{"id":"92233186-eea4-4092-89d5-a3d7033a5caf","arxiv_id":"2607.08969","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.5,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A tuned 3DCNN recovers Reynolds stress tensor components from filtered quiet-Sun velocity and density fields with ~31% lower RMSE on diagonals and ~8% on off-diagonals versus physics baselines.","lead":"A 3D convolutional neural network predicts subgrid Reynolds stress components in quiet-Sun simulations more accurately than Gradient or Smagorinsky models. This offers a data-driven path to cheaper lower-resolution solar convection runs that still capture essential turbulence transport.","discovery_kind":"new_application","skeptic_critique":{"model":"grok-4.5","headline":"Offline a-priori RMSE gains do not establish that the CNN is a viable live SGS closure inside time-evolving StellarBox runs.","rationale":"The reader correctly isolates the offline-to-online gap as the single load-bearing assumption. All quantitative claims (Table 5, Figs. 9–12, density-binned errors) are solid a-priori results; the paper itself flags the missing coupled demonstration. No stronger internal inconsistency appears in the architecture, the log-transform justification, or the baseline comparisons. The CONDITIONAL verdict with HIGH confidence is therefore the appropriate standing judgment; the concrete online test above would be the natural next step that could move the paper to ACCEPT or expose a deeper failure.","tokens_in":22959,"tokens_out":527,"duration_ms":23672,"concrete_test":"Couple 3DCNN1.1 into StellarBox as the sole SGS term, run a 50 km quiet-Sun simulation for several convective turnover times, and compare the resulting horizontally-averaged kinetic-energy spectra and vertical-velocity PDFs against both the original 12.5 km reference and the same low-resolution run with Smagorinsky (Cs=Cc=0.1). If the CNN spectra deviate from the high-resolution reference by more than the Smagorinsky spectra (or become unstable), the offline gains do not transfer.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim (Table 5 / abstract) rests on a-priori tests: filtered high-resolution (12.5 km) cubes supply both the 3\times3\times3 velocity+density inputs and the exact Reynolds-stress targets (Eq. 5). When the same network is fed genuine low-resolution (~50 km) fields (Sec. 5.4), the predicted stress PDFs systematically compress in the tails relative to the high-resolution targets (Figs. 13–14). Because the paper never inserts the network into the right-hand side of the momentum/energy equations (explicitly deferred, Secs. 1 and 5.4), it remains untested whether those compressed stresses produce stable, statistically faithful low-resolution evolution or merely reproduce the offline mapping. In classical LES literature this a-priori / a-posteriori gap is known to be decisive; the reported 31 % / 8 % RMSE reductions therefore cannot yet be read as evidence that the CNN is a drop-in replacement for Smagorinsky.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.5","summary":"The paper develops 3D convolutional neural network (3DCNN) surrogates, along with an MLP baseline, to predict the six components of the Reynolds stress tensor τ_ij from local 3×3×3 averaged velocity fields and central density in quiet-Sun StellarBox radiative hydrodynamic simulations. High-resolution (∼12.5 km) cubes are filtered to ∼50 km effective resolution to generate both inputs and exact targets via the definition τ_ij = 〈u_i u_j〉 − ũ_i ũ_j. After log/signed-log transforms and standardization of the heavily skewed targets, the best architecture (3DCNN1.1 with LeakyReLU) is shown via held-out RMSE, R², PDFs, error histograms, density-binned diagnostics, and low-resolution snapshot tests to outperform the Gradient model, Smagorinsky (Cs=Cc=0.1 and 0.001), and MLP, with average RMSE reductions of ∼31 % on diagonal and ∼8 % on off-diagonal components relative to Gradient (Table 5). The authors conclude that CNNs are a viable candidate for subgrid-scale modeling, while explicitly deferring online coupling into StellarBox.","tokens_in":23250,"tokens_out":1171,"duration_ms":19004,"significance":"If the offline gains translate under live coupling, the work would supply a coefficient-free, data-driven alternative to classical Smagorinsky/Gradient closures for solar convection-zone LES, with clear practical value for multi-scale quiet-Sun modeling. Even as a pure a-priori study the manuscript is carefully executed: physical units are restored after inverted transforms, multiple physics baselines and architecture ablations are reported, and density-binned plus cluster diagnostics expose regime-dependent performance. These elements constitute a solid, reproducible foundation for subsequent a-posteriori tests and are already useful to the solar and LES communities.","major_comments":[{"comment":"Abstract, §1 and §5.4: the central