{"id":"6d3c720c-7b4e-4412-ab8e-078ed8dbd702","arxiv_id":"1908.09332","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Driven disordered spring networks settle into attractor states fine-tuned to absorb unusually little work from the specific drive, and these states are more stable than expected for their energy absorption.","lead":"Simulations of disordered mechanical networks of bistable springs show that, when driven by a repeating force, the networks settle into states that absorb unusually little energy from that specific drive. The finding suggests a general physical mechanism by which driven multistable systems adapt to their environment, and it could help design materials that 'discover' useful response properties.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Fig. 3's permutation control conflates steady-state adaptation with transient response to a sudden drive switch, weakening the key evidence for atypically low work absorption.","rationale":"The reader's weakest assumption concerned metastability: that observed attractors might escape on longer timescales. That is a real limitation, but it is secondary because the paper's central assertion is about selection bias and atypical work absorption, and the finite-time observation of an attractor can be operationally meaningful even if strict asymptotic stability is not proven. My identified concern is different and more load-bearing: the main statistical control in Fig. 3 is confounded by the transient that follows the drive permutation. The paper's own Figs. 5 and 12 show that switching frequency or direction causes a spike in work absorption and barrier crossings, so the maximum over the post-switch interval in Fig. 3 is very likely a measure of that transient spike, not of the steady-state absorption of the selected state. This undermines the quantitative claim that the attractor is fine-tuned to have atypically low work absorption relative to random drives. The normal-mode analysis in Fig. 4 provides supporting evidence that the final state has reduced coupling to the drive, but it does not quantify atypicality against a null distribution; Fig. 3 is the key statistical test. The proposed check—discarding the transient and measuring steady-state absorption after re-equilibration—would settle whether the effect is real. Until then, the current CONDITIONAL verdict is appropriate, since the paper is promising but its central quantitative evidence is not yet conclusive. I therefore keep the reader's CONDITIONAL verdict unchanged, while identifying a different weakness than the reader did.","tokens_in":15252,"tokens_out":8636,"duration_ms":97293,"concrete_test":"Repeat the Fig. 3 protocol (601 networks/drives, A=5.5, same initial conditions) but after the drive permutation discard the first 20% of the second half to allow the transient to decay, and compute both the time-averaged and the maximum work absorption rate over the remaining cycles. In addition, run the permuted trajectories until the barrier-crossing rate per cycle falls to zero (i.e., a new stable configuration is found) and then measure the steady-state work absorption rate. Compare these steady-state values with the unchanged-drive trajectories. If the steady-state absorption of permuted-drive attractors is not significantly higher than unchanged-drive attractors, the claim of drive-specific fine-tuning must be revised; if it remains higher, the transient confound is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The principal quantitative evidence for the claim that selected attractor states are atypically low work absorbers is Fig. 3, which compares the maximum of the average work absorption rate over the second half of trajectories for unchanged drives versus drives permuted at the midpoint. Because the permutation is a sudden switch in forcing frequency and direction, the second half of the permuted trajectories begins with a transient that the paper itself shows produces a large work-absorption spike (Figs. 5 and 12). The maximum over the entire second half is therefore likely dominated by this transient, not by the steady-state absorption of the attractor. The comparison thus demonstrates that a sudden change of drive causes a transient increase in absorption, rather than establishing that the selected attractor has atypically low work absorption relative to random states under the same drive. To support the headline claim, the permuted trajectories must be allowed to settle, or the transient must be discarded, before comparing absorption. As it stands, Fig. 3's statistics conflate transient response with attractor selection, so the central claim's quantitative support is weaker than presented.