{"id":"5651d271-d1ac-4d23-b97d-02fa818c1017","arxiv_id":"2501.06093","paper_version":3,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":13,"one_line_summary":"A reduced streamer model represents channels as conducting cylinders and uses fits from 1000 axisymmetric simulations to reproduce branching discharges with time steps about 1000 times larger than fluid models.","lead":"This paper presents a fast computational model for filamentary electric discharges, representing streamers as conducting cylinders on a grid. The model is fitted to 1000 detailed simulations and can grow branching streamer trees in 3D in minutes on a desktop computer.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 3D branching results rest on an untested transfer of the LE-based closure (Eqs. 16–18) from isolated axisymmetric streamers to interacting, small-radius branches in a different geometry.","rationale":"The reader's weakest_assumption is exactly the same load-bearing point I identify: the LE-based fits are trained on axisymmetric, isolated, moderate-field streamers and then applied without recalibration to 3D branching, small-radius, interacting streamers in a different gap. I agree with the conditional verdict because the paper is honest about this limitation, provides code and data, and demonstrates within-distribution accuracy and step-size insensitivity. The missing piece is a quantitative check that the closure survives the regime change. Without that check, the 4–8 minute 3D simulations are an impressive engineering result but not yet a validated predictive model for branching discharges. No other concern seems more load-bearing: the axisymmetric comparison is partially in-sample but the train/test split shows no overfitting; the time-step and grid-spacing tests are credible; and the branching parameters are explicitly left as adjustable. The proposed fluid-vs-reduced 3D comparison is the single test that would settle whether the central transfer assumption holds.","tokens_in":18615,"tokens_out":3234,"duration_ms":37269,"concrete_test":"Run a small 3D drift-diffusion fluid simulation (e.g., with afivo-streamer) of a single streamer and of a short branching event in a geometry close to Sec. 5.3, using the same definitions as Sec. 3.2 to extract LE, v, R_E, and sigma_h from the fluid data. Then run cocydimo with the same electrode geometry, voltage, and initial conditions, and compare time-resolved head positions, radii, and line conductivities. If the reduced-model predictions fall within the R2 scatter of the axisymmetric fits (roughly ±20–30% for v and R_E), the transfer is supported; if systematic offsets appear (e.g., velocities biased by more than 30% or radii consistently wrong as branches shrink), the 3D simulations should be presented as qualitative demonstrations rather than quantitative predictions.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The practical claim that matters for the paper is the 3D capability: 20+ streamers in 4–8 minutes with realistic morphologies. That claim depends entirely on Eqs. (16)–(18) being universal functions of LE. The dataset used to fit those expressions covers only isolated, axisymmetric positive streamers in a 30 mm gap, rod radii 0.5–1.5 mm, and voltages 36–60 kV (Sec. 3.2). Section 3.4 explicitly lists the missing regimes: no branching, no streamer interactions, and no low-background-field cases with small radii. The 3D simulations in Sec. 5.3 operate precisely in those missing regimes: branches can approach the stagnation threshold R_E,min = 0.15 mm (Sec. 4.2), and each head sees a field superposition from neighboring channels that the single-head axisymmetric fits never encountered. The only 3D comparison is qualitative morphology against experiments with a different voltage rise time, and Sec. 5.3 states that the branching parameters cannot be determined from it. Therefore, if the closure is biased in the branching/low-field/small-radius regime, the simulated velocities, radii, and conductivities are biased, and the impressive speedup produces fast but potentially incorrect 3D predictions. This is not an internal inconsistency; it is a load-bearing extrapolation that the paper acknowledges but does not quantitatively close.