Version of Record: https://www.sciencedirect.

com/science/article/pii/S2352431620302455
Manuscript_eead8550e11bb335b3302e5534906ada

Nonlocality in Granular Complex Networks:

Linking Topology, Kinematics and Forces
K. Karapiperisa , J. E. Andradea,∗
a
Division of Engineering and Applied Science, California Institute of Technology, Pasadena, CA 91125,
USA

Abstract
Dry granular systems respond to shear by a process of self-organization that is nonlocal
in nature. This study reveals the interplay between the topological, kinematical and force
signature of this process during shear banding in an sample of angular sand. Using Level-
Set Discrete Element simulations of an in-situ triaxial compression experiment, and complex
networks techniques, we identify communities of similar topology (cycles), kinematics (vortex
clusters) and kinetics (force chains), and study their cooperative evolution. We conclude
by discussing the implications of our observations for continuum modeling, including the
identification of mesoscale order parameters, and the development of nonaffine kinematics
models.
Keywords: granular materials, complex networks, nonlocality

1. Introduction
The study of nonlocality in granular materials can be traced back to the pioneering
experiments of Roscoe [1] and, later, Mühlhaus and Vardoulakis [2], establishing the char-
acteristic width of a shear band in sand at 8-10 particle diameters. Further evidence of
5 nonlocality has been identified in the dynamic flow regime [3] for example in the form of
nozzle jamming in silos [4] and thickness-dependent repose angles in surface flows [5, 6].
Later, advances in experimental techniques [7–9] as well as discrete element (DEM) [10] and
contact dynamics (CD) [11] simulations inspired grain-scale studies in an effort to explain
these emergent phenomena. Most notably, photoelastic experiments and particle simula-
10 tions helped identify the heterogeneous nature of force transmission in an assembly in the
form of force chains [12–14]. Subsequent experiments conjectured force chain buckling [8] as
a mechanism for shear localization, which was later investigated through structural stability
analyses enabled by DEM [15, 16]. Measurements of force correlations were used in [17] to
quantify the heterogeneous nature of force networks under shear, yielding a consistent cor-
15 relation length of about 10 particle diameters. Similarly, the study of velocity correlations
has revealed the nonaffine nature of granular kinematics, termed granulence [18, 19]. As a

∗
Corresponding author
Email address: [email protected] (J. E. Andrade)
Preprint submitted to Extreme Mechanics Letters October 27, 2020

© 2020 published by Elsevier. This manuscript is made available under the Elsevier user license
https://www.elsevier.com/open-access/userlicense/1.0/
 signature of these correlations, vortices of characteristic sizes emerge [20–22], accompanied
by intense particle rotations [23–27], eventually leading to the development of a shear band
[28–31]. Finally, the use of complex networks techniques [28, 32–38] has contributed to the
20 identification of stable [35] and unstable [39] mesoscale features and communities [40] and
their topological transformations.
Alongside these micromechanical studies, it was recognized that standard continuum
theories failed to capture nonlocal effects [41, 42]. As a result, two major families of theo-
ries have emerged: enhanced continua [43] and nonlocal theories [44]. Enhanced (or weakly
25 nonlocal [45]) continua depart from the standard Cauchy assumption of affine deformation,
by introducing higher-order kinematics and their conjugate kinetics. Most notably, the
micropolar theory [46, 47], which equips the material point with rotational degrees of free-
dom, has succesfully captured several aspects of shear bands in sands [2, 48]. On the other
hand, (strongly) nonlocal theories introduce an additional field that represents the (solid-
30 like or fluid-like) state of the material locally [44, 49]. Characteristic examples include the
Landau-type [50] order parameter formulation termed partial fluidization theory [49], gra-
dient plasticity [51, 52], and the nonlocal granular fluidity model [53]. In this family, the
length scale is typically identified as the characteristic length scale of the diffusion process
of a local microstructural event, such as a shear transformation [54], due to the correlated
35 motion of its neighbors. Recently, the hypotheses inherent to some of these formulations
have been supported by micromechanics, through advanced homogenization techniques [55–
59], kinematic models [31] and direct micromechanical descriptions of order parameters [60].
Yet, the micromechanical description of nonlocality within a sound theoretical framework
still remains largely an open question.
40 In this Letter, we investigate the emergent length scale in the quasistatic flow of sand.
To this end, we rely on three-dimensional Level-Set Discrete Element Method (LS-DEM)
simulations of triaxial compression [61] of a sample of angular sand characterized by X-ray
computed tomography [62]. We utilize complex network techniques, which have not re-
ceived proper attention in 3D systems, to reveal stable and unstable mesoscale topological
45 structures, vortex clusters and force chains, which depart from earlier observations in ide-
alized and predominantly two-dimensional systems. Particular emphasis in placed on the
cooperative evolution of these features through distinct stages of the experiment.
This Letter is organized as follows: Section 2 details the experiments and simulations
furnishing the micromechanical data to be subsequently analyzed. In Sections 3-5, we present
50 the methods used to analyze the topology, kinematics and forces in the system respectively,
and discuss the outcome of each analysis. We conclude by summarizing our main findings,
and discussing their implications for continuum theory in Section 6.

