Minh, N. H. & Cheng, Y. P. (2013). Geotechnique 63, No. 1, 4453 [http://dx.doi.org/10.1680/geot.10.P.

058]

A DEM investigation of the effect of particle-size distribution on

one-dimensional compression
N. H . M I N H  a n d Y. P. C H E N G 

The effect of particle-size distribution on the one-dimensional compressive behaviour of granular soil
materials was investigated using the discrete-element method (DEM), and the results were compared
with published experimental data with similar gradations. The particles used in this study were
spherical, and their size range mimicked various power-law distributions. The contact formulation was
calibrated such that the different compressibility and initial specific volume of non-uniform assemblies
depended purely on the interaction of different particle sizes of the non-uniform gradations. The
compressive behaviour of these assemblies can be classified as either big-particle-dominated or smallparticle-dominated. The underlying micromechanical explanations for the effect of the particle-size
distribution on the packing characteristics and on the compressibility are presented. Particles involved
in the strong force transmission, which carried larger-than-average contact forces, were found to
contribute to the stiffness of granular samples. When loose and dense samples with the same grading
were compressed to a stage where their stiffnesses were identical, an identical size distribution was
found for those particles involved in the strong force network. For some materials, there existed
regions of particles that did not carry any strong force, and the non-affine movements of these small
particles partially filling the void space between the bigger particles were especially significant. This
behaviour was visualised in the DEM simulations. As these particles rearranged themselves, these
regions contributed to a higher volumetric compression. It was also found that the coordination
number of DEM particles significantly increased with better gradations. A change in the particle-size
distribution of real sand through particle breakage would therefore lead to a higher coordination
number of particles, and it would gradually reduce the probability of particle breakage during the later
stages of the compression, as has been observed in other laboratory experiments.
KEYWORDS: compressibility; discrete-element modelling; fractals; particle-scale behaviour; sands; silts

INTRODUCTION
The behaviour of granular materials is affected by their
particle-size distribution (PSD). Different PSDs can be
obtained in the laboratory as a result of particle breakage,
which occurs in the form of catastrophic splitting during the
compression or shearing of uniformly graded samples
(Cheng et al., 2004, 2005; Bolton et al., 2008). An original
PSD usually evolves toward a more well-graded final distribution. Fig. 1(a) shows that the PSDs of real sands (G1,
G2) have different levels of uniformity, and that these initial
distributions evolve toward different final PSDs after onedimensional compression, as shown in Fig. 1(b). For granular soils subjected to many thousand percent of shear
strain, a critical grading was also found at which no
possible further particle breakage occurred (Coop et al.,
2004). The critical grading can be plotted as a straight line
in a log-log scale, thus indicating a power-law distribution
with a fractal dimension where D 2.57 (Coop et al.,
2004). Altuhafi & Coop (2011) conducted oedometer tests
on granular soils artificially prepared with critical grading
(CG in Fig. 1), and did not observe significant particle
breakage for these soils; rather, the particle surface roughness was instead reduced. For these soils, non-converging
normal compression lines were occasionally observed. This
paper describes DEM simulations conducted using PFC3D

version 4.0 (Itasca, 2008). Assemblies of spherical particles

with different PSDs but no particle breakage were subjected
to one-dimensional compressions. The effect of PSD on the
packing characteristics and compressive behaviour was investigated through micromechanical analyses.
DEM SIMULATION OF DIFFERENT GRANULAR
MATERIALS
The one-dimensional compression simulations were conducted in three stages: (a) calculation of the specific PSDs;
(b) preparation of dense and loose samples using different
frictional coefficients; and (c) one-dimensional compression
until the compression curves of the dense and loose samples
converged.
Calculations of specific PSDs
The uniform and non-uniform DEM PSDs shown in Fig.
1 were created, and the data are listed in Table 1. The PSDs
of the non-uniform samples used in this DEM study satisfy
a power law between the number of particles and the particle
size, with 0.092 mm < d < 1.18 mm. They are calculated
using equation (1) and described in terms of the cumulative
mass (Tyler & Wheatcraft, 1992), which can be plotted as a
straight line in a log-log scale, as in Fig. 1(b).

3 D
M ( L,d)
d

(1)
d max
MT

Manuscript received 25 May 2010; revised manuscript accepted 11

May 2012. Published online ahead of print 14 September 2012.
Discussion on this paper closes on 1 June 2013, for further details see
p. ii.
 Department of Civil, Environmental and Geomatic Engineering,
University College London, UK.

where M( L, d ) is the mass of all particles smaller than d;

dmax is the size of the largest particle; MT is the total mass
44

Non-uniform
Uniform
Real sand

70

Uniform sand

60
G1*
50

D 25

0
001

D 14

01
Particle diameter, d: mm
(a)

Uniform
sand
1

100

D 27

2.4

D 25
D 24

D 21

10
CG

2.3

D 23
S2.0

Fig. 1. Particle-size distributions (PSDs) of DEM samples plotted

in: (a) semi-log scale; (b) log-log scale. G1, G2 and CG are real
silica sand samples from Altuhafi & Coop (2011); S2.0 is a silica
sand sample from Nakata et al. (2001);  samples at beginning and
**samples at end of compression; CG and S2.0 are PSDs that did
not show any significant particle breakage in compression test

of all particles; and D is the fractal dimension (Turcotte,

1986). In this study, D 1.42.7, dmax 1.18 mm, and the
distribution is truncated at a minimum particle size,
dmin 0.092 mm. As shown in Fig. 1(a), two uniformly
graded materials were also created for this study, with one
representing a uniform sand (d50 0.7 mm, Cu 1.4) and
the other representing a uniform silt (only one particle size
d 0.1 mm, Cu 1.0). For comparison, PSDs of some real
sands before and after oedometer tests (after Nakata et al.,
2001; Altuhafi & Coop, 2011) are also plotted in Fig. 1. The
two soils (CG and S2.0) do not show any significant breakage, and hence their PSDs remain the same. Particle breakage does occur in the G1 and G2 soils, however: thus their
final PSDs are then comparable to the PSD of D 2.3 in
Fig. 1(b).
These DEM samples are different from the perfect, selfsimilar, space-filling Apollonian packing. For a perfect pack-

