{"id":"dcbae4c6-8e3c-48e2-8f2f-9df9dab71695","arxiv_id":"1908.09235","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Two modified lattice Boltzmann boundary schemes conserve mass in two-phase simulations, preventing artificial evaporation of droplets on curved cylinders.","lead":"This computational physics paper fixes a numerical flaw that made simulated droplets on curved surfaces slowly disappear. It presents two boundary-condition modifications for lattice Boltzmann simulations that keep two-phase droplets from losing mass at curved walls.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The mass-conservation metric may be confounded by the contact-angle implementation's acknowledged unphysical mass-transfer layer, so the modified schemes' conservation is not yet isolated from the wetting model.","rationale":"The paper is a well-executed numerical study with a clear diagnosis, a reasonable correction strategy, and validation against single-phase benchmarks as well as a two-phase mass-conservation test. The strength of the evidence is genuinely moderate: the modified schemes achieve about 0.01% mass variation over 20,000 steps while the original schemes lose tens of percent. The single most load-bearing weakness is the confound between the boundary-scheme mass leakage and the contact-angle implementation. The paper explicitly acknowledges that the constant virtual wall density scheme usually produces an unphysical mass-transfer layer near the solid boundary, and it offers no quantitative argument that this layer does not contribute to the measured mass changes. Since the central claim is about the boundary schemes' intrinsic mass conservation, this missing isolation is important. A targeted test with an alternative, mass-conservative wetting boundary condition would resolve the issue cheaply. Because the paper's evidence is still substantial and the concern is testable rather than a demonstrated failure, the appropriate verdict remains CONDITIONAL, unchanged from the reader's assessment.","tokens_in":16041,"tokens_out":6612,"duration_ms":70420,"concrete_test":"Re-run the droplet-on-cylinder test (Sec. 4.2, θ = 30°, 90°, 120°) with modified schemes A and B, but replace the constant virtual wall density with a mass-conservative curved-boundary contact angle scheme, for example the one from Li et al. (2019), keeping all other parameters identical. If the normalized mass variation remains at or below 0.01% over 20,000 time steps, the wetting confound is ruled out. If mass loss reappears, the apparent mass conservation is not intrinsic to the modified boundary schemes.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim (Sec. 4.2) that modified schemes A and B 'are capable of conserving mass' is supported solely by the total normalized mass in a droplet-on-circular-cylinder test. The contact angle is imposed via a constant virtual wall density (Sec. 3.2), which the paper itself states 'usually leads to an unphysical mass-transfer layer near the solid boundary.' The paper asserts this does not affect the conclusions, but no supporting evidence is given. If the wetting implementation creates or destroys mass near the wall, then the measured change is Δm_total = Δm_boundary + Δm_wetting. Schemes A and B enforce Δm_boundary ≈ 0 by construction (Eqs. 18 and 19–22), but the observed residual of about 0.01% (and possibly part of the original schemes' 40% loss) could be dominated by Δm_wetting. Moreover, Eq. (15) counts only distribution functions crossing the curved wall; the wetting force enters through the forcing term and can alter post-collision values at boundary nodes, so it is not captured by the leakage diagnostic. Without isolating Δm_boundary, the comparison does not establish that the modified schemes are intrinsically mass-conservative curved boundary schemes; they may only compensate the boundary-scheme leakage in combination with this particular wetting model.