2. Triaxial compression experiments and simulations

The data analyzed in this work are obtained from a high-fidelity discrete element simu-
55 lation of a quasi-static in-situ triaxial compression experiment reported in an earlier publi-
cation [61]. In the experiment, a cylindrical specimen of angular Hostun sand, encased in a
flexible latex membrane, is subjected to a triaxial loading protocol [62]. Fig. 1 a) shows a
2
 2-dimensional slice of the XRCT scanned specimen, which measures 11 mm in diameter and
22 mm in height, and is comprised of 53,939 angular grains. The specimen is first compressed
60 isotropically to 100 kPa. Next, keeping the radial pressure constant, a freely rotating platen
in contact with the top part of the sample enforces a vertical compression until failure. Air
can escape through a hole in the loading platen, giving rise to drained conditions.
The experiment is computationally replicated using a variant of DEM [10], termed LS-
DEM [63]. Similarly to the original formulation, LS-DEM resolves the kinematics of particles
65 interacting through frictional contacts, but also accounts for accurate particle morphology.
In particular, for each physical grain in the triaxial sample, a virtual grain is generated
through a level set imaging algorithm (Fig. 1 a)). The resulting virtual specimen is subjected
to identical boundary conditions, by modeling the membrane as well as the kinematics of
the platen. The deformed virtual specimen is shown in Fig. 1 a), where the formation of
70 a shear band can be identified. Note that the friction coefficient (µ = 0.55) was calibrated
against the experiment, but falls within the range of experimental values [64]. For more
details regarding the LS-DEM simulation, the interested reader is referred to [61].
Figures 1 b) and c) compare the macroscopic response of the sample in experiment and
simulation, in terms of ratio of major to minor principal stress σ1 /σ3 and volumetric strain
75 εv respectively. The sample exhibits a peak in the macroscopic stress ratio (equivalently,
friction angle), only to decay later to a constant critical state value. In accordance with
earlier experiments and theory [65, 66], the peak state coincides with the maximal rate of
dilation of the shear band, while the volume remains constant at critical state. Note that the
volumetric strain in the simulation plateaus slightly earlier than in the experiment, which
80 can be attributed to the slightly premature attainment of the peak state in the simulation,
and the different method of measuring the volumetric strain in the experiment (3D DIC)
and the simulation (change of volume enclosed by the membrane). The peak and critical

(a) (b)
X-Ray CT LS-DEM 7 Peak Crit. State
Segmented Grain 5
1/ 3

Grain Level Set

3 Experiment
LS-DEM
1
(c)
0.00 0.05 0.10 0.15
150 μm a
Level Set
Characterization (ii)
0.08
0.04
v

Experiment
0.00 LS-DEM

(i) (iii) 0.00 0.05 0.10 0.15

a
Figure 1: a) LS-DEM simulation of a sample of Hostun sand characterized by XRCT.
b) Principal stress ratio plotted against axial strain. The peak and critical state are
indicated with dashed lines. c) Volumetric strain plotted against axial strain.

3
 state regimes are of particular interest to this study, and are highlighted in the figures.

3. Mesoscale topological evolution

85 3.1. Complex networks
The nonlocal response of the granular assembly to externally applied loads is encoded in
its evolving contact network structure. Motivated by the success of complex network analysis
in identifying the nonlocal topological evolution of idealized two-dimensional systems [35], we
proceed by considering the assembly as a graph, where particles serve as nodes, and contacts
90 or contact forces serve as edges connecting the nodes (Fig. 2 a)). In particular, we study two
types of networks that are defined as follows. In the unweighted or binary network B, edge
weights Bij are zero-valued if no contact exists between particles i and j, and unit-valued
if the particles interact through at least one contact. On the other hand, in the weighted
network W, edge weights are given by the normalized interparticle force Wij = fij /hfi, where
95 fij is the total force arising from potentially multiple contact points. The unweighted and
weighted contact networks have been recently used to extract interesting features [67] such
as cycles [33–35] and communities [40], and to pinpoint shear band nucleation [36, 37] and
force chain development [68].
In this section, we restrict our attention to the unweighted network, while the weighted
100 network will be considered later in Section 5. In particular, we focus on identifying and
characterizing the evolution of mesoscale structures during the shear-banding or unjamming
transition. This is achieved through a minimal cycle analysis of the network [69]. A cycle
is a closed walk along the graph edges, i.e. one that starts and ends at the same node. A
cycle basis is a set of simple (non-intersecting) cycles that forms a basis of the cycle space
105 of the graph. A minimum cycle basis is a basis with minimal total length of cycles. Fig. 2
b) shows examples of minimal 3-,4-,5- and 6-cycles that pass through a given particle in the
system. In a two-dimensional system, the minimal cycles would simply correspond to the
faces of the contact graph, i.e. the regions formed by drawing the graph.

3.2. Results
110 The size of our network allows us to study the statistics of cycle sizes, since even rare
occasions of long minimal cycles can be accounted for. Fig. 2 c) shows the probability
density function (PDF) of cycle sizes Nc in the whole sample at peak and critical state.
Interestingly, the density appears to decay super-exponentially with cycle size and may be
3
well approximated by P (Nc ) ∼ e−αNc , as shown in the same figure. In accordance with
115 previous observations [35, 37], the density of longer cycles increases at critical state, which
could be a manifestation of dilatation.
We now turn our attention to the density of individual cycles inside the shear band,
where significant topological changes occur. Fig. 2 d) shows the evolution of the density of
3-, 4-, 5- and 6-cycles (constituting ∼80% of the total number of cycles) as a function of the
120 shear strain s in the localized band. We observe that 3-,4- and 5-cycles are almost equally
populous in the initially jammed configuration, hinting on their importance as stabilizing
mesoscale features. This lies in contrast to previous studies on idealized two-dimensional
4
 [35] and three-dimensional [37] assemblies which showed that 3-cycles constitute the clear
majority in a jammed state, due to their high rotational frustration. We conjecture that
125 this is due to the pronounced asphericity and irregularity of the sand grains, that enhance
stability, in line with recent evidence of the effect of particle morphology on the mechanical
response [70]. Upon further deformation, 3-, 4- and 5-cycles exhibit a power-law decay to a
common critical state density, also shared by 6-cycles, hinting on the existence of a common
steady state in cycle birth-and-death dynamics.
130 Inspired by the apparent importance of several classes of cycles in our system, and in
order to describe the ensuing phase P transition,
P we introduce the average cycle membership
of a particle i, given by N̄c (i) = j jcj (i)/ j cj (i), where j ∈ [3, ∞) is the cycle size and
cj (i) is the number of minimal cycles of size j that pass through node i. This measure in
turn gives rise to the minimal cycle coefficient D(i) = 3/N̄c (i). A highly stable particle
135 entirely surrounded by 3-cycles will have D = 1, while a highly unstable particle will be

