{"id":"1d31c396-3d88-4dc0-af68-ab71f4a8c08e","arxiv_id":"2511.06354","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A binomial-code controlled-Z gate between two superconducting cavities is demonstrated with 97.4 ± 0.8% post-selected, reference-normalized process fidelity using GRAPE-optimized geometric phase control.","lead":"Researchers built a controlled-Z gate between two quantum-error-correcting 'binomial' microwave cavities and report 97.4% process fidelity, the highest claimed for such two-logical-qubit gates. The number is obtained after discarding runs where a coupler qubit leaked, and is normalized against reference operations, so it is a conditional rather than raw fidelity.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Headline '97.4% process fidelity' is a post-selected, reference-normalized estimate; the paper never directly characterizes the CZ channel or validates the additivity/post-selection assumptions, so the claimed comparison is not yet established.","rationale":"The reader's weakest-assumption correctly flags the post-selection problem: the 97.4% figure is conditional on a 92.2% survival probability, and the text itself admits QC excitations arise partly from the CZ gate. I agree this is a serious concern. However, I find an equally load-bearing issue that the reader mentions but does not center: even the un-post-selected 94.9% figure is an inferred quantity, obtained by normalizing a full-sequence QPT result by an encode/decode reference. This normalization assumes that infidelities of concatenated processes add, which is not generally valid. The paper provides no direct characterization of the CZ process alone — no standalone logical QPT, no randomized benchmarking, and no data release that would allow re-analysis. The same concern applies to the post-selected number: the rejected runs are excluded under the assumption that they are external, correctable failures, but the observed 3.5% leakage during encode/decode alone versus 7.8% for the full sequence suggests an additional 4.3% leakage appears when the CZ is inserted. Without quantifying the logical state in the rejected runs, one cannot know whether post-selection is a valid heralding procedure or a way of hiding gate-induced errors. The proposed geometric phase engineering method is plausible and the experimental effort is substantial; the error budget and Wigner tomography provide supporting evidence. But the headline 'process fidelity' is not yet established as a true process fidelity, so the verdict should remain conditional pending a direct verification.","tokens_in":18170,"tokens_out":7397,"duration_ms":78530,"concrete_test":"Perform interleaved randomized benchmarking (IRB) of the logical CZ gate using the same encode/decode pulses and QC post-selection, and separately without post-selection. Compare the IRB gate fidelity to the QPT-derived 97.4±0.8%. If either IRB estimate deviates beyond the quoted error bar, the reported 'process fidelity' is not a reliable standalone estimate and the factorization/post-selection assumptions are falsified.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is the abstract's 'process fidelity of 97.4±0.8%' for the CZ gate. This number is produced by two unvalidated steps. First, the CZ fidelity is not measured directly: the experiment QPTs the concatenated encode-CZ-decode channel, and the CZ fidelity is extracted using the encode/decode reference (Fig. 2(c)). This implicitly assumes that infidelities add under composition. Average process infidelity is not additive for general CPTP maps, and coherent errors or leakage can violate the assumption. The paper provides no independent check of this factorization. Second, the quoted 97.4% is conditional on rejecting 7.8% of runs in which QC leaves |g>. The text states these excitations 'originate not only from the CZ gate itself but also intrinsically from the two-cavity system,' and the supplement attributes them to 'transmon ionization-like processes.' No evidence distinguishes CZ-induced leakage from external ionization, nor shows the discarded runs have intact logical states. If rejected runs are counted as failures, an unconditional bound is roughly 0.974×0.922≈0.90. Thus the headline overstates what was established: a conditional, reference-normalized estimate rather than a standalone process fidelity. This does not invalidate the geometric engineering method, but it undermines the comparison 'surpassing all previously reported two-logical-qubit gates.'","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a 'geometric phase engineering' approach for implementing controlled-Z (CZ) gates between two binomial logical qubits in a circuit-QED architecture. The method uses a single coupler transmon drive, with the waveform optimized via GRAPE under constraints that the coupler returns