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper investigates mass leakage of three types of curved boundary schemes in two-phase lattice Boltzmann (LB) simulations by simulating a droplet resting on a circular cylinder. The authors show that the MLS, Bouzidi, and Zhao-Yong schemes all cause the droplet to 'evaporate' under isothermal no-slip conditions, and they attribute the loss to an imbalance between outgoing and incoming distribution functions at boundary nodes. Building on the MLS scheme, they propose two modified schemes: Scheme A adds the computed mass leakage to the rest distribution at each boundary node, and Scheme B redefines the ghost-node density so that the incoming and outgoing mass sums balance exactly. They validate the modified schemes against published results for steady flow past a cylinder and report that in droplet-on-cylinder tests the normalized system mass varies by only about 0.01% over 20,000 time steps, whereas the original schemes lose substantial mass (up to 40% for the Zhao-Yong scheme).","tokens_in":16277,"tokens_out":7332,"duration_ms":65763,"significance":"The paper addresses a real and practically important issue: mass conservation of curved boundary treatments in two-phase LB, which matters for phase-change and wetting simulations. The derivations of the two modifications are explicit and transparent, and the comparison with the scheme of Bao et al. is a useful clarification of the role of the q-dependent parameter chi. The numerical evidence for mass conservation is visually and quantitatively striking. However, the central empirical claim rests on a single test configuration, and the paper acknowledges but does not isolate the unphysical mass-transfer layer of the contact-angle model; the recommendation below therefore requires additional diagnostics to remove the confounding risk.","major_comments":[{"comment":"The contact-angle implementation via a constant virtual wall density is acknowledged in Sec. 3.2 to 'usually lead to an unphysical mass-transfer layer near the solid boundary,' and the paper asserts that this does not affect the conclusions without providing supporting evidence. Because Eq. (15) counts only distribution functions crossing the curved wall, while the wetting force enters through the forcing term and can alter post-collision values at boundary nodes, the residual 0.01% mass variation observed for schemes A and B could in principle be dominated by the wetting model's own mass transfer. The comparison between original and modified schemes therefore does not by itself isolate the boundary-scheme contribution. Please provide an explicit test that isolates the boundary contribution, for example by tracking the cumulative sum of Eq. (15) at boundary nodes for schemes A and B, or by repeating the droplet test with a different (e.g., geometric) contact-angle implementation and showing that the same near-exact mass conservation is obtained.","section":"Sec. 3.2 and Sec. 4.2"},{"comment":"Scheme A as defined by Eq. (18) adds the entire mass leakage to the rest distribution f0, and the authors note that f0 'may become negative in certain cases.' The manuscript does not report whether negative rest distributions occurred in any of the simulations of Sec. 4.2, nor how such values were handled (e.g., clipping, renormalization, or leaving them). If negative populations or clipping are present, the mass-conservation result is no longer a clean demonstration of Eq. (18). Please report the minimum value of f0(x_b, t+delta_t) encountered at boundary nodes in the droplet simulations and state explicitly how any negative values were treated.","section":"Sec. 4.1, Eq. (18)"},{"comment":"The two-phase mass-conservation claim is supported by a single grid resolution (300x350 with R=70, r=50) and three contact angles, and the single-phase validation in Table 1 is at a single resolution. Because the boundary fraction q varies over the circular cylinder, the exactness of schemes A and B in conserving mass is an algebraic property only if the sums in Eqs. (18) and (22) cancel exactly; a numerical demonstration at one resolution does not establish the property as independent of grid and geometry. Please add a resolution study (at least one finer and one coarser grid) or report a per-node mass-balance diagnostic showing that the cumulative sum of Eq. (15) at boundary nodes is zero to machine precision for the modified schemes in the two-phase tests.","section":"Sec. 4.2"}],"minor_comments":[{"comment":"Typos: 'ultilized' should be 'utilized', and later in Sec. 3.2 'doplet' should be 'droplet'.","section":"Sec. 3.2"},{"comment":"The legend in Fig. 5 uses 'I-Bouzidi scheme' while the text and Fig. 4 use 'L-Bouzidi scheme'; these should be consistent.","section":"Fig. 5"},{"comment":"The notation in Eq. (7) introduces a 'post-collision density distribution function at the solid node' f_alpha_eq(rho_s, u_s; x_s); since x_s is a ghost node, the definition should be stated more explicitly to avoid implying that actual post-collision data exist