53535
261
21 671
21 932
53535
278
14 217
14 495

53535
292
7688
7980

01
Particle diameter, d: mm
(b)

53535
292
5142
5434

1
001

Box size (mm3 )

Number of big particles (d . 0.4 mm)
Number of small particles (d < 0.4 mm)
Total number of particles

Real sand

G1**

1.9

Non-uniform
D 14

1.6

D 16

1.4

G2**

2.1

D 19

Table 1. Summary of sample size and total number of particles for different DEM materials

Percent equal and finer: %

D 26

53535
195
51 020
51 215

G2*

D 21
D 19
D 16

53535
219
41 066
41 285

S2.0

53535
239
33 145
33 384

CG
10

2.6

D 24
D 23

20

53535
169
63 436
63 605

2.7

D 26

30

53535
139
78 921
79 060

D 27
40

2.5

Percent equal and finer: %

80

15 3 15 3 15
10 680
0
10 680

Uniform silt

90

33333
0
25 783
25 783

Uniform silt

EFFECT OF PARTICLE-SIZE DISTRIBUTION ON ONE-DIMENSIONAL COMPRESSION

100

45

MINH AND CHENG

46

ing, zero porosity can be created by controlling the position

of each particle precisely; the fractal dimension of perfect
packing was calculated to be 2.474 (Borkovec et al., 1994).
However, in this study, the locations of the DEM particles
were randomly packed, and there is a minimum particle-size
limit. The value of dmin was chosen to limit the total number
of the DEM particles, and hence the computational time. It
is, however, much larger than the true comminution limit, as
under compression of real soils it is approximately 1 m
(Kendall, 1978).
The input parameters are listed in Table 2. A linear elastic
contact model (Itasca, 2008) was used, and the stiffness kn
of a particle with size d was calculated using the same
elastic contact modulus, Ec
k n 2Ec d

(2)

Preparation of dense and loose samples

Numerical particles were generated within a cubical space
bounded by rigid walls (100 times the average particle
stiffness of k n ) with an initial porosity of 0.5, and the box
sizes shown in Table 1 were chosen such that the behaviour
was size independent. The randomly generated assemblies
were then isotropically loaded until the mean pressure
reached approximately 100 kPa without gravity. Stress and
strain were calculated from forces and displacements at the
wall boundaries, and differential density scaling (Itasca,
2008) was adopted to reduce the computational time.
The loading was applied in a strain-controlled manner,
corresponding to an overall strain rate of 106 /s. For each
small loading increment, each particle and wall were set to
move towards the centre by assigning a position-dependent
velocity calculated according to its current position and the
specified strain rate. The applied velocity was then set to
zero, and the system was cycled to equilibrium before the
next loading increment. During this sample preparation
stage, the dense samples were prepared using frictionless
particles with 0.0, whereas the loose samples were
prepared using particles with 0.5. Fig. 2 plots the
internal view of some dense samples at the end of this
sample preparation stage. As D increases, the number of
small particles in an assembly increases, and they then
gradually fill the void spaces in Fig. 2(b) and separate the
big particles in Fig. 2(c).
One-dimensional compression
During one-dimensional compression, the interparticle
friction coefficient was reset to 0.5. Vertical compression
was applied in a similar strain-controlled manner with an
overall strain rate of 105 /s; however, the lateral walls were
fixed, and the particles and horizontal walls were moved
towards the horizontal mid-plane. The compression of these
randomly packed granular assemblies is due to the rearrangement of the spherical particles and elastic compression at the contact points. In our simulations, the average
Table 2. Input parameters for three-dimensional DEM simulation
Parameter
Particle density: kg/m3
Particle friction,
Elastic contact modulus, Ec : N/m2
Particle stiffness ratio, ks /kn
Wall friction, wall
Wall stiffness, kwall

Value
2650
0.5
1 3 109
1.0
0.0
100k n (k n : average particle
stiffness)