to the ground state and that a π geometric phase is accumulated only on the |22> Fock component. This reduces the optimization Hilbert space to three Fock levels per cavity plus the coupler. The authors experimentally demonstrate a process fidelity of 97.4±0.8% for the CZ gate, extracted from QPT of an encode-CZ-decode sequence normalized by an encode/decode reference, with post-selection on the coupler ground state (92.2% retention). They also show joint Wigner tomography of a logical Bell state and provide a simulated error budget.","tokens_in":18578,"tokens_out":6909,"duration_ms":67331,"significance":"If the 97.4% figure is robust, it would be a substantial advance over previous two-logical-qubit gates in bosonic codes, and the idea of using geometric phase structure to reduce the GRAPE search space is a useful contribution. The paper is transparent about the post-selection and the leakage issue, and it includes an error budget. However, the headline fidelity is a conditional, reference-normalized estimate: the extraction assumes that fidelities factor through composition, and the post-selection discards 7.8% of runs. These issues directly affect the comparison with prior work. The method itself is plausible and worth publishing after the extraction is validated or the claims are appropriately qualified.","major_comments":[{"comment":"The CZ fidelity is obtained by 'normalizing the resulting fidelity with respect to the encode/decode reference' (main text). For general CPTP maps, average process fidelity is not multiplicative under composition; coherent errors can add or interfere. The paper provides no proof or numerical check that F_seq / F_ref equals the CZ process fidelity. This is load-bearing because the headline 97.4% rests on this step. Please validate the factorization with a simulation of the full sequence using the independently characterized CZ and encode/decode channels, or use a reference-free method (e.g., interleaved randomized benchmarking, or direct QPT with separate state prep/measurement) to corroborate the number.","section":"§Results, Fig. 2(c)"},{"comment":"Post-selection on QC ground state retains 92.2% of runs; the quoted 97.4% is conditional. The paper states the discarded excitations 'originate not only from the CZ gate itself but also intrinsically from the two-cavity system,' but offers no quantitative decomposition. The evidence of leakage during encode/decode with no QC drive establishes a background floor, not that the additional leakage is external. If discarded runs are counted as failures, the unconditional worst-case fidelity is at most ≈0.974×0.922 ≈ 0.90. Please report the unconditional process fidelity (or a lower bound) and explicitly discuss how the 'surpassing all previously reported' claim depends on the same post-selection metric as prior work.","section":"§Results and Supplementary Sec. VII"}],"minor_comments":[{"comment":"The phrase 'process fidelity' should be qualified as 'conditional on the coupler remaining in its ground state' to avoid ambiguity about the post-selection.","section":"Abstract"},{"comment":"Provide a table comparing the gate fidelities and metrics of Refs. [15-19] (conditional/unconditional, process vs. entanglement fidelity) to support the 'surpassing all previously reported' claim.","section":"Introduction/Results"},{"comment":"Define 'origin data' and 'exclude leakage' in the caption; clarify whether the reference and full sequence use the same post-selection threshold and how the fidelities are computed.","section":"Fig. 2 caption"},{"comment":"The error budget does not include a bar for the 7.8% post-selection loss. Please explain how leakage is handled in the simulation or add a bar for the heralding failure.","section":"Fig. 4"},{"comment":"The Bell state section is qualitative; consider reporting a quantitative fidelity or concurrence for the prepared logical Bell state to support the 'excellent agreement' statement.","section":"Fig. 3"}],"recommendation":"major_revision","confidential_remarks":"The central issue is the unsupported factorization of fidelities in the CZ fidelity extraction. The paper is otherwise a solid experimental demonstration of a useful control technique, but the headline claim needs either a rigorous validation of the factorization or a softened, properly qualified statement."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Worth reading, but don't take the abstract at face value. The 97.4% CZ fidelity is a conditional, reference-normalized number, and the paper does not supply enough evidence to back the headline comparison against prior bosonic two-logical-qubit gates.