{"id":"eabbba5b-667e-4443-819d-156d00abe69c","arxiv_id":"1908.07637","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"Optimal control can generate high-fidelity, charge-robust microwave pulses for heavy-fluxonium and 0-pi protected superconducting qubits, in simulation, by temporarily routing population through excited states.","lead":"Using computer optimization, this paper designs microwave pulse shapes for performing quantum logic gates on two kinds of protected superconducting qubits that are normally hard to control. A smart generalist would read it because it offers a concrete route toward high-fidelity gates on qubits that promise long coherence, though the results are so far only simulations.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"0-π gate fidelities rely on the dispersive drive model (Eq. 8) without the resonator-occupation validation that is reported for heavy fluxonium; with |v(t)| up to ~1.5 GHz this is a load-bearing unverified assumption.","rationale":"The heavy-fluxonium portion of the paper is reasonably supported: pulses are smooth, the resonator-occupation check is reported, the CZ gate compares favorably with an independent calculation, and closed/open fidelities are consistently above 99%. The central weakness is the 0-π section, which is also where the paper's own language is most hedged: the parameters are called 'optimistic,' the ζ-mode coupling is checked only after optimization on the ideal system, and the conclusion acknowledges that ζ-mode shot noise, offset-charge fluctuations, and dielectric loss are the main barriers. The reader's weakest assumption captured the experimental-realism side of this. I add a more internal concern: the 0-π simulations are performed in an effective dispersive model, but the paper does not demonstrate that this model remains valid for the actual optimized pulses, which are substantially stronger than the fluxonium pulses. This is not an accusation of error; it is an unverified assumption at the very point where the numerical claim is least supported. The proposed full-Jaynes-Cummings simulation would settle it. If it passes, the 0-π claim remains conditional on the optimistic parameter set and on the ζ-mode assumptions, which is exactly the reader's CONDITIONAL verdict. If it fails, the 0-π results would need to be recomputed in a larger model before the claim can be evaluated. Either way, the reader's verdict is not moved; it is reinforced with a sharper condition.","tokens_in":17159,"tokens_out":10596,"duration_ms":143136,"concrete_test":"Re-simulate the optimized 0-π X-gate pulse in the full generalized Jaynes-Cummings model including one resonator mode, using the same g and ω_r as the effective model (or, if not stated, the fluxonium values g/2π = 300 MHz, ω_r/2π = 7.5 GHz as a representative check), and compute the time-dependent resonator photon number ⟨a†a⟩(t) and the closed-system gate fidelity in the full Hilbert space. If the peak photon number is ≳ 0.01 or the full-model fidelity deviates from the reported 98.6% by more than ~0.5%, the dispersive reduction Eq. (8) is not a valid proxy for these pulses. Repeating at a higher resonator frequency would bound the sensitivity.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The 0-π single-qubit results are computed with the dispersively filtered drive Hamiltonian Eq. (8) inside H(t) = H_0-π(ng) + V(t) (Eq. 14), but the validity of that reduction is not checked for 0-π. For heavy fluxonium the authors explicitly verify in a separate simulation that the resonator occupation obeys ⟨a†a⟩ ≪ 1 (Sec. III.A); Section IV contains no analogous check, no resonator frequency/coupling values for 0-π, and no statement of the Hilbert-space truncation or leakage penalty C2. This matters because the optimized 0-π X/H pulses reach |v(t)| ≈ 1.5 GHz, about five times the fluxonium amplitude, and because the high-lying delocalized states used for transfer have transition frequencies that may approach the resonator frequency. If these pulses create non-negligible resonator photons or higher-order corrections beyond the second-order Schrieffer-Wolff expansion, then the optimized unitary U_f in C1 is not the physical gate and the reported closed/open fidelities (98.6% average and 95.8% lower bound for X) do not describe the actual device. The paper's own conclusion identifies ζ-mode coupling, offset charge, and dielectric loss as barriers, but it does not address this more basic model-validity issue.