value of contact overlap was less than 6% throughout the

compression.
GRADINGS AND COMPRESSIBILITY
The one-dimensional compression curves are plotted in
Fig. 3. In these simulations, dense and loose samples of each
DEM material converge at approximately 150200 MPa; the
plots are shown on a semi-log scale (Fig. 3(a)) and on a
natural scale (Fig. 3(b)). The linear part of the curves in the
semi-log plot, where the lines converge, is the normal compression line (NCL), and the slope is the compression index,
Cc : Because of equation (2), the simulated compression
curves for the uniform samples (uniform sand and uniform
silt) are significantly similar, in terms of both the initial void
ratio and the compressibility. This means that any difference
in the compressibility of all other non-uniform DEM samples
is purely the result of the interactions of different particle
sizes. Fig. 3(b) shows that there is a convergence of the
compression curves for the dense and loose samples, but not
for samples with different gradings. The DEM results are in
agreement with experimental data (e.g. Altuhafi & Coop,
2011), which do not show a unique converged NCL for
materials of different PSDs, such as that described by BenNun et al. (2010). This is because the excessive shear strain
required to reach a critical grading could not be satisfied in
a laboratory oedometer test condition (Coop et al., 2004).
The NCLs of some real sands that are slightly less compressible than the DEM materials are shown in Fig. 3(a), but the
data show the same overall trend, as the more uniformly
graded materials that start with a higher specific volume are
more compressible than the better-graded materials.
Impact of grading and initial specific volume
The initial specific volume (v 1 + e) of the numerically
dense samples at the beginning of one-dimensional compression, and their compression index (Cc ) at the final state,
were calculated. These data are plotted in Fig. 4(a) against
the percentage of small particles, as defined in Table 1. The
minimum particle size, dSmin 0.4 mm, of the uniform sand
was used as a delimiting size that separated the notional big
and small particles. These small particles represent the
breakage fragments from a uniform sand sample (e.g. G1) in
the laboratory. The label of the data points shows the
D-value of these samples. As D increases, the specific
volume v first decreases and approaches a minimum value
when D 2.3 (equivalent to 47% of small particles). It then
increases again as D changes from 2.3 to 2.7, implying that
these materials are looser. The DEM materials with D > 2.3
consist of a large number of minimum-size particles, and
their behaviour is more similar to that of gap-graded materials. Lade & Yamamuro (1997) referred to the first packing
regimes as sand dominated, as only a small quantity of silt
particles partially fills the void space (see Fig. 2(a)). The
second packing regime is silt dominated, as the large
quantity of silt particles separates the bigger sand particles
(see Fig. 2(c)), and the silt particles play a more prominent
role in controlling the behaviour of the whole system. These
particle interactions can be seen explicitly in Fig. 2. The
lowest degree of porosity is achieved when the sand particles
are still in contact with each other and the void spaces are
reasonably filled with the dense randomly packed silt particles (see Fig. 2(b)). The transition is also controlled by the
PSDs of sand and silt, and the characteristics of these
particles. Thus a perfect Apollonian packing of zero porosity
is not always applicable to real soils.
Experimental data for sandsilt mixtures show similar
transitions when the silt content is between 25% and 45%

EFFECT OF PARTICLE-SIZE DISTRIBUTION ON ONE-DIMENSIONAL COMPRESSION

47

18
G1
Uniform
sand

17

x
z

Specific volume, ( 1 e)

16

Uniform silt

D 14
D 27

15

14

D 23

13

12

Non-uniform

11

Uniform

G2

Real sand
10
001

01

CG
1
10
Vertical stress: MPa
(a)

(a)

100

1000

18

17
Uniform
sand

x
z

Specific volume, ( 1 e)

16

Uniform silt

15

14
D 27

13

D 14

12
Non-uniform

D 23

11
Uniform
10
(b)

50

200
100
150
Vertical stress: MPa
(b)

250

300

Fig. 3. Simulated one-dimensional compression curves of DEM

samples plotted in: (a) semi-log scale; (b) natural scale. G1, G2
and CG are normal compression lines of real sand samples from
Altuhafi & Coop (2011)

x
z

(c)

Fig. 2. Internal views of some numerical samples at end of the

sample preparation stage (dense samples): (a) D 1.4;
(b) D 2.3; (c) D 2.7

(Polito & Martin, 2001). Fig. 4(a) also shows the data for
N50/200 silty sand (Lade & Yamamuro, 1997), with a mean
grain size (d50 )sand of 0.2 mm and (d50 )silt of 0.06 mm, and
the data for the Stava tailing (Carrera et al., 2011), with
(d50 )sand 0.19 mm and (d50 )silt 0.03 mm.
For these simulations, the variation of the compression
index in Fig. 4(a) follows the same trend as that observed
for the specific volume, but attains its minimum value when
D 2.1. A similar trend was found in the experimental
results for the Stava tailings (Carrera et al., 2011). This type
of transition is not the same as the non-converging transitional behaviour defined by Altuhafi & Coop (2011). The
value of Cc would be better correlated to the uniformity
coefficient Cu of the PSD, as shown in Fig. 4(b). The
experimental data for the G1, G2 and CG soils (of the
Leighton Buzzard sand by Altuhafi & Coop, 2011) in Fig.
4(b) show that soils of better gradations (larger Cu ) would
be denser (smaller ini ) and less compressible (smaller Cc ),
similar to the DEM results when D < 2.3. For an increasing

MINH AND CHENG

48
06

18
Cc (DEM, D 1421)
Cc (DEM, D 2327)
17

(Label: D-value)
14
16

04

27
19

14

16

26
24

03

25
15

21 23
16
27

02
25
21 23 24

v: DEM samples

01

14

26

19

Specific volume, v ( 1 e)

Compression index, Cc

05

13

v: real soil (Stava)

v: real soil (N50/200)
0

10

20

30 40 50 60
% of small particles
(a)

70

80

12

07
DEM, uniform samples

Compression index, Cc

163

05

DEM, D 1421

161

06

DEM, D 2327
Sand

Real soil (S2.0)

Silt

Real soil (G1, G2, CG)

159
G1

04

03

155
143

(Label: v, initial specific volume)

