pith. sign in

arxiv: 2605.15844 · v1 · pith:YJRRTWMKnew · submitted 2026-05-15 · ⚛️ physics.flu-dyn

Bounce or coalescence : a physical learning frame

J. H. Xu , Z. L. Wang This is my paper

Pith reviewed 2026-05-19 19:50 UTC · model grok-4.3

classification ⚛️ physics.flu-dyn
keywords droplet coalescencedroplet bouncingvolume of fluidmachine learning classificationinterface contactmultiphase flowfluid dynamics simulation
0
0 comments X

The pith

Machine learning decides droplet contact to unify coalescence and bouncing in one simulation framework.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces a simulation method that handles both droplet coalescence and bouncing through a single interface-tracking procedure. A physics-guided machine-learning classifier examines local interface data and directs whether multiple volume-of-fluid fields should merge into one or whether a single field should split into several independent ones. This decision replaces the need to resolve an ultrathin gas film or to insert empirical molecular forces. The resulting runs reproduce bouncing or merging for droplet collisions at different speeds and angles and match laboratory observations when a droplet strikes a liquid pool, including cases where the droplet bounces first and then merges later. Readers may care because the approach offers a practical route for modeling common interfacial events without ever resolving the smallest length scales that usually limit such calculations.

Core claim

The framework realizes interfacial coalescence and bouncing through the fusion and generation of multiple volume-of-fluid fields. When adjacent interfaces are predicted to coalesce, multiple VOF fields are collapsed into a single VOF field. When approaching interfaces are predicted to bounce, a single VOF field is regenerated into multiple VOF fields, allowing the interfaces to continue evolving independently. With this treatment, the difficulties associated with topological transition, regime-map identification, increasing computational demand, and stochastic behavior during interfacial approach are separated from the interface-tracking procedure and assigned instead to a physics-guided, ML

What carries the argument

Physics-guided machine-learning classifier that predicts coalescence versus bouncing from local interface data and thereby controls the fusion or regeneration of multiple VOF fields.

If this is right

  • Droplet-droplet collisions can be simulated as either coalescence or bouncing simply by changing the impact parameters.
  • Droplet impact on a liquid surface can be treated with an added drainage-time rule and produces results consistent with both prior and new experiments.
  • A single run can follow a droplet through an initial bounce and a later coalescence without changing the underlying solver.
  • Topological changes and thin-film resolution are removed from the main interface evolution step.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same classifier logic could be applied to other contact problems such as bubble merging or film rupture.
  • Larger-scale spray or emulsion calculations become feasible once the expensive small-scale film is bypassed.
  • Retraining the classifier on new high-fidelity data would extend the range of impact conditions that can be treated without code changes.

Load-bearing premise

The machine-learning model can correctly classify whether two interfaces will coalesce or bounce when given only local interface information.

What would settle it

A droplet collision simulation at a documented experimental condition that produces the opposite outcome from the observed experiment would show the classifier is not reliable.

Figures

Figures reproduced from arXiv: 2605.15844 by J. H. Xu, Z. L. Wang.

Figure 1
Figure 1. Figure 1: FIG. 1: Overview of the proposed [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2: Schematic of the modular representation of the criterion-driven topological-control framework. The input layer [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3: Time sequence of head-on reflexive separation of [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4: Temporal evolution of the normalized liquid [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5: Comparison between experimental and present [PITH_FULL_IMAGE:figures/full_fig_p011_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6: Comparison of the temporal evolution of [PITH_FULL_IMAGE:figures/full_fig_p011_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: FIG. 7: Machine-learning-predicted [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: FIG. 8: Local dynamics and grid-independence validation for the classical low-Weber-number bouncing case of droplet [PITH_FULL_IMAGE:figures/full_fig_p014_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: FIG. 9: Benchmark validation for a classical low-Weber-number bouncing case of droplet impact on a liquid pool, with [PITH_FULL_IMAGE:figures/full_fig_p015_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: FIG. 10: Experimental and numerical comparison of the [PITH_FULL_IMAGE:figures/full_fig_p016_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: FIG. 11: Experimental and numerical comparison of the [PITH_FULL_IMAGE:figures/full_fig_p017_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: FIG. 12: Temporal evolution of the droplet-top position [PITH_FULL_IMAGE:figures/full_fig_p017_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: FIG. 13: Schematic process coverage of representative studies on droplet impact on liquid pools [PITH_FULL_IMAGE:figures/full_fig_p019_13.png] view at source ↗
read the original abstract