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper uses quantum optimal control with automatic differentiation to design microwave pulses for two protected superconducting qubits, heavy fluxonium and 0-pi, whose computational states have disjoint support. For heavy fluxonium, optimized 60 ns pulses realize X, H, and T gates with closed-system fidelities above 99.9% and open-system fidelities around 99.6%, and a two-qubit controlled-Z gate with 99.4% closed and 99.0% open fidelity. For the 0-pi qubit, the authors introduce a stochastic-gradient scheme in which the offset charge is randomized during optimization, yielding X, H, and T gates with average closed-system fidelities of 98.6%, 99.4%, and 99.95%, and conservative open-system lower bounds of 95.8%, 97.9%, and 99.7% when coupling to the zeta-mode is included. The paper argues that optimal control naturally exploits higher-lying delocalized states to overcome protection, and that including multiple cost targets yields realistic pulse shapes.","tokens_in":17444,"tokens_out":4582,"duration_ms":56468,"significance":"If the reported results hold, the paper provides a practical route toward gate sets for protected superconducting qubits, a problem that is important because disjoint-support protection suppresses exactly the matrix elements needed for direct gates. The heavy-fluxonium results appear internally consistent: the optimized pulses are characterized in both time and frequency domain, the resonator occupation is checked, and the open-system model includes flux noise and dielectric loss with clearly stated parameters. The idea of randomizing the offset charge during optimization to obtain charge-insensitive pulses is a useful and transferable methodological contribution. The 0-pi results, however, are less secure: they rely on an explicitly 'optimistic' parameter set and on a dispersive drive reduction whose validity is not verified for the large drive amplitudes used. The paper does not ship code or pulse data, and omits several optimization hyperparameters, which limits reproducibility. Overall, the central approach is sound and the heavy-fluxonium part is convincing; the 0-pi part needs additional validation before the reported fidelities can be considered predictive for experiments.","major_comments":[{"comment":"The 0-pi optimization relies on the dispersively filtered drive Hamiltonian of Eq. (8) without the resonator-occupation check that is explicitly performed for heavy fluxonium in Section III.A. Section IV gives no resonator frequency, no coupling strength g, no verification that the drive leaves the resonator near vacuum, and no statement of the Hilbert-space truncation or leakage penalty C2. This is load-bearing because the optimized X and H pulses reach |v(t)| ~ 1.5 GHz, about five times the fluxonium amplitude, and because the protocol intentionally populates high-lying delocalized states whose transition frequencies may approach the resonator frequency. The second-order Schrieffer-Wolff reduction in Appendix B can fail if such transitions are near resonant. The authors should provide the 0-pi resonator parameters and a numerical check of <a†a> << 1 for the optimized pulses, ideally by comparing the effective model against the full generalized Jaynes-Cummings Hamiltonian for representative offset-charge values.","section":"Section IV, Eq. (8)"},{"comment":"The 0-pi gate fidelities are computed with the parameter set EL/h = EC/h = 40 MHz, EJ/h = 10 GHz, and ECJ/h = 20 GHz, which the paper itself calls 'optimistic' and which has not been demonstrated experimentally. Consequently, the reported lower bounds of 95.8% (X) and 97.9% (H) do not transfer to currently realizable devices. The manuscript should either present these results explicitly as a design study for future devices or include a sensitivity analysis showing how the fidelities degrade when realistic parameter variations and disorder beyond the 5% level are included.","section":"Section IV, parameter set"},{"comment":"The central results are numerical, but the paper does not provide the optimized pulse data, the cost-functional weights (C1, C2, C3, C4), the optimizer hyperparameters (learning rate, iteration count, batch size for the offset-charge randomization), or the Hilbert-space cutoffs used for the fluxonium and 0-pi simulations. Without these, the reported fidelities cannot be reproduced or checked by independent groups. At minimum, the authors should list all hyperparameters and supply the optimized pulse arrays as supplementary material or a public repository.","section":"Sections II-IV, reproducibility"},{"comment":"The abstract states that 'Closed-system fidelities obtained are 99% or higher,' but the 0-pi X gate has an average closed-system fidelity of 98.6% (Fig. 6 and Section IV). This is a direct quantitative inconsistency in the abstract. The sentence should be qualified, for example by noting that most gates exceed 99% while the charge-randomized X gate reaches only 98.6%.","section":"Abstract and Fig. 6"}],"minor_comments":[{"comment":"The symbol for the low-frequency cutoff appears as 'ωir' in the text and 'ω_ir' in the equation; please unify the notation.","section":"Eq. (9) and text around it"},{"comment":"The caption reports a fidelity of '99.933%' for the Hadamard gate while the text gives '99.93%'; these values should be made consistent.","section":"Fig. 2 caption"},{"comment":"When the authors state that nφ is replaced by nθ for the 0-pi qubit, they should explicitly rewrite Eq. (8) with the 0-pi eigenbasis and charge operator so that the notation is self-contained.","section":"Section IV, Eq. (8) usage"},{"comment":"The