148
140
139
137
134
134
133
128
137

02

G2

S2.0
123
CG

01

10
Cu ( d60 /d10)

100

(b)

Fig. 4. (a) Effect of percentage of small particles on compression

index Cc and on minimum initial specific volume for DEM
samples; the label shows the value of D. The N50/200 sandsilt
mixture (Lade & Yamamuro, 1997) and Stava tailing (Carrera et
al., 2011) show similar variation of with silt content.
(b) Variation of compression index with coefficient of uniformity
Cu for real sand samples (G1, G2 and CG, Altuhafi & Coop, 2011;
S2.0, Nakata et al., 2001) and for DEM samples. Label shows
value of minimum initial specific volume of samples

value of D > 2.3, the DEM materials produce an inversion

of the trend observed (smaller Cu and larger Cc ). This
implies that these samples are actually more uniformly
graded, which is the effect of the truncation of the PSD. The
material with, for example, D 2.6 is not directly comparable to the CG soil with the critical grading in the experiment. However, it would be relevant to compare the results
from the microscopic analysis of the uniform sand material
and those materials where D < 2.3 with the behaviours of
the real G1 and G2 soils.
Impact of strong and weak networks
Owing to the discrete nature of granular materials, particles carrying larger-than-average contact forces form a

force network that has been found to be responsible for the

shear strength of such materials (Radjai et al., 1998). This
strong force network is also considered to constitute the
backbone of the system (Thornton & Antony, 2000). In
this DEM study, with its wide range of PSDs, the present
authors propose to describe the strong network at a stress
state in terms of the size distribution of the particles carrying strong forces. Thus it is called the strong-network
particle-size distribution, or SN-PSD. Fig. 5(a) shows the
calculated SN-PSDs at various vertical stresses of the dense
(DS) and loose (LS) samples with D 1.9. Fig. 5(a) shows
that at relatively small stresses with v , 50 MPa (where the
compression curves have not yet converged), there is a
significant difference between the SN-PSD curves for loose
and dense samples. As the vertical stress increases during
one-dimensional compression, more particles carry strong
forces and become involved in strong force transmission. At
relatively large stresses with v  200 MPa, the SN-PSD
curves overlap. This happens at the same time that the dense
and loose samples show an identical stiffness, thus resulting
in the same gradient (Cc ) at the NCL. In other words, the
NCL stiffness is closely linked to the size distribution of
particles involved in the strong force network. In another
series of DEM simulations of gap-graded mixtures (Minh &
Cheng, 2010), it was found that granular materials have the
same final SN-PSD at high stress levels even though different total PSDs yielded the same value of Cc :
The PSDs and final SN-PSDs of the various materials are
plotted together in Fig. 5(b). The difference between each
PSD curve and its corresponding SN-PSD curve in the large
particle size range is usually small. This implies that the
large particles are usually involved in strong force transmission, regardless of the size distribution of the assembly. To
illustrate this point for D 1.4, the distance between the
two distribution curves at d50 is only 0.05 mm
(d50(PSD)  0.82 mm; d50(SN-PSD)  0.77 mm), and this distance is even smaller for other non-uniform materials, as
shown in Fig. 5(b). This agrees with the results for a
polydisperse granular medium published by Voivret et al.
(2009), who suggested that strong force chains preferentially
pass through larger particles.
This then implies that the area between a pair of PSD
and SN-PSD curves (Aw ), similar to that shown in Fig. 5(a)
for the case of D 1.9, represents the small particles of
the material that are excluded from the strong force transmission network. These particles are either lightly loaded
between adjacent strong force chains or are those that
partially fill the void space, which carry only weak forces
or do not transmit any force. The uninvolved particles are
different from other small particles that allow the transmission of both the strong and weak forces. It was proposed
previously by Radjai (2008) that regions containing weak
contacts are more susceptible to failure, and that such
regions encourage the local rearrangement of particles bearing weak forces. Particles represented by the area Aw are
therefore more likely to rearrange, and are considered to be
the source of weakness in the system. The higher compressive volumetric strain of a loose sample as compared with
a dense sample could be explained by the larger area Aw of
the loose sample at the beginning of compression, as shown
in Fig. 5(a), where v , 50 MPa. As the difference in the
specific volume of dense and loose samples disappears
when the compression curves converge (Fig. 3), the areas
Aw between their corresponding PSD curves and their SNPSD curves also become the same. If the void ratio (and
hence the specific volume) of a granular material indicates
the available void space for the volumetric compression of
a sample, then the area Aw of the distribution curves is a
property of the solid phase that represents the weak regions

EFFECT OF PARTICLE-SIZE DISTRIBUTION ON ONE-DIMENSIONAL COMPRESSION

100

10

D 23 (initial)

SN-PSD, v 233 MPa (LS)

80

SN-PSD, v 50 MPa (DS)

70

SN-PSD, v 31 MPa (LS)

60

D 23 (final)

SN-PSD, v 203 MPa (DS)

Particle-size-related coordination number, Zd

Percent equal and finer: %

90

49

SN-PSD, v 48 MPa (DS)

SN-PSD, v 45 MPa (LS)

50
40
30

PSD (D 19)

20

Final Aw
Final SN-PSD

10

102

D 21

D 27 (final)
D 27 (initial)

101

100

Uniform
sand

D 27

D 23

Sand (final)
101

Sand (initial)

D 14

D 14 (final)
D 14 (initial)

