PaperlayerView exact revision on arXiv

Source-linked AI summary

Extraction of Force-Chain Network Architecture in Granular Materials Using Community Detection

Danielle S. Bassett, Eli T. Owens, Mason A. Porter, M. Lisa Manning, Karen E. Daniels

arXiv:1408.3841v2cond-mat.softnlin.AOnlin.PSphysics.data-an

TL;DR

Force chains matter for mechanical stability and sound transmission, but there is no agreed quantitative definition, complicating comparisons across materials and conditions. The paper uses community detection with a contact-constrained geographical null model to extract chain-like structures and characterize them with three diagnostics. Across laboratory and frictionless numerical packings, force-chain sizes are consistent with an exponential distribution, while their pressure-related geometry differs between the two systems.

  • Problem

    Force chains lack an agreed quantitative definition, and it remains unclear which properties are universal or depend on material conditions.

    Source: §I · p. 1

  • Method

    The authors apply community detection with a geographical null model to force-weighted contact networks and characterize the extracted structures by size, network force, and gap factor.

    Source: §I · p. 3, A. Force-Chain Extraction · p. 4

  • Results

    Force-chain sizes are consistent with an exponential distribution in both systems; laboratory communities become more compact at higher pressure, while numerical force-chain geometry appears less pressure-sensitive.

    Source: §V · p. 11

  • Takeaways & Limitations

    The three diagnostics are used to identify and characterize force-chain architectures across laboratory and numerical packings and to compare particulate materials.

    Source: §I · p. 3, §V · p. 10

Abstract

from arXiv · show

Force chains form heterogeneous physical structures that can constrain the mechanical stability and acoustic transmission of granular media. However, despite their relevance for predicting bulk properties of materials, there is no agreement on a quantitative description of force chains. Consequently, it is difficult to compare the force-chain structures in different materials or experimental conditions. To address this challenge, we treat granular materials as spatially-embedded networks in which the nodes (particles) are connected by weighted edges that represent contact forces. We use techniques from community detection, which is a type of clustering, to find sets of closely connected particles. By using a geographical null model that is constrained by the particles' contact network, we extract chain-like structures that are reminiscent of force chains. We propose three diagnostics to measure these chain-like structures, and we demonstrate the utility of these diagnostics for identifying and characterizing classes of force-chain network architectures in various materials. To illustrate our methods, we describe how force-chain architecture depends on pressure for two very different types of packings: (1) ones derived from laboratory experiments and (2) ones derived from idealized, numerically-generated frictionless packings. By resolving individual force chains, we quantify statistical properties of force-chain shape and strength, which are potentially crucial diagnostics of bulk properties (including material stability). These methods facilitate quantitative comparisons between different particulate systems, regardless of whether they are measured experimentally or numerically.

I. INTRODUCTION

Force chains are heterogeneous force-bearing structures, but their quantitative definition and the properties shared across particulate systems remain unresolved. The paper proposes a network-based, threshold-free extraction method and diagnostics for comparing force-chain architectures across materials and conditions.

I. INTRODUCTION

  • The field lacks a quantitative definition of force chains, and it remains unclear which chain properties are universal or sensitive to friction, adhesion, boundaries, and gravity.This limits consistent characterization of chain size and shape across particulate systems.Source: §I · p. 1
  • Strong-force thresholds can omit neighboring particles needed to keep a linear chain stable against buckling.The authors therefore motivate methods that do not require a minimum force threshold for network membership.Source: §I · p. 2
  • The method is applied to laboratory experiments and idealized numerical packings to examine pressure-dependent structure and compare force-chain architectures.The two case studies are a wet laboratory system and a computational model of jammed packings.Source: §I · p. 3
  • The paper represents particles as nodes and contact forces as network edges, then uses community detection with a geographical null model to extract chain-like structures.The null model accounts for spatial embeddedness when identifying particle groups linked by strong forces.Source: §I · pp. 2–3, Figure 1 · p. 2
  • Three diagnostics—gap factor, size, and network force—are proposed to characterize and classify force-chain structures across materials.The gap factor measures how closely a community's geometry resembles a branched force chain.Source: §I · pp. 2–3

II. METHODS

The methods compare two-dimensional laboratory and frictionless numerical packings at approximately matched pressures. They construct force-weighted contact networks and identify communities using modularity with a geographical null model.

