Pith. sign in

REVIEW 2 major objections 7 minor 24 references

Local particle refinement in terramechanical simulations

T0 review · 2 major / 7 minor · reviewed 2026-08-10 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper claims that soil beds with fine particles at the surface and coarser particles below reproduce a uniform fine-particle reference within 3.4%–11% aggregate error, cutting particle count 2.3–25x and simulation time 3.1–43x.

desk verdict Useful, systematic study of local particle refinement for DEM terramechanics; the cost–accuracy trend holds, but the headline error numbers rest on an ambiguous aggregate metric. read the letter →

arxiv 2501.05300 v1 pith:L5PPLCWD submitted 2025-01-09 cs.CE physics.comp-ph

Markus Pogulis , Martin Servin This is my paper · ORCID
classification cs.CEphysics.comp-ph
keywords discreteelementmethodgranularmaterialsparticlerefinementsizescalingterramechanicspressure-sinkageshear-displacementcomputationalefficiency
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper attempts to make discrete element method (DEM) simulations of vehicle–terrain interaction much cheaper by grading particle size with depth: fine particles at the soil surface, where wheels or tracks engage, and progressively larger particles below. The authors claim that this local refinement preserves the soil's bulk mechanical response—triaxial tests on mildly graded beds keep the internal friction angle within 3% of uniform beds—and that in pressure-sinkage and shear-displacement tests, refined beds deviate from a uniform 8.5 mm reference bed by aggregate normalized errors of only 3.4% to 11%. These same beds contain 2.3 to 25 times fewer particles and run 3.1 to 43 times faster, with better-than-linear speedup because the shorter contact network requires fewer solver iterations. If the claim holds, it gives terramechanics a practical route to full-vehicle simulations that previously required impractically long run times.

What carries the argument

The load-bearing object is the particle refinement profile, d(z) = dmin + γz for 0 < z < zmax and d = dmax below, realized as discrete layers of thickness ηdn with diameters dn = r dn−1, so the scaling aggressiveness γ = (r − 1)/(rη) is the single dimensionless control parameter. This parameter determines the particle count, the vertical contact-network length nd = η log(dmax/dmin)/log r + (h − zmax)/dmax, and hence the number of projected Gauss–Seidel iterations Nit = 0.1 nd/ε per timestep. Because nd shrinks faster than the particle count, the method achieves better-than-linear speedup; the timestep stays tied to the smallest (surface) particles, preserving resolution exactly where grouser–particle interaction is strongest.

What would settle it

Run DEM triaxial compression tests on beds with scaling aggressiveness γ ≈ 0.5 to 1.0 and compare the internal friction angle to uniform beds; a deviation beyond the 3% tolerance seen for mild gradients would mean the plate-test errors at high γ mix a change in soil constitutive behavior with discretization error, invalidating the accuracy claim for aggressive refinement.

Watch

Extended reading notes

Core claim

The central claim is that a linearly graded particle size profile, d(z) = dmin + γz for 0 < z < zmax and d = dmax below, can substitute for a uniformly fine bed in terramechanical simulations. The paper evaluates 36 refined beds with scaling aggressiveness γ from 0.012 to 0.96, plus uniform control beds of 8.5, 15, 22.5, and 30 mm particles, each repeated five times. Against the uniform 8.5 mm bed as the high-resolution reference, refined beds reproduce static sinkage, dynamic sinkage, peak traction, and average traction with a combined normalized error of 3.4% at low γ, 6.2% at medium γ, and 11.3% at high γ, while uniform coarse beds show errors of 31% to 39%. The paper interprets the gradual size transition as mechanically meaningful: small surface particles fill grouser cavities and launch force chains through the size gradient, mobilizing a deeper soil mass and producing a characteristic double-peak traction response.

Load-bearing premise

The paper verifies that bulk soil strength and stiffness survive particle refinement only for mild size gradients, and assumes the same invariance holds at the aggressive gradients (up to γ = 0.96) that produce the largest speedups.