102

D 21 (final)
D 21 (initial)

0
001

01
Particle diameter, d: mm
(a)

10

100

01

Particle diameter, d: mm

Fig. 6. Particle-size-related coordination number Zd of dense

samples at initial and final states

PSD

90

SN-PSD

Percent equal and finer: %

80
70
60
50
D 27
40
30

D 25

Uniform sand

D 23

SN-PSD

PSD

20
10
D 14

D 19
0

01

1
Particle diameter, d: mm
(b)

Fig. 5. (a) Evolution of strong-network particle-size-distribution

(SN-PSD) with stress level of dense (DS) and loose (LS) samples,
where D 1.9; (b) particle-size distribution (PSD) and final SNPSD of DEM samples

of instability inside the assembly (Staron et al., 2002;

Radjai, 2008).
GRADINGS AND PACKING
Particle-size-related coordination number
To examine the impact of having a wide range of particle
sizes, particles of these DEM granular materials were sorted
according to their size. For each size increment from d i to
d i1 (d i /d i1 1.2), the number of particles and their contacts were counted. The number of contacts divided by the
number of particles within the size increment becomes the
particle-size-related coordination number (Zd ). The values of
Zd for several dense systems at the initial and final states
were calculated, and are plotted against the mid-point of size
increments in Fig. 6.
As each sample compresses and the volume decreases
(Fig. 3), the DEM spherical particles rearrange, and the
particles in the void space may come into contact with other
particles. This results in new contacts, as shown in Fig. 6,

where the final Zd curve is above the initial Zd curve. For

example, the uniform sand sample has the coordination
number Zd 310 initially and reaches Zd 515 at the
final state of compression. The increase of Zd with stress
level is the most significant for the material with D 1.4,
because this sample has the most significant number of
particles with very few initial contacts (which can be seen in
Fig. 7) and a correspondingly large Aw , between the specific
PSD (solid) line and the SN-PSD (dotted) line of this
material in Fig. 5(b). Because the particles of other materials
connect better with one another, even in the initial state, the
effect on Zd of increasing the stress level for these materials
is therefore less significant, as shown in Fig. 6.
In general, Zd increases with particle diameter, which
means that the value of Zd of the bigger particles is
generally greater than that of the smaller particles for any
non-uniform assembly. However, this also depends on the
size distribution. For the case where D 2.7, the value of
Zd of the biggest particles can be as large as 400, whereas
that of the smallest particles (d 0.092 mm) is only approximately 6. For materials with an abundance of small
particles, their coordination numbers are bounded such that
those of the biggest size particles should approach a constant
value when they are fully covered by the smallest particles,
whereas those of the smallest particles should approach that
of the randomly packed uniform silt sample.
The present data can be compared with those gathered
from the oedometer experiments of Altuhafi & Coop (2011).
For example, the PSD of their G2 sample moves from G2
PSD (corresponding to PSD with D 1.4) to a better-graded
G2** PSD (corresponding to PSD with D 2.3). This shift
of the PSD curve implies a similar shift of the coordination
number profile, as shown in Fig. 6. Nakata et al. (2001) also
observed that the frequency of the breakage events decreases
in the later stages of the compression experiments, which
should increase the coordination number due to breakage.
As the particle coordination numbers increase, particles are
subjected to an increased isotropic stress condition, and are
therefore less prone to breakage (Tsoungui et al., 1999).
Although the size distribution of the present studys DEM
material with D 2.7 represents only part of the critical
graded CG soil (Fig. 1(b)), the value of Zd in this case is
already much higher than other cases. Continuously adding

MINH AND CHENG

50
6

Contact-type-related coordination number, Zc

Overall

Overall

b_b

(3)

b_s
s_s
s_s

b_b
1

b_s

14

16

18

20

22

24

26

28

D
(a)
8
Overall

Contact-type-related coordination number, Zc

Overall

b_b
b_s

s_s

s_s

b_s
b_b

14

16

2C
2Cb_b
2Cb_s
2Cs_s
; Z b_b
; Z s_s

; Z b_s
c
c
c
N
Nb
N
Ns

18

20

22

24

26

28

D
(b)

Fig. 7. Effect of PSD on variation of contact-type-related coordination numbers of dense samples: (a) at initial state; (b) at
final state

finer particles to such a system would further improve the

effect of the coordination number, and hence bring the
system to a saturation regime (Tsoungui et al., 1999): this is
the regime where fragmentation of the particles would not
occur. The condition of a perfect fractal distribution may not
be required, as neither the S2.0 soil nor the CG soil showed
any significant breakage under compression tests (Nakata et
al., 2001; Altuhafi & Coop, 2011).
Contact-type-related coordination numbers
To study further the interaction between different particle
sizes, three types of force-transmitting contact are defined:
bigbig (Cb_b ), smallsmall (Cs_s ) and bigsmall (Cb_s ),
where the subscripts b and s represent the big and small
particles respectively, as shown in Table 1. The coordination
number can be related to the contact type, and calculated in
similar fashion to the conventional overall coordination number (Z), as