II. METHODS

  • The study compares two-dimensional bidisperse disk packings of about 600 particles across approximately seven matched pressure values, despite differences in friction, gravity, and boundary conditions.The experiments include friction and gravitational pressure; the simulations are frictionless and use periodic boundaries.Source: §II · p. 3

A. Granular Experiments

  • Laboratory packings use photoelastic disks between Plexiglas sheets, with particle contacts and forces inferred from images taken with and without polarizers.The experimental configurations are rearranged by hand and loaded with brass weights.Source: A. Granular Experiments · p. 3

B. Frictionless Simulations

  • The numerical study analyzes mechanically stable frictionless bidisperse disks in periodic, zero-gravity boxes, selecting packing fractions to match experimental mean pressures.Twenty packings are analyzed at each of seven packing fractions.Source: B. Frictionless Simulations · pp. 3–4

A. Force-Chain Extraction

  • Each contact network is weighted by normal force relative to the mean force, and modularity maximization with a contact-constrained geographical null model groups particles into force-chain communities.The resulting hard partition assigns each particle to one community whose forces are stronger than expected under the null model.Source: A. Force-Chain Extraction · pp. 4–5

B. Properties of Force Chains

The paper characterizes communities using size, network force, and a gap factor that captures gaps and branching. It also defines systemic gap-factor measures for comparing communities across a network.

B. Properties of Force Chains

  • Community size counts particles, while network force measures a community’s contribution to modularity and is larger for densely packed chains with stronger forces.Sparse chains with smaller inter-particle forces have lower network force.Source: B. Properties of Force Chains · p. 5
  • The gap factor uses the correlation between hop and physical distances to characterize gaps and branching; compact or linear communities have perfect correlation.Gaps, branches, and rings increase hop distance relative to physical distance and reduce the correlation.Source: B. Properties of Force Chains · pp. 5–6
  • The systemic gap factor weights communities in proportion to their size, whereas the alternative uniform measure gives every community equal weight.The paper focuses on the size-weighted measure because it varies less across packings; weighted and uniform measures tend to be larger for branched and linear communities, respectively.Source: B. Properties of Force Chains · p. 6

IV. CHARACTERIZING FORCE CHAINS

The authors detect force-chain-like communities by maximizing modularity across resolutions and selecting a resolution that approximately maximizes the systemic gap factor. In a laboratory packing, that factor peaks at γ = 0.7; the resulting communities are characterized by size, network force, and gap factor.

IV. CHARACTERIZING FORCE CHAINS

  • The method maximizes modularity with a geographical null model across γ values, then selects a resolution that approximately maximizes the systemic gap factor.The resulting communities tend to be chain-like; communities obtained with the Newman–Girvan null model instead tend to be compact.Source: §IV · p. 6
  • Small γ favors compact communities and large γ favors linear ones, while maximizing the systemic gap factor selects structures most similar to visually observed force chains.The resolution parameter sets community spatial scale: smaller values yield larger communities and larger values yield smaller ones.Source: §IV · pp. 6–7
  • For one laboratory packing, the systemic gap factor peaks at γ = 0.7 among tested resolutions, and the detected communities vary in size, network force, and gap factor.Single-particle communities are common near γ = 1 and become more common as γ increases beyond 1.Source: Figure 4 · p. 7, §IV · p. 7
  • The analysis compares community size, network force, and gap-factor distributions across pressure for laboratory experiments and frictionless simulations.Figure 5 also displays laboratory community structure at three pressures and plots gap factor against resolution and pressure.Source: §IV · p. 7, Figure 5 · p. 7

A. Granular Experiments

In the experiments, increasing pressure makes communities more compact and force-chain networks more homogeneous and less chain-like. The resolution maximizing the systemic gap factor identifies the most branched structures, while size and force distributions become narrower at high pressure.

A. Granular Experiments

  • As pressure rises, communities become more compact at low γ and shrink toward two-particle chains at high γ, indicating more homogeneous, less chain-like networks.At γ = 0.7, communities shift from branched structures to compact domains; at γ = 1.9, many-particle chains tend to shrink to two-particle chains.Source: A. Granular Experiments · pp. 7–8
  • The systemic gap factor has a maximum at every pressure, with γopt shifting from 0.7 at 2.7 × 10−4E to 0.9 at 5.9 × 10−3E.The maximum approximately selects the most branch-like communities; high-pressure packings have a smaller systemic gap factor at both small and large γ.Source: A. Granular Experiments · p. 8
  • Community size and network force are approximately exponentially distributed across all seven pressures, while most communities have large gap factors.At high pressure, size and force distributions are less broad, consistent with force-chain homogenization.Source: A. Granular Experiments · p. 8, Figure 6 · p. 8

B. Frictionless Simulations

In frictionless simulations, resolution changes communities from compact domains to small branch-like structures, while their qualitative structure changes little with pressure. At γ = 0.9, the gap factor is maximized; community size and force are approximately exponentially distributed, with pressure-related shifts in their means.