statement that states |4⟩ and |5⟩ do not contribute 'due to the lack of connecting matrix elements' would benefit from a brief symmetry explanation, since the reader might otherwise wonder whether this is a numerical artifact.","section":"Section IV, paragraph after Fig. 6"},{"comment":"Table I introduces the leakage penalty C2, but the text describing the fluxonium optimizations mentions only C1, C3, and C4. The authors should clarify whether C2 was used and, if not, why leakage into the truncated Hilbert space is not a concern.","section":"Section III.A, cost function"}],"recommendation":"major_revision","confidential_remarks":"This is a solid numerical study with a useful methodological idea (offset-charge randomization), and the heavy-fluxonium part seems convincing. My main concern is the unvalidated dispersive reduction for the 0-pi qubit, which is load-bearing for the main 0-pi claims. This is fixable in revision by adding resonator parameters, an occupation check, and a comparison against the full model. I do not see problems with novelty or scope. The lack of pulse data and hyperparameters is a reproducibility issue that should be addressed."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's the short version. The genuinely new thing in this paper is the offset-charge randomization during optimization: drawing a fresh ng each iteration steers the pulses toward solutions that work across the full offset-charge range. That is a clever, transferable technique and the main reason to read the paper. The heavy-fluxonium work is also a clean demonstration of optimal control on a protected qubit: >99.9% closed-system fidelities for X, H, T, a CZ gate at 99.4%, and a careful check that the resonator stays essentially unoccupied. The open-system numbers there, around 99.5% or better, are believable. The 0-pi section is more fragile than the abstract suggests. First, the abstract says closed-system fidelities are 99% or higher, but the X gate averages 98.6%. That's a straightforward overstatement. Second, the 0-pi simulations use the dispersively filtered drive of Eq. (8) without the resonator-occupation check the authors perform for heavy fluxonium. Given that the maximal drive amplitude reaches about 1.5 GHz, that omission is not minor. If the drive populates the resonator or the second-order Schrieffer-Wolff reduction breaks down, the optimized unitaries don't represent the physical gate. Third, the 0-pi parameters are explicitly called \"optimistic.\" That's acceptable for a design study, but the noise-model lower bounds of 95.8% and 97.9% are not device-level predictions. Finally, no code, pulse data, or optimizer hyperparameters are given, so reproducing the 0-pi results would require guesswork. I checked the stress-test worry about the missing 0-pi dispersive validation, and it holds up. The paper's own conclusion lists zeta-mode coupling and offset charge as barriers, but it does not address this more basic model-validity question. Citations are fine: the authors cite the relevant experimental and theoretical work on fluxonium, 0-pi, and optimal control, and the main self-citation is to their own automatic-differentiation optimizer, which is appropriate. Who is this for? Researchers working on gates for protected superconducting qubits, and to a lesser extent optimal-control practitioners. It deserves a serious referee. I would send it to review with a recommendation for major revision: fix the abstract, add the missing 0-pi dispersive validation and parameter values, and release the pulse data or code.","headline":"The offset-charge randomization trick is a genuine new idea, and the heavy-fluxonium gates are solid; the 0-pi numbers are conditional and the abstract overreaches.","tokens_in":786,"tokens_out":822,"would_cite":true,"duration_ms":746332,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Optimal control can produce high-fidelity gates for protected superconducting qubits by using higher-lying states and, for the 0-pi qubit, randomizing offset charge during optimization to make pulses charge-insensitive.","keywords":["superconducting qubits","protected qubits","heavy fluxonium","0-pi qubit","optimal control","automatic differentiation","offset charge","gate fidelity"],"falsifier":"Build a 0-pi device with the paper's simulated parameters, run the optimized X, H, and T pulses while sweeping the offset charge $n_g$, and compare the measured average process fidelities to the closed-system values of 98.6%, 99.4%, and 99.95% and to the open-system lower bounds of 95.8%, 97.9%, and 99.7% (for X, H, and T respectively). If the measured fidelities fall clearly below those bounds, or if they vary strongly with $n_g$, the charge-randomization claim fails.","tokens_in":16927,"feed_emoji":"⚛️","tokens_out":11406,"duration_ms":101881,"temperature":0.7,"pith_summary":"Protected superconducting qubits such as heavy fluxonium and the 0-pi circuit store quantum information in computational states that live in separate potential wells, so ordinary microwave pulses cannot directly move population