(a) (b)

(c) (d) (e)

100 0.4 3-cycles 0.8
4-cycles 6
Cycle density

5-cycles
0.3 5
Z

6-cycles 4
10 2
0.6
P(Nc)

0.0 0.2 0.4 0.6

0.2
D

10 4 Peak 0.1
Critical state 0.4
100 101 0.0 0.2 0.4 0.6 0.0 0.2 0.4 0.6
Nc s s

Figure 2: a) Graph representation of the sample at critical state, with nodes colored
by their minimal cycle coefficient. b) Example minimal 3-,4-,5- and 6-cycles passing
through a given center node. c) PDF of cycle size Nc in the sample, and fitted
distribution. d) Evolution of density of 3-, 4-, 5- and 6-cycles inside the shear band.
e) Evolution of the average minimal cycle coefficient inside the shear band. Inset:
Evolution of the average coordination number.

5
 surrounded by long cycles such that D → 0. The proposed coefficient is a generalization of
the local clustering coefficient C(i), which measures the density of 3-cycles surrounding a
particle, and similar higher order coefficients for longer cycles [71].
We note that the minimal cycle coefficient D also bears resemblance to two other mea-
140 sures in the complex networks literature: the loop coefficient P [72, 73] and the subgraph
centrality [74]. The former is given by L(i) = 1/(ki (ki − 1)) j,k∈Γi 1/djk/i , where Γi is the
local subgraph of neighbors of i where node i has been excluded, and djk/i is the shortest
path length between particles
P j and k within Γi . The second relevant measure is the sub-
graph centrality SC(i) = ∞ B
k=0 ii
k
/k!, which measures the number of closed walks starting
145 and ending in node i. The difference with D is that SC includes trivial even-sized walks, and
cycles that are not simple. It has been successfully used as a proxy for fluctuating kinetic
energy during failure of a granular assembly [75].
Fig. 2 e) shows the evolution of the average minimal cycle coefficient hDi within the shear
band, and compares it to that of the average coordination number hZi, the prototypical
150 order parameter for the jamming transition [76]. In both cases, we observe a similar decay
to a critical state value. The large Pearson’s correlation coefficient (∼0.6) between the
particle-scale D and Z implies that the minimal-cycle coefficient could serve as an equivalent
mesoscale order parameter for this transition. In the next sections, we will address its
relevance to the kinematics and kinetics of the system.

155 4. Nonaffine kinematics

In this section, we characterize for the first time the kinematics that accompany the topo-
logical changes within a three dimensional shear band by studying the formation and evo-
lution of vortex clusters. Earlier kinematics studies have either focused on two-dimensional
systems [27, 30, 31], have instead relied on simplified scalar measures of nonaffine deforma-
160 tion [39, 77, 78] or have not addressed shear bands [79].

4.1. Vortex identification

We analyze the nonaffine particle displacements δui ≡ ui −  · x̃i , where ui is the dis-
placement, and x̃i is the position of particle i with respect to a local coordinate system
aligned with the shear band. The affine (approximately simple-shear) strain  dominating
the band’s deformation is found in a least-squares sense [77] as:
X
min ||ui −  · x̃i ||2 (1)

i∈S

where the summation takes place among the set of particles S within the band.
We proceed to identify mesoscale vortex structures formed by the nonaffine displacements.
As opposed to two-dimensional systems[20, 22], where vorticity is parallel to the out-of-
plane axis, vortices can freely rotate in 3D systems and form clusters [79]. To identify those
clusters, we employ the methodology of [79], by first computing the vorticity field:
XX
ω(x) = ρ(x)−2 φi (x) [∇φj (x) × δuij ] (2)
i∈S j∈S

6
 2 ¯2
where φ(x) = e−||x−xi || /d is the coarse-graining kernel[80], ρ(x) is the coarse-grained number
density field, and δuij = δui − δuj is the relative nonaffine displacement between particles
i and j. To identify a cluster, we choose a particle within the shear band at random and
165 traverse its contact network using a breadth-first search algorithm. For contacting particles
i, j, their normalized vorticities, ω̃ i , ω̃ j respectively, are compared by computing the angle
of their cosine similarity θij = cos−1 (ω̃ i · ω̃ j ). The particles are included in the same cluster if
θij < θc = π/6, and the search continues until no more particles are included in the cluster1 .

