{"id":"77eaa255-ecde-44ff-9196-368e038de5e8","arxiv_id":"2607.18215","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"An active mechanical oscillator lattice experimentally realizes tuneable non-Abelian gauge fields and demonstrates direction-dependent non-Hermitian Wilson loops and switchable non-Hermitian skin modes.","lead":"By linking pairs of mechanical oscillators with real-time feedback, this paper builds a tabletop lattice whose couplings mimic particles with spin moving through a non-Abelian gauge field. It verifies the field by measuring Wilson loops and shows that the same platform can realize non-Hermitian effects such as switchable skin-mode localization.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Wilson-loop extraction assumes lossless unitary bonds; uncalibrated per-bond loss/damping can suppress |W4| below 2, mimicking the non-Abelian signature.","rationale":"The reader's weakest assumption identifies exactly the same load-bearing point: the Wilson-loop extraction treats each bond transfer as an ideal unitary rotation and ignores systematic loss/incomplete transfer, but the sharp criterion |W4|≠2 versus |W4|=2 is vulnerable to a loss-induced suppression. I agree with this assessment. The concern is concrete and is not an artifact of disagreeing with the consensus: the paper's own supplement explicitly describes intentional damping of remnant amplitude and acknowledges incomplete transfer and oscillator-energy-loss differences, yet provides no quantitative error analysis. The extraction formula (S10) contains no loss or nonunitarity terms, so a global or stepwise amplitude loss scales the inferred Wilson-loop magnitude downward. A control measurement with a known Abelian configuration would directly calibrate this systematic effect. I do not think this concern overturns the central claim, because the reported numerical-experimental agreement and the ability to run such a control make the issue addressable; the appropriate disposition remains CONDITIONAL, which is the reader's verdict. Therefore no adjustment is recommended. I also note the SM has blank section references and an unedited thesis fragment, which support the need for additional detail but do not independently change the verdict.","tokens_in":18898,"tokens_out":8395,"duration_ms":77794,"concrete_test":"Re-run the same Wilson-loop protocol with the link variables replaced by a known commuting/Abelian configuration (e.g., set X=I, Y=e^{iβσ_x}, so exact |W4|=2). If the extracted |W4| is significantly below 2, the per-bond loss/damping systematically suppresses |W| and the Fig. 1(d) data cannot distinguish Abelian from non-Abelian. Additionally, perform full process tomography on each individual bond to reconstruct the actual 2-site transfer map and compare its fidelity to the ideal unitary X/Y; if the reconstruction error or measured nonunitarity exceeds the observed deficit |2−|W4|| in the claimed non-Abelian regions, the claim is unsupported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central non-Abelian claim rests on measuring |W4|≠2 for U4=U2†U1†U2U1 in Fig. 1(d). The extraction formula (S10) is derived under the assumption that each feedback-imposed bond is an ideal unitary rotation realized by one calibrated tunneling time. But the SM's own protocol intentionally damps remnant amplitude at the original site after each step, and the text lists 'incomplete transfer of energy during a single rotation' and 'accumulated differences in energy loss' as error sources, without quantifying them or giving error bars. Any per-bond amplitude loss η enters the tomographic combinations A,B,C,D quadratically (e.g., A=a0^2), so the inferred |W| is suppressed. Thus an Abelian configuration with true |W4|=2 would be reported as |W4|<2—exactly the criterion claimed to demonstrate non-Abelianness. The missing sections/unedited fragments in the SM mean the derivation of (S10) and the basis-loop reduction cannot be checked from the manuscript. Unless the systematic loss is below the gap between measured |W4| and 2, the headline experiment does not establish a genuinely non-Abelian gauge field.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript reports an experimental realization of tunable SU(2) gauge fields in a measurement-and-feedback-coupled classical mechanical oscillator array. Each lattice site is encoded by two oscillators, and feedback imposes spin-dependent hopping matrices X=e^{iασ_y}, Y=e^{iβσ_x}. Wilson-loop magnitudes are extracted by a sequence of four initial-state/tomography-like measurements and used to infer a genuinely non-Abelian field from |W4|=|Tr(U2†U1†U2U1)|≠2. The same platform is then used to realize a non-Hermitian 2D model with direction-dependent Wilson loops and a 1D non-Abelian Hatano-Nelson model in which the gauge angles (θ_R,θ_L) switch the non-Hermitian skin effect between the two ends of an open 7-site chain.","tokens_in":19220,"tokens_out":10785,"duration_ms":105716,"significance":"If the measurement systematics are under control, this would be a notable experimental milestone: a programmable