Editorial extensions

If this is right

  • Full-vehicle DEM runs over extended terrain become practical: choosing γ in the low-to-medium range keeps aggregate errors at a few percent while cutting run times by roughly an order of magnitude.
  • The layer construction transfers directly to other refinement geometries, such as radial zones around a cone penetrometer or hemispherical zones around a load point, carrying the same accuracy–cost trade-off.
  • Because the coarse lower region preserves stress distribution without adding cost, deep soil beds can be simulated at full height, avoiding the boundary artifacts the paper documents for truncated beds.
  • The reported error bands give practitioners a quantitative basis for choosing γ: if a 3–6% error is acceptable, a speedup of 3–8x is available; accepting ~11% error unlocks up to 43x.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The bulk-property invariance is verified only for mild gradients (γ up to 0.072); if aggressive refinement changes the constitutive response, part of the 11% error at high γ would reflect altered soil strength rather than discretization error—a triaxial test at high γ would separate the two.
  • The linear-with-depth profile assumes the deformation is surface-dominated; for excavation or deep loading, an adaptive or radial profile would likely need far fewer particles for the same accuracy, but would require new invariance and percolation checks.
  • The double-peak traction response hints that graded interfaces engage more soil mass than sharp layer transitions, which could inform grouser and tread design; the paper observes but does not exploit this.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 7 minor

Summary. The paper proposes and systematically evaluates a local particle refinement strategy for DEM terramechanics simulations: fine particles are kept at the soil surface where they interact with grousers, while particle size increases linearly with depth. The authors first run triaxial tests on uniform and mildly refined beds to show that the internal friction angle is approximately invariant under refinement. They then simulate combined pressure-sinkage and shear-displacement tests for 36 refined beds with scaling aggressiveness γ from 0.012 to 0.96, comparing against uniform control beds, with the uniform 8.5 mm bed as the high-resolution reference. They report particle-count reductions of 2.3–25×, simulation-time reductions of 3.1–43×, and normalized aggregate errors of 3.4–11% relative to the reference. The paper concludes that local particle refinement offers substantial computational savings with controlled accuracy loss, and it discusses limitations such as particle percolation and the need for application-specific refinement profiles.

Significance. If substantiated, the result is practically valuable: it offers a concrete route to much cheaper DEM simulations for vehicle-terrain interaction while keeping resolution where it matters. The paper has notable strengths: a systematic sweep over 36 configurations, five independent replicates per configuration including bed generation, comparison against several uniform-size control beds, and openly available simulation scripts and data. The claim is empirical and falsifiable, and the authors are candid about method limitations. The main caveats are that the printed aggregate-error formula is incomplete and that the invariance validation covers only mild refinement gradients, so the headline error range needs either a definitional fix or a restriction in scope.

major comments (2)
  1. [Sec. 4.3] The definitions ε_sinkage = (z − z_ref)/h_grouser and ε_traction = (FT − FT,ref)/FN,max, combined with the statement that the aggregate error is the arithmetic mean of the four normalized errors, do not specify absolute values. As printed, the signed arithmetic mean is not what the paper reports: using the component deviations in Secs. 4.1–4.2, the signed aggregate for the low, medium, and high γ groups would be approximately 2.2%, 3.6%, and 7.6%, whereas the reported values are 3.4%, 6.2%, and 11.3%. The reported numbers are consistent only if absolute values are taken before averaging. Because the abstract and conclusion quote the 3.4–11% range directly, the metric definition must be corrected, either by writing ε = |Δ|/norm or by stating the signed convention explicitly and reporting the mean absolute value.
  2. [Sec. 3.2 and Sec. 4] The triaxial validation of bulk-property invariance under local particle refinement covers only γ = 0.050 and 0.072 (Table 1), while the pressure-sinkage and shear-displacement tests use γ up to 0.96, and the headline error range includes the high-γ group with 11.3% aggregate error. The paper's abstract states that triaxial tests verify that bulk mechanical properties are preserved under local particle refinement, which is a broader claim than the data support. If aggressive refinement changes the constitutive response — an outcome the authors themselves associate with size mixing and percolation in Secs. 3.1 and 5 — then the high-γ plate-test errors conflate physical property changes with discretization error. Please either add triaxial or comparable bulk-property checks at representative high-γ values, or explicitly limit the invariance and accuracy claims to low/medium γ and re-derive the error range accordingly.
minor comments (7)
  1. [Sec. 3.1] The text says that cases with γ ∼ 1 are excluded from the simulations because of notable size mixing between layers, but Table 2 lists configurations with γ up to 0.96 and Sec. 5 reports percolation at large scaling; please clarify which configurations were actually excluded and how the exclusion criterion was applied.
  2. [Sec. 4.3] The traction normalization uses FN,max = 2550 N while the reported quantity is the ratio FT/FN. Because FN equals FN,max in these tests, ε_traction numerically equals the change in FT/FN, but this equivalence should be stated explicitly to avoid unit confusion.
  3. [Sec. 2.2] The symbol ε is used both for constraint compliance in Eq. (3) and for the error tolerance in Eqs. (9) and (10); please disambiguate, for example by using ε_tol for the solver tolerance.
  4. [Sec. 2.2] There is a typo in the sentence on friction direction: 'maximumum dissipation' should read 'maximum dissipation'.
  5. [Sec. 4.1] The relative static-sinkage deviations for uniform beds (up to 349% for d = 30 mm) are large only because the reference static sinkage is 1.26 mm; the absolute deviations are already given and should be emphasized in the text, since the relative percentages alone can mislead readers about the mechanical significance of the differences.
  6. [Sec. 4.4] The reported speed-up is acknowledged to be implementation-dependent; reporting a particle-count-normalized speed-up in addition to wall-clock speed-up would make the computational claim more transferable across DEM codes.
  7. [Sec. 3.3] The 8.5 mm uniform bed is used as the high-resolution reference, but no convergence check against a finer uniform bed is reported; a single finer reference simulation would strengthen the interpretation of the errors as discretization errors relative to a converged solution.