4.2. Results
170 The nonaffine displacement field along with the identified vortex clusters are shown in
Fig. 3 a). In contrast to two-dimensional systems where there is a clear geometrically defined
length scale in the form of a vortex radius [22, 31], here the complex shape of vortices requires
an alternative definition of length scale. To this end, we compute the distribution of vortex
cluster size Nv throughout the stages of the experiment, and find, in accordance with [79],
175 that it is well described by a power law with exponential cutoff, P (Nv ) ∼ Nv−α e−Nv /νv , as
shown in Fig. 3 b). By analyzing the exponential tails, a characteristic vortex length scale
1/3
`v ≡ νv of about 4 grain diameters is obtained. Its evolution throughout the experiment
is shown in the inset Fig. 3 b), where a slight increase with shear strain is identified.
Next, we characterize the vortex strength ωv , which we define as the average vorticity
180 in each cluster. As shown in Fig. 3 c), the average vortex strength increases to a steady
state value, while its density is well approximated by the Boltzmann-Maxwell distribution
2
P (ωv ) ∼ ωv2 e−ωv /2 . Finally, we characterize the directionality of these vortex clusters. This is
achieved by computing the average normalized vorticity of each cluster ω̃ c , and comparing it
to a macroscopic 'director' Ω that is orthogonal to both the direction of shear and the normal
185 to the shear band plane, as shown in Fig. 3 a). The distribution of their cosine similarity
cos χ̄ = ω̃ c · Ω is plotted is Fig. 3 d). We observe a slightly anisotropic distribution with
some degree of preferential alignment of the vorticity with ±Ω, which would correspond
to the primary (homothetic) and secondary (antithetic) vortices observed in 2D systems
[27, 30, 31]. Most vortices appear to be arbitrarily oriented in space.

190 5. Forces
To shed light on the kinetics that accompany the kinematical (Section 4) and topological
(Section 3) transition, we will now characterize the evolving force chain architecture. Driven
by the lack of general agreement on what constitutes a force chain, we first reconcile the
two major identification techniques in the literature: network community detection [81]
195 and ‘direct’ identification [82]. We, then, proceed with observations on chain stability and
establish direct links with topology and kinematics.

1
The critical angle is chosen as θc = π/6, following [79]. A sensitivity analysis showed that cluster sizes
decrease with increasing θc , yet their distribution consistently follows a power law with exponential cutoff.

7
 (a)

(b) (c) (d)

100 Peak 80 Peak 0.8
Critical state
Critical state
60 0.6
10 2

P(cos )
P( v)
P(Nv)

5
40 Increasing
shear strain 0.4
4
20
v /d

10 4 3 0.2 Peak
Critical State
0.00 0.25 0.50
0 0.0
100 101 102 0.0 0.1 0.2 1 0 1
Nv v cos
Figure 3: a) Nonaffine displacement field and identified vortex clusters within the
shear band. b) PDF of cluster size Nv and fitted power laws with exponential cutoff.
Inset: Evolution of characteristic length `v as a function of the shear strain within the
band. c) PDF of the vortex strengths and fitted Maxwell-Botzmann distribution. d)
PDF of the orientational order parameter cos χ̄.

5.1. Force chain extraction via community detection

Our point of departure is the characterization of the weighted and unweighted contact
networks outlined in Section 3. Following [81], we seek communities {si } of grains, strongly
connected via intergranular forces of similar magnitude, by maximizing the modularity func-
tion:
X
Q= (Wij − γPij )δ(si , sj ) (3)
i,j

where Wij is the weighted adjacency matrix, γ is the resolution parameter controlling the size
of communities, Pij is the so-called null model representing the expected weight of the edge
200 connecting nodes i and j, and δ(si , sj ) is the Kronecker delta. We adopt the geographical
null model [81], given by the unweighted adjacency matrix, Pij = Bij , in order to respect
the spatial connectivity constraints in the granular system.
8
 5.2. Direct force chain identification
For the ‘direct’ extraction of force chains, we employ a three-dimensional extension of the
detection algorithm described in [82]. Hereby, chains are identified as quasilinear sequences
of particles that reside in the strong-force network [13]. More specifically, let σ3 , n3 denote
the minor (most compressive) principal stress and its direction respectively, obtained
P c from
p c
a spectral decomposition of the particle stress. The latter is given by σ = 1./V cf ⊗x ,
p c c
where V is the particle volume, and x is the location of the contact force f with respect to
the particle centroid. For a sequence of particles to constitute a force chain S, its members
must i) exhibit a compressive stress that is higher than the sample average:
N
1 X j
|σ3i | > |σ |, ∀i ∈ S (4)
N j=1 3

where N is the number of particles in the sample, and ii) be sufficiently colinear:
lij · ni3
> cos α, ∀i, j ∈ S (5)
||lij || · ||ni3 ||
where lij is the branch vector connecting consecutive particles i, j in the chain. The angle α
205 represents the maximum allowable angle between chain segments which we take as α = 45◦ .

5.3. Results
Fig. 4 a) shows the contact force network at critical state, along with the resulting force
chains obtained via direct identification and community detection respectively. The latter is
based on a resolution parameter γ ∗ = 3.0 (recall its definition in Section 5.1), optimized to
210 generate chains maximally similar to those identified ‘directly’. Similarity is assessed based
on the coincidence of member particles of chains determined by the two methods. We find
that, for γ = γ ∗ , the chains determined by the two methods share the majority (∼ 60%) of
their participating members.
Fig. 4 c) shows the PDF of community sizes Ncom for varying resolution parameter γ.
−α
215 The sizes are found to follow a power law distribution P (Ncom ) ∼ Ncom , where α is almost
linearly correlated with γ. On the other hand, in accordance with earlier studies [82], the
size of chains obtained by direct identification follows an exponential distribution, as shown
in Fig. 4 d). This is further evidence that for large enough γ, the two identification methods
are reconciled.
220 Of particular importance to force transmission is the structural characterization is force
chains. Fig. 4 b) shows examples of a jammed (stable) and a buckled force chain, along with
several of their topological and kinematical attributes. A chain is assumed to have buckled
when two criteria are met [15]: i) an increase in local chain curvature beyond a critical
threshold, identified via sensitivity analysis, and ii) a reduction of the potential energy U
225 stored in the deformed chain contacts. The latter is given by:
1 X ||fnc ||2 ||ftc ||2
 
