-
The underlying RHD simulation is insufficiently specified, limiting reproducibility and physical interpretation. Sec. 2.1 mentions an “advanced radiation-hydrodynamics code” and a snapshot file, but does not clearly state the code name/version; coordinate system and grid geometry (including whether $\theta$–$\phi$ spacing is uniform and whether AMR/refinement is present); gravity/orbital implementation (rotating vs inertial frame, companion treatment/sink vs potential); equation of state and opacities; radiation transport method/closure (e.g., FLD/M1/VET) and coupling terms; boundary/initial conditions; and the mapping from code units to physical units (density/velocity/time/$E_r$). These details directly affect the meaning of: the 53-cell “convective scale,” the ROI sizes in $R_\odot$, radiation anisotropy and alignment statistics, and clump masses/volumes (Sec. 2.1, Sec. 3.2–3.5).
Recommendation: Expand Sec. 2.1 (or add a dedicated “Simulation setup” subsection) to provide a concise but self-contained model description: code name/version; grid geometry and resolution (including $\theta$–$\phi$ spacing, radial spacing/refinement, AMR treatment if any); orbital setup (separation, mass ratio, frame, rotation state), donor/companion parameters; gravity and companion implementation; EOS ($\gamma$, mean molecular weight) and opacity sources; radiation transport scheme and closure and how $P_{\rm rad}/F_r/E_r$ are defined in the code; boundary/initial conditions. Add a short “Units and scaling” table mapping code units to physical units and, where you quote code-unit values in Sec. 3, give approximate physical values. If details are in a prior paper, cite it but still summarize the essentials here.
-
Single-snapshot inference is over-extended in places. The analysis uses one instantaneous snapshot of one configuration, but the Abstract/Sec. 1/Sec. 4.2 contain generalized statements (e.g., “fundamentally clumpy” transfer; radiation not being the primary mechanism shaping the bulk stream; L1 being dominated by infall). Given the intrinsic time variability of RSG convection and intermittent RLOF/nozzle dynamics, it is unclear whether the reported PDFs, the $\sim 53$-cell correlation length, the $\sim 60^\circ/110^\circ$ alignment peaks, and the clump counts/masses are typical or transient (Sec. 1, Sec. 3.1–3.5, Sec. 4.2–4.3).
Recommendation: Either (i) add a minimal temporal robustness check using a small set of nearby snapshots/orbital phases (even $3$–$10$ outputs) and show stability (or variability ranges) for the headline metrics: convective correlation length (Sec. 3.2), L1 $v_r$ sign balance (Sec. 3.1.1), alignment/anisotropy PDFs (Sec. 3.3), and clump statistics (Sec. 3.4); or (ii) if more outputs are not available, systematically qualify claims throughout (Abstract, Sec. 1, Sec. 3, Sec. 4.2–4.3) as applying to “this snapshot/model” and add a concise limitations paragraph in Sec. 4.3 describing how results might change with time, mass ratio, separation, or evolutionary stage.
-
ROI/mask definitions and thresholds are not sufficiently justified or stress-tested and may bias the main conclusions. Examples include: photosphere defined as $r=640$–$660\ R_\odot$ (Sec. 2.2.1) without an optical-depth/$\tau$-based justification; L1 vicinity as a $100\ R_\odot$ sphere around an approximate L1 location (Sec. 2.2.2); stream mask $r>700\ R_\odot$, $v_r>0$, and a $45^\circ$ angular cut (Sec. 2.2.3), which can pre-select outward flow and potentially exclude deflected/returning stream material; and clumps defined as connected components above the $95$th percentile of $|\dot{m}|$ within the stream (Sec. 2.6, Sec. 3.4).
Recommendation: In Sec. 2.2.1–2.2.3 and Sec. 2.6, add physical justification for each mask and perform a small sensitivity study. Concretely: (a) photosphere—if $\tau$ is available, consider defining a $\tau\approx 2/3$ surface or show that the $r$-shell captures comparable material; (b) L1 vicinity—justify $100\ R_\odot$ as a fraction of separation/nozzle scale and report how key PDFs/anisotropy change for e.g. $50$ and $150\ R_\odot$ (Sec. 3.1.1, Sec. 3.3); (c) stream—repeat key results for alternative angular/radial cuts and one definition not explicitly imposing $v_r>0$ (e.g., density-enhanced tube along the binary axis or flux-based selection) to show conclusions are not selection artifacts (Sec. 3.1.2–3.3); (d) clumps—repeat for at least two other thresholds (e.g. $90$th and $98$th/$99$th percentiles) and report how the number of structures, total clump mass, and clump-carried mass flux vary (Sec. 3.4).
-
Potential bug/dimensional inconsistency in the L1-vicinity definition: the text uses a squared-distance expression but compares it to “100” rather than $(100\ R_\odot)^2$, implying the ROI could be mis-defined (Sec. 2.2.2, p.3). This would propagate into Sec. 3.1.1 and Sec. 3.3 conclusions about the L1 neighborhood.
Recommendation: Correct Sec. 2.2.2 to use either $(\Delta x^2+\Delta y^2+\Delta z^2) < (100\ R_\odot)^2$ or $\sqrt{\Delta x^2+\Delta y^2+\Delta z^2} < 100\ R_\odot$, with explicit units. Re-run the L1 ROI analysis if the implemented mask differs from the written one, and briefly note the impact (if any) on L1 PDFs and anisotropy/alignment results.
