.","section":"Section IV, Eq. (12)"},{"comment":"Large-circuit results are obtained by DC-Inst, which extrapolates a pattern observed on circuits with at most six qubits, and Figure 10 explicitly labels the large-size depth curves as 'predictions derived from observed patterns.' Tables I and II report GHZ7/GHZ8 and W6 results from this extrapolation, but no direct synthesis or hardware validation is provided for those sizes. The scaling claims should be clearly labeled as extrapolations, and ideally validated by at least one intermediate-size directly synthesized circuit or by executing the predicted circuit on a simulator.","section":"Section VII A and Figure 10"}],"minor_comments":[{"comment":"There is a typo in 'allows us to reproduce all previosuly dynamic circuit published results'; 'previosuly' should be 'previously.'","section":"Section VI"},{"comment":"The caption contains the typo 'DC achievs the shortest circuit depth'; 'achievs' should be 'achieves.'","section":"Table III caption"},{"comment":"The text states that 'Each blue block is explained in Eq. (19),' but Eq. (19) is empty in the manuscript; the block's unitary decomposition (one CNOT plus two U3 gates) should be written explicitly.","section":"Equation (19) and Figure 5"},{"comment":"The unitaries Ub0 and Ub1 used in the independent-measurement expression are not defined; please define them or relate them to the general Ub notation used elsewhere.","section":"Section V, Eq. (15)"},{"comment":"The equivalence argument between Cdyn1 and Cdyn2 would be clearer if it stated the monotonic relation Cdyn2 = 1 - (1 - Cdyn1)^2, which makes the optimization equivalence immediate; as written, 'both are aiming to maximize |T|' is somewhat informal.","section":"Appendix B"},{"comment":"The caption panel labels are inconsistent: panel (c) is described as 'CNOT between (q0, q3),' but the figure appears to show a three-qubit circuit; please reconcile the labels and the described gates.","section":"Appendix D, Figure 15"}],"recommendation":"major_revision","confidential_remarks":"The paper is technically solid in its derivations and demonstrations, but the universality claim in the abstract and the 'first practical procedure' claim should be reconciled with the acknowledged limitations in Sections VI and VIII and with the existing variational approach in Ref. [1]. The editor may wish to ask the authors to soften the novelty claim and to separate proven results from pattern-based extrapolations. This is within the scope of a major revision rather than a rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's my take. The paper gives something genuinely useful: a general automated synthesis route to dynamic circuits, with MCM-aware cost functions derived from Hilbert-Schmidt and channel fidelity, integrated into BQSKit. The faithfulness proofs in Appendices A-C are real and check out. Demonstrations cover GHZ/W/Dicke states, long-range gates, Toffoli/Fanout/Fredkin, and a lattice simulation, validated in simulation and on hardware. That is a solid engineering contribution. I'd also credit the authors for reporting the hardware fidelity degradation transparently instead of hiding it.\n\nThe real new thing is Cdyn1 and Cdyn2 for unitary preparation, plus the fact that numerical optimization replaces hand-crafted teleportation protocols for the demonstrated targets. It is not the first dynamic state preparation — Alam and Clark's variational approach is cited — but the unitary case and the BQSKit integration are new. The state cost function is a clean multi-branch fidelity generalization.\n\nThe soft spots are real but localized. The abstract says the framework prepares 'any state or unitary', yet the implemented generator restricts each branch Ui to a single layer of single-qubit U3 gates, and the paper gives no expressiveness or convergence proof for that ansatz. An isometry-extension argument suggests local branch unitaries are sufficient in principle, so the 'any' claim is plausible but not established for the actual algorithm. The stress-test note has this right, and the paper's own discussion flags the restriction as a simplification. Still, the abstract should be qualified. The large-circuit depth predictions in Figure 10 are extrapolations from observed patterns, not synthesized circuits; useful as projections, but they should not be read as measured dynamic circuit performance. And there is no random dense state or unitary tested, only structured targets, so we have no empirical evidence for the 'any' claim beyond those cases.\n\nData and citation pattern look fine. The BQSKit integration is real and reproducible, and self-citation is not an issue here. I'd want a random-target test and a clear statement about ansatz completeness before the universality claim is accepted. Even without that, the framework is valuable for compiler developers, MCM hardware benchmarking, and algorithm designers weighing dynamic-circuit trade-offs.