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript studies numerically driven disordered mechanical networks of 20 particles connected by 50 bistable springs, each spring having two stable rest lengths described by a double-well potential. For an intermediate range of forcing amplitudes, the authors observe that after an initial period of barrier crossings the network settles into a metastable configuration with a low cycle-averaged work absorption rate. They argue that this low absorption is drive-specific: permuting drives after half the simulation increases the maximum later-half work absorption (Fig. 3), changing the forcing frequency or direction produces absorption spikes and renewed exploration (Fig. 5 and Appendix Fig. 12), and normal-mode analysis shows reduced coupling between the forcing and near-resonant modes in the selected configurations (Fig. 4). They also report that the selected configurations are more stable against escape under the original drive than under perturbed drives with comparable or higher work absorption (Fig. 9), and they interpret the results as a form of dissipative adaptation: driven exploration of configuration space is biased toward states with atypically low work absorption from the specific drive.","tokens_in":15473,"tokens_out":7317,"duration_ms":75348,"significance":"If the central claim holds, the paper provides a clear numerical demonstration in a generic classical many-body system that drive-specific selection can produce states with atypically low work absorption, extending earlier ideas about dissipative adaptation to deterministic, low-noise, strongly driven multistable mechanical networks. The main strengths are the use of direct control comparisons (unchanged vs. permuted drives, frequency/direction switches) and the normal-mode analysis that identifies a concrete linear-response mechanism (reduced coupling to resonant modes) consistent with Eq. (5). The qualitative conclusion is plausible, and the evidence in Figs. 2, 4, 5, 6, 7, and 9 is mutually consistent. The paper is not circular: work absorption is measured, not defined into the conclusion, and the comparison with permuted drives provides an independent control. The main weakness is that the key quantitative evidence for atypicality rests on a control that may conflate transient response with steady-state attractor selection, and the manuscript omits several simulation parameters needed for reproducibility.","major_comments":[{"comment":"The main quantitative evidence for the claim that the selected configurations have atypically low work absorption is the permutation control in Fig. 3. The comparison uses the maximum of the cycle-averaged work absorption rate over the later half of the trajectory, with drives permuted at the midpoint. Because the permutation is a sudden change in forcing frequency and direction, the later half begins with a transient; the paper itself shows in Fig. 5 and Appendix Fig. 12 that such switches produce a large spike in the work absorption rate. The maximum over the entire later half is therefore likely dominated by this transient, so the comparison conflates the transient response to a drive switch with the steady-state absorption of the putative attractor. To support the headline claim, the authors should either discard an initial transient after the permutation and compare late-time average work absorption, or allow the permuted systems to settle into their new attractors before measuring. As it stands, Fig. 3 establishes that a sudden drive change causes a transient absorption increase, but not that the selected states are atypically low absorbers.","section":"Section III, Fig. 3"},{"comment":"The numerical study is not reproducible as written because the double-well potential U(d) is only described by a figure (Fig. 1(a)) and no analytic form is given, and the simulation parameters (mass m, damping gamma, noise strength k_B T, forcing amplitude A, integration timestep, number of drive cycles, and the precise relaxation/ramping protocols used in Section III.B and Fig. 9) are not specified in the text. These details are needed to check the reported results and to understand the claimed range of intermediate amplitudes. The authors should add a complete model specification, including all parameter values and the Verlet integrator settings.","section":"Sections II and III"},{"comment":"The inference that the network has reached a stable attractor is based on observing that barrier crossings vanish over a simulation of a few thousand drive cycles (Fig. 2(c) and Fig. 3 caption). The paper does not provide an estimate of the escape time from the selected configurations or a test of their stability under longer runs or different noise realizations. If the low-work-absorbing states are only long-lived metastable states that would eventually escape, the central claim of permanent selection would need to be qualified. The authors should either report escape-time statistics or soften the terminology from 'attractors' to 'long-lived metastable states' where appropriate.","section":"Section III (attractor inference)"}],"minor_comments":[{"comment":"The caption phrase 'Density of normal modes, stays unchanged' is ungrammatical; it should be 'Density of normal modes is unchanged'.","section":"Fig. 4 caption"},{"comment":"The notation in Eq. (5) is not fully defined in the text; in particular, the relationship between F(t), the 2N-dimensional force vector, and the scalar