{"id":"4d4a4ff7-cba6-4754-8187-c8198c1c790e","arxiv_id":"2506.12788","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"high","formal_verification":"none","parameter_count":6,"one_line_summary":"Adding controlled noise from a simulated time crystal improved fitting accuracy for two quantum neural network variants while degrading quantum reservoir computing, in small numerical tests.","lead":"The authors test a proposed time-crystal-based computing scheme on three quantum machine learning tasks and report that the added noise hurts reservoir computing but improves two neural-network-style fitting methods. They suggest this is a step toward quantum error mitigation that exploits noise, but the evidence rests on small simulations with no code, no error bars, and an uneven comparison protocol.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Reported QTCC advantage is not a controlled result: loss averaging differs between arms, no error bars are given, and the authors concede noiseless loss may overtake with more trials.","rationale":"The load-bearing condition is that the observed QTCC advantage is caused by the time-crystal noise and persists under a fair comparison. Both components are unsecured. First, the paper itself flags that the noiseless loss is still decreasing and might overtake, so the headline effect may be an artifact of comparing before convergence. Second, QTCC's 10-time loss averaging is a different objective-evaluation protocol from the noiseless arm; this is close to the reader's weakest assumption, but if noiseless is deterministic the averaging alone would not change that arm, so the more precise objection is that the arms are not matched and the contribution of averaging or variance reduction is never isolated. Third, no error bars or repeated-seed statistics accompany Table II, so the roughly 9-point QTCC advantage cannot be distinguished from run-to-run fluctuation. I agree with the reader's rejection but only partially with the stated weakest assumption: the fixed test is not repeated-loss averaging per se; it is the absence of a matched protocol with convergence and uncertainty quantification. The reader's verdict REJECT remains appropriate; the central claim is not established.","tokens_in":7973,"tokens_out":7380,"duration_ms":92337,"concrete_test":"Rerun the Table II QNN comparison under matched protocols: noiseless and QTCC, each with the per-point loss computed once and averaged over 10 evaluations, run for at least 50 CMA-ES trials, and repeated over 10 independent seeds with mean plus/minus standard deviation reported. If noiseless matches or beats QTCC at convergence, or if QTCC with a single noise realization no longer beats noiseless, the central claim is unsupported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that time-crystal rotation noise improves QNN/VQKAN accuracy rests on Table II and Fig. 3, which compare QTCC and noiseless propagation at a single, early optimization budget. The authors explicitly state that the noiseless loss lowers gradually and 'hence, the values may be smaller than those of case (c) and (d) after the trials' (Sec. III). A claim of improvement therefore requires showing the ordering persists at convergence, not merely at trial 15. The comparison is also not protocol-matched: the paper says the loss function of each point is calculated 10 times for QTCC, with no equivalent averaging for the noiseless arm. If the simulator is deterministic this particular averaging is not itself a confound, but it still means the two arms use different objective-evaluation procedures, and no repeated-seed statistics or error bars are reported for the QTCC-versus-noiseless gaps in Table II. With stochastic QTCC noise and stochastic CMA-ES, the observed advantage is consistent with a variance-reduction or optimization-budget artifact rather than with 'noise improves accuracy.'","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes Quantum Time Crystal Computing (QTCC), a method in which a time-crystal Floquet Hamiltonian is supplemented with controlled rotation noise, and applies it to three quantum machine learning tasks: echo generation with Quantum Reservoir Computing (QRC), function fitting with a Quantum Neural Network (QNN), and function fitting with a Variational Quantum Kolmogorov-Arnold Network (VQKAN). The numerical results in Section III show that QTCC worsens QRC performance (average loss 35.4215 vs 14.4402 in Table I) but improves QNN and VQKAN prediction accuracy (sum of absolute distances 9.5243 and 10.4943 vs 18.6153 and 22.1747 in Table II). The authors interpret this as evidence that controlled time-crystal noise can improve quantum machine learning and as a potential