claim that the 3DCNN is “a viable candidate for modeling subgrid processes and a promising alternative to traditional turbulence models” rests exclusively on a-priori regression accuracy (Table 5, Figs. 9–12). When the same network is applied to genuine low-resolution (∼50 km) fields the predicted stress PDFs systematically compress in the tails (Figs. 13–14). Classical LES literature treats the a-priori/a-posteriori gap as decisive; without at least one online StellarBox integration (or a clear, quantitative statement that the present results do not yet establish live-closure viability) the abstract and conclusion over-reach the evidence that is actually supplied.","section":"Abstract, §1, §5.4"},{"comment":"Table 5 and §5.1: the headline ∼31 % / ∼8 % RMSE reductions are computed solely against the Gradient model. Against Smagorinsky (Cs=Cc=0.001) the gains are larger, yet that coefficient set is known to under-dissipate; against Cs=Cc=0.1 the diagonal gains shrink. The abstract and conclusion should report the full range of relative improvements (or at least both Smagorinsky settings) so that the claimed superiority is not tied to a single, relatively weak baseline.","section":"Table 5, §5.1"},{"comment":"§3.1 and Eq. (5): targets are obtained by direct spatial averaging of the high-resolution velocity products. While this is the standard a-priori procedure, the manuscript never quantifies how sensitive the learned mapping is to the precise filter kernel or to the non-overlapping 4×4×4 sub-cube sampling. A short sensitivity test (or an explicit statement that the reported RMSE is filter-specific) is needed before the numbers can be treated as robust estimates of subgrid stress.","section":"§3.1, Eq. (5)"}],"minor_comments":[{"comment":"Figure 1 and Figure 2 captions: “normalized units” are defined only later in the text; a one-sentence reminder in the captions would improve readability.","section":"Figs. 1–2"},{"comment":"§5.7: the K-means analysis is performed on a single data cube and uses an arbitrary k=5; the section title already labels it “preliminary,” but the body still presents cluster-wise R² values as if they generalize. Soften the language or move the entire subsection to an appendix.","section":"§5.7"},{"comment":"Table 6: RMSLE is introduced without definition; a brief formula or reference would help non-ML readers.","section":"Table 6"},{"comment":"Throughout: occasional typographical inconsistencies (e.g., “Th eAstrophysical Journal”, missing spaces around ×10^n) should be cleaned in production.","section":null}],"recommendation":"major_revision","confidential_remarks":"The a-priori results are solid and the authors are transparent about the deferred coupling; the manuscript is therefore salvageable with clearer claim language and a fuller baseline comparison. I would not reject, but I would not accept until the over-statement of “viable SGS candidate” is corrected. Scope is appropriate for an astrophysical methods journal."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The paper’s real result is a careful offline regression: a modest 3D CNN (3DCNN1.1) beats Gradient, two Smagorinsky coefficient settings, and an MLP on the six Reynolds-stress components extracted from StellarBox quiet-Sun cubes. Table 5 and the restored physical units make the ~31 % / ~8 % RMSE claim checkable; the log / signed-log transforms, density-channel ablations, and density-binned diagnostics are done cleanly. That is useful engineering for anyone already running solar LES.\n\nWhat is new is the solar dataset and the dual-coefficient Smagorinsky benchmark, not the architecture. They follow Karpov et al. (2022) closely and apply standard 3-D convolutions. The math is just the definition of τij plus ordinary MSE training; citations look appropriate and the baselines are fair.\n\nThe soft spot is exactly the one the stress-test flags, and the authors already flag it themselves: everything is a-priori. When the same network sees genuine ~50 km fields the stress PDFs compress in the tails (Figs. 13–14). No online coupling into the momentum/energy equations is shown, so we still do not know whether those stresses produce stable, statistically faithful low-resolution evolution. That is a real gap for any claim of “viable candidate for modeling subgrid processes,” but it is an engineering gap, not a circularity or data-fabrication problem. Data availability is also limited to “upon request.”