forcing amplitude A is implicit and should be stated explicitly.","section":"Eq. (5)"},{"comment":"There is a typo in 'classical many-body systenm' (should be 'system').","section":"Section IV"},{"comment":"No data or code availability statement is included; for a purely numerical study, providing the code or a clear statement of availability would improve reproducibility.","section":"General"},{"comment":"The caption says 'a few thousand drive cycles' without giving the exact simulation duration; the precise number of cycles is important for assessing the convergence of the trajectories.","section":"Fig. 3 caption"}],"recommendation":"major_revision","confidential_remarks":"The paper addresses a timely and interesting question, and the qualitative conclusion is supported by several independent analyses. The main issue is the interpretation of Fig. 3, which is the key quantitative evidence for atypicality; it should be re-analyzed after discarding or settling the transient following the drive permutation. The authors should also supply full simulation parameters. With these revisions, the paper would be suitable for publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague, here's my take on Kedia et al. The new thing is the demonstration that the known resonant-destabilization mechanism can operate in high-dimensional disordered mechanical networks: driven networks settle into configurations whose normal-mode couplings near the drive frequency are reduced, and these states are more stable at equal work absorption than states reached by other drives. That's a real step beyond the single-oscillator and low-dimensional examples in the prior literature. The normal-mode analysis (Fig. 4) and the multiple driving protocols (force, displacement, rotating direction) make the claim concrete, and the stability comparison in Fig. 9 with a drive matched in linear-response work absorption is a good control. The paper is honest about the distinction from Prigogine's minimum entropy production.\n\nThe soft spots are in proportion. First, the Fig. 3 permutation control, which is the central evidence for atypically low steady-state work absorption, conflates two things: the permuted trajectories undergo a sudden drive switch at the midpoint, and the paper itself shows that such switches produce a large transient spike in work absorption (Fig. 5 and Appendix Fig. 12). Taking the maximum over the entire second half therefore likely measures the transient, not the steady-state absorption of the attractor. The authors need to either discard an initial transient after the permutation or compare settled averages. Until that's done, the quantitative support for 'atypically low' is weaker than presented. Second, the claim that the network has reached an attractor rests on barrier crossings vanishing over a few thousand drive cycles; there is no estimate of escape time or test against longer runs, so the selected states might be long-lived metastable. That is a minor-to-moderate concern depending on how one reads 'attractor.' Third, the paper gives no simulation parameters (the double-well potential, damping, noise strength, integration details) and no code/data, so independent verification is not possible. For a numerical study that makes a general claim, that's a real deficiency.\n\nThe citation pattern looks fair: prior work on attractor annihilation and dissipative adaptation is acknowledged, and the paper is careful to say what is new. The generality claim ('captures the essential physical properties of a broad class') does outrun the demonstrated parameter range, but that's a typical overreach in a numerical study.\n\nWho is this for? Researchers working on driven multistable systems, dissipative adaptation, and mechanical metamaterials. It's a credible numerical demonstration with a fixable flaw in the key figure. I'd send it to peer review, with the clear request that the permutation analysis be redone to separate transient from steady state, and that simulation details be made available. My own verdict would be conditional until that is addressed.","headline":"A credible numerical demonstration that driven disordered mechanical networks fine-tune normal-mode couplings to reduce work absorption, but the key permutation control in Fig. 3 conflates transient response with steady-state adaptation and needs fixing.","tokens_in":15960,"tokens_out":2354,"would_cite":false,"duration_ms":23048,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Driven disordered networks of bistable springs settle into attractor states that absorb unusually little work from the drive, and these states are more stable than expected.","keywords":["disordered mechanical networks","bistable springs","multistability","attractor selection","work absorption","dissipative adaptation","normal mode analysis","driven nonequilibrium dynamics"],"falsifier":"Run the drive-permutation experiment of Fig. 3 on a large ensemble: if the distribution of maximum work absorption rates in the second half is statistically indistinguishable between unchanged-drive and permuted-drive