between them. This paper shows that quantum optimal control can nonetheless produce fast, high-fidelity universal gates by routing population through higher-lying delocalized states, temporarily lifting the protection for part of the pulse. The enabling tool is an automatic-differentiation optimizer that can simultaneously minimize gate infidelity, leakage into forbidden states, pulse roughness, and pulse power. For the 0-pi qubit, the key step is to randomize the offset charge $n_g$ at every optimizer iteration, steering the search toward pulses insensitive to charge fluctuations. If these pulses work on real devices, protected qubits—chosen for their long coherence—gain a practical route to the high-fidelity operations quantum error correction requires.","feed_headline":"Protected qubits clear 99% fidelity with optimal-control pulses","feed_subtitle":"Universal 60-ns gates work even though computational states never overlap; charge randomization keeps 0-pi stable.","key_machinery":"The motor of the argument is a multi-objective cost functional minimized with automatic differentiation: the gate infidelity $C_1=1-\\frac{1}{n^2}|\\mathrm{Tr}(U_t^\\dagger U_f)|^2$, a penalty $C_2$ for occupation of forbidden states, a penalty $C_3$ on pulse derivatives, and a penalty $C_4$ on pulse power. Because gradients are computed by automatic differentiation rather than hand-derived formulas, these targets can be added or modified flexibly. The drive enters dispersively through a resonator, producing an effective qubit Hamiltonian $V(t)=2g\\omega_r v(t)\\sum_{l,l'}\\frac{\\langle l|n_\\varphi|l'\\rangle}{(\\epsilon_l-\\epsilon_{l'})^2-\\omega_r^2}|l\\rangle\\langle l'|$ for fluxonium, with $n_\\theta$ replacing $n_\\varphi$ for 0-pi. The decisive addition for 0-pi is to draw a fresh random offset charge $n_g$ uniformly from $[0,1)$ at every optimizer iteration, turning the gradient descent into stochastic gradient descent and converging to a pulse that works on average across all offset charges. Physically, the common enabler in both circuits is the set of higher-lying states that delocalize across the potential wells and carry population between computational states while the protection is momentarily lifted.","core_discovery":"The central claim is that optimal control with automatic differentiation can synthesize realistic, smooth drive pulses that realize high-fidelity gates for protected superconducting qubits even though the computational states have exponentially suppressed transition matrix elements. For heavy fluxonium at $\\Phi_{\\text{ext}}=0.45\\Phi_0$ with $E_C/h=0.5$ GHz, $E_L/h=0.25$ GHz, and $E_J/h=4.0$ GHz, optimized 60-ns pulses realize X, Hadamard, and T gates with closed-system fidelities of 99.94%, 99.93%, and 99.93%, and open-system fidelities of 99.66%, 99.60%, and 99.59%, together with a resonator-mediated controlled-Z gate at 99.4% closed and 99.0% open. For the 0-pi qubit, using the optimistic parameter set from [33], the same approach gives offset-charge-averaged closed-system fidelities of 98.6% for X, 99.4% for H, and 99.95% for T, with conservative worst-case open-system lower bounds of 95.8%, 97.9%, and 99.7% once zeta-mode shot noise, charge noise, and dielectric loss are included. The underlying mechanism is temporary occupation of delocalized higher levels: X and H gates move population out of the computational subspace, through states such as $|2\\rangle,|3\\rangle$ for fluxonium and $|13\\rangle,|14\\rangle$ for 0-pi, and back, while the T gate only makes a brief excursion to $|3\\rangle$ to accumulate its phase.","pith_inferences":["The same randomize-a-parameter-per-iteration trick could be applied to any qubit whose gate performance drifts with a slowly varying parameter such as flux bias, qubit frequency, or coupling strength; the cost would be a trade-off between peak fidelity and robustness.","Because the optimized gates deliberately occupy unprotected high-energy states, gate operation and qubit protection are in tension: the same mechanism that opens the gate window also exposes the system to relaxation and charge noise during the pulse, so even shorter gate times are unlikely to make fidelity improve monotonically.","If the optimistic 0-pi parameters cannot be realized in the lab, the 0-pi fidelities are best read as upper bounds for this method; the heavy-fluxonium results, which use demonstrated parameter values, are the more immediately actionable predictions.","An experiment on existing heavy-fluxonium hardware running the optimized 60-ns X, H, and T pulses described in the paper and measuring process fidelities would directly test whether the predicted roughly 0.3% gap between closed- and open-system fidelity is correct, or whether additional noise channels are missing."],"forward_implications":["Heavy-fluxonium devices can be controlled by a complete gate set—X, H, T, and CZ—at 60 ns per gate with open-system fidelities above 99%, without changing the protected circuit design.","The 0-pi qubit, whose protection is stronger but whose ideal parameters are harder