where C is the number of force-transmitting contacts and N

( Nb + Ns ) is the total number of particles. Classifying the
coordination number by the contact type would actually lead
to four different contact-type-related coordination numbers
(Martin & Bouvard, 2003). In principle, the number of
contacts between big and small particles can be divided
differently by Nb and Ns : For simplification, this is divided
by the total number of particles N, thus yielding only one
bigsmall coordination number Z b_s
c :
The coordination numbers in equation (3) were calculated
for each dense sample at the initial state (Fig. 7(a)) and at
the final state (Fig. 7(b)) of the one-dimensional compression. For materials where D 1.41.6 in Fig. 7(a), small
particles that partially fill the void space in between the big
particles have very few contacts. This results in small values
of the overall coordination number (Z) and the smallsmall
b_b
is approxicoordination number ( Z s_s
c ). The value of Z c
mately 3 at the initial state (Fig. 7(a)), which is still smaller
than the overall coordination number of a uniform sample
(Z  5.2), and also smaller than the minimum stability
requirement (Z 4) for frictional spheres (Hecke, 2010).
This implies that removing the small particles from these
systems will still lead to a disturbance of the bigbig
contact network. Although small particles do not have a
dominant role, they do contribute to the overall stability for
systems where D 1.41.6.
In Figs 7(a) and 7(b), the bigbig contact coordination
decreases with increasing D as the granular
number Z b_b
c
material becomes small-particle dominated, while the small
shows a reverse
small contact coordination number Z s_s
c
trend. The bigsmall contact coordination number Z b_s
c
reaches a peak near where Z b_b
and Z s_s
cross over. The
c
c
present studys DEM results support the two packing regimes
proposed by Lade & Yamamuro (1997), in which particlepacking systems transit from a big-particle-dominated behaviour to a small-particle-dominated behaviour. Using DEM
simulations, the transitional behaviour can be explained in
terms of the crossover of the micromechanical contact-typerelated coordination numbers. The micromechanical transition occurs for those materials with D 2.3 at the initial
state (Fig. 7(a)) and those materials with D 2.1 at the end
of compression (Fig. 7(b)). Both of these findings match the
macroscopic observations in Fig. 4(a), yielding a minimum
initial specific volume when D 2.3, and a minimum
compression index Cc when D 2.1. The shift of the
transitional point between the initial and final states is due
to the added participation of the small particles at higher
stress levels.
Movement of small particles in polydisperse assemblies
Movement of small particles can be important for the
behaviour of sand-dominated materials, and for the transition
mentioned in the previous section. Upon shearing, the
smaller grains positioned between two bigger sand particles
tend to slide into the still largely empty void spaces, hence
leading to more contractive behaviour (Lade & Yamamuro,
1997; Yamamuro et al., 2008). The movement of these small
particles is non-affine, as it deviates from the overall applied
strain field. Non-affine movement usually occurs at low-level
stress ranges when granular material is in a loose state with
a low coordination number (Storakers et al., 1999; Hecke,
2010), and this is due to the tendency of the particles to roll
past each other as they rearrange. Non-affine movement can
also be constrained by a large number of neighbouring

EFFECT OF PARTICLE-SIZE DISTRIBUTION ON ONE-DIMENSIONAL COMPRESSION

particles that prevent local rearrangement. This means that
these non-affine movements would be less significant for
granular materials with a higher overall coordination number, that is, a higher D-value in this study (see Figs 6 and
7). In DEM simulations, particle movements can be visualised. Fig. 8 plots the displacement vectors of some of the
small particles with diameters ranging from 0.19 mm to
0.23 mm inside the assemblies of three different materials at
the final state. It is clear that for the granular material with
D 1.4 the movement is non-affine, as the displacement
vectors are randomly oriented. For the granular materials
with D 2.3 and D 2.7, the orientation of the vectors
becomes more affine, thus following the direction of the
applied vertical strain field. In particular, for D 2.7
the displacement vectors are almost perfectly aligned with
the vertical compressive strain.
Non-affinity in particle movement also leads to less
uniformity of contact forces (Storakers et al., 1999; Martin
et al., 2003; Hecke, 2010). Therefore the movement of the
small particles in these DEM simulations was investigated,
in detail, by checking the uniformity of the smallsmall
contact forces at different stress levels, and the following
micro-parameter by Makse et al. (2000) was adopted,
!1
M
X
2
qi
(4)
M

06

D 27

05

D 23

04

03

02
001

D 14

01

1
10
Vertical stress: MPa

100

1000

Fig. 9. Variation in the uniformity of smallsmall normal contact

forces with stress level of dense samples

In Fig. 9, there is also a clear difference among these

three materials in terms of the variation of the participation
number () with the stress level. For D 2.7, which is
small-particle dominated, the initial change in the value of
at v , 3 MPa implies particle movements and rearrangements to attain a more optimal configuration before it
reaches a constant value of . For D 2.3, a continuous
increase in reflects a gradual reduction of particle
rearrangement. For D 1.4, however, the value of decreases greatly and continuously after reaching approximately 1 MPa. This indicates that non-affine particle
rearrangement continues until the end of the compression,
unlike that in the other two materials. This is because the
final area Aw of this sample (see Fig. 5(b)) remains relatively
large, even at the end of the compression: hence this
confirms that the existence of the weak regions encourages
particle rearrangement, even at high stress levels.