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents a reduced modeling framework (cocydimo) for filamentary discharges, in which streamer channels are represented as conducting cylindrical segments moving on a numerical mesh. A 1000-run axisymmetric fluid simulation dataset is used to fit simple expressions for streamer head radius, velocity, and line conductivity as functions of the high-field length scale L_E (Eqs. 16-18). These expressions are then embedded in a reduced model with a stochastic branching rule, and the model is tested against the axisymmetric dataset, studied for time-step and grid-spacing sensitivity, and demonstrated in 3D simulations with branching in a 4 cm gap.","tokens_in":19057,"tokens_out":6847,"duration_ms":70782,"significance":"If the closure and the 3D transfer hold, this would be a substantial practical advance: 3D simulations with 20+ interacting streamers in minutes on a desktop computer, with public code and dataset, would enable parameter studies and large-scale discharge morphologies that full fluid simulations cannot reach. The time-step insensitivity (Fig. 10) and the grid-spacing correction (Fig. 11) are convincing for the tested case, and the open-source availability of both the code and the dataset is a clear strength. However, the current evidence does not yet establish the predictive accuracy of the model in the regimes where it is most useful: branching, streamer interaction, low background fields, and small-radius branches are all outside the dataset used to fit the closure, and the 3D experimental comparison is qualitative with a ~30% velocity discrepancy and undetermined branching parameters.","major_comments":[{"comment":"The validation in §5.1 is not an out-of-sample test of Eqs. (16)-(18). Section 3.3 states that the data were split into 70% training and 30% test sets, but it also says there was 'essentially no overfitting', and §5.1 does not state whether the eight displayed runs in Figs. 7-9 belong to the training or test portion. Since the same dataset was used both to fit and to assess the closure, the R² values in Fig. 5 and the qualitative agreement in Figs. 7-9 overstate predictive skill. Given that the R² values are only 0.80 for σ_h and 0.79 for R_E, this distinction matters. Please report test-set metrics separately, identify the displayed runs, and ideally validate on parameter ranges or geometries excluded from the training set.","section":"§3.3, §5.1"},{"comment":"The central 3D claim rests on applying Eqs. (16)-(18) outside their training regime. The dataset contains only isolated axisymmetric positive streamers in a 30 mm gap, and Section 3.4 explicitly lists the missing regimes: no branching, no streamer interactions, and no low-background-field or stagnating cases with small radii. The 3D simulations of Section 5.3 operate in exactly these regimes: branches can approach the stagnation radius R_E,min = 0.15 mm, and each head sees fields from neighboring channels. This is a load-bearing extrapolation that the paper acknowledges but does not quantitatively close. A concrete test would be to extract L_E, v, R_E, and σ_h from a full 3D fluid simulation of a branched discharge (or from a two-head interaction setup) and compare them with Eqs. (16)-(18); alternatively, run the reduced model and a full fluid reference on the same small 3D case and quantify errors in velocity, radius, and conductivity.","section":"§4.2, §5.3, §3.4"},{"comment":"The experimental comparison does not currently constrain the predictive accuracy of the 3D model. The fastest simulated streamer velocity is about 1.1 mm/ns versus 0.8 ± 0.2 mm/ns in the experiments, a roughly 30% difference that is attributed to the voltage rise time, but no simulation with a finite rise time is presented to test that explanation. The branching parameters c_b and L_b are varied and judged only qualitatively, and Section 5.3 concludes that their values 'could not accurately be determined'. Please provide quantitative morphology metrics (branch counts, branch angles, velocity distributions, channel radii) and a calibration or sensitivity statement for c_b and L_b, or explicitly present Section 5.3 as a feasibility demonstration rather than a validation.","section":"§5.3"}],"minor_comments":[{"comment":"The caption reads 'a) c1 = 0x' for the uncorrected case; this appears to be a typo for 'c1 = 0' or 'c1 = 0.0'.","section":"Fig. 11 caption"},{"comment":"The axes of Fig. 5 give σ_h in units of A m/MV, while the text describing Eq. (16) lists units of A m/V; please unify the notation so that the fit coefficients and the plotted quantities are immediately comparable.","section":"Fig. 5 and Eq. (16)"},{"comment":"The text states that L_E,min = 0.1 mm corresponds to R_E,min = 0.15 mm, but Eq. (17) with L_E = 0.1 mm gives R_E ≈ 0.16 mm; please check the rounding or clarify how the threshold was obtained.","section":"§4.2"},{"comment":"The sentence introducing Eqs. (16)-(18) says the quantities are 'made dimensionless using the following units'; this wording is confusing because the expressions remain dimensionful. Consider saying that the fits are written in SI units.","section":"§3.3"}],"recommendation":"major_revision","confidential_remarks":"This is a methods-oriented paper that would fit a journal like Computer Physics Communications if the claims are aligned with the evidence. The dataset and code availability are strong assets. The main risk is not an internal inconsistency but an unclosed extrapolation from axisymmetric training data to branched 3D predictions; I would not reject if the authors either add quantitative out-of-sample tests or substantially soften the validation claims."