B. Frictionless Simulations

  • Simulated networks change from one compact community at γ = 0.7 to many small branch-like communities at γ = 1.3, with no strong qualitative structural change across pressures.Figure 7 presents community structure at 6.7×10−4E, 2.2×10−3E, and 5.9×10−3E; color encodes log10 network force.Source: B. Frictionless Simulations · p. 8, Figure 7 · p. 9
  • The systemic gap factor has a clear maximum at γ = 0.9, and its curve shape is more consistent across pressures than in experiments.The maximum is observed among tested γ values from 0.1 to 2.1.Source: B. Frictionless Simulations · pp. 8–9
  • Community size and network force have approximately exponential distributions across pressures, while most communities have large gap factors.Most communities are relatively small and weak, but a few are relatively large and strong; only a few have small gap factors.Source: B. Frictionless Simulations · p. 9
  • Unlike in experiments, distribution shapes change little with pressure, although mean community size and network force vary systematically with pressure.The size, force, and gap-factor distributions do not change dramatically across the simulated pressures.Source: B. Frictionless Simulations · p. 9, Figure 8 · p. 9

C. Comparison Between Laboratory and Numerical Force Chains

Laboratory and frictionless numerical packings differ in force-chain size, strength, and gap factor. Their responses to pressure also differ, with laboratory chains becoming less branch-like while simulated geometry changes less.

C. Comparison Between Laboratory and Numerical Force Chains

  • Numerical packings have larger and stronger force chains on average than laboratory packings, whose chains have smaller mean size and network force.The comparison summarizes the diagnostics across equal-pressure runs.Source: C. Comparison Between Laboratory and… · pp. 9–10, Figure 9 · p. 10
  • As pressure rises, laboratory chain size and network force increase while systemic gap factor decreases; numerical size and network force also increase, but gap factor does not decrease.In laboratory packings, increasing pressure is associated with more homogeneous, densely connected force structure and less branching; simulated geometry is affected less by pressure.Source: C. Comparison Between Laboratory and… · p. 10
  • Laboratory packings have smaller average gap factor than frictionless packings, a difference that may reflect both structural homogeneity and gravity-related vertical chains.The numerical packings are gravity-free, whereas laboratory packings are influenced by gravity.Source: C. Comparison Between Laboratory and… · p. 10
  • Force-chain community size decreases as pressure approaches zero, and the authors postulate that these communities are negatively correlated with mechanically unstable regions.At zero temperature, zero pressure marks the jamming transition; the proposed connection predicts shrinking force chains as unstable regions grow.Source: C. Comparison Between Laboratory and… · p. 10

V. CONCLUSIONS AND DISCUSSION

The paper applies a geographical-null-model community-detection method to characterize force-chain architecture in laboratory and numerical packings. The results show both a shared size-distribution pattern and differences in chain geometry and pressure response.

V. CONCLUSIONS AND DISCUSSION

  • The method detects chain-like communities using modularity maximization with a geographical null model and characterizes them with size, network force, and gap factor.The authors choose a resolution parameter that approximately maximizes gap factor in each scenario.Source: §V · p. 10
  • Force-chain size distributions are consistent with an exponential distribution in both laboratory and numerical packings.Laboratory chains are smaller and weaker on average, while numerical chains tend to be larger and stronger.Source: §V · p. 11
  • Laboratory packings develop compact rather than branching communities at high pressure, whereas frictionless numerical force-chain geometry appears less affected by pressure.Together, these differences support the conclusion that the systems differ in force-chain structure, topology, and geometry.Source: §V · p. 11

Methodological Insights

The geographical null model preserves contact-network topology while scrambling force-weighted geometry, incorporating more particulate physics than the commonly used Newman-Girvan null model. The framework is proposed for comparing systems and examining how community arrangements relate to material response.