conditions, the attractor states are not drive-specific; likewise, in the Fig. 9 stability protocol, if the original drive does not show a lower escape probability than drives matched in linear-response work absorption, the extra-stability claim fails.","tokens_in":1723,"feed_emoji":"⚙️","tokens_out":6386,"duration_ms":121766,"temperature":0.7,"pith_summary":"This paper uses simulations of disordered networks of bistable springs to ask whether a multistable system that is driven by a time-varying force settles into states that bear a specific signature of that drive. It finds that after an initial exploratory phase the network becomes trapped in a metastable configuration whose response properties are fine-tuned to absorb unusually little work from that particular drive, and that this fine-tuning is specific to the drive frequency and direction. The same configuration absorbs more work when the drive is changed, and the configuration is more stable under the original drive than under other drives that deliver the same amount of work. If correct, driven exploration of a vast configuration space is biased toward states with a special relationship to the driving environment, offering a route to 'discover' materials with desired response properties.","feed_headline":"Spring networks adapt to their drive, absorbing less work","feed_subtitle":"Simulations show final states are matched to the drive's frequency and direction, not random.","key_machinery":"The load-bearing object is the cycle-averaged work absorption rate of a metastable configuration in the linear-response regime, $\\langle F \\cdot V \\rangle_\\tau = \\frac{\\gamma A^2 \\omega^2}{2} \\sum_{\\lambda} \\frac{(\\hat{F}\\cdot\\hat{\\omega}_\\lambda)^2}{m^2(\\omega_\\lambda^2 - \\omega^2)^2 + \\gamma^2 \\omega^2}$, where $\\hat{\\omega}_\\lambda$ and $\\omega_\\lambda$ are the normal modes and natural frequencies of the dynamical matrix of the configuration. A configuration absorbs work efficiently when a normal mode lies near the drive frequency $\\omega$ and the forcing $\\hat{F}$ couples strongly to it. The simulations show that over time the network changes its configuration so that the coupling $(\\hat{F}\\cdot\\hat{\\omega}_\\lambda)^2$ of modes near resonance drops significantly, while the mode density near resonance is unchanged. The second piece of machinery is contraction analysis, which shows that within a concave-up potential well the damped driven dynamics converge to a single one-dimensional periodic trajectory; this is what makes the discovered state more stable than its work-absorption level alone would predict.","core_discovery":"The central discovery is that driven disordered multistable mechanical networks settle into attractor states that are fine-tuned to the external forcing to have low work absorption from it, and these states are even more stable than expected for that level of work absorption. The mechanism is resonant destabilization: configurations with normal modes strongly excited by the drive absorb work efficiently and become unstable, causing the network to hop over energy barriers until it finds a configuration whose normal modes near the drive frequency have weak coupling to the forcing. A normal-mode analysis shows the density of modes near resonance does not change, but the coupling of those modes to the forcing is significantly reduced in the final configuration. In addition, contraction analysis shows that within a potential well the driven motion converges to a one-dimensional periodic orbit, and experiments with drives of equal linear-response work absorption show that the drive that discovered the well is atypically stable because the shape of the trajectory matters, not just the amount of work absorbed.","pith_inferences":["Inference: A testable extension is to ramp the drive amplitude in small steps after the system settles, which might reveal a cascade of drive-specific states, each stable over a range of amplitudes; the paper does not study amplitude ramps.","Inference: The mechanism implies a form of 'drive-specific immunity': a network adapted to one drive should be comparatively resistant to that drive but not to other drives of the same power, which could be used to design materials with tunable vibrational response by pre-cycling them under a chosen drive.","Inference: The normal-mode coupling signature (reduced projection onto resonant modes) could serve as an experimental diagnostic, since measuring the vibrational eigenmodes of a material before and after a period of driving should show the same reduction in coupling near the drive frequency."],"forward_implications":["For intermediate forcing amplitudes, the same network with the same initial condition and same drive can settle into different attractors depending on noise, but nearly all have low work absorption from that drive.","Switching the drive frequency, direction, or protocol (force versus displacement) causes a spike in work absorption and renewed exploration, ending in a new low-work-absorbing configuration, so the adaptation is reversible and