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nI read the paper on the reduced streamer model. The genuinely new piece is the mesh-based conducting-cylinder framework (cocydimo) plus a 1000-run axisymmetric dataset from which they fit simple closures for velocity, radius, and line conductivity using a single feature, L_E, the size of the high-field region ahead of the head. The framework separates the growth model (Python) from the field solver (Fortran/afivo), and both code and data are public. That is real, checkable value.\n\nWhat's done well: the time-step insensitivity (Fig. 10) and grid-spacing correction (Fig. 11) are convincing. The model really does allow ns time steps and sub-mm spacing while reproducing the axisymmetric cases decently. The fit scores (R2 0.8-0.93) are reported honestly, and the dataset's limits are spelled out in Sec. 3.4.\n\nThe soft spots, in proportion. The main axisymmetric validation is in-sample: Eqs. (16)-(18) were fit on the same dataset used for comparison. The train/test split shows no overfitting, but it only proves the fits represent that axisymmetric family; it doesn't test transfer. The 3D results in Sec. 5.3 run exactly in the regimes the dataset excludes: branching, streamer interactions, and small-radius branches near the stagnation threshold. The paper acknowledges this, but the speedup claim for 3D is only as good as the closure's universality. The experimental comparison in 3D is qualitative, with velocities about 30% high and branching parameters not determined. There is also a factor-of-two line-conductivity underestimate near the rod in some axisymmetric cases, and several hand-chosen parameters (c_ahead, tau_delay, branching c_b/L_b) are arbitrary.\n\nIs the central argument sound? For the axisymmetric regime, yes. For 3D branching, it's plausible but unproven. That is a conditional result, not a failure. The stress-test note is fair.\n\nWho this is for: anyone modeling streamer or leader dynamics who needs a fast tool for parameter studies or coupling to gas dynamics. The framework and dataset are useful even if the 3D closure turns out to need recalibration. It deserves a serious referee; the public code and data make the claims checkable, and the limitations are stated rather than hidden. I'd send it to review, with the expectation that the 3D transfer be framed as a hypothesis to test, not a validated prediction.","headline":"The reduced-model framework and public dataset are a real contribution; the 3D branching claim is an acknowledged extrapolation, not a validation.","tokens_in":19498,"tokens_out":4296,"would_cite":true,"duration_ms":36015,"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 single length scale ahead of a streamer tip predicts its velocity, radius, and channel conductivity, so reduced models can skip picosecond steps and run 3D branching discharges on a desktop.","keywords":["electric discharge","streamer discharge","reduced model","data-driven model","conducting cylinders","branching","Poisson equation","positive streamers"],"falsifier":"Run the reduced model in a regime the training set did not cover - for example, a positive streamer propagating in a background field below about 10 kV/cm or a gap over 30 mm - and compare its predicted head velocity, radius, and line conductivity against a full 3D drift-diffusion simulation or an experiment. If the $v=1.78\\times10^9 L_E$ relation or the radius and conductivity fits systematically miss outside the trained range, the transferability assumption is falsified.","tokens_in":2066,"feed_emoji":"⚡","tokens_out":4366,"duration_ms":122187,"temperature":0.7,"pith_summary":"This paper tries to show that the expensive microscopic physics of positive streamers in air can be coarse-grained into a single measurable quantity: the size $L_E$ of the high-field zone just ahead of a streamer tip. From 1000 axisymmetric fluid simulations, the paper derives simple fit expressions for streamer velocity, radius, and line conductivity (conductance per unit length) as functions of $L_E$, and builds a reduced model in which each streamer is a chain of conducting cylindrical segments on a numerical mesh. The reduced model reproduces the axisymmetric simulations' velocity, radius, conductivity, and on-axis field profiles while using time steps up to about 1 ns instead of about 2 ps and grid spacings up to hundreds of micrometers instead of a few micrometers. Because the electron dynamics no longer have