Circularity Check

0 steps flagged · score 1.0 of 10

No circularity: the refinement profiles are prescribed inputs, all error measures compare against an independent uniform-resolution benchmark, and the self-citations are tooling and parameter provenance rather than load-bearing inference.

full rationale

The derivation chain is self-contained and comparative. Each refined bed is constructed from the prescribed profile d(z)=dmin+γz (Eq. 1) and the layer rule dn=r d_{n-1}, so γ is an input, not a parameter fitted to the sinkage or traction outputs. Accuracy is judged by normalized differences between refined beds and a uniform 8.5 mm bed (Sec. 4); no error term enters the construction of the beds or the solver settings. The triaxial tests (Sec. 3.2, Table 1) independently probe whether the friction angle changes with γ; they are limited to mild gradients, so extending invariance to γ≈1 is an extrapolation or assumption, but not a circular one. The cited prior work [16,17,18] supplies the nonsmooth DEM formulation and AGX solver, and [23] supplies material parameters; these self-citations are transparent tooling and calibration provenance. The target claim, that refinement preserves accuracy while reducing cost, is not an input to those citations, and the cited works are themselves falsifiable elsewhere. The aggregate-error definition in Sec. 4.3 does not state absolute values before averaging and may allow signed cancellation; that is a quantification or correctness risk, not circularity. Accordingly, no step reduces by construction to its own input.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

No new physical entities are introduced; all costs and benefits derive from the chosen particle size distribution, which is controlled by user-selected parameters, plus solver settings. The central claim thus rests on the representational adequacy of DEM and on the invariance of bulk properties over the refinement range.

free parameters (2)
  • Refinement profile parameters [dmin, gamma, dmax] and layer ratios [r, eta] = dmin=8.5 mm (fixed); gamma=0.012 to 0.96; dmax=15, 22.5, or 30 mm; r, eta not explicitly reported
    Chosen by hand to define the particle size gradient; they are the independent variables of the parametric study rather than fitted to a target result. The performance map (error vs. gamma) is the paper's main output.
  • PGS solver iterations Nit and timestep dt = Nit=215-275 for refined beds, 125-375 for uniform beds; dt=0.10-0.35 ms
    Set per bed using the guiding rules in Eqs. (9)-(11), but the error tolerance epsilon in those rules is not specified, so the values are effectively tuned to achieve convergence. The reported speed-ups depend on these choices, and better-than-linear scaling is tied to Nit being reduced.
assumptions (4)
  • domain assumption The nonsmooth DEM contact model with Hertz compliance, Coulomb friction, and rolling resistance adequately represents dry sand for the tests considered.
    Stated in Sec. 2.2 and Sec. 3.1; material parameters taken from Ref. [23]. The paper does not validate against physical experiments.
  • domain assumption Bulk mechanical properties, particularly the internal friction angle, are invariant under local particle refinement for the ranges tested.
    Tested only for gamma=0.050 and 0.072 in Sec. 3.2 (Table 1), with a 3% tolerance; extended by assumption to gamma up to 0.96 in the plate tests.
  • domain assumption The uniform bed with d=8.5 mm particles is a valid high-resolution reference for measuring refinement error.
    Used throughout Sec. 4 as the reference; no experimental ground truth is available, so 'accuracy' is relative to this simulation.
  • domain assumption A linear depth profile d(z)=dmin+gamma*z and discrete layers of thickness eta*dn sufficiently capture the refinement behavior.
    Introduced in Sec. 2.1 as a modeling choice for simplicity; the method's conclusions may not extend to radial or adaptive refinement profiles.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Local particle refinement in terramechanical simulations." pith.science (2026). https://pith.science/paper/L5PPLCWD