there.","section":"Sec. 2.2, Eq. (7)"},{"comment":"The denominator in Eq. (22) is a sum of chi-weighted equilibrium terms; the paper does not discuss configurations in which this denominator could vanish or become very small, which may affect the robustness of Scheme B.","section":"Sec. 4.1, Eq. (22)"},{"comment":"Reference Li et al. (2019) is cited as 'In press'; if the final published version is now available, the citation should be updated.","section":"References"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a useful, well-executed numerical-methods paper, and the genuinely new item is scheme B. The authors point out that Bao et al.'s mass-conserving curved boundary scheme moves the position-dependent coefficient chi outside the sum when computing the ghost-node density, which breaks the nodal mass balance in two-phase simulations. Keeping chi inside the sum (their Eq. (22)) corrects the balance. That is a small, concrete, and credible correction, and the derivation in Sec. 4.1 is algebraically explicit. Scheme A, adding the leaked mass to the rest distribution, is less novel—the idea originates with Aidun and Lu—and the authors themselves note the rest distribution can go negative.\n\nThe paper also earns credit for the diagnosis. The single-phase static-gravity tests quantify mass loss across q values for three curved-boundary families, and the droplet-on-cylinder tests show the original MLS scheme loses a large fraction of droplet mass while both modified schemes hold total mass variation to about 0.01%. The steady cylinder-flow benchmarks at Re = 20 and 40 provide independent evidence that the corrections do not degrade accuracy.\n\nThe soft spots are modest but real. No code or data are released, there is no grid-convergence study, no error bars, and the two-phase validation is essentially one geometry with three contact angles. Also, mass conservation is enforced by construction at each boundary node, so the conservation test is not a free empirical prediction; what it demonstrates is that the enforcement is compatible with the forcing scheme and wetting model in practice. The stress-test worry about the constant-virtual-wall-density contact angle creating a spurious mass-transfer layer is a fair confound to raise in principle, but it does not sink the comparison: the same wetting model is used for original and modified schemes, and swapping the boundary scheme changes the outcome from 40% mass loss to roughly 0.01%, so the boundary treatment is clearly the dominant factor. The residual could include a wetting-layer contribution, and a sentence acknowledging that would be appropriate, but the central claim holds.\n\nWho this is for: LB practitioners doing two-phase flow, wetting, or boiling on curved surfaces. Citation pattern is fine; self-citations point to the relevant pseudopotential and forcing-scheme literature. I would bring it to a reading group and would cite it if I worked in the area. It deserves a serious referee. My recommendation: send it to review, and ask the authors for code/data and one grid-convergence check before publication.","headline":"A solid, modest methods paper: Scheme B's correction of Bao et al.'s position-dependent coefficient is the real contribution, and the two-phase mass-leakage diagnosis is credible; send it to review, ask for code and a convergence check.","tokens_in":16788,"tokens_out":3615,"would_cite":true,"duration_ms":36439,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["47.11.Qr"],"model":"deepseek-v4-flash","headline":"Two modified curved boundary schemes conserve mass in two-phase lattice Boltzmann simulations.","keywords":["lattice Boltzmann method","curved boundary scheme","mass leakage","mass conservation","two-phase flow","pseudopotential multiphase model","droplet on cylinder","interpolated bounce-back"],"falsifier":"A concrete check is a single-node budget: record the outgoing and incoming distribution sums at every boundary node of the droplet-on-cylinder test and compare their difference with Eq. (15). If the node-level balance is restored in the modified schemes but negative rest densities appear, or if the residual mass drift scales with the virtual-wall density rather than with the link fraction q, then the proposed leakage mechanism and mass-conservation claim are not the whole story.","tokens_in":15839,"feed_emoji":"💧","tokens_out":5916,"duration_ms":52680,"temperature":0.7,"pith_summary":"This paper studies why droplets sitting on a circular cylinder “evaporate” in