to reach, can be driven by microwave pulses that realize a universal single-qubit gate set with worst-case open fidelities of 95.8% or higher.","Penalizing forbidden-state occupation and pulse roughness keeps the optimized pulses smooth and bounded, so the predicted fidelities are not tied to idealized unbounded drives.","For 0-pi, zeta-mode shot-noise dephasing, offset-charge dephasing, and dielectric loss are the dominant open-system error channels, and the paper identifies active cooling or open-system optimization as the natural next step to push these fidelities higher.","The offset-charge randomization scheme directly removes the main obstacle that earlier 0-pi gate proposals faced—the strong $n_g$ dependence of matrix elements among high-lying states—without requiring charge control in the experiment."],"supporting_citations":[{"why":"Supplies the automatic-differentiation quantum optimizer that the paper extends to multiple cost targets and iteration-dependent Hamiltonians.","marker":"[22]"},{"why":"Demonstrates heavy-fluxonium protection and Raman-assisted gates, providing the experimental context and baseline for the optimized pulses.","marker":"[28]"},{"why":"Defines the fluxonium Hamiltonian and parameter regime used in all fluxonium simulations.","marker":"[30]"},{"why":"Supplies the optimistic 0-pi parameter set and the zeta-mode shot-noise dephasing channel used for open-system bounds.","marker":"[33]"},{"why":"Proposes DC-pulse gates for 0-pi and derives zeta-mode coupling, the scheme this work extends and contrasts with charge-randomized microwave pulses.","marker":"[34]"},{"why":"Reports a comparable 60-ns fluxonium CZ gate via direct qubit coupling, serving as the benchmark for the resonator-mediated CZ result.","marker":"[54]"},{"why":"Introduces gradient-ascent pulse engineering, the prior art whose hand-derived gradients are replaced here by automatic differentiation.","marker":"[1]"}],"fun_headline_variants":["Optimal control beats overlap barrier for protected qubits","Autodiff pulses deliver 99% fidelity to protected qubits","Charge-randomized optimal control stabilizes 0-pi qubit gates","Universal gates for fluxonium and 0-pi qubits at 99%","Higher-level excursions enable protected qubit gate pulses"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The 0-pi results hold only if the optimistic, not-yet-demonstrated circuit parameters ($E_L/h=E_C/h=40$ MHz, $E_J/h=10$ GHz, $E_{CJ}/h=20$ GHz) are physically realizable and if the noise model with 5% disorder in the zeta-mode coupling captures the dominant error sources.","fun_headline_variants_meta":{"raw":{"variants":["Optimal control beats overlap barrier for protected qubits","Autodiff pulses deliver 99% fidelity to protected qubits","Charge-randomized optimal control stabilizes 0-pi qubit gates","Universal gates for fluxonium and 0-pi qubits at 99%","Higher-level excursions enable protected qubit gate pulses"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000666,"raw_usage":{"total_tokens":3092,"prompt_tokens":1054,"completion_tokens":2038,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":670,"completion_tokens_details":{"reasoning_tokens":1950}},"tokens_in":670,"tokens_out":2038,"duration_ms":17633,"temperature":1.0,"reasoning_tokens":1950,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T12:01:32.964435+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Build a 0-pi device with the paper's simulated parameters, run the optimized X, H, and T pulses while sweeping the offset charge $n_g$, and compare the measured average process fidelities to the closed-system values of 98.6%, 99.4%, and 99.95% and to the open-system lower bounds of 95.8%, 97.9%, and 99.7% (for X, H, and T respectively). If the measured fidelities fall clearly below those bounds, or if they vary strongly with $n_g$, the charge-randomization claim fails.","supporting_citations":[{"cited_title":"Earnest, S","cited_arxiv_id":null,"evidence_quote":"Demonstrates heavy-fluxonium protection and Raman-assisted gates, providing the experimental context and baseline for the optimized pulses."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines the fluxonium Hamiltonian and parameter regime used in all fluxonium simulations."},{"cited_title":"Groszkowski, A","cited_arxiv_id":null,"evidence_quote":"Supplies the optimistic 0-pi parameter set and the zeta-mode shot-noise dephasing channel used for open-system bounds."},{"cited_title":"Di Paolo, A","cited_arxiv_id":null,"evidence_quote":"Proposes DC-pulse gates for 0-pi and derives zeta-mode coupling, the scheme this work extends and contrasts with charge-randomized microwave pulses."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Reports a comparable 60-ns fluxonium CZ gate via direct qubit coupling, serving as the benchmark for the resonator-mediated CZ result."},{"cited_title":"Khaneja, T","cited_arxiv_id":null,"evidence_quote":"Introduces gradient-ascent pulse engineering, the prior art whose hand-derived gradients are replaced here by automatic differentiation."}],"review_version":1}