repeatable.","The final configuration is fine-tuned not only to the drive frequency but also to its direction; a rotated forcing couples more strongly to resonant modes.","The low-work configurations are more stable than expected for their work-absorption level: at matched work absorption, the original drive escapes the potential well less often than perturbed drives or a thermal bath.","The mechanism uses generic ingredients (many particles, a rugged landscape, and a diversity of response properties), and the qualitative behavior is reported to be similar across different network topologies."],"supporting_citations":[{"why":"Supplies the dissipative-adaptation framework that the paper extends and contrasts with, where selection favors high work absorption.","marker":"[34]"},{"why":"Formalizes the statistical physics of dissipative adaptation, providing the theoretical context for drive-specific selection.","marker":"[35]"},{"why":"Demonstrates resonant annihilation of an attractor in a bistable system, the single-state mechanism the paper generalizes to networks.","marker":"[37]"},{"why":"Shows how a resonant perturbation controls multistability, further establishing the attractor-annihilation mechanism.","marker":"[38]"},{"why":"Reviews control of multistability and provides the broader framework for resonant destabilization.","marker":"[39]"},{"why":"Supplies the bistable-spring network model from glass physics that this paper adopts.","marker":"[36]"},{"why":"Provides the contraction analysis used to argue convergence to a one-dimensional periodic attractor and trajectory-shape stability.","marker":"[44]"},{"why":"Presents the least-rattling selection mechanism that the paper distinguishes from its own.","marker":"[47]"}],"fun_headline_variants":["Networks of bistable springs tune to their drive","Driven spring networks find states that absorb least work","Bistable spring networks adapt to external forcing","Networks self-organize to minimize work from drive"],"cache_read_input_tokens":18176,"weakest_assumption_plain":"The claim that the system is permanently trapped in a fine-tuned state rests on the observation that barrier crossings vanish over a few thousand drive cycles, leaving open the possibility that the state would escape on much longer timescales or under a different noise realization.","fun_headline_variants_meta":{"raw":{"variants":["Networks of bistable springs tune to their drive","Driven spring networks find states that absorb least work","Bistable spring networks adapt to external forcing","Networks self-organize to minimize work from drive"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000542,"raw_usage":{"total_tokens":2598,"prompt_tokens":945,"completion_tokens":1653,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":561,"completion_tokens_details":{"reasoning_tokens":1591}},"tokens_in":561,"tokens_out":1653,"duration_ms":12223,"temperature":1.0,"reasoning_tokens":1591,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:14:56.885905+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the drive-permutation experiment of Fig. 3 on a large ensemble: if the distribution of maximum work absorption rates in the second half is statistically indistinguishable between unchanged-drive and permuted-drive conditions, the attractor states are not drive-specific; likewise, in the Fig. 9 stability protocol, if the original drive does not show a lower escape probability than drives matched in linear-response work absorption, the extra-stability claim fails.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the dissipative-adaptation framework that the paper extends and contrasts with, where selection favors high work absorption."},{"cited_title":"Marsland, and Jeremy L","cited_arxiv_id":null,"evidence_quote":"Formalizes the statistical physics of dissipative adaptation, providing the theoretical context for drive-specific selection."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Demonstrates resonant annihilation of an attractor in a bistable system, the single-state mechanism the paper generalizes to networks."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Shows how a resonant perturbation controls multistability, further establishing the attractor-annihilation mechanism."},{"cited_title":"Pisarchik and Ulrike Feudel","cited_arxiv_id":null,"evidence_quote":"Reviews control of multistability and provides the broader framework for resonant destabilization."},{"cited_title":"Why glass elasticity affects the thermodynamics and fragility of su- percooled liquids","cited_arxiv_id":null,"evidence_quote":"Supplies the bistable-spring network model from glass physics that this paper adopts."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the contraction analysis used to argue convergence to a one-dimensional periodic attractor and trajectory-shape stability."},{"cited_title":"Least-rattling feed- back from strong time-scale separation","cited_arxiv_id":null,"evidence_quote":"Presents the least-rattling selection mechanism that the paper distinguishes from its own."}],"review_version":1}