to be resolved, 3D simulations with 20+ branching streamers run in 4-8 minutes on a desktop computer and produce morphologies that resemble experimental discharges. If the approach holds, it opens a path to simulating large multi-streamer systems, leaders, and sprites that are currently out of reach of direct fluid models.","feed_headline":"Streamer model skips picosecond steps, simulates branching in minutes","feed_subtitle":"Data-driven fits from 1000 fluid runs let a desktop 3D model match full simulations with nanosecond time steps.","key_machinery":"The load-bearing object is the length scale $L_E$: the distance from the streamer head to the point where the on-axis electric field falls below 50 kV/cm. All three fitted quantities - velocity, radius, and line conductivity - are functions of $L_E$ alone, making it the single state variable that carries the reduced model. The computational carrier is the framework in which each channel is grown as cylindrical segments with a semi-spherical cap; the segment conductivity is mapped onto a tree-structured adaptive mesh, and the next potential comes from solving $\\nabla\\cdot[(\\varepsilon_0+\\Delta t\\,\\sigma)\\nabla\\phi]=-\\rho/\\varepsilon_0$ with geometric multigrid. Branching is modeled as a memoryless Poisson process with mean time $\\bar{\\tau}_{\\mathrm{branch}} = c_b (R_\\sigma/v)(1+L_b^2/R_\\sigma^2)$, whose two parameters control branch frequency and the suppression of thin channels.","core_discovery":"The central claim is that the dynamics of a positive streamer head in air are determined, to a good approximation, by the extent $L_E$ of the region in front of the head where the electric field exceeds 50 kV/cm. Using this single feature, the paper fits closed-form expressions (Eqs. 16-18) for the streamer's head velocity $v$, its electrodynamic radius $R_E$ (the radius at which the radial electric field peaks), and the line conductivity $\\sigma_h$, and then embeds these fits in a mesh-based model of conducting cylinders with semi-spherical caps. The resulting reduced model agrees well with the drift-diffusion simulations it was trained against, and it is numerically stable for time steps up to about 1 ns and grid cells hundreds of micrometers wide, because the potential update solves an implicit variable-coefficient Poisson equation rather than tracking electron density. This speedup is what makes branching 3D simulations of 20+ channels practical on a desktop computer. The paper also shows that a simple grid-spacing correction removes most of the dependence of the measured $L_E$ on resolution, and that a two-parameter Poisson branch model produces experimentally plausible discharge trees.","pith_inferences":["If $L_E$ is truly a sufficient predictor, the same fitting pipeline should transfer to negative streamers, other gas mixtures, or sprite discharges, provided the training dataset is regenerated; the paper lists these as future work, but the transferability is a direct consequence of the single-feature assumption.","A sharper, testable consequence is that the reduced model should fail precisely where the $L_E$-scaling breaks down: in low background fields where streamers become thin and stagnate, a regime the dataset excludes; comparing predicted radius and velocity against full 3D fluid simulations in such fields would probe the boundary of the method.","The branching parameters $c_b$ and $L_b$ are only qualitatively calibrated; a quantitative check would compare the model's branch-angle and branch-spacing distributions against high-speed imaging statistics from experiments.","Since the conductivity field lives on the mesh, the model is naturally positioned to be coupled to gas dynamics for ohmic heating, making a streamer-to-leader transition simulation a plausible near-term extension rather than a separate framework."],"forward_implications":["Time steps in the reduced model can be up to about 1 ns, roughly three orders of magnitude larger than the ~2 ps steps of the fluid simulations, so multi-streamer discharges can be evolved over much longer physical times.","Grid spacings of hundreds of micrometers suffice, compared with a few micrometers for fluid models, and a one-parameter correction for grid resolution makes the predicted streamer velocity nearly independent of $\\Delta x$.","3D simulations with 20+ branching streamers in a 4 cm gap reproduce qualitative experimental features - stagnation of overtaken branches, near-horizontal propagation near the electrode, fastest vertical propagation around 1.1 mm/ns - and complete in 4-8 minutes on a desktop computer.","Because the framework only needs a rule for advancing position, radius, and line conductivity, it can accept other growth models, such as physics-based reduced models or machine-learned surrogates, without