U= + (6)
2 c kn kt
9
 where fnc and ftc are the resolved normal and tangential force at a contact c, while kn and
kt are the normal and tangential contact stiffness respectively. A jammed chain is one
that persists through loading without buckling. Fig. 4 e) shows the PDF of buckled chain
segment sizes Nb throughout the experiment, where we can also identify the longest segment
230 that is prone to buckling (Nb = 10). It appears that longer buckling wavelengths are not
energetically favorable. Note that, in contrast to earlier studies, the stability of force chains
is assessed here without recourse to numerical proxies such as rolling friction, but rather as
an immediate consequence of morphology and interlocking.
Finally, we address the stability of chains in relation to the measures of topology and
235 kinematics investigated in Sections 3 and 4. To do so, we compare these measures for buckled
chains (occurring almost exclusively within the band) and persistent stable chains. Under
(a) (b)

(c) (d) (e)

100 = 1.0 100 Peak
= 1.5 Critical state 0.2
10 1 = 2.0
= 2.5
10 1
P(Ncom)

P(Nb)
P(Nf)

= 3.0
10 2 10 2 0.1
10 3 10 3
0.0
0 10 20 30 0 10 20 2 4 6 8 10
Ncom Nf Nb
Figure 4: a) Force network, directly identified chains and detection of communities
at critical state b) Characterization of a jammed and a buckled chain using various
descriptors c) PDF of community size Ncom and fitted power law distribution for
different values of the resolution parameter γ. d) PDF of force chain size Nf and
fitted exponential distribution. e) PDF of buckled force chain size Nb and fitted
distribution.

10
 an increase in the stored elastic energy, and a loss of lateral support due to dilatation, a
chain becomes increasingly susceptible to buckling. With the help of nonconvex Voronoi
tesselations (Appendix A), we can identify the critical local chain packing fraction φc that
240 induces instability in a chain, as shown in Fig. 5 a). Similarly, Fig. 5 b) compares the
average minimal cycle coefficient Dc , in stable and buckled chains, showing that buckling is
associated with a significant increase in Dc . Finally, Fig. 5 c) shows the minimum distance
(in grain diameters) of stable and buckled chains to the surface of the nearest vortex cluster.
Interestingly all chains avoid forming in the interior of vortices, and most buckling events
245 happen on the surface of vortices. This is because the kinematics in the interior of vortex
clusters are unfavorable for force chain stability. At the same time, there is no space for
force chains to form further away from the surface of clusters, since the latter densely occupy
the shear band (Fig. 3 a)). Overall, this confirms that, along with topology, vortices also
govern the formation of chains, in accordance with recent observations in two-dimensional
250 systems [31].

(a) (b) (c)

15 Jammed
7.5 Jammed Jammed
Buckled Buckled 1.0 Buckled

10 5.0 Vortex Interior Vortex Exterior

P(Nf v)
P(Dc)
P( c)

5 0.5
2.5
0 0.0 0.0
0.5 0.6 0.7 0.8 0.4 0.6 0.8 3 2 10 1 2 3
c Dc Nf v
Figure 5: PDFs of a) Packing fraction, b) Minimal cycle coefficients and c) Minimal
topological distance to the surface of the nearest vortex cluster, for jammed and
buckled chains.

6. Conclusion
Nonlocality is inherently linked to pattern formation such as shear banding. The objec-
tive of this study is to reveal the topological, kinematical and force signature of shear banding
of a sample of angular sand. We based our investigation on high-fidelity three-dimensional
255 simulations using the Level-Set Discrete Element Method and relied on complex networks
techniques to characterize the emergent length scale. Several implications for improved
predictive capabilities arise from this investigation.
Regarding topology, we found that 4- and 5-cycles emerge as equally stabilizing mesoscale
structures alongside the shorter well-documented 3-cycles [37]. We conjecture that this is
260 due to particle asphericity and angularity, which enhances topological interlocking, reduces
rotations, and thus increases structural stability. By uncovering the evolution of the density
of minimal cycles, the prediction of shear banding becomes possible given information about
11
 the current density and its rate of decay. We presented the first evidence of a characteristic
signature of cycle evolution that is determined by particle morphology. Such a quantitative
265 analysis has not been possible with methods available so far. We also introduced the minimal
cycle coefficient D, in order to collectively account for the apparent importance of several
families of cycles.
With regard to kinematics, we have provided evidence that dilatancy is the result of the
attrition of shorter (3-,4-,5-) cycles and the associated build up of longer cycles. Furthermore,
270 we have revealed the nonaffine nature of mesoscale kinematics by characterizing, for the
first time, the strength and orientational order of vortex clusters within the shear band,
while extending previous findings regarding their size distribution. Interestingly, the vortex
strength of these arbitrarily shaped clusters follows a Maxwell-Boltzmann distribution, that
converges to a well-defined critical state. In terms of their orientational order, we identified
275 a significant departure from the primary and secondary vortices observed in earlier two-
dimensional studies. These are all essential descriptors of nonlocal kinematics in enhanced
continuum theories.
To delineate the conjugate kinetics, we relied once again on complex networks. First,
we reconciled the definition of force chains as fundamental units of force transmission, by
280 comparing the two major identification techniques. We characterized for the first time the
distribution of the critical buckling wavelength of force chains, which emerged from a high-
fidelity representation of particle morphology, and without recourse to proxies such as rolling
friction. The width of a shear band naturally arises as the maximum such wavelength. We
also found that buckling chains were characterized by significantly lower average minimal
285 cycle coefficient compared to stable chains, which confirms the high relevance of this new
coefficient.
There are several ways to incorporate these new observations into improved predictive
nonlocal continuum descriptions. The coarsest description would rely on a mesoscale order
parameter that collectively accounts for topological rearrangements, within a Landau-type
290 framework [49]. The proposed minimal cycle coefficient appears to be a promising candidate.
Work towards this direction would require systematically investigating its diffusive spatial
coupling and its relation to the local rheology. Perhaps a more detailed description would
rely on modeling, and subsequently coarse-graining, the coupled birth-and-death dynamics of
force chains, cycles and vortices. We identified several constraints that these dynamics must
295 satisfy, such as their critical state densities and rates of decay. Additional work is required
in order to understand the dissipation that accompanies these birth-and-death processes.
A final avenue would be to explicitly model the nonaffine kinematic field within the shear
band [31]. Our characterization of the size, strength and orientation of vortices, can form the
basis for constructing admissible nonaffine fields. Which theoretical framework can provide
300 a consistent closure of these kinematics in terms of their conjugate kinetics remains an open
question. Could this be a modified micropolar framework or, perhaps, a gradient theory of
self-organization [83]? The road to a unified nonlocal continuum theory remains long and
challenging.