milestone for quantum error mitigation.","tokens_in":8235,"tokens_out":4383,"duration_ms":49081,"significance":"If established, the claim that a specific, physically motivated noise mechanism improves the accuracy of variational quantum learning would be an interesting and nonobvious result, and the paper's emphasis on time-crystal coherence is a plausible route to explore. The paper does provide quantitative tables and figures rather than only qualitative statements, and it explicitly acknowledges in the concluding remarks that only one problem was solved. However, the evidence is currently too preliminary and methodologically confounded to support the broad claim: it rests on a single toy target, a small number of attempts (10), no error bars or statistical tests, a comparison that differs in how loss evaluations are aggregated, and an explicit admission that the noiseless baselines may overtake the QTCC results after further optimization. The methods themselves also rely on the authors' own unpublished preprints, which limits independent verification.","major_comments":[{"comment":"The comparison between QTCC and noiseless propagation is not protocol-matched: the manuscript states in Section II that for QTCC \"the loss function of each point is calculated 10 times,\" while no such averaging is described for the noiseless arm. This means the two arms differ not only in the presence of time-crystal noise but also in the number of loss evaluations used to form the optimizer's objective. The reported advantage in Table II (9.5243 and 10.4943 vs 18.6153 and 22.1747) could therefore be due to variance reduction from repeated loss evaluation rather than to the time-crystal noise. To support the central claim, the authors must include a noiseless arm with the same 10-fold averaging, or a QTCC arm without averaging, and show that the advantage persists.","section":"Section II, Table II, Fig. 3"},{"comment":"The claimed improvement is evaluated at trial 15, yet the authors themselves write that the noiseless losses \"lower gradually, hence, the values may be smaller than those of case (c) and (d) after the trials.\" This is a direct admission that the observed ordering may not persist at convergence. A claim that noise improves accuracy requires demonstrating that the advantage is stable over the optimization trajectory or at least that the reported budget reflects a reasonable stopping point. As it stands, the result is compatible with QTCC merely accelerating early optimization rather than improving the final achievable accuracy.","section":"Section III, Fig. 3"},{"comment":"The paper reports averages over 10 attempts but provides no error bars, confidence intervals, or statistical tests for the QNN/VQKAN comparisons, despite the QTCC arm being stochastic (Gaussian rotation noise) and the optimizer (CMA-ES) also being stochastic. The observed differences in Table II could be within the range of run-to-run and optimizer variance. The authors should report the distribution of the sum of absolute distances across independent seeds, and ideally a paired statistical test, before drawing any conclusion about accuracy improvement.","section":"Section III, Tables I and II, Figs. 4-13"},{"comment":"The entire positive claim rests on a single target function, f_aim(x) = exp(sin(x0^2+x1^2)+sin(x2^2+x3^2)), with one set of hyperparameters (2 layers, T=10, Nq=4, and the specific Hamiltonian Z0Z1+Z2Z3). This is too narrow to support the abstract's general statement that QTCC \"improved the accuracy of Quantum Neural Network and Variational Quantum Kolmogorov-Arnold Network.\" The concluding remark partially acknowledges this, but the abstract and introduction present the result as a general finding. Additional tasks and hyperparameter settings, or a substantial re-scoping of the claims, are necessary.","section":"Section III, Eq. (3)"},{"comment":"The Hamiltonian in Eq. (1) and the accompanying description of H1 are not sufficiently specified for reproducibility. The all-to-all Ising term and the rotation noise are described in prose, but the precise form of H1, the exact noise realization procedure, the time-evolution operator (e.g., Trotterization order), and the meaning of T in the period condition versus the \"number of time frames\" are ambiguous. Without a complete specification, the numerical results cannot be independently reproduced, which is especially problematic because the central claim is purely numerical.","section":"Section II, Eq. (1); Section II, paragraph on H1"}],"minor_comments":[{"comment":"The phrase \"may be the one of milestones\" is grammatically awkward; it should be \"may be one of the milestones.