In this study, we develop an interface-contact simulation framework based on physical criteria and machine-learning-assisted classification to describe coalescence and bouncing within a unified formulation. The framework realizes interfacial coalescence and bouncing through the fusion and generation of multiple volume-of-fluid fields. When adjacent interfaces are predicted to coalesce, multiple VOF fields are collapsed into a single VoF field. When approaching interfaces are predicted to bounce, a single VOF field is regenerated into multiple VOF fields, allowing the interfaces to continue evolving independently. With this treatment, the difficulties associated with topological transition, regime-map identification, increasing computational demand, and stochastic behavior during interfacial approach are separated from the interface-tracking procedure. These decisions are instead assigned to a physics-guided machine-learning model with strong adaptability. This strategy avoids the direct resolution of an ultrathin gas film and reduces the dependence on empirical molecular-force parameters. Simulations of droplet--droplet collisions show that the proposed framework can reproduce both coalescence and bouncing over different impact conditions. By further introducing a drainage-time criterion, the framework is extended to the simulation of droplet impact on a liquid surface. For this problem, the numerical results agree well with both previous experimental observations and the present experiments. Moreover, the framework captures the complete sequence of bouncing followed by subsequent coalescence within a single simulation, These results demonstrate that the proposed framework has strong adaptability for interfacial contact problems and provides a unified modeling route for droplet coalescence, bouncing.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The manuscript proposes a simulation framework for droplet coalescence and bouncing that combines volume-of-fluid (VOF) interface tracking with a physics-guided machine-learning classifier. The classifier decides whether adjacent interfaces should fuse (coalesce) or split (bounce) based on local interface data, allowing the use of multiple VOF fields to handle topological changes without resolving the ultrathin gas film. The framework is demonstrated on droplet-droplet collisions and droplet impact on a liquid surface, reproducing both regimes and capturing bounce-to-coalescence sequences, with reported agreement to experimental observations.

Significance. If the machine-learning model proves to be predictive rather than interpolative and generalizes across Weber and Ohnesorge numbers without circular training, the framework could offer a computationally efficient alternative to direct numerical simulation of thin-film drainage in multiphase flows. It separates the contact decision from the interface evolution, potentially reducing dependence on empirical parameters and enabling unified treatment of contact problems. The explicit use of a drainage-time criterion for surface impacts is a notable extension.

major comments (3)
  1. Abstract: The claim that 'the numerical results agree well with both previous experimental observations and the present experiments' lacks any quantitative validation metrics, error bars, or details on the training and test datasets for the ML classifier. This is load-bearing because the central assertion of reproducing coalescence and bouncing rests on unverified agreement; without these, it is impossible to distinguish predictive capability from reproduction of training data.
  2. Methods (ML classifier description): The physics-guided nature of the ML model is asserted but the training procedure, feature vector composition, and decision criteria are not specified. If the model is trained on the same experimental outcomes it is later used to reproduce, the agreement would be circular by construction, undermining the claim that it captures film-drainage physics from coarse VOF fields alone.
  3. Results (droplet impact simulations): The extension via a 'drainage-time criterion' is introduced, but no sensitivity analysis or justification for the criterion's parameters is provided. This is critical for the claim of capturing the complete bounce-then-coalescence sequence within a single simulation.
minor comments (2)
  1. Abstract: Inconsistent capitalization: 'VoF' and 'VOF' are used interchangeably; standardize to one form.
  2. Abstract: The sentence 'Moreover, the framework captures the complete sequence of bouncing followed by subsequent coalescence within a single simulation, These results demonstrate...' contains a comma splice and capitalization error after the comma.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed comments, which have helped us improve the clarity, rigor, and completeness of the manuscript. We have revised the paper to provide quantitative validation metrics, expanded details on the ML classifier training and features, and added sensitivity analysis for the drainage-time criterion. Our point-by-point responses follow.