12
 Appendix A. Set Voronoi tesselation for arbitrarily shaped particles
305 To calculate the local packing fraction in Section 5.3,
we need to compute the volume of the cell associated
to each particle. To do so, we employ a generaliza-
tion of the standard Voronoi tesselation [84] for the case
of arbitrarily-shaped and nonconvex particles. This in-
310 volves a standard Voronoi tesselation of the particles’
discretized surface points. In other words, the points
at the surface mesh of all particles are used as the in-
put of a standard Voronoi computation [84], generating
the example subcells shown in Fig. A.6. Then, all sub-
315 cells belonging to the same particle are conglomerated
into the nonconvex Voronoi super-cell attached to that Figure A.6: Two nonconvex
particle. The accuracy of the scheme depends on the Voronoi cells each comprised
density of the surface discretization of each particle. of multiple convex subcells.

References
320 [1] K. H. Roscoe, The influence of strains in soil mechanics, Gotechnique 20 (1970) 129–170.
[2] H. B. Mühlhaus, I. Vardoulakis, The thickness of shear bands in granular materials, Géotechnique 37
(1987) 271–283.
[3] G. MiDi, On dense granular flows, The European Physical Journal E 14 (2004) 341–365.
[4] I. Zuriguel, D. R. Parisi, R. C. Hidalgo, C. Lozano, A. Janda, P. A. Gago, J. P. Peralta, L. M. Ferrer,
325 L. A. Pugnaloni, E. Clment, D. Maza, I. Pagonabarraga, A. Garcimartn, Clogging transition of many-
particle systems flowing through bottlenecks, Scientific Reports 4 (2014) 7324.
[5] O. Pouliquen, Scaling laws in granular flows down rough inclined planes, Physics of Fluids 11 (1999)
542–548.
[6] L. Silbert, J. W. Landry, G. S. Grest, Granular flow down a rough inclined plane: Transition between
330 thin and thick piles, Physics of Fluids 15 (2003) 1–10.
[7] J. Desrues, R. Chambon, M. Mokni, F. Mazerolle, Void ratio evolution inside shear bands in triaxial
sand specimens studied by computed tomography, Géotechnique 46 (1996) 529–546.
[8] M. Oda, H. Kazama, J. Konishi, Effects of induced anisotropy on the development of shear bands in
granular materials, Mechanics of Materials 28 (1998) 103 – 111.
335 [9] G. Viggiani, M. M. Küntz, J. Desrues, An experimental investigation of the relationships between
grain size distribution and shear banding in sand, Springer Berlin Heidelberg, Berlin, Heidelberg, pp.
111–127.
[10] P. A. Cundall, O. D. L. Strack, A discrete numerical model for granular assemblies, Géotechnique 29
(1979) 47–65.
340 [11] J. J. Moreau, Unilateral contact and dry friction in finite freedom dynamics, in: J. J. Moreau, P. D.
Panagiotopoulos (Eds.), Nonsmooth Mechanics and Applications, Springer Vienna, 1988, pp. 1–82.
[12] C. h. Liu, S. R. Nagel, D. A. Schecter, S. N. Coppersmith, S. Majumdar, O. Narayan, T. A. Witten,
Force fluctuations in bead packs, Science 269 (1995) 513–515.
[13] F. Radjai, M. Jean, J. Moreau, S. Roux, Force distributions in dense two-dimensional granular systems,
345 Phys. Rev. Lett. 77 (1996) 274–277.
[14] F. Radjai, D. E. Wolf, M. Jean, J.-J. Moreau, Bimodal character of stress transmission in granular
packings, Phys. Rev. Lett. 80 (1998) 61–64.
[15] A. Tordesillas, M. Muthuswamy, On the modeling of confined buckling of force chains, Journal of the
Mechanics and Physics of Solids 57 (2009) 706 – 727.
13
350 [16] A. Tordesillas, G. Hunt, J. Shi, A characteristic length scale in confined elastic buckling of a force
chain, Granular Matter 13 (2011) 215–218.
[17] T. Majmudar, R. Behringer, Contact force measurements and stress-induced anisotropy in granular
materials, Nature 435 (2005) 1079–1082.
[18] F. Radjai, S. Roux, Turbulentlike fluctuations in quasistatic flow of granular media, Phys. Rev. Lett.