\"","section":"Abstract"},{"comment":"There is a typo \"conputing\" in the sentence \"chaotic phenomena are exploited for conputing\"; it should be \"computing.\"","section":"Section I"},{"comment":"The term \"Quantum Resavoir Computing\" is a misspelling; it should be \"Quantum Reservoir Computing.\"","section":"Section II"},{"comment":"In the sentence \"The value of Loss functions of cases (c) and (d) are smaller han those of case (a) and (b) at all trials,\" \"han\" should be \"than.\"","section":"Section III"},{"comment":"The symbol T is used both as the period in the modulo condition and as the \"number of time frames\" later in the text; please disambiguate these two uses.","section":"Section II, Eq. (1)"},{"comment":"The text refers to these figures as histograms, but the figure captions describe \"time propagation\" of absolute distances; please clarify what is actually plotted and what the vertical axis represents.","section":"Section III, Figs. 4, 5, 8, 9"},{"comment":"Reference [22] is the original QTCC proposal and is an author preprint; please include a full citation with an arXiv identifier or DOI, and specify which parts of the method are taken from it, so that readers can verify the numerical setup.","section":"References, Ref. [22]"}],"recommendation":"major_revision","confidential_remarks":"The paper's central claim rests on a comparison that is confounded by different loss-evaluation procedures in the two arms, and the authors themselves concede the ordering may reverse after more trials. These are fixable with additional numerical experiments, but they are load-bearing. I also note that the method and the baseline ansatze come largely from the authors' own unpublished preprints, which makes independent assessment difficult; a more self-contained description would help. The manuscript is very short and contains numerous typos and unclear passages; even after the technical concerns are addressed, a thorough editorial revision is needed. I would not recommend acceptance in the current form."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The one thing to know: this paper does not show that time-crystal noise improves quantum machine learning. It shows a specific numerical comparison where a Floquet Hamiltonian with added rotation noise beats clean time propagation on two fitting tasks after 15 optimization trials, and loses on a third task. The comparison is not controlled enough to support the advertised claim.\n\nThe genuinely new part is empirical: nobody has run these three QML methods under the QTCC protocol before. The authors also report the negative QRC result honestly, and they explicitly state the noiseless loss may overtake QTCC with more trials. That admission alone undercuts the central claim. It is a preliminary empirical observation, not a demonstrated effect.\n\nThe soft spots are load-bearing. First, the paper says the loss function of each point is calculated 10 times for QTCC, with no equivalent averaging in the noiseless arm. Even if the simulator is deterministic, this is a protocol mismatch, and it could easily explain the apparent accuracy gain. Second, there are no error bars or significance tests; Table II is a single summary number per method. Ten attempts is a small sample, and the paper shows average curves without spread. Third, only one toy target function is used. Fourth, there is no code or data shipped, so the numbers cannot be independently reproduced. The stress-test note is correct on all these points.\n\nThe self-citation pattern is not a problem by itself. QTCC and VQKAN come from the authors' own preprints, and the paper is transparent about that. The concern is that the comparison baseline is internal: the noiseless protocol is what they chose, and the choice of 15 trials, 10 loss evaluations, and noise standard deviation 1/3 are all free parameters. The reader's circularity score of 2/10 is fair, but I would not call it circular; it is just weak evidence.\n\nWho gets value from this? People working on time-crystal applications in quantum computing or on noise-induced robustness in variational circuits might want a quick read to know the idea is out there. But it is not a result to build on without replication.\n\nMy recommendation: send it to a serious referee. The surprising claim deserves scrutiny, and a referee can demand a controlled comparison: identical objective-evaluation procedures, error bars, more trials, and shipped code. The paper may contain a real kernel, but it is not there yet.","headline":"The paper reports a new but uncontrolled numerical comparison: QTCC beats noiseless propagation on two fitting tasks at trial 15, but the claim that noise helps QML is not supported as presented.","tokens_in":8743,"tokens_out":1751,"would_cite":false,"duration_ms":21483,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P68"],"pacs":["03.67.