\n\nWhat is actually new: the 'geometric phase engineering' idea—start from the fact that a binomial-code CZ is natively a geometric phase gate, then use GRAPE to compress the pulse and absorb nonlinearities while keeping the optimization in a small logical subspace (three Fock levels per cavity). That is a sensible and potentially useful synthesis of existing tools. The experiment is substantial: two cavities, a coupler transmon, QPT, Bell-state Wigner tomography, and an error budget. The authors are also transparent about the post-selection and the reference normalization, which is more than many experimental papers bother to do.\n\nThe soft spots are real. First, the CZ fidelity is not measured directly. The experiment QPTs the concatenated encode-CZ-decode channel and divides by the encode/decode reference. That implicitly assumes the infidelities compose in a simple way, but average process infidelity is not generally additive, and coherent errors or leakage can invalidate the step. No independent check is shown. Second, the quoted number keeps only the 92.2% of runs where the coupler ends in |g>. The paper's own text says the discarded excitations originate 'not only from the CZ gate itself but also intrinsically from the two-cavity system,' and the supplement attributes them to transmon-ionization-like processes—but never distinguishes CZ-induced errors from external leakage. If a substantial fraction are gate-induced, the unconditional fidelity is bounded well below 97.4%, roughly 0.90 in the worst case. Third, the comparison to earlier work (CNOT, CZ, ECD-based gates) is not on a common metric, since previous numbers are often unconditional and directly characterized. The Bell-state Wigner is nice, but it does not by itself rescue the fidelity claim.\n\nThe method itself seems credible, and the paper is honest about its limitations—the discuss open-GRAPE, error-transparent extensions, and coupler design. But the central quantitative assertion needs rework. For me, this is a paper for people working on bosonic QEC and optimal control who want to see a plausible new control recipe and are willing to dig past the headline. It deserves a serious referee, not a desk reject, but the referee should demand unconditional and directly characterized fidelities, validation of the normalization assumption, and ideally release of the pulse waveforms and data.\n\nI would bring it to a reading group to talk about what counts as a demonstrated gate fidelity, but I would not cite the 97.4% number as established.","headline":"A promising control method whose headline fidelity is a post-selected, reference-normalized estimate — not yet a demonstrated 97.4% process fidelity.","tokens_in":19059,"tokens_out":2169,"would_cite":false,"duration_ms":23430,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P68","81V80"],"pacs":["03.67.Lx","42.50.Pq","85.25.-j"],"model":"deepseek-v4-flash","headline":"Geometric phase engineering realizes a 97.4% controlled-Z gate between two binomial-code logical qubits, surpassing previous bosonic two-qubit gates.","keywords":["binomial code","geometric phase","controlled-Z gate","bosonic code","transmon","GRAPE","cavity QED","quantum error correction"],"falsifier":"Measure the logical CZ process fidelity without post-selecting on the coupler qubit (i.e., including all runs in quantum process tomography), and compare against the conditional 97.4% figure. If the unconditional fidelity is substantially lower, and a control experiment with the CZ drive turned off shows that the drive itself produces coupler excitation, the paper's central performance claim fails.","tokens_in":18075,"feed_emoji":"⚛️","tokens_out":4157,"duration_ms":37324,"temperature":0.7,"pith_summary":"The paper proposes geometric phase engineering, a hybrid gate scheme that implements a controlled-Z gate between two binomial-code logical qubits using a single drive on a coupler transmon. Because the binomial code's CZ operation is natively a geometric phase gate on the |22> component, the optimization can be restricted to a tiny effective Hilbert space—three Fock levels per cavity plus the coupler's two levels. The authors experimentally measure a process fidelity of 97.4±0.8%, after post-selecting on runs where the coupler ends in its ground state, surpassing previously reported two-logical-qubit gate fidelities in bosonic codes. If correct, the method offers a practical route to fast, high-fidelity multi-qubit logical operations while retaining insensitivity to drive timing delays.","feed_headline":"Geometric pulse engineering hits 97.4% for two-logical-qubit gate","feed_subtitle":"A single optimized drive yields a binomial-code controlled-Z gate beyond earlier bosonic two-qubit gates.","key_machinery":"The central object is the geometric phase gate: a drive resonant with the coupler only when both cavities are in |2> (the |22> component) induces a full rotation on the Bloch sphere, accumulating a phase of π on that component. The GRAPE optimization constrains the solid