355 89 (2002) 064302.
[19] O. Pouliquen, Velocity correlations in dense granular flows, Phys. Rev. Lett. 93 (2004) 248001.
[20] J. Peters, L. Walizer, Patterned nonaffine motion in granular media, Journal of Engineering Mechanics
139 (2013).
[21] K. Kamrin, C. Rycroft, M. Z. Bazant, The stochastic flow rule: a multi-scale model for granular
360 plasticity, Modelling and Simulation in Materials Science and Engineering 15 (2007) S449.
[22] P. Rognon, T. Miller, I. Einav, A circulation-based method for detecting vortices in granular materials,
Granular Matter 17 (2015) 177–188.
[23] M. Oda, J. Konishi, S. Nemat-Nasser, Experimental micromechanical evaluation of strength of granular
materials: Effects of particle rolling, Mechanics of Materials 1 (1982) 269 – 283.
365 [24] M. Oda, K. Iwashita, Study on couple stress and shear band development in granular media based on
numerical simulation analyses, International Journal of Engineering Science 38 (2000) 1713 – 1740.
[25] J. Bardet, Numerical simulations of the incremental responses of idealized granular materials, Inter-
national Journal of Plasticity 10 (1994) 879 – 908.
[26] M. R. Kuhn, K. Bagi, Contact rolling and deformation in granular media, International Journal of
370 Solids and Structures 41 (2004) 5793 – 5820. Granular Mechanics.
[27] F. Alonso-Marroquı́n, I. Vardoulakis, H. J. Herrmann, R. Weatherley, P. Mora, Effect of rolling on
dissipation in fault gouges, Phys. Rev. E 74 (2006) 031306.
[28] M. R. Kuhn, Structured deformation in granular materials, Mechanics of Materials 31 (1999) 407 –
429.
375 [29] A. L. Rechenmacher, Grain-scale processes governing shear band initiation and evolution in sands,
Journal of the Mechanics and Physics of Solids 54 (2006) 22 – 45.
[30] S. Abedi, A. Rechenmacher, A. Orlando, Vortex formation and dissolution in sheared sands, Granular
Matter 14 (2012) 695–705.
[31] A. Tordesillas, S. Pucilowski, Q. Lin, J. F. Peters, R. P. Behringer, Granular vortices: Identification,
380 characterization and conditions for the localization of deformation, Journal of the Mechanics and
Physics of Solids 90 (2016) 215 – 241.
[32] M. Satake, New formulation of graph-theoretical approach in the mechanics of granular materials,
Mechanics of Materials 16 (1993) 65 – 72.
[33] A. Smart, J. M. Ottino, Evolving loop structure in gradually tilted two-dimensional granular packings,
385 Phys. Rev. E 77 (2008) 041307.
[34] R. Arévalo, I. Zuriguel, D. Maza, Topology of the force network in the jamming transition of an
isotropically compressed granular packing, Phys. Rev. E 81 (2010) 041302.
[35] A. Tordesillas, D. Walker, Q. Lin, Force cycles and force chains, Phys. Rev. E 81 (2010) 011302.
[36] D. M. Walker, A. Tordesillas, Topological evolution in dense granular materials: A complex networks
390 perspective, International Journal of Solids and Structures 47 (2010) 624 – 639.
[37] D. Walker, A. Tordesillas, S. Pucilowski, M. Hopkins, J. Peters, A complex network analysis of granular
fabric evolution in three-dimensions, Dynamics of Continuous, Discrete and Impulsive Systems, Series
B: Applications and Algorithums 19 (2012) 471–495.
[38] D. M. Walker, A. Tordesillas, J. Zhang, R. P. Behringer, E. And, G. Viggiani, A. Druckrey, K. Alshibli,
395 Structural templates of disordered granular media, International Journal of Solids and Structures 54
(2015) 20 – 30.
[39] Y. Cao, J. Li, B. Kou, C. Xia, Z. Li, R. Chen, H. Xie, T. Xiao, W. Kob, L. Hong, J. Zhang, Y. Wang,
Structural and topological nature of plasticity in sheared granular materials, Nature Communications
9 (2018).
400 [40] C. Giusti, L. Papadopoulos, E. T. Owens, K. E. Daniels, D. S. Bassett, Topological and geometric