changing the field solver.","The implicit potential update removes the dielectric-relaxation time restriction, so the model is stable for time steps much larger than $\\tau_{\\mathrm{drt}}=\\varepsilon_0/\\sigma$."],"supporting_citations":[{"why":"Supplies the parallel tree-structured AMR and multigrid Poisson solver that carries the framework.","marker":"[24]"},{"why":"Provides the tree model whose Poisson-branching and line-conductivity concepts the reduced model adapts.","marker":"[22]"},{"why":"Experiments in a 4 cm gap at +40 kV against which the 3D morphology and velocity are compared.","marker":"[37]"},{"why":"The fluid model and AMR code used to generate the 1000-run axisymmetric dataset.","marker":"[17]"},{"why":"Provides the reaction and transport data used in the fluid simulations that produced the training data.","marker":"[30]"},{"why":"Cross-section data for N2 and O2 from which ionization coefficients and electron mobility, and hence the CFL limits, are computed.","marker":"[12, 13]"},{"why":"Six-code comparison cited for the time-step restriction that motivates the reduced model.","marker":"[11]"},{"why":"Stagnation studies used to justify the $L_{E,\\min}$ threshold and identify the missing low-field regime.","marker":"[33, 34, 35]"}],"fun_headline_variants":["Streamer speed from one field length: 3D branching in minutes","Reduced streamer model: 20+ branches in 8 minutes on a desktop","From 1000 fluid runs to a fast reduced streamer model","Predict streamers with one length scale, branch in minutes"],"cache_read_input_tokens":21504,"weakest_assumption_plain":"The fitting formulas come from axisymmetric simulations in one geometry and are applied, without recalibration, to 3D branching discharges in a different geometry; everything rests on the assumption that the size of the high-field zone ahead of a streamer tip is the only information needed to predict how fast it moves, how thick it grows, and how conductive it becomes.","fun_headline_variants_meta":{"raw":{"variants":["Streamer speed from one field length: 3D branching in minutes","Reduced streamer model: 20+ branches in 8 minutes on a desktop","From 1000 fluid runs to a fast reduced streamer model","Predict streamers with one length scale, branch in minutes"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000243,"raw_usage":{"total_tokens":1573,"prompt_tokens":1032,"completion_tokens":541,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":648,"completion_tokens_details":{"reasoning_tokens":463}},"tokens_in":648,"tokens_out":541,"duration_ms":5071,"temperature":1.0,"reasoning_tokens":463,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T21:05:53.448648+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the reduced model in a regime the training set did not cover - for example, a positive streamer propagating in a background field below about 10 kV/cm or a gap over 30 mm - and compare its predicted head velocity, radius, and line conductivity against a full 3D drift-diffusion simulation or an experiment. If the $v=1.78\\times10^9 L_E$ relation or the radius and conductivity fits systematically miss outside the trained range, the transferability assumption is falsified.","supporting_citations":[{"cited_title":"Comparison of six simulation codes for positive streamers in air","cited_arxiv_id":null,"evidence_quote":"Six-code comparison cited for the time-step restriction that motivates the reduced model."},{"cited_title":"Afivo: A framework for quadtree/octree AMR with shared-memory parallelization and geometric multigrid meth- ods","cited_arxiv_id":null,"evidence_quote":"Supplies the parallel tree-structured AMR and multigrid Poisson solver that carries the framework."},{"cited_title":"Growing discharge trees with self- consistent charge transport: The collective dynamics of streamers","cited_arxiv_id":null,"evidence_quote":"Provides the tree model whose Poisson-branching and line-conductivity concepts the reduced model adapts."},{"cited_title":"Positive and negative streamers in ambient air: Measuring diameter, ve- locity and dissipated energy","cited_arxiv_id":null,"evidence_quote":"Experiments in a 4 cm gap at +40 kV against which the 3D morphology and velocity are compared."},{"cited_title":"Simulating streamer discharges in 3D with the parallel adaptive Afivo framework.Journal of Physics D: Applied Physics, 50(47):474001, October 2017","cited_arxiv_id":null,"evidence_quote":"The fluid model and AMR code used to generate the 1000-run axisymmetric dataset."},{"cited_title":"Comparing simulations and experiments of positive streamers in air: Steps toward model validation","cited_arxiv_id":null,"evidence_quote":"Provides the reaction and transport data used in the fluid simulations that produced the training data."}],"review_version":1}