\n\nThis is for computational solar physicists and LES method people who need a quantitative starting point for data-driven SGS. It is not a turbulence-theory paper. I would send it to peer review; the offline evidence is solid enough to deserve referee time, with the clear expectation that the a-posteriori test is the next required step. Worth reading if you work on solar convection or ML closures; not urgent otherwise.","headline":"Solid offline a-priori CNN gains on quiet-Sun Reynolds stresses, but the live-closure claim is still untested.","tokens_in":23848,"tokens_out":467,"would_cite":false,"duration_ms":6332,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.5","headline":"A 3D convolutional network predicts solar subgrid Reynolds stresses more accurately than standard physics-based closures.","keywords":["subgrid-scale turbulence","Reynolds stress tensor","3D convolutional neural networks","quiet Sun","radiative hydrodynamics","Large Eddy Simulation","Smagorinsky model"],"falsifier":"Insert the trained network as the active subgrid model inside a low-resolution StellarBox run and check whether the resulting large-scale velocity and density statistics, plus the emergent stress distributions, match those of a high-resolution reference simulation better than the same run performed with a Smagorinsky closure.","tokens_in":23888,"feed_emoji":"☀️","tokens_out":602,"duration_ms":7415,"temperature":0.7,"pith_summary":"Solar plasma simulations cannot resolve every scale of turbulent motion, so they need a cheap estimate of the missing small-scale stresses that still transport momentum and dissipate energy. This paper shows that a carefully designed three-dimensional convolutional neural network can learn those stresses directly from high-resolution quiet-Sun radiative-hydrodynamic runs and then reconstruct the six components of the Reynolds stress tensor more faithfully than either a classical Gradient model or the Smagorinsky eddy-viscosity model. The network is trained on local 3-by-3-by-3 velocity neighborhoods plus density, after a signed-log transform that tames the heavily skewed stress distributions. On held-out data the best architecture cuts root-mean-square error by roughly thirty percent on the diagonal stresses and eight percent on the off-diagonal stresses relative to the Gradient baseline. The result matters because it opens a practical route to cheaper yet more faithful large-eddy simulations of the solar convection zone and lower atmosphere.","feed_headline":"Neural net cuts solar subgrid stress error by ~30%","feed_subtitle":"3D CNN beats Gradient and Smagorinsky closures on quiet-Sun Reynolds stresses","key_machinery":"The 3DCNN1.1 architecture: two 3-by-3-by-3 convolutional layers (32 then 64 filters, LeakyReLU) that extract spatial features from the three velocity channels, followed by concatenation of the central density and SoftSign fully-connected layers that output the six stress components; trained after signed-log transformation of the targets.","core_discovery":"A 3D convolutional neural network that ingests local averaged velocity cubes and density can reconstruct the six Reynolds stress tensor components of quiet-Sun turbulence more accurately than the Gradient model, the Smagorinsky model (under two coefficient choices), and a multilayer perceptron, delivering average RMSE reductions of about 31 percent on the diagonal components and 8 percent on the off-diagonal components.","pith_inferences":[],"forward_implications":[],"fun_headline_variants":["3D CNN cuts quiet-Sun Reynolds stress RMSE ~31% on diagonals","CNN beats Gradient and Smagorinsky on solar subgrid stresses","3DCNN reconstructs quiet-Sun stress tensor better than MLP","Log transform plus 3D CNN improves solar turbulence closures","Deep net lowers off-diagonal solar stress error by ~8%"],"cache_read_input_tokens":16512,"weakest_assumption_plain":"That accurate offline predictions on filtered high-resolution cubes (and static low-resolution snapshots) will remain accurate once the same network is inserted as a live subgrid closure inside a time-evolving solar simulation.","fun_headline_variants_meta":{"raw":{"variants":["3D CNN cuts quiet-Sun Reynolds stress RMSE ~31% on diagonals","CNN beats Gradient and Smagorinsky on solar subgrid stresses","3DCNN reconstructs quiet-Sun stress tensor better than MLP","Log transform plus 3D CNN improves solar turbulence closures","Deep net lowers off-diagonal solar stress error by ~8%"]},"model":"grok-4.5","effort":"low","cost_usd":0.00575,"raw_usage":{"total_tokens":1581,"prompt_tokens":838,"num_sources_used":0,"completion_tokens":93,"cost_in_usd_ticks":57500000,"prompt_tokens_details":{"text_tokens":838,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":650,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":838,"tokens_out":93,"duration_ms":6727,"temperature":1.0,"reasoning_tokens":650,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-13T05:24:28.742335+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Insert the trained network as the active subgrid model inside a low-resolution StellarBox run and check whether the resulting large-scale velocity and density statistics, plus the emergent stress distributions, match those of a high-resolution reference simulation better than the same run performed with a Smagorinsky closure.","supporting_citations":[],"review_version":1}