-a"],"model":"deepseek-v4-flash","headline":"This paper claims that deliberately injecting rotation noise through a Floquet time-crystal Hamiltonian improves the test accuracy of quantum neural networks and variational quantum Kolmogorov-Arnold networks, and treats this as evidence…","keywords":["quantum time crystal computing","time crystals","Floquet Hamiltonian","quantum machine learning","quantum neural network","quantum reservoir computing","variational quantum Kolmogorov-Arnold network","noise-induced accuracy improvement"],"falsifier":"Run the same QNN and VQKAN function-fitting protocol with the QTCC Hamiltonian but set the rotation-noise standard deviation to 0 while keeping the 10-fold loss averaging; if the sum of absolute test distances returns to roughly 18.6 and 22.2, the reported improvement comes from evaluating each loss point ten times rather than from time-crystal noise.","tokens_in":7792,"feed_emoji":"🕰️","tokens_out":8823,"duration_ms":89118,"temperature":0.7,"pith_summary":"The paper argues that controlled noise from a time-crystal phase can improve the accuracy of some quantum machine learning methods. It implements Quantum Time Crystal Computing (QTCC) as a Floquet Hamiltonian whose noise term is a rotation error with standard deviation 1/3, and compares QTCC against noiseless time propagation on three tasks. The result is mixed: QTCC worsens echo generation in quantum reservoir computing, but lowers the sum of absolute test errors in function fitting from 18.6153 to 9.5243 for a quantum neural network and from 22.1747 to 10.4943 for a variational quantum Kolmogorov-Arnold network. The paper presents this as evidence that noise can be a resource for quantum machine learning and as a possible step toward quantum error mitigation. A sympathetic reader would care because the result challenges the default assumption that noise is always the enemy of quantum computation.","feed_headline":"Time-crystal noise halves quantum network fitting error","feed_subtitle":"Controlled rotation noise improves two quantum learning methods but hurts reservoir computing.","key_machinery":"The central object is the Floquet Hamiltonian of a discrete time crystal, $$H = \\begin{cases} \\sum_{j=0}^{N} 0.5(1-d) X_j & (0 \\le t \\bmod 2T \\le T),\\\\ H_1 & (T < t \\bmod 2T \\le 2T), \\end{cases}$$ where $H_1$ contains an all-to-all Ising interaction and rotation noise on the parameters, $\\theta_i = \\theta^0_i + \\theta^r_i \\, \\mathrm{Err}(0,1/3)$, with $d=0.001$. QTCC runs the system through alternating coherent and noisy periods rather than trying to remove the noise. For QRC the learned filter is $W = V^{-1} y$ and predictions are $\\tilde{y} = V W$; for QNN and VQKAN the coefficients of $H_1$ are the trainable parameters, optimized by CMA-ES. A protocol detail that distinguishes the QTCC runs is that each loss point is evaluated 10 times and averaged, whereas the noiseless runs are not averaged in this way.","core_discovery":"On the paper's own terms, the central discovery is that Quantum Time Crystal Computing changes quantum machine learning accuracy in a task-dependent way: it degrades echo generation in quantum reservoir computing (average loss 35.4215 vs 14.4402 for noiseless propagation) but improves function fitting by a quantum neural network (sum of absolute test distances 9.5243 vs 18.6153) and by a variational quantum Kolmogorov-Arnold network (10.4943 vs 22.1747). The paper interprets this as evidence that rotational noise can improve the accuracy of quantum machine learning, and it proposes QTCC as a step toward quantum error mitigation. The proposed mechanism is that the noise turns each loss point into a random variable, so the sum of averaged losses can fall below the noiseless value.","pith_inferences":["The 10-fold loss averaging in QTCC is a confound: if repeated evaluation alone reduces variance, the observed accuracy gain may be a Monte Carlo averaging effect rather than a property of time-crystal order, and a matched equal-evaluation comparison would separate the two.","One testable extension would sweep the noise standard deviation, for instance 0, 1/6, 1/3, 2/3, and 1, and look for a non-monotonic accuracy peak; the paper only tests one noise level.","If the benefit is real, a classical analogue by injecting rotation-like noise into classical neural network optimization would predict similar improvements, connecting the result to known noise-regularization effects in classical machine learning."],"forward_implications":["QTCC can be added to existing variational quantum machine learning pipelines without