i1

where is the participation number, M is the

P total number
of force-transmitting contacts, and q i f i = M
j1 f j , with fi
being the magnitude of total force at contact i. Hence 1
indicates a spatially homogeneous force distribution (i.e.
q i 1/M, 8i; Makse et al., 2000). As the dependence of the
tangential force distribution on the stress level is much
weaker than that of the normal contact force distribution
(Zhang & Makse, 2005), only the normal contact forces of
smallsmall contacts are considered. Parameters M and fi in
equation (4) were substituted by the total number of small
small contacts Cs_s and the magnitude of the normal contact
n
) i at the ith smallsmall contact respectively. The
force ( f s_s
result, plotted in Fig. 9, shows the level of uniformity of the
smallsmall normal contact forces for different materials at
the considered stress level. While a non-affine movement of
these particles is associated with a lower uniformity of these
contact forces, and vice versa, the increasing quantity of
small particles restricts the non-affine rearrangement, and
hence improves the uniformity of smallsmall contact
forces. This explains the different levels of uniformity
observed at the final states of the three samples shown in
Fig. 9: D 1:4 , D 2:3 , D 2:7 :

(a)

51

07

CONCLUSIONS
As the packing and compressibility of a granular assembly
are significantly dependent on the distribution of particle
sizes, the authors attempted to investigate these issues
through numerical DEM simulations using Itasca PFC3D.

(b)

(c)

Fig. 8. Displacement vectors of particles with diameter ranging from 0.19 mm to 0.23 mm at final state of dense samples: (a) D
(b) D 2.3; (c) D 2.7

1.4;

MINH AND CHENG

52

The simulation results showed that materials of better gradations are both denser and less compressible than uniformly
graded materials, which agrees with the experimental results
reported by others. These DEM samples reported herein are,
however, different from some special real soil samples with
critical grading that show non-converging NCLs for loose
and dense samples (e.g. Altuhafi & Coop, 2011).
Compressibility is linked to particles carrying larger-thanaverage contact force inside the assembly. For each granular
material with a specific PSD, as the compression curves of
loose and dense samples converge, the PSDs of the strong
network for these two samples also converge to the same
distribution. This could be related to the compression index.
In another dataset (Minh & Cheng, 2010), the present
authors found that materials with different total PSDs but
the same final SN-PSD would produce the same value of Cc :
The higher compressive volumetric strain of the loose samples can also be explained through the particles that are not
involved in the strong force network, as these weak regions
inside a granular assembly may encourage particle rearrangement. The authors found that a loose sample has a larger
proportion of these uninvolved particles compared with
that of a dense sample at the beginning of the compression.
Furthermore, significant non-affine displacements associated
with particle rearrangement were shown visually for the
material with D 1.4, where weak regions still exist even in
the final states. For materials with D 2.3 to D 2.7,
because most of their particles carry a strong force, rearrangement is limited and particle movements are affine, thus
following the direction of the overall applied strain field.
Because of the truncation of the power-law distribution in
the present studys simulations, the behaviour of the DEM
materials is similar to the behaviour of the sandsilt mixtures reported by Lade & Yamamuro (1997), which involved
a mechanism of two different packing regimes dominated by
two different particle-size ranges. The transition from the
sand-dominated regime to the silt-dominated regime occurs
at a minimum specific volume, which can be calculated
explicitly through the micromechanical analysis of the present studys DEM systems. At the transitional point, the
coordination number of the smallsmall contacts becomes
greater than that of the bigbig contacts, and the bigsmall
coordination number also reaches a maximum value. The
uniformity of contact forces thereby increases and rearrangement decreases.
The coordination number of the DEM systems increases
greatly for better gradations. This implies that a change in
the PSD of real sand through particle breakage would lead
to a higher coordination number of the particles, hence
reducing the probability of particle breakage. This should
explain why there was no observable breakage for the
critical grading in some experiments (e.g. Altuhafi & Coop,
2011).

ACKNOWLEDGEMENTS
The authors gratefully acknowledge the sponsor of an
EPSRC grant (EP/F036973/1) for this research project. They
also thank Professor Matthew Coop and Dr Colin Thornton
for their valuable suggestions, which have helped to improve
the quality of the paper significantly.

NOTATION
Aw
C
Cb_b
Cb_s

area between distribution curves representing uninvolved

particles
number of force-transmitting contacts
number of force-transmitting bigbig contacts
number of force-transmitting bigsmall contacts

number of force-transmitting smallsmall contacts

compression index
coefficient of uniformity
fractal dimension
particle size
size of largest particle
minimum particle size
minimum particle size of uniform sand
particle size at 10% finer of size distribution curve
particle size at 50% finer of size distribution curve
particle size at 60% finer of size distribution curve
elastic contact modulus
void ratio
magnitude of total force at contact i
magnitude of total force at contact j
magnitude of normal contact force at ith smallsmall
contact
kn particle stiffness
k n average particle stiffness
kwall wall stiffness
M total number of force-transmitting contacts
M( L, d ) mass of all particles smaller than d
MT total mass of all particles
N total number of particles
Nb number of big particles
Ns number of small particles
v specific volume
vini initial specific volume
Zc contact-type-related coordination number
Z b_b
bigbig contact coordination number
c
Z b_s
bigsmall contact coordination number
c
Z s_s
smallsmall contact coordination number
c
Zd particle-size-related coordination number
particle friction
wall wall friction
participation number
v vertical stress
Cs_s
Cc
Cu
D
d
dmax
dmin
dSmin
d10
d50
d60
Ec
e
fi
fj
n
( f s_s
)i