Methodological Insights

  • The geographical null model fixes binary contact-network topology while scrambling force-weighted geometry, incorporating more particulate physics than the Newman-Girvan null model.The authors identify adding constraints such as force balance as a future direction.Source: Methodological Insights · p. 11

Practical Utility

  • The diagnostics can quantify similarity between simulations and experiments, or compare experiments and simulations with different particles or boundary conditions.The authors also propose using diagnostic distributions to help predict differences in macroscopic behavior.Source: Practical Utility · p. 11
  • Community-level particle connections and spatial arrangements may help investigate links between force-chain organization and linear or nonlinear packing response.The authors suggest that large, strong communities could indicate relatively stable regions and community boundaries could indicate weakness; these possibilities remain for future investigation.Source: Practical Utility · p. 11

APPENDIX 1: EFFECT OF COMMUNITY WEIGHTING ON FORCE-CHAIN EXTRACTION

The appendix compares force-chain extraction using a uniformly weighted systemic gap factor with the main text’s size-weighted measure. The weighting choice changes both the selected resolution and the resulting chain morphology.

APPENDIX 1: EFFECT OF COMMUNITY WEIGHTING ON FORCE-CHAIN EXTRACTION

Gap Factor Versus Resolution Parameter

  • For laboratory packings, guniform peaks at γ = 0.9 for high pressure and γ = 1.5 for low pressure; numerical packings peak at γ = 1.1 across pressures.The laboratory maxima vary with pressure, whereas the numerical maximum varies little with pressure.Source: Gap Factor Versus Resolution Parameter · p. 12, Figure 10 · p. 12

Optimal Value of the Resolution Parameter

Resolution Parameter

  • At γ = 0.9, chains are larger and more branched than at γ = 1.1; size weighting favors these larger communities, while uniform weighting reveals smaller, more linear communities.The γ = 1.1 communities also include more singletons.Source: Resolution Parameter · p. 12

APPENDIX 2: METHODOLOGICAL CONSIDERATIONS

Community assignments—and the force chains inferred from them—remain statistically similar after adding Gaussian force noise with width fij/3 in both experimental and numerical packings. The appendix also compares chain structures identified at two gap-factor-based resolutions.

the Estimation of Contact Forces

  • Laboratory force-measurement errors could reach ±30% of contact force, with high variability attributed to nonlinear fitting.Contact forces are estimated by solving the inverse photoelastic problem from images taken with polarizers.Source: the Estimation of Contact Forces · p. 12
  • At γ = 0.9, the size-weighted gap factor identifies larger, more branched chains; at γ = 1.1, the uniformly weighted factor identifies smaller, less branched chains.The contrast occurs across pressures, but is most evident at high laboratory pressures and least evident at low laboratory pressures; numerical packings vary little with pressure.Source: Figure 11 · p. 13
  • To test robustness, the authors added Gaussian noise with width a fij/3 to each contact in 20 simulated networks for each of 21 experimental packings across seven pressures.They reevaluated the community structure for each simulated network.Source: the Estimation of Contact Forces · p. 13
  • Community assignments remain statistically similar after noise of width fij/3: z-scores exceed 18 in laboratory packings and 9 in numerical packings.The authors interpret these results as robustness to force-measurement errors of about ±30% of contact forces.Source: the Estimation of Contact Forces · p. 13

Laboratory Packings

In laboratory packings, community size, network force, and gap factor relate to communities’ vertical position at low pressures, but not at higher pressures. The authors interpret the low-pressure pattern as consistent with gravity’s role in force-chain shape.

Laboratory Packings

  • For each pressure, the authors averaged network force, gap factor, and mean vertical position across communities of each unique size to reduce degrees of freedom and potential false positives.They then tested correlations between the network diagnostics and vertical position using Spearman rank correlation to increase robustness to outliers.Source: Laboratory Packings · p. 14
  • At low pressures, community size, gap factor, and strength each significantly correlate with mean vertical position.The correlations use uncorrected p-values below 0.05.Source: Laboratory Packings · p. 14, Figure 13 · p. 14
  • At 2.7 × 10−4E and 4.1 × 10−4E, larger, stronger communities tend to lie lower; at 4.1×10−4E, communities with larger gap factors tend to lie higher.Between 22×10−4E and 59×10−4E, vertical position shows no relationship with community size, network force, or gap factor; the authors conclude gravity can affect force-chain shape, particularly at low pressures.Source: Laboratory Packings · p. 14, Figure 13 · p. 14
Loading 1408.3841v2…