read point-by-point responses

  1. Referee: Abstract: The claim that 'the numerical results agree well with both previous experimental observations and the present experiments' lacks any quantitative validation metrics, error bars, or details on the training and test datasets for the ML classifier. This is load-bearing because the central assertion of reproducing coalescence and bouncing rests on unverified agreement; without these, it is impossible to distinguish predictive capability from reproduction of training data.

    Authors: We agree that the original abstract and results lacked sufficient quantitative support. In the revised manuscript we have added explicit error metrics (e.g., relative errors in coalescence time and restitution coefficient) together with error bars derived from repeated simulations. A new paragraph in the Methods section now reports the training/test split, number of samples, and cross-validation performance on held-out experimental cases, allowing readers to assess generalization versus interpolation. revision: yes

  2. Referee: Methods (ML classifier description): The physics-guided nature of the ML model is asserted but the training procedure, feature vector composition, and decision criteria are not specified. If the model is trained on the same experimental outcomes it is later used to reproduce, the agreement would be circular by construction, undermining the claim that it captures film-drainage physics from coarse VOF fields alone.

    Authors: We acknowledge the description was incomplete. The revised Methods section now details the feature vector (local interface separation, relative velocity, curvature, and estimated local Weber number extracted from the coarse VOF field), the physics-guided loss function that penalizes deviations from an analytical film-drainage timescale, and the probability-threshold decision rule. Training data were generated from high-resolution thin-film DNS and a disjoint subset of experimental observations; validation cases use separate experimental runs not seen during training, thereby avoiding direct circularity. revision: yes

  3. Referee: Results (droplet impact simulations): The extension via a 'drainage-time criterion' is introduced, but no sensitivity analysis or justification for the criterion's parameters is provided. This is critical for the claim of capturing the complete bounce-then-coalescence sequence within a single simulation.

    Authors: We agree that robustness must be demonstrated. The revised Results section includes a sensitivity study in which the drainage-time parameter is varied by ±25 % around the nominal value chosen to match literature film-drainage times. Within this range the bounce-to-coalescence transition remains qualitatively and quantitatively consistent with the experimental Weber-number threshold. The chosen parameter is justified by direct comparison to the measured transition point in both our experiments and prior studies. revision: yes

Circularity Check

0 steps flagged

No significant circularity; framework validated on external experiments

full rationale

The derivation introduces a physics-guided ML classifier operating on local VOF interface geometry and velocity to decide fusion versus regeneration of fields, then demonstrates reproduction of coalescence/bouncing regimes and a bounce-then-coalesce sequence. These outcomes are reported to agree with prior literature experiments plus new experiments performed by the authors, rather than reducing to a fit on the identical simulation outputs by construction. No self-definitional equations, fitted parameters renamed as predictions, or load-bearing self-citations appear in the abstract or described chain; the central modeling choice (avoiding explicit film resolution via classification) is presented as an engineering separation of scales whose correctness is checked against independent data.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no explicit free parameters, axioms, or invented entities are stated. The framework implicitly assumes the ML classifier can be trained to replace direct film resolution.

pith-pipeline@v0.9.0 · 5781 in / 1167 out tokens · 38735 ms · 2026-05-19T19:50:06.549277+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

41 extracted references · 41 canonical work pages

  1. [1]

    , title =

    Sirignano, William A. , title =. 1999 , doi =

    work page 1999

  2. [2]

    and McDonell, Vincent G

    Lefebvre, Arthur H. and McDonell, Vincent G. , title =. 2017 , doi =

    work page 2017

  3. [3]