@misc{pith2026250105300,
  author       = {Pith},
  title        = {Pith review of: Local particle refinement in terramechanical simulations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/L5PPLCWD}},
  note         = {Machine review of arXiv:2501.05300}
}
abstract

The discrete element method (DEM) is a powerful tool for simulating granular soils, but its high computational demand often results in extended simulation times. While the effect of particle size has been extensively studied, the potential benefits of spatially scaling particle sizes are less explored. We systematically investigate a local particle refinement method's impact on reducing computational effort while maintaining accuracy. We first conduct triaxial tests to verify that bulk mechanical properties are preserved under local particle refinement. Then, we perform pressure-sinkage and shear-displacement tests, comparing our method to control simulations with homogeneous particle size. We evaluate $36$ different DEM beds with varying aggressiveness in particle refinement. Our results show that this approach, depending on refinement aggressiveness, can significantly reduce particle count by $2.3$ to $25$ times and simulation times by $3.1$ to $43$ times, with normalized errors ranging from $3.4$\% to $11$\% compared to high-resolution reference simulations. The approach maintains a high resolution at the soil surface, where interaction is high, while allowing larger particles below the surface. The results demonstrate that substantial computational savings can be achieved without significantly compromising simulation accuracy. This method can enhance the efficiency of DEM simulations in terramechanics applications.

Figures

Figures reproduced from arXiv: 2501.05300 by the authors.

Figure 1
Figure 1. Illustration of the idea of local particle refinement for accelerating vehicle-terrain simulations. Fine particles are used at the terrain [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Illustration of a soil bed with vertical particle refinement [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. The container and the plate with their dimensions. Parti [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (7 more)
Figure 5
Figure 5. Figure 5: Coefficient of resistance, FT /FN , with the identified region marked. In the three later zones, a chosen peak is marked in each region. The red curve is the reference case for uniform bed d = 8.5 mm. 4. Results The sinkage and traction behavior of 36 different par￾tic…
Figure 4
Figure 4. Figure 4: Track sinkage in simulations using particle beds with dif [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 6
Figure 6. Figure 6: Sample images of particle velocities at fixed times during Phase II and Phase III. Particle colors indicate velocity magnitude [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Sinkage in both phase I and phase II-III depending on [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 8
Figure 8. Figure 8: Initial peak traction in phase II depending on particle size [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 10
Figure 10. Figure 10: Aggregated error depending on particle size and scaling. [PITH_FULL_IMAGE:figures/full_fig_p008_10.png]
Figure 11
Figure 11. Figure 11: Effect of particle size and scaling aggressiveness on total [PITH_FULL_IMAGE:figures/full_fig_p009_11.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

24 extracted references · 23 canonical work pages

  1. [1]

    P. A. Cundall, O. D. Strack, A discrete numerical model for granular assemblies, geotechnique 29 (1979) 47–65

    1979

  2. [2]

    Miyai, M

    S. Miyai, M. Kobayakawa, T. Tsuji, T. Tanaka, Influence of particle size on vertical plate penetration into dense cohesion- less granular materials (large-scale dem simulation using real particle size), Granular Matter 21 (2019) 1–21

    work page 2019

  3. [3]

    Coetzee, Calibration of the discrete element method: Strate- gies for spherical and non-spherical particles, Powder Technol- ogy 364 (2020) 851–878

    C. Coetzee, Calibration of the discrete element method: Strate- gies for spherical and non-spherical particles, Powder Technol- ogy 364 (2020) 851–878

    work page 2020

  4. [4]

    Coetzee, Particle upscaling: Calibration and validation of the discrete element method, Powder technology 344 (2019) 487–503

    C. Coetzee, Particle upscaling: Calibration and validation of the discrete element method, Powder technology 344 (2019) 487–503

    work page 2019

  5. [5]

    Y. Feng, K. Han, D. Owen, J. Loughran, On upscaling of dis- crete element models: similarity principles, Engineering Com- putations 26 (2009) 599–609

    work page 2009

  6. [6]

    Y. Feng, D. Owen, Discrete element modelling of large scale particle systems—i: exact scaling laws, Computational Particle Mechanics 1 (2014) 159–168

    work page 2014

  7. [7]

    W. Ge, L. Wang, J. Xu, F. Chen, G. Zhou, L. Lu, Q. Chang, J. Li, Discrete simulation of granular and particle-fluid flows: from fundamental study to engineering application, Reviews in Chemical Engineering 33 (2017) 551–623

    work page 2017

  8. [8]