lattice Boltzmann (LB) simulations even under isothermal, no-slip conditions, and it identifies the cause as mass leakage at the curved boundary rather than physical evaporation. The authors show that three standard curved boundary schemes all lose droplet mass over time, and that the leak emerges from an imbalance between distribution functions streaming out of the system and those streaming back in at boundary nodes. They then build two modified schemes on the MLS curved boundary treatment: scheme A returns the leaked mass through the rest distribution function, and scheme B redefines the fictitious density at ghost nodes so that incoming and outgoing masses balance node by node. In droplet-on-cylinder tests, both modified schemes hold the normalized system mass to within about 0.01% over 20,000 time steps, whereas the original schemes lose tens of percent of droplet mass.","feed_headline":"Droplets on curved walls stop shrinking in two-phase LB","feed_subtitle":"Two corrected schemes keep droplet mass within 0.01 percent over 20,000 steps, making curved surfaces practical in two-phase LB.","key_machinery":"The load-bearing object is the boundary-node mass-leakage identity, Eq. (15): the change in mass at a boundary node is the difference between the mass streamed out by post-collision outgoing distributions and the mass streamed in by incoming distributions. Scheme A cancels that leak by adding the computed deficit to the rest distribution function. Scheme B enforces a zero deficit by choosing a fictitious density in the ghost-node equilibrium distribution, with the crucial detail that χ depends on the link fraction q and therefore cannot be pulled out of the summation. This distinction is what separates scheme B from the earlier Bao et al. correction, which treats χ as a common factor and still leaks mass in two-phase tests.","core_discovery":"Under isothermal conditions, a droplet placed on a circular cylinder with three representative curved boundary schemes—the MLS scheme, the Bouzidi interpolated bounce-back scheme, and the Zhao-Yong single-node scheme—shrinks continuously as if evaporating. The paper attributes this to a node-level mass imbalance: the post-collision outgoing distribution functions carry more mass out of the fluid domain than the incoming distribution functions return. Guided by that mechanism, the authors modify the MLS scheme in two ways. Scheme A adds the computed mass leakage back to the rest distribution function at each boundary node. Scheme B instead adjusts the fictitious density in the ghost-node equilibrium distribution so that the outgoing and incoming sums match exactly, taking care that the interpolation-dependent parameter χ remains inside the summation. Both modified schemes reproduce the flow-past-a-cylinder benchmark accurately and conserve droplet mass in the two-phase test, while the original schemes and the previously proposed Bao et al. correction still leak mass.","pith_inferences":["The same node-level balance condition could be applied to other curved boundary schemes, since any scheme whose unknown distributions depend on the link fraction should carry that fraction inside the mass-balance sum.","A natural testable extension is three dimensions: on D3Q19 or D3Q27 lattices, the scheme-B density correction could be averaged over all boundary links of a node, and one could check whether node-wise conservation still holds or only total mass.","Because scheme A modifies only the rest distribution, it is easier to implement in existing codes, but it may break positivity when the leakage is large; a positivity monitor on f0 would tell how far the fix can be pushed.","The constant virtual-wall density used for contact angles is an acknowledged possible source of spurious mass transfer, so testing the two schemes with an alternative contact-angle implementation would isolate boundary-scheme leakage from wetting artifacts."],"forward_implications":["Two-phase LB simulations with curved solid walls no longer need the stair-stepped halfway bounce-back approximation, so boiling on cylinders or other curved surfaces can be simulated without artificial nucleation sites.","Both modified schemes keep system mass variation to about 0.01% over 20,000 steps for contact angles of 30°, 90°, and 120°, a level small enough to attribute residual fluctuations to diffuse-interface initialization.","Scheme B provides the more principled conservation fix of the two because it enforces node-level balance by