14
 measurements of force-chain structure, Phys. Rev. E 94 (2016) 032909.
[41] I. V. J. Sulem, Bifurcation analysis in geomechanics, Blackie Academic & Professional, 1st ed edition,
1995.
[42] O. Pouliquen, Y. Forterre, A non-local rheology for dense granular flows, Philosophical Transactions of
405 the Royal Society of London A: Mathematical, Physical and Engineering Sciences 367 (2009) 5091–5107.
[43] A. Eringen, Microcontinuum Field Theories: I. Foundations and Solids, Springer New York, 1999.
[44] K. Kamrin, Non-locality in granular flow: Phenomenology and modeling approaches, Frontiers in
Physics 7 (2019) 116.
[45] D. Rogula, Introduction to Nonlocal Theory of Material Media, Springer Vienna, pp. 123–222.
410 [46] E. Cosserat, F. Cosserat, Theorie des corps deformables, Paris: Herman, 1909.
[47] A. Eringen, Theory of Micropolar Elasticity, Springer New York, New York, NY, pp. 101–248.
[48] J. Tejchman, W. Wu, Numerical study on patterning of shear bands in a cosserat continuum, Acta
Mechanica 99 (1993) 61–74.
[49] I. Aranson, L. Tsimring, Continuum description of avalanches in granular media, Phys. Rev. E 64
415 (2001) 020301.
[50] L. Landau, E. Lifshitz, Statistical physics, Pergamon, New York, 1980.
[51] I. Vardoulakis, E. Aifantis, A gradient flow theory of plasticity for granular materials, Acta Mechanica
87 (1991) 197–217.
[52] K. C. Valanis, A gradient theory of internal variables, Acta Mechanica 116 (1996) 1 – 14.
420 [53] K. Kamrin, G. Koval, Nonlocal constitutive relation for steady granular flow, Phys. Rev. Lett. 108
(2012) 178301.
[54] M. L. Falk, M. Toiya, W. Losert, Shear transformation zone analysis of shear reversal during granular
flow, ArXiv e-prints (2008).
[55] W. Ehlers, E. Ramm, S. Diebels, G. DAddetta, From particle ensembles to cosserat continua: homog-
425 enization of contact forces towards stresses and couple stresses, International Journal of Solids and
Structures 40 (2003) 6681 – 6702.
[56] W. Ehlers, Homogenisation of discrete media towards micropolar continua: A computational approach,
AIP Conference Proceedings 1227 (2010) 306–313.
[57] J. Bardet, I. Vardoulakis, The asymmetry of stress in granular media, International Journal of Solids
430 and Structures 38 (2001) 353 – 367.
[58] A. Tordesillas, M. Muthuswamy, A thermomicromechanical approach to multiscale continuum modeling
of dense granular materials, Acta Geotechnica 3 (2008) 225–240.
[59] Q. Liu, A new version of hill’s lemma for cosserat continuum, Archive of Applied Mechanics 85 (2015)
761–773.
435 [60] Q. Zhang, K. Kamrin, Microscopic description of the granular fluidity field in nonlocal flow modeling,
Phys. Rev. Lett. 118 (2017) 058001.
[61] R. Kawamoto, E. Andò, G. Viggiani, J. E. Andrade, All you need is shape: Predicting shear banding
in sand with LS-DEM, Journal of the Mechanics and Physics of Solids 111 (2018) 375 – 392.
[62] E. Andò, G. Viggiani, S. Hall, J. Desrues, Experimental micro-mechanics of granular media studied by
440 x-ray tomography: recent results and challenges, Gotechnique Letters 3 (2013) 142–146.
[63] R. Kawamoto, E. Andò, G. Viggiani, J. E. Andrade, Level set discrete element method for three-
dimensional computations with triaxial case study, Journal of the Mechanics and Physics of Solids 91
(2016) 1 – 13.
[64] C. S. Sandeep, K. Senetakis, Exploring the micromechanical sliding behavior of typical quartz grains
445 and completely decomposed volcanic granules subjected to repeating shearing, Energies 10 (2017) 1–16.
[65] M. Schofield, C. Wroth, Critical State of Soil Mechanics, McGraw-Hill: London, 1968.
[66] M. D. Bolton, The strength and dilatancy of sands, Géotechnique 36 (1986) 65–78.
[67] L. Papadopoulos, M. Porter, K. Daniels, D. Bassett, Network Analysis of Particles and Grains, ArXiv
e-prints (2017).
450 [68] J. E. Kollmer, K. E. Daniels, Betweenness centrality as predictor for forces in granular packings, Soft
Matter 15 (2019) 1793–1798.

15
 [69] N. Deo, G. Prabhu, M. S. Krishnamoorthy, Algorithms for generating fundamental cycles in a graph,
ACM Trans. Math. Softw. 8 (1982) 2642.
[70] K. Karapiperis, J. Harmon, E. And, G. Viggiani, J. E. Andrade, Investigating the incremental behavior
455 of granular materials with the level-set discrete element method, Journal of the Mechanics and Physics
of Solids 144 (2020) 104103.
[71] Caldarelli, G., Pastor-Satorras, R., Vespignani, A., Structure of cycles and local ordering in complex
networks, Eur. Phys. J. B 38 (2004) 183–186.
[72] I. Vragović, E. Louis, A. Dı́az-Guilera, Efficiency of informational transfer in regular and complex
460 networks, Phys. Rev. E 71 (2005) 036122.
[73] I. Vragović, E. Louis, Network community structure and loop coefficient method, Phys. Rev. E 74
(2006) 016105.
[74] E. Estrada, J. A. Rodrguez-Velzquez, Subgraph centrality in complex networks, Physical Review E 71
(2005).
465 [75] D. M. Walker, A. Tordesillas, G. Froyland, Mesoscale and macroscale kinetic energy fluxes from granular
fabric evolution, Phys. Rev. E 89 (2014) 032205.
[76] A. J. Liu, S. R. Nagel, The jamming transition and the marginally jammed solid, Annual Review of
Condensed Matter Physics 1 (2010) 347–369.
[77] M. L. Falk, J. S. Langer, Dynamics of viscoplastic deformation in amorphous solids, Phys. Rev. E 57
470 (1998) 7192–7205.
[78] A. Tordesillas, M. Muthuswamy, S. Walsh, Mesoscale measures of nonaffine deformation in dense
granular assemblies, Journal of Engineering Mechanics 134 (2008).
[79] N. Oyama, H. Mizuno, K. Saitoh, Avalanche interpretation of the power-law energy spectrum in
three-dimensional dense granular flow, Phys. Rev. Lett. 122 (2019) 188004.
475 [80] J. Zhang, R. P. Behringer, I. Goldhirsch, Coarse-Graining of a Physical Granular System, Progress of
Theoretical Physics Supplement 184 (2010) 16–30.
[81] D. S. Bassett, E. T. Owens, M. A. Porter, M. L. Manning, K. E. Daniels, Extraction of force-chain
network architecture in granular materials using community detection, Soft Matter 11 (2015) 2731–
2744.
480 [82] J. Peters, M. Muthuswamy, J. Wibowo, A. Tordesillas, Characterization of force chains in granular
material, Phys. Rev. E 72 (2005) 041307.
[83] K. C. Valanis, A gradient thermodynamic theory of self-organization, Acta Mechanica 127 (1998) 1–23.
[84] C. Rycroft, Multiscale modeling in granular flow, Doctoral dissertation, Massachusetts Institute of
Technology, 2007.

16