active-mechanical platform that realizes non-Abelian holonomies and, simultaneously, non-Hermitian effects, with the Wilson-loop magnitude as a gauge-invariant observable. The finite-size basis-loop reduction in the Supplemental Material is an elegant way to reduce the infinite set of loop operators to U1 and U2 and to justify the |W4|≠2 criterion. The independent calibration of J via two-site tunneling, the use of four tomographic initial states, and the direct time-resolved observation of skin-mode switching are all strengths. However, the central experimental evidence is Wilson-loop magnitudes, and these are currently presented without uncertainty bars, repetition statistics, or quantitative control measurements; the extraction formula assumes ideal unitary bond rotations. The non-Abelian claim therefore needs additional quantitative support before it can be considered established.","major_comments":[{"comment":"The load-bearing criterion for a genuinely non-Abelian field is the sharp threshold |W4|≠2 versus |W4|=2. The extraction formula (S10) is derived for ideal unitary bond rotations; any per-bond transfer loss, incomplete rotation, or spin-dependent damping enters the tomographic combinations A,B,C,D and suppresses the inferred |W|. The SM explicitly lists 'incomplete transfer of energy during a single rotation' and 'accumulated differences in energy loss' as error sources, but no error bars, repetition statistics, or per-bond efficiency calibration are reported. A configuration with true |W4|=2 (e.g. α=0 or β=0, where one link variable is the identity) could then be measured as |W4|<2, exactly the signature claimed for non-Abelianness. Please supply uncertainty estimates, state whether final states are renormalized before applying Eq. (S10), report control measurements on lines where |W4|","section":"SM 'Wilson Loop Measurement Protocol'; Eq. (S10); Fig. 1(d)"},{"comment":"The Supplemental Material is not in a complete, checkable state. The Overview has empty section references ('In Section , ...'), the derivation of the central extraction formula Eq. (S10) is not given, and the numerical-simulation model is not fully specified (ideal model versus model including dissipation/feedback). An unedited thesis fragment remains in the SM (the caption 'Figure 5.8: Selected dynamics... used to produce Fig. 5.7' and the text '5.3 Conclusions ... Here I have shown ...'). The basis-loop reduction is presented, but because the Wilson-loop extraction is the basis for the main non-Abelian claim, the absence of a complete derivation and the presence of unedited fragments compromise reproducibility. This must be fixed in revision.","section":"SM 'OVER VIEW', 'Wilson Loop Measurement Protocol', and 'Derivation of basis loops'"},{"comment":"The 1D skin-effect demonstration relies on long-time bias measurements while a nonlinear energy-stabilizing feedback of the form Hstab=-igD Σ c†c is applied, with D depending on the total squared amplitude. The text states that this feedback does not affect the normalized energy distribution and that numerical simulations confirm this, but no supporting data or derivation are shown. The experimental bias map in Fig. 3(f) is also presented without error bars and is systematically offset from the numerics, an effect attributed to damping. Please quantify the sensitivity of the measured bias to the stabilizing gain g, provide repetition statistics, and either justify more rigorously or demonstrate numerically that the normalized spatial distribution is independent of this feedback.","section":"SM Eq. (8); Fig. 3(e,f)"}],"minor_comments":[{"comment":"Typos: 'geninuely' appears repeatedly and should be 'genuinely'; 'tuneable' is used inconsistently.","section":"General"},{"comment":"Ref. [43] and Ref. [49] are the same work (Chuang and Nielsen, J. Mod. Opt. 44, 2455 (1997)); one duplicate should be removed.","section":"References"},{"comment":"The caption of the supplemental dynamics figure still uses thesis-chapter numbering ('Figure 5.8') and refers to 'Fig. 5.7'; renumber and remove the fragment beginning '5.3 Conclusions'.","section":"SM Figure 6"}],"recommendation":"major_revision","confidential_remarks":"The central theoretical reasoning is sound, and the platform is original and timely. My main concern is that the experimental signature of genuine non-Abelianness—|W4|≠2—is directly vulnerable to the acknowledged but unquantified loss and incomplete-transfer systematics in the tomographic extraction. This is fixable with control measurements and error bars, but it is load-bearing and must be addressed before publication. The supplementary material also contains unedited thesis fragments and missing derivations, which suggests the manuscript needs a full editorial pass. I do not see a fundamental correctness error in the finite-size loop argument itself."