-
Two-point correlation methodology needs clearer definition, physical scaling, and uncertainty estimates. The headline convective scale is reported as “$\sim 53$ grid cells” (Sec. 3.2), but without converting to an angular/linear length at $r\approx 649.6\ R_\odot$ (and accounting for spherical metric factors and any non-uniform $\theta$–$\phi$ spacing). FFT-based correlations also implicitly assume periodicity and require careful handling of poles/edges; mean subtraction and normalization are ambiguous (autocorrelation described as a “correlation coefficient” with $C(0)=1$ despite an FFT expression that appears unnormalized), and the $v_r$–$F_r$ cross-correlation does not clearly state whether both fields are mean-subtracted and normalized (Sec. 2.4, Sec. 3.2).
Recommendation: Revise Sec. 2.4 and Sec. 3.2 to: (a) precisely define the slice geometry ($\theta$–$\phi$ at fixed $r$ vs Cartesian patch), domain used (e.g., excluding polar caps), and boundary treatment for FFT (periodic in $\phi$; what about $\theta$?); (b) state explicitly whether autocorrelation is normalized by $C(0)$ or by the variance so that $C(0)=1$, and provide the exact cross-correlation definition (e.g., $\langle \delta v_r \ \delta F_r \rangle/(\sigma_{v_r} \sigma_{F_r})$); (c) convert the $53$-cell lag to an angular scale (degrees/radians) and linear scale ($R_\odot$) using local grid spacing at the analyzed radius, and report an uncertainty (e.g., from varying nearby shells/patches or bootstrap resampling). This will make the convective scale comparable to theory (e.g., pressure scale height / mixing-length expectations) and other simulations/observations.
-
The inference that radiation is not dynamically important for the bulk stream is currently based mainly on anisotropy and misalignment between the principal eigenvector of $P_{\rm rad}$ and the gas velocity (Sec. 3.3, Sec. 4.2). Misalignment alone is not a force diagnostic, and the principal axis of $P_{\rm rad}$ is not necessarily the direction of radiative acceleration. In addition, eigenvectors have a sign ambiguity; without enforcing a sign convention, angle distributions can be mirrored (e.g., 110° vs 70°). The manuscript also does not compare against flux-direction alignment (often more directly tied to radiative momentum transport in moment methods) nor quantify the magnitude of radiative acceleration relative to gravity/inertial terms.
Recommendation: Strengthen Sec. 3.3 by adding: (a) a force-based comparison in ROIs 2–3—e.g., PDFs of $|a_{\rm rad}|/|a_{\rm grav}|$ and/or $|a_{\rm rad}|/|v\cdot\nabla v|$, using available code outputs for $\nabla\cdot P_{\rm rad}$, $\kappa \rho F_r/c$, or the radiation–matter coupling term; (b) a companion alignment diagnostic using the radiation flux vector (e.g., angle between $F$ and $v$, and between $F$ and the stream direction), clearly stating how $F_r$ is defined; (c) an explicit convention to remove eigenvector sign ambiguity (e.g., flip the eigenvector to have positive dot with $+r$ or with $F$ before computing the angle), and clarify whether angles are reported in $[0,180^\circ]$ or folded to $[0,90^\circ]$. Update Sec. 4.2 language to reflect the quantitative force ratios: either support “sub-dominant” with numbers or qualify the claim to “misaligned/complex coupling” where radiation is non-negligible.
-
Clump/feature identification and clustering results depend on underspecified methodological choices and have internal inconsistencies in reported structure properties. Connected-component labeling does not specify 3D connectivity (6/18/26) nor boundary handling (Sec. 2.6), and clumps defined by high $|\dot{m}|$ may capture shocks/shear rather than coherent overdensities unless cross-checked with density-based criteria. Cell volumes in spherical grids are generally non-uniform; computing clump volume as $\rm num\_cells \times $(average cell volume) can be inconsistent with mass integrals (Sec. 2.6). Table 1 appears inconsistent (volume/num_cells varies by large factors; units/column definitions such as $r_c$ are unclear; some radii look truncated vs the text) (Sec. 3.4, Table 1). For clustering, $k=5$ is justified mainly by a visual elbow and “physical meaningfulness,” with limited validation, unclear sampling/imbalance handling across ROIs, and incomplete hyperparameter reporting (Sec. 2.7, Sec. 3.5).
Recommendation: For Sec. 2.6–2.7 and Sec. 3.4–3.5: (a) specify connected-component connectivity (6/18/26) and any filtering; compute clump volume as $\sum dV$ per cell (not average-volume approximations) and report typical clump sizes relative to grid spacing; (b) add robustness checks across multiple $|\dot{m}|$ thresholds and at least one alternative clump definition involving density (or joint criteria), and quantify what fraction of total stream mass flux is carried by clump cells vs diffuse cells; (c) correct and fully define Table 1 (column names, units, consistent radii/masses) and explain any non-uniform cell volumes if volume/num_cells varies; (d) for clustering, report the sampling strategy (number of cells, weighting, ROI balancing), feature scaling, and MiniBatchKMeans parameters (batch size, n_init, random_state), provide at least one internal validation metric (silhouette or Davies–Bouldin) across $k$, and quantify cluster contributions to volume/mass/mass flux plus overlap with ROIs and clump masks to support the physical interpretation.