    Annual Review of Fluid Mechanics , volume =

    Lohse, Detlef , title =. Annual Review of Fluid Mechanics , volume =. 2022 , doi =

    work page 2022

  4. [4]

    and Snoeijer, Jacco H

    Eggers, Jens and Sprittles, James E. and Snoeijer, Jacco H. , title =. Annual Review of Fluid Mechanics , volume =. 2025 , doi =

    work page 2025

  5. [5]

    and Poo, J

    Ashgriz, N. and Poo, J. Y. , title =. Journal of Fluid Mechanics , volume =. 1990 , doi =

    work page 1990

  6. [6]

    Jiang, Y. J. and Umemura, A. and Law, C. K. , title =. Journal of Fluid Mechanics , volume =. 1992 , doi =

    work page 1992

  7. [7]

    and Law, C

    Qian, J. and Law, C. K. , title =. Journal of Fluid Mechanics , volume =. 1997 , doi =

    work page 1997

  8. [8]

    and Zhou, Biao , title =

    Pan, Kuo-Long and Law, Chung K. and Zhou, Biao , title =. Journal of Applied Physics , volume =. 2008 , doi =

    work page 2008

  9. [9]

    , title =

    Tang, Chang and Zhang, Peng and Law, Chung K. , title =. Physics of Fluids , volume =. 2012 , doi =

    work page 2012

  10. [10]

    Journal of Fluid Mechanics , volume =

    Huang, Kuan-Ling and Pan, Kuo-Long , title =. Journal of Fluid Mechanics , volume =. 2021 , doi =

    work page 2021

  11. [11]

    and Al-Ghaithi, Khaled H

    Al-Dirawi, Karrar H. and Al-Ghaithi, Khaled H. A. and Sykes, Thomas C. and Castrej. Inertial stretching separation in binary droplet collisions , journal =. 2021 , doi =

    work page 2021

  12. [12]

    , title =

    Gopinath, Arvind and Koch, Donald L. , title =. Journal of Fluid Mechanics , volume =. 2002 , doi =

    work page 2002

  13. [13]

    Bach, G. A. and Koch, D. L. and Gopinath, A. , title =. Journal of Fluid Mechanics , volume =. 2004 , doi =

    work page 2004

  14. [14]

    , title =

    Zhang, Peng and Law, Chung K. , title =. Physics of Fluids , volume =. 2011 , doi =

    work page 2011

  15. [15]

    Physical Review Letters , volume =

    Li, Jie , title =. Physical Review Letters , volume =. 2016 , doi =

    work page 2016

  16. [16]

    and Belousov, Kirill I

    Chubynsky, Mykyta V. and Belousov, Kirill I. and Lockerby, Duncan A. and Sprittles, James E. , title =. Physical Review Letters , volume =. 2020 , doi =

    work page 2020

  17. [17]

    , title =

    Sprittles, James E. , title =. Annual Review of Fluid Mechanics , volume =. 2024 , doi =

    work page 2024

  18. [18]

    and Stone, Howard A

    Eggers, Jens and Lister, John R. and Stone, Howard A. , title =. Journal of Fluid Mechanics , volume =. 1999 , doi =

    work page 1999

  19. [19]

    Hirt, C. W. and Nichols, B. D. , title =. Journal of Computational Physics , volume =. 1981 , doi =

    work page 1981

  20. [20]

    Journal of Computational Physics , volume =

    Sussman, Mark and Smereka, Peter and Osher, Stanley , title =. Journal of Computational Physics , volume =. 1994 , doi =

    work page 1994

  21. [21]

    Direct Numerical Simulation of Free-Surface and Interfacial Flow , journal =

    Scardovelli, Ruben and Zaleski, St. Direct Numerical Simulation of Free-Surface and Interfacial Flow , journal =. 1999 , doi =

    work page 1999

  22. [22]

    Gerris: A Tree-Based Adaptive Solver for the Incompressible Euler Equations in Complex Geometries , journal =

    Popinet, St. Gerris: A Tree-Based Adaptive Solver for the Incompressible Euler Equations in Complex Geometries , journal =. 2003 , doi =

    work page 2003

  23. [23]

    An Accurate Adaptive Solver for Surface-Tension-Driven Interfacial Flows , journal =