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is the first time someone has put a tunable SU(2) gauge field on a feedback-coupled mechanical oscillator lattice and measured Wilson loops directly. The platform is real, and the experimental maps in Figs. 1–3 reproduce the numerics well enough to take seriously. The theoretical criterion — that |W4| ≠ 2 for the commutator loop implies noncommuting loop operators in a finite SU(2) lattice — is correct, and the supplement's spanning-tree basis argument actually proves the loop basis is complete for the six-site ladder. That is a real contribution.\n\nThe 2D non-Hermitian Wilson-loop direction dependence and the 1D gauge-angle switching of the NHSE are also convincing at the level of qualitative agreement with simulations. The non-Hermitian part is less novel (the model and prediction are from Pang et al. [33]), but the experimental control is the point.\n\nNow the soft spots, in order of size. The biggest is the Wilson-loop extraction. The supplement's Eq. S10 assumes each feedback bond is an ideal unitary rotation executed in exactly one calibrated tunneling time, and the protocol intentionally damps remnant amplitude at the starting site after each step. The paper lists 'incomplete transfer' and 'accumulated differences in energy loss' as error sources but gives no per-bond loss, no repetition statistics, and no error bars on any |W|. The measured quantity enters quadratically in amplitude, so even a few percent loss per bond suppresses |W4| and can push an Abelian configuration below 2. That is exactly the kind of systematic that could fake the non-Abelian signature. The maps in Fig. 1 are visually reassuring — the deviations from 2 are structured, not uniformly suppressed — but the sharp threshold claim needs numbers. This is fixable, but it is not just a minor presentational issue.\n\nSecond, the supplement is unfinished in places: blank section names, and a leftover thesis-style fragment with 'Figure 5.8' and '5.3 Conclusions' embedded in the file. That should never go to referee as-is.\n\nThird, the 1D energy-stabilizing feedback is a nonlinear term whose only justification is numerical confirmation. I believe the claim that it doesn't affect normalized distributions, but it deserves a more careful statement in an experiment that is otherwise linear.\n\nThe citation practice is fine; the prior self-citations are to the platform papers, and the theory claims are traced to Refs. [9,10,22,33].\n\nBottom line: this is a serious experimental paper with a real platform and a correct theoretical framing. The central claim is plausible but not yet proven on the written record. I would send it to referees, and the first referee report should ask for quantified loss/error analysis and a cleaned supplement.","headline":"A genuinely new experimental platform and a mostly sound non-Abelian Wilson-loop measurement, but the headline claim lacks the error analysis needed to rule out loss-induced false positives.","tokens_in":19659,"tokens_out":2648,"would_cite":true,"duration_ms":26053,"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":"A feedback-controlled lattice of paired mechanical oscillators implements a tunable SU(2) gauge field, and Wilson-loop measurements show the loop operators fail to commute.","keywords":["non-Abelian gauge field","Wilson loop","synthetic gauge field","active mechanical lattice","feedback control","non-Hermitian skin effect","SU(2) holonomy","two-oscillator pseudo-spin"],"falsifier":"A controlled-loss test: choose an intentionally Abelian configuration where |W4| should equal 2, then deliberately increase oscillator damping and see whether the measured |W4| falls below 2. If it does, the magnitude-only threshold is not a clean non-Abelian signature. The decisive check is to reconstruct the full 2×2 matrices U1 and U2 (not just traces) from the four-state tomography and verify directly that [U2,U1]≠0.","tokens_in":18836,"feed_emoji":"⚙️","tokens_out":7357,"duration_ms":67499,"temperature":0.7,"pith_summary":"This paper tries to show that a purely classical, feedback-controlled mechanical lattice can carry a genuinely non-Abelian gauge field—a field whose holonomies depend on the order in which paths are traversed—and that the same platform can be used to explore non-Hermitian versions of such fields. Each lattice site is a pair of oscillators encoding a pseudo-spin, and real-time measurement and feedback impose spin-dependent hopping matrices. The authors extract Wilson-loop magnitudes and argue that a measured value |W4|≠2 for a specially chosen commutator loop proves the loop operators do not commute, hence the gauge field is genuinely non-Abelian. They further show that non-reciprocal couplings make the Wilson loop direction-dependent and that tuning gauge angles can switch skin-mode localization between the two ends of a one-dimensional chain. A sympathetic reader would care because this offers a programmable tabletop setting for gauge-field physics usually studied in quantum or photonic systems.","feed_headline":"Mechanical oscillators host a genuinely non-Abelian gauge field","feed_subtitle":"Pairs of springs-and-magnets oscillators act as pseudo-spins; Wilson-loop traces show loop order changes the outcome.","key_machinery":"The load-bearing device is the