construction rather than adding mass back after the fact.","The finding that the earlier Bao et al. correction leaks in two-phase cases follows directly from treating χ as a common factor; moving χ inside the summation is what supplies the missing conservation.","The modified schemes also improve accuracy over the halfway bounce-back scheme in the steady flow-past-a-cylinder benchmark, removing the nonphysical penetration of velocity below the cylinder surface."],"supporting_citations":[{"why":"Supplies the MLS curved boundary scheme that both modified schemes are built on.","marker":"Mei et al., 1999"},{"why":"The earlier mass-conservation correction whose fictitious-density idea scheme B extends and corrects by keeping χ inside the summation.","marker":"Bao et al., 2008"},{"why":"Provides the interpolated bounce-back baseline whose linear and quadratic versions are compared for mass leakage.","marker":"Bouzidi et al., 2001"},{"why":"Provides the single-node scheme baseline that leaks the most mass in the two-phase tests.","marker":"Zhao and Yong, 2017"},{"why":"Introduces the original curved boundary formulation and ghost-node equilibrium idea underlying the MLS family.","marker":"Filippova and Hänel, 1998"},{"why":"Origin of the add-the-leak-back idea used in scheme A.","marker":"Aidun and Lu, 1995"}],"fun_headline_variants":["Two LB fixes stop droplet shrinkage on curved walls","Curved boundaries no longer leak mass in two-phase LB","Droplet on cylinder stays put with corrected LB scheme","Mass-conserving LB schemes keep droplets on curved surfaces"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument rests on the assumption that the mass change at a curved boundary is completely captured by the difference between outgoing and incoming distribution sums, and that the constant virtual-wall density used to set the contact angle does not itself create or destroy mass.","fun_headline_variants_meta":{"raw":{"variants":["Two LB fixes stop droplet shrinkage on curved walls","Curved boundaries no longer leak mass in two-phase LB","Droplet on cylinder stays put with corrected LB scheme","Mass-conserving LB schemes keep droplets on curved surfaces"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00058,"raw_usage":{"total_tokens":2739,"prompt_tokens":957,"completion_tokens":1782,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":573,"completion_tokens_details":{"reasoning_tokens":1717}},"tokens_in":573,"tokens_out":1782,"duration_ms":13527,"temperature":1.0,"reasoning_tokens":1717,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:16:57.603850+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A concrete check is a single-node budget: record the outgoing and incoming distribution sums at every boundary node of the droplet-on-cylinder test and compare their difference with Eq. (15). If the node-level balance is restored in the modified schemes but negative rest densities appear, or if the residual mass drift scales with the virtual-wall density rather than with the link fraction q, then the proposed leakage mechanism and mass-conservation claim are not the whole story.","supporting_citations":[{"cited_title":"An accurate curved boundary treatment in the lattice Boltzmann method","cited_arxiv_id":null,"evidence_quote":"Supplies the MLS curved boundary scheme that both modified schemes are built on."},{"cited_title":"A mass conserving boundary condition for the lattice Boltzmann 24 equation method","cited_arxiv_id":null,"evidence_quote":"The earlier mass-conservation correction whose fictitious-density idea scheme B extends and corrects by keeping χ inside the summation."},{"cited_title":"Momentum transfer of a Boltzmann-lattice fluid with boundaries","cited_arxiv_id":null,"evidence_quote":"Provides the interpolated bounce-back baseline whose linear and quadratic versions are compared for mass leakage."},{"cited_title":"Single-node second-order boundary schemes for the lattice Boltzmann method","cited_arxiv_id":null,"evidence_quote":"Provides the single-node scheme baseline that leaks the most mass in the two-phase tests."},{"cited_title":"Grid Refinement for Lattice-BGK Models","cited_arxiv_id":null,"evidence_quote":"Introduces the original curved boundary formulation and ghost-node equilibrium idea underlying the MLS family."},{"cited_title":"Lattice Boltzmann simulation of solid particles suspended in fluid","cited_arxiv_id":null,"evidence_quote":"Origin of the add-the-leak-back idea used in scheme A."}],"review_version":1}