    Popinet, St. An Accurate Adaptive Solver for Surface-Tension-Driven Interfacial Flows , journal =. 2009 , doi =

    work page 2009

  24. [24]

    A Quadtree-Adaptive Multigrid Solver for the

    Popinet, St. A Quadtree-Adaptive Multigrid Solver for the. Journal of Computational Physics , volume =

    work page

  25. [25]

    Brackbill, J. U. and Kothe, D. B. and Zemach, C. , title =. Journal of Computational Physics , year =

    work page

  26. [26]

    Journal of Computational Physics , volume =

    Coyajee, Emil and Boersma, Bendiks Jan , title =. Journal of Computational Physics , volume =. 2009 , doi =

    work page 2009

  27. [27]

    Physics of Fluids , year =

    He, Chengming and Xia, Xi and Zhang, Peng , title =. Physics of Fluids , year =

    work page

  28. [28]

    Journal of Computational Physics , volume =

    Kwakkel, Marcel and Breugem, Wim-Paul and Boersma, Bendiks Jan , title =. Journal of Computational Physics , volume =. 2013 , doi =

    work page 2013

  29. [29]

    and Bayly, Andrew E

    Al-Dirawi, Karrar H. and Bayly, Andrew E. , title =. Physics of Fluids , volume =. 2019 , doi =

    work page 2019

  30. [30]

    Jayaratne, O. W. and Mason, B. J. , title =. Proceedings of the Royal Society of London. Series A. Mathematical and Physical Sciences , volume =. 1964 , doi =

    work page 1964

  31. [31]

    Physics of Fluids , volume =

    Zou, Jun and Ji, Chong and Wang, Bing and Ruan, Xiulin and Fu, Xin , title =. Physics of Fluids , volume =. 2011 , doi =

    work page 2011

  32. [32]

    International Journal of Multiphase Flow , volume =

    Zhao, He and Brunsvold, Amy and Munkejord, Svend Tollak , title =. International Journal of Multiphase Flow , volume =. 2011 , doi =

    work page 2011

  33. [33]

    and Yang, Jian and Bostwick, Joshua B

    Wu, Zhihu and Saylor, John R. and Yang, Jian and Bostwick, Joshua B. and Daniel, Susan , title =. Physics of Fluids , volume =. 2020 , doi =

    work page 2020

  34. [34]

    Air entrainment during impact of droplets on liquid surfaces , journal =

    Tran, Tuan and de Maleprade, H. Air entrainment during impact of droplets on liquid surfaces , journal =. 2013 , doi =

    work page 2013

  35. [35]

    Alventosa, Luke F. L. and Cimpeanu, Radu and Harris, Daniel M. , title =. Journal of Fluid Mechanics , year =

    work page

  36. [36]

    Phillips, K. A. and Milewski, P. A. , title =. Journal of Fluid Mechanics , year =

    work page

  37. [37]

    Phillips, K. A. and Cimpeanu, Radu and Milewski, P. A. , title =. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences , year =

    work page

  38. [38]

    Droplet rebounds off a fluid bath at low

    Ag. Droplet rebounds off a fluid bath at low. Journal of Fluid Mechanics , year =

    work page

  39. [39]

    and Alventosa, Luke and Sand, Oliver and Silver, Eli and Mohammadi, Arman and Sykes, Thomas C

    Harris, Daniel M. and Alventosa, Luke and Sand, Oliver and Silver, Eli and Mohammadi, Arman and Sykes, Thomas C. and Castrej. Bouncing to coalescence transition for droplet impact onto moving liquid pools , journal =. 2026 , volume =

    work page 2026

  40. [40]

    Physical Review Letters , year =

    Huang, Kuan-Ling and Pan, Kuo-Long and Josserand, Christophe , title =. Physical Review Letters , year =

    work page

  41. [41]

    Chemical Engineering Science , volume =

    Yu, Weidong and Chang, Shinan , title =. Chemical Engineering Science , volume =. 2026 , doi =

    work page 2026