    Ahmadi, S

    A. Ahmadi, S. Larsson, C. Wers¨ all, Scaling granular material with polygonal particles in discrete element modeling, Partic- uology 75 (2023) 151–164

    work page 2023

Show all 24 references
  1. [9]

    Roessler, A

    T. Roessler, A. Katterfeld, Scalability of angle of repose tests for the calibration of dem parameters, in: International Conference on Bulk Materials Storage, Handling and Transportation (12th: 2016: Darwin, NT), Engineers Australia Barton, ACT, 2016, pp. 201–211

    2016

  2. [10]

    Roessler, A

    T. Roessler, A. Katterfeld, Scaling of the angle of repose test and its influence on the calibration of dem parameters using upscaled particles, Powder technology 330 (2018) 58–66

    2018

  3. [11]

    Coetzee, Calibration of the discrete element method, Powder Technology 310 (2017) 104–142

    C. Coetzee, Calibration of the discrete element method, Powder Technology 310 (2017) 104–142

    2017

  4. [12]

    J. W. Rudnicki, J. Rice, Conditions for the localization of de- formation in pressure-sensitive dilatant materials, Journal of the Mechanics and Physics of Solids 23 (1975) 371–394

    1975

  5. [13]

    G. R. Mcdowell, O. Falagush, H.-S. Yu, A particle refinement method for simulating dem of cone penetration testing in gran- ular materials, G´ eotechnique Letters 2 (2012) 141–147

    2012

  6. [14]

    Sharif, M

    Y. Sharif, M. Ciantia, M. J. Brown, J. A. Knappett, J. D. Ball, Numerical techniques for the fast generation of samples using the particle refinement method, in: Proceedings of proceedings of the 8th international conference on discrete element methods, 2019. 10

    2019

  7. [15]

    H. Pan, C. Zhao, J. Xing, R. Zhang, C. Morita, An attempt on the application of the gradient particle size discrete element specimen in rock engineering, in: IOP Conference Series: Earth and Environmental Science, volume 861, IOP Publishing, 2021, p. 072086

    2021

  8. [16]

    Servin, D

    M. Servin, D. Wang, C. Lacoursi` ere, K. Bodin, Examining the smooth and nonsmooth discrete element approach to granular matter, Int. J. Numer. Meth. Engng. 97 (2014) 878–902

    2014

  9. [17]

    Wiberg, M

    V. Wiberg, M. Servin, T. Nordfjell, Discrete element modelling of large soil deformations under heavy vehicles, Journal of Ter- ramechanics 93 (2021) 11–21

    2021

  10. [18]

    URL: https://www

    Algoryx Simulations, AGX dynamics, 2024. URL: https://www. algoryx.se/agx-dynamics

    2024

  11. [19]

    Radjai, V

    F. Radjai, V. Richefeu, Contact dynamics as a nonsmooth dis- crete element method, Mechanics of Materials 41 (2009) 715– 728

    2009

  12. [20]

    Lacoursi` ere, Ghosts and machines: regularized variational methods for interactive simulations of multibodies with dry frictional contacts, Ph.D

    C. Lacoursi` ere, Ghosts and machines: regularized variational methods for interactive simulations of multibodies with dry frictional contacts, Ph.D. thesis, Ume ˚ a University, SE-901 87 Ume ˚ a, 2007

    2007

  13. [21]

    Lacoursi´ ere, M

    C. Lacoursi´ ere, M. Linde, O. Sabelstr¨ om, Direct sparse factor- ization of blocked saddle point matrices, Para 2010: State of the Art in Scientific and Parallel Computing, Reykjavik, June 6-9 (2010)

    2010

  14. [22]

    D. Wang, M. Servin, T. Berglund, Warm starting the projected gauss–seidel algorithm for granular matter simulation, Compu- tational Particle Mechanics 3 (2016) 43–52

    2016

  15. [23]

    Servin, T

    M. Servin, T. Berglund, S. Nystedt, A multiscale model of ter- rain dynamics for real-time earthmoving simulation, Advanced Modeling and Simulation in Engineering Sciences 8 (2021) 11

    2021

  16. [24]

    Da Cruz, S

    F. Da Cruz, S. Emam, M. Prochnow, J.-N. Roux, F. Chevoir, Rheophysics of dense granular materials: Discrete simulation of plane shear flows, Physical Review E—Statistical, Nonlinear, and Soft Matter Physics 72 (2005) 021309. 11

    2005

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.