REFERENCES
Altuhafi, F. N. & Coop, M. R. (2011). Changes to particle characteristics associated with the compression of sands. Geotechnique
61, No. 6, 459 471, http://dx.doi.org/10.1680/geot.9.P.114.
Ben-Nun, O., Einav, I. & Tordesillas, A. (2010). Force attractor in
confined comminution of granular materials. Phys. Rev. Lett.
104, No. 10, 1013.
Bolton, M. D., Nakata, Y. & Cheng, Y. P. (2008). Micro and macro
mechanical behaviour of DEM crushable materials. Geotechnique 58, No. 6, 471480, http://dx.doi.org/10.1680/geot.
2008.58.6.471.
Borkovec, M., Paris, W. D. & Peikert, R. (1994). The fractal
dimension of the Apollonian sphere packing. Fractals 2, No. 4,
521526.
Carrera, A., Coop, M. & Lancellotta, R. (2011). Influence of
grading on the mechanical behaviour of Stava tailings. Geotechnique 61, No. 11, 935946, http://dx.doi.org/10.1680/geot.
9.P.009.
Cheng, Y. P., Bolton, M. D. & Nakata, Y. (2004). Crushing and
plastic deformation of soils simulated using DEM. Geotechnique
54, No. 2, 131141, http://dx.doi.org/10.1680/geot.2004.54.
2.131.
Cheng, Y., Bolton, M. D. & Nakata, Y. (2005). Grain crushing and
critical states observed in DEM simulations. Proc. 5th Int. Conf.
on Micromechanics of Granular Media, Stuttgart, 13931397.
Coop, M. R., Sorensen, K. K., Freitas, T. B. & Georgoutsos, G.
(2004). Particle breakage during shearing of a carbonate sand.
Geotechnique 54, No. 3, 157163, http://dx.doi.org/10.1680/
geot.2004.54.3.157.
Hecke, M. V. (2010). Jamming of soft particles: geometry, mechanics, scaling and isostaticity. J. Phys.: Condens. Matter 22,
033101.
Itasca (2008). Particle flow code in 3 dimensions (PFC3D) version
4. Minneapolis, MN, USA: Itasca Consulting Group, Inc.
Kendall, K. (1978). The impossibility of comminuting small particles by compression. Nature 272, 710711.

EFFECT OF PARTICLE-SIZE DISTRIBUTION ON ONE-DIMENSIONAL COMPRESSION

Lade, P. V. & Yamamuro, J. A. (1997). Effects of nonplastic fines
on static liquefaction of sands. Can. Geotech. J. 34, No. 6,
918928.
Makse, H. A., Johnson, D. L. & Schwartz, L. M. (2000). Packing
of compressible granular materials. Phys. Rev. Lett. 84, No. 18,
41604163.
Martin, C. L. & Bouvard, D. (2003). Study of the cold compaction
of composite powders by the discrete element method. Acta
Mater. 51, No. 2, 373386.
Martin, C., Bouvard, D. & Shima, S. (2003). Study of particle
rearrangement during powder compaction by the discrete element method. J. Mech. Phys. Solids 51. No. 4, 667693.
Minh, N. H. & Cheng, Y. P. (2010). One dimensional compression
of gap-graded mixtures in DEM. Proceedings of the international symposium on geomechanics and geotechnics: From micro
to macro, Shanghai, pp. 727731.
Nakata, Y., Hyde, A. F. L., Hyodo, M., Kato, Y. & Murata, H.
(2001). Microscopic particle crushing of sand subjected to high
pressure one-dimensional compression. Soils Found. 41, No. 1,
6982.
Polito, C. P. & Martin, J. R. (2001). Effects of nonplastic fines on
the liquefaction resistance of sands. J. Geotech. Geoenviron.
Engng 127, No. 5, 408415.
Radjai, F. (2008). Particle-scale origins of shear strength in granular
media. arXiv, 0801.4722v1.
Radjai, F., Wolf, D., Jean, M. & Moreau, J.-J. (1998). Bimodal

53

character of stress transmission in granular packings. Phys. Rev.

Lett. 80, No. 1, 6164.
Staron, L., Vilotte, J.-P. & Radjai, F. (2002). Preavalanche instabilities in a granular pile. Phys. Rev. Lett. 89, No. 20, 14.
Storakers, B., Fleck, N. A. & McMeeking, R. M. (1999). The
viscoplastic compaction of composite powders. J. Mech. Phys.
Solids 47, No. 4, 785815.
Thornton, C. & Antony, S. J. (2000). Quasi-static shear deformation
of a soft particle system. Powder Technol. 109, No. 13, 179
191.
Tsoungui, O., Vallet, D. & Charmet, J.-C. (1999). Numerical mode
of crushing of grains inside two-dimensional granular materials.
Powder Technol. 105, No. 13, 190198.
Turcotte, D. L. (1986). Fractals and fragmentation. J. Geophys. Res.
91, No. B2, 19211926.
Tyler, S. W. & Wheatcraft, S. W. (1992). Fractal scaling of soil
particle size distributions: analysis and limitations. Soil Sci. Soc.
Am. J. 56, No. 2, 362369.
Voivret, C., Radjai, F., Delene, J. Y. & El Youssoufi, M. S. (2009).
Multiscale force networks in highly polydisperse granular media.
Phys. Rev. Lett. 102, No. 17, 178001.
Yamamuro, J. A., Wood, F. M. & Lade, P. V. (2008). Effect of
depositional method on the microstructure of silty sand. Can.
Geotech. J. 45, No. 11, 15381555.
Zhang, H. & Makse, H. (2005). Jamming transition in emulsions
and granular materials. Phys. Rev. E 72, No. 1, 112.