\n\nIt deserves a serious referee. The formal core is sound and the empirical work is honest; the needed changes are calibration of claims and a bit more evidence, not a rethink.","headline":"Useful synthesis framework and honest hardware study; the 'any' claim outruns the implemented U3-only branch ansatz.","tokens_in":28769,"tokens_out":1510,"would_cite":true,"duration_ms":17730,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P68"],"pacs":["03.67.Lx"],"model":"deepseek-v4-flash","headline":"The paper presents an automated framework that synthesizes dynamic quantum circuits—circuits with mid-circuit measurements and feed-forward operations—to prepare arbitrary quantum states or implement arbitrary unitary operators, using…","keywords":["dynamic quantum circuits","mid-circuit measurement","feed-forward","circuit synthesis","numerical optimization","state preparation","unitary preparation","cost function"],"falsifier":"Run the generator on a Haar-random three-qubit unitary with one ancilla and compare the optimized cost under the single-U3-layer branch restriction against a version that allows two-qubit gates in branches; if the restricted version cannot reach the same near-zero cost that the unrestricted version reaches, the expressiveness assumption underlying \"any unitary\" is false.","tokens_in":27834,"feed_emoji":"⚛️","tokens_out":7263,"duration_ms":75386,"temperature":0.7,"pith_summary":"Dynamic quantum circuits—circuits that measure some qubits mid-execution and choose later gates based on the outcome—have so far been hand-designed, mostly from teleportation protocols. This paper claims to replace that manual design with numerical optimization: given a target state or unitary, a search routine finds a circuit of the form \"prepare with $U_b$, measure ancillas, then apply branch unitary $U_i$ depending on outcome\" that realizes the target. The paper introduces two families of cost functions, one for state preparation and one for unitary preparation, and proves they are faithful. It implements these in an open-source synthesis framework and shows the resulting dynamic circuits match or beat static circuits in depth across GHZ, W, and Dicke states, long-range entangling gates, multi-qubit gates, block-level optimization of quantum algorithms, and lattice simulations. Hardware and noisy-simulation validation show the depth savings are real but currently offset by mid-circuit measurement errors; the paper's noise model quantifies the measurement fidelity needed for dynamic circuits to win.","feed_headline":"Automated search designs dynamic circuits for any quantum target","feed_subtitle":"Mid-circuit measurement and feed-forward, tuned numerically, cut circuit depth on states, gates, and simulation blocks.","key_machinery":"The mechanism is the branched circuit identity $U = \\sum_i (U_i \\otimes |i\\rangle\\langle i|) U_b$, interpreting the measurement as a projection-valued decomposition of the ancilla. The new cost functions turn the non-unitary measurement process into an objective a numerical optimizer can drive to zero: for unitary preparation the key trick is deferring the measurement, so the task becomes making the pre-measurement unitary $U$ equal to $V\\otimes W$ on the ancilla-zero subspace, and $C_{\\mathrm{dyn1}}$ and $C_{\\mathrm{dyn2}}$ are the resulting Hilbert-Schmidt distances. Type 2 is also derived independently from a quantum-channel composition, giving it an operational meaning as the trace of the channel implementing $V^\\dagger$ after the dynamic circuit. The implementation combines a topology-aware search for the pre-measurement block with instantiation of one layer of single-qubit $U_3$ gates per branch, and the paper documents circuit patterns for GHZ, W, Dicke, long-range gates, and Trotter steps that allow scaling via fixed-layer growth or circuit partitioning.","core_discovery":"Dynamic circuits have mostly been designed by hand using teleportation and stabilizer reasoning. The paper's central claim is that they can instead be found automatically: for a target state or unitary, search over a pre-measurement unitary $U_b$ on system-plus-ancilla qubits and over outcome-conditioned branch unitaries $U_i$, guided by cost functions that are faithful—zero exactly when the circuit realizes the target. For state preparation the cost is $C_T = 1 - \\sum_i |\\langle T \\otimes i| (U_i \\otimes |i\\rangle\\langle i|) U_b |0^{\\otimes(s+a)}\\rangle|^2$. For unitary preparation, the paper defers the measurement and demands the pre-measurement state factorize as $(V|\\phi\\rangle)\\otimes|\\alpha\\rangle$, yielding two costs: $C_{\\mathrm{dyn1}}$ with an extra ancilla unitary $W$, and $C_{\\mathrm{dyn2}}$ with $W$ optimized away, $C_{\\mathrm{dyn2}}(U)=1-\\frac{1}{4^s}\\sum_i |\\mathrm{Tr}(V^\\dagger U_{i,0})|^2$, where $U_{i,0}$ is a partial trace over ancillas. The same costs extend to independent and nested measurement placements by changing the form of the total unitary $U$. The implementation uses a topology-aware search for $U_b$ and one layer of single-qubit $U_3$ gates per branch; on the demonstrated targets the resulting circuits are shallower than static-circuit baselines, and hardware validation confirms the protocol works, with fidelity currently limited by mid-circuit measurement errors.","pith_inferences":["An implication the authors leave implicit is that allowing two-qubit gates inside branch unitaries—not just one layer of single-qubit U3 gates—would likely reduce depth further, since the cost functions themselves do not constrain branch structure; the paper flags this as future work rather than claiming it.","If mid-circuit measurement fidelity reaches the thresholds their model computes, dynamic-circuit synthesis becomes a general depth-reduction tool for NISQ algorithms, not just for the demonstrated families; this is an extrapolation from the paper's trade-off curves.","The same generator could be used to build algorithm-level hardware benchmarks for mid-circuit measurement and feed-forward, since it produces many circuits with tunable depth, width, and measurement trade-offs; the paper mentions benchmarks as a motivation but does not build the