feedback-imposed link: each bond is engineered as a prescribed 2×2 SU(2) rotation on the two-oscillator pseudo-spin, switched on for one calibrated tunneling time. The diagnostic is the Wilson-loop magnitude criterion—in SU(2), |tr(U)|=2 forces U=±1, so the commutator loop U4=U2†U1†U2U1 has |W4|≠2 exactly when the elementary loop operators fail to commute. A spanning-tree argument shows that on the six-site lattice all loops are generated by U1, U2, and their inverses, so checking these traces decides whether the whole finite lattice is Abelian or genuinely non-Abelian. For the non-Hermitian models, non-reciprocal link matrices break the Hermitian relation betw","core_discovery":"On its own terms, the paper establishes that an active mechanical lattice—oscillators coupled only through real-time measurement and feedback—can implement a spin-1/2 hopping model with tunable non-Abelian link variables X=e^{iασ_y}, Y=e^{iβσ_x}. Using four initial pseudo-spin states and single-bond transfer sequences, the authors extract the magnitudes of Wilson loops for the elementary plaquette, a double plaquette, their product, and the commutator loop U4=U2†U1†U2U1. Because an SU(2) loop operator with |tr(U)|=2 must be ±identity, a measured |W4|≠2 directly implies U2U1≠U1U2, i.e., genuine non-Abelianness. They then break reciprocity and use the same apparatus to demonstrate direction-de","pith_inferences":["The same feedback protocol should transfer to other classical wave platforms, such as acoustic, photonic, or electronic arrays, because it requires only real-time measurement and actuation rather than quantum coherence.","The |W4|≠2 test is magnitude-only; reconstructing the full 2×2 loop matrices from the same four-state tomography would provide a stronger, direct check of U2U1≠U1U2 and would help separate loss-induced suppression from true noncommutation.","Direction-dependent Wilson loops in non-Hermitian settings may serve as a gauge-invariant order parameter for non-Hermitian topology, potentially connecting to point-gap winding in higher dimensions.","In larger lattices, iterating these feedback-imposed SU(2) links could simulate lattice gauge theories or topological band structures with position-dependent gauge fields, including dynamical back-action."],"forward_implications":["Active mechanical lattices become a reconfigurable tabletop platform for non-Abelian gauge fields, with arbitrary rotation angles and non-reciprocal hopping strengths.","The Wilson-loop magnitude criterion provides a practical gauge-invariant test for genuine non-Abelianness in finite synthetic lattices.","Non-Hermitian non-Abelian gauge potentials can make Wilson loops direction-sensitive, a signature unavailable in Hermitian systems.","Tuning the gauge angles gives direct control over the direction and localization of energy transport via the non-Hermitian skin effect in one dimension.","The platform's programmability extends to nonlinear or amplitude-dependent couplings, opening a route toward dynamical gauge fields where the wave itself alters the holonomy."],"fun_headline_variants":["Oscillator pairs reveal non-Abelian gauge fields in a mechanical lattice","Feedback-driven oscillators realize non-Abelian gauge fields","Wilson loops show noncommuting order in a mechanical lattice","Mechanical lattice with pseudo-spins maps non-Abelian gauge fields","Non-Hermitian effects on non-Abelian gauge fields in a mechanical array"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The Wilson-loop extraction treats every single-bond feedback transfer as an ideal unitary rotation completed in one calibrated tunneling time, with losses and transfer errors small enough that a measured |W4|≠2 is read as noncommutation rather than as systematic suppression.","fun_headline_variants_meta":{"raw":{"variants":["Oscillator pairs reveal non-Abelian gauge fields in a mechanical lattice","Feedback-driven oscillators realize non-Abelian gauge fields","Wilson loops show noncommuting order in a mechanical lattice","Mechanical lattice with pseudo-spins maps non-Abelian gauge fields","Non-Hermitian effects on non-Abelian gauge fields in a mechanical array"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.0013,"raw_usage":{"total_tokens":5159,"prompt_tokens":780,"completion_tokens":4379,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":524,"completion_tokens_details":{"reasoning_tokens":4286}},"tokens_in":524,"tokens_out":4379,"duration_ms":31422,"temperature":1.0,"reasoning_tokens":4286,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T15:39:29.072313+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A controlled-loss test: choose an intentionally Abelian configuration where |W4| should equal 2, then deliberately increase oscillator damping and see whether the measured |W4| falls below 2. If it does, the magnitude-only threshold is not a clean non-Abelian signature. The decisive check is to reconstruct the full 2×2 matrices U1 and U2 (not just traces) from the four-state tomography and verify directly that [U2,U1]≠0.","supporting_citations":[],"review_version":1}