angle to π for the target state and exactly zero for all other Fock states, while leaving photon-number distributions unchanged. This preserves a fixed phase reference across Fock states and confines the optimization to a 3×3×2-dimensional Hilbert space, dramatically reducing computational cost compared to full Hamiltonian engineering.","core_discovery":"For lowest-order binomial codes with codewords |0L>=(|0>+|4>)/√2 and |1L>=|2>, the controlled-Z gate reduces to adding a π phase only to the |1L1L>=|22> component. The paper shows this can be realized by a single frequency-selective pulse on a coupler transmon that executes a full Bloch-sphere rotation—a geometric phase—for the |22> state while returning all other Fock components to zero phase. A gradient-ascent pulse engineering (GRAPE) optimization enforces this condition while incorporating self- and cross-Kerr nonlinearities and compressing the gate to 1 μs. In a circuit-QED experiment with two storage cavities and three transmons, the authors report a conditional process fidelity of 97.","pith_inferences":["The headline 97.4% is conditional: only the 92.2% of runs in which the coupler stays in its ground state are kept. If the CZ drive itself is responsible for a meaningful share of the discarded excitations, the unconditional gate fidelity is lower—at worst roughly 0.974×0.922≈0.90—so a circuit-level benchmark that cannot afford such post-selection may see a smaller advantage.","The observed coupler leakage even during encode/decode (when no drive is applied to it) points to a multi-mode transmon-ionization phenomenon; this may be a general obstacle for multi-cavity processors and deserves dedicated study beyond the scope of this paper.","One testable extension: run the same geometric-phase CZ gate without coupler post-selection and compare unconditional vs conditional process fidelities, or add a leakage-detection-and-correct pulse to convert discarded runs into detectable errors.","The scheme's reliance on a native geometric decomposition may not hold for arbitrary logical gates (e.g., a general controlled-phase with variable angle), so its generality beyond CZ-like gates remains to be demonstrated."],"forward_implications":["Two-logical-qubit gate fidelities in bosonic codes can exceed 90%, a regime previously unreached; simulation indicates >99% would follow if the shorter-coherence cavity matched the other.","The method is readily transferable to other bosonic codes, code orders, and more than two logical qubits, since it only requires a native geometric decomposition of the target gate.","Because self-Kerr and cross-Kerr commute with geometric phase operations, timing delays among drive lines become less consequential, simplifying circuit-level control.","The reduced optimization dimensionality (three Fock levels per cavity) makes optimal-control design scalable to multi-mode, high-dimensional systems."],"fun_headline_variants":["Record 97.4% fidelity for bosonic two-qubit gate","Geometric phase hits 97.4% on binomial-code CZ gate","97.4% two-qubit gate for bosonic codes via geometric drive","Geometric pulse sets bosonic two-qubit gate record at 97.4%","Single geometric pulse yields 97.4% CZ for binomial qubits"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The 97.4% gate fidelity is obtained by discarding the 7.8% of experimental runs in which the coupler qubit leaves its ground state; if a significant fraction of that leakage is caused by the CZ pulse itself rather than by unrelated background effects, the post-selected number overstates the gate's real-world performance.","fun_headline_variants_meta":{"raw":{"variants":["Record 97.4% fidelity for bosonic two-qubit gate","Geometric phase hits 97.4% on binomial-code CZ gate","97.4% two-qubit gate for bosonic codes via geometric drive","Geometric pulse sets bosonic two-qubit gate record at 97.4%","Single geometric pulse yields 97.4% CZ for binomial qubits"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000678,"raw_usage":{"total_tokens":2893,"prompt_tokens":693,"completion_tokens":2200,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":437,"completion_tokens_details":{"reasoning_tokens":2098}},"tokens_in":437,"tokens_out":2200,"duration_ms":14491,"temperature":1.0,"reasoning_tokens":2098,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T23:17:23.858752+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the logical CZ process fidelity without post-selecting on the coupler qubit (i.e., including all runs in quantum process tomography), and compare against the conditional 97.4% figure. If the unconditional fidelity is substantially lower, and a control experiment with the CZ drive turned off shows that the drive itself produces coupler excitation, the paper's central performance claim fails.","supporting_citations":[],"review_version":1}