suite.","The constant-depth long-range gate pattern suggests the technique extends naturally to distributing entanglement across arbitrary pairs in limited-connectivity processors, potentially aiding distributed quantum computing; the paper only demonstrates linear topologies."],"forward_implications":["For GHZ and W states, dynamic circuits reach depth 2–4 for up to 8 qubits, shallower than analytic or compiled static circuits, with depth growing more slowly with qubit count.","Long-range two-qubit gates such as CNOT, CZ, CS, Rzz, Rxx, and Ryy are realized in constant depth, with specific gates at depth 2 and generic two-qubit gates at depth 6, using one measurement per skipped qubit and cutting the number of mid-circuit measurements by a factor of three relative to Bell-state teleportation.","Partitioning large algorithms into three-qubit blocks and replacing blocks with one-ancilla dynamic circuits reduces circuit depth by about 30% on Grover, TFIM, and QAOA circuits at the cost of about 13% more CNOTs, with a mixed static/dynamic choice recovering the CNOT overhead.","For lattice-simulation Trotter steps mapped to limited-connectivity hardware, dynamic long-range Rzz circuits halve the CNOT count relative to structure-aware unitary synthesis, at constant circuit depth.","The noise model predicts a threshold mid-circuit measurement fidelity above which dynamic circuits beat static ones; the threshold drops as circuit width grows, and for lattice simulation it is already below current readout fidelity under the model's assumptions."],"supporting_citations":[{"why":"Supplies the topology-aware optimal search algorithm used as the synthesis backbone for the pre-measurement unitary.","marker":"[13]"},{"why":"Provides the Quantum Shannon Decomposition baseline whose CNOT counts and depths the dynamic circuits are compared against.","marker":"[44]"},{"why":"The synthesis toolkit whose workflows and partitioning engine the implementation extends.","marker":"[58]"},{"why":"The prior variational method for dynamic state preparation that this work generalizes to arbitrary unitaries.","marker":"[1]"},{"why":"Establishes the teleportation-based long-range entanglement circuits whose resource counts the paper matches or improves on.","marker":"[3]"},{"why":"Provides the hardware roofline and noise model used to compute the required mid-circuit measurement fidelity.","marker":"[29]"},{"why":"Supplies the circuit partitioning algorithm used to scale dynamic-circuit optimization to large circuits.","marker":"[56]"}],"fun_headline_variants":["Automated synthesis crafts dynamic circuits for any quantum state","First practical method auto-generates dynamic quantum circuits","Optimization-driven compilation cuts depth in dynamic circuits","Dynamic circuits get an auto-pilot: numerical search does the design"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The generated circuits only allow a single layer of single-qubit U3 gates in each branch after the measurement, and the paper does not prove that this restricted branch ansatz can prepare every state or unitary; if some targets require entangled or deeper branch unitaries, the \"any target\" claim would fail as implemented.","fun_headline_variants_meta":{"raw":{"variants":["Automated synthesis crafts dynamic circuits for any quantum state","First practical method auto-generates dynamic quantum circuits","Optimization-driven compilation cuts depth in dynamic circuits","Dynamic circuits get an auto-pilot: numerical search does the design"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000813,"raw_usage":{"total_tokens":3661,"prompt_tokens":1139,"completion_tokens":2522,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":755,"completion_tokens_details":{"reasoning_tokens":2458}},"tokens_in":755,"tokens_out":2522,"duration_ms":19942,"temperature":1.0,"reasoning_tokens":2458,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T18:21:07.615483+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the generator on a Haar-random three-qubit unitary with one ancilla and compare the optimized cost under the single-U3-layer branch restriction against a version that allows two-qubit gates in branches; if the restricted version cannot reach the same near-zero cost that the unrestricted version reaches, the expressiveness assumption underlying \"any unitary\" is false.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the topology-aware optimal search algorithm used as the synthesis backbone for the pre-measurement unitary."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the Quantum Shannon Decomposition baseline whose CNOT counts and depths the dynamic circuits are compared against."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"The synthesis toolkit whose workflows and partitioning engine the implementation extends."},{"cited_title":"We use DC-QSearch for circuits smaller than 6 qubits and an DC-Inst for larger circuits","cited_arxiv_id":null,"evidence_quote":"The prior variational method for dynamic state preparation that this work generalizes to arbitrary unitaries."},{"cited_title":"The results of the output distributions are shown in Figure 7, where the correct output states are highlighted with red arrows","cited_arxiv_id":null,"evidence_quote":"Establishes the teleportation-based long-range entanglement circuits whose resource counts the paper matches or improves on."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the circuit partitioning algorithm used to scale dynamic-circuit optimization to large circuits."}],"review_version":1}