changing the network architecture, since it only changes how the Hamiltonian noise term is generated.","The QRC results imply that reservoir-style quantum machine learning must be adapted before it can benefit from time-crystal noise; running it unmodified makes predictions worse.","Because the noiseless loss curves fall gradually, the advantage of QTCC may shrink or reverse with more optimization trials, as the paper itself notes.","A successful demonstration on more than one problem is needed before QTCC can be called a general quantum error mitigation technique."],"supporting_citations":[{"why":"Provides the quantum reservoir computing method whose echo-generation accuracy is compared with and without QTCC.","marker":"[12]"},{"why":"Second quantum reservoir computing reference supplying the filter-matrix method used in the echo task.","marker":"[13]"},{"why":"Describes the measurement-feedback mechanism used to produce chaotic behavior in the reservoir.","marker":"[14]"},{"why":"Introduces the time-crystal concept whose persistent oscillation and time-reversal symmetry QTCC exploits.","marker":"[18]"},{"why":"Shows discrete time crystals extend qubit lifetime, a premise for using the TC phase to preserve coherence.","marker":"[19]"},{"why":"Second demonstration of discrete time crystals supporting the coherence-lifetime premise.","marker":"[20]"},{"why":"Shows time crystal systems can act as qubits with long coherence, motivating QTCC.","marker":"[21]"},{"why":"Proposes Quantum Time Crystal Computing, the controlled-noise framework this paper applies to quantum machine learning.","marker":"[22]"},{"why":"Provides the QNN and VQKAN architectures and parameterization used in the function-fitting demonstrations.","marker":"[35]"},{"why":"Supplies the CMA-ES optimizer used to train the QNN and VQKAN parameters.","marker":"[38]"}],"fun_headline_variants":["Time-crystal noise improves quantum function fitting","Crystal noise boosts two quantum ML tasks, not reservoir","Quantum time crystals: noise aids some learning methods","Time-crystal computing: mixed impact on quantum machine learning","Noisy time-crystal qubits help QNN, VQKAN, not reservoir"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The comparison assumes QTCC and noiseless runs differ only by the controlled time-crystal noise, but the QTCC protocol additionally evaluates each loss point ten times and averages the losses, while the noiseless protocol does not, so the repeated-loss averaging alone could explain the smaller QTCC errors.","fun_headline_variants_meta":{"raw":{"variants":["Time-crystal noise improves quantum function fitting","Crystal noise boosts two quantum ML tasks, not reservoir","Quantum time crystals: noise aids some learning methods","Time-crystal computing: mixed impact on quantum machine learning","Noisy time-crystal qubits help QNN, VQKAN, not reservoir"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001159,"raw_usage":{"total_tokens":4740,"prompt_tokens":828,"completion_tokens":3912,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":444,"completion_tokens_details":{"reasoning_tokens":3829}},"tokens_in":444,"tokens_out":3912,"duration_ms":29416,"temperature":1.0,"reasoning_tokens":3829,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T00:42:17.765896+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same QNN and VQKAN function-fitting protocol with the QTCC Hamiltonian but set the rotation-noise standard deviation to 0 while keeping the 10-fold loss averaging; if the sum of absolute test distances returns to roughly 18.6 and 22.2, the reported improvement comes from evaluating each loss point ten times rather than from time-crystal noise.","supporting_citations":[{"cited_title":"Integrable systems and complex geometry","cited_arxiv_id":"0706.1579","evidence_quote":"Provides the quantum reservoir computing method whose echo-generation accuracy is compared with and without QTCC."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Describes the measurement-feedback mechanism used to produce chaotic behavior in the reservoir."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Second demonstration of discrete time crystals supporting the coherence-lifetime premise."},{"cited_title":"Chaotic Roots of the Modular Multiplication Dynamical System in Shor's Algorithm","cited_arxiv_id":"2306.16446","evidence_quote":"Shows time crystal systems can act as qubits with long coherence, motivating QTCC."},{"cited_title":"Genetic-Multi-initial Generalized VQE: Advanced VQE method using Genetic Algorithms then Local Search","cited_arxiv_id":"2109.02009","evidence_quote":"Supplies the CMA-ES optimizer used to train the QNN and VQKAN parameters."}],"review_version":1}