REPORT NO.

UCD/CGM-23/01 CENTER FOR GEOTECHNICAL MODELING

PM4SAND (VERSION 3.3):

A SAND PLASTICITY MODEL FOR
EARTHQUAKE ENGINEERING
APPLICATIONS

BY

R. W. BOULANGER

K. ZIOTOPOULOU

DEPARTMENT OF CIVIL & ENVIRONMENTAL ENGINEERING

COLLEGE OF ENGINEERING
UNIVERSITY OF CALIFORNIA AT DAVIS

June 2023
(revised October 2023)
 PM4Sand (Version 3.3):
A Sand Plasticity Model for Earthquake Engineering Applications

by

Ross W. Boulanger

and

Katerina Ziotopoulou

Report No. UCD/CGM-23/01

Center for Geotechnical Modeling

Department of Civil and Environmental Engineering
University of California
Davis, California

June 2023
(revised October 2023)
First Edition (PM4Sand Version 1) June 2010

Second Edition (PM4Sand Version 2) May 2012

Revised July 2012

Third Edition (PM4Sand Version 3) March 2015

Revised (PM4Sand Version 3.1) October 2017

Revised January 2018

Revised March 2018

Revised July 2018

Revised September 2018

Revised (PM4Sand Version 3.2) July 2022

Revised (PM4Sand Version 3.3) June 2023

Compiled as modelpm4sand005_64.dll for FLAC 8.1 (Itasca 2019)
Compiled as cmodelPM4Sand2D009.dll for FLAC2D 9.00 (Itasca 2023)

Revised October 2023

ABSTRACT

The sand plasticity model PM4Sand for geotechnical earthquake engineering applications is
presented. The model follows the basic framework of the stress ratio-controlled, critical state-based,
bounding surface plasticity model for sand presented by Dafalias and Manzari (2004). Modifications
to the model were developed and implemented by Boulanger (2010, Version 1), Boulanger and
Ziotopoulou (2012, Version 2), Boulanger and Ziotopoulou (2015, Version 3; 2017, Version 3.1; 2022,
Version 3.2) and further herein (Version 3.3) to improve its ability to approximate the stress-strain
responses important to geotechnical earthquake engineering applications; in essence, the model was
calibrated at the equation level to provide for better approximation of the trends observed across a set
of experimentally- and case history-based design correlations. These constitutive modifications
included: revising the fabric formation/destruction to depend on plastic shear rather than plastic
volumetric strains; adding fabric history and cumulative fabric formation terms; modifying the plastic
modulus relationship and making it dependent on fabric; modifying the dilatancy relationships to
include dependence on fabric and fabric history, and to provide more distinct control of volumetric
contraction versus expansion behavior; providing a constraint on the dilatancy during volumetric
expansion so that it is consistent with Bolton’s (1986) dilatancy relationship; modifying the elastic
modulus relationship to include dependence on stress ratio and fabric history; modifying the logic for
tracking previous initial back-stress ratios (i.e., loading history effect); recasting the critical state
framework to be in terms of a relative state parameter index; simplifying the formulation by restraining
it to plane strain without Lode angle dependency for the bounding and dilatancy surfaces; incorporating
a methodology for improved modeling of post-liquefaction reconsolidation strains; and providing
default values for all but three primary input parameters. The model is coded as a user defined material
in a dynamic link library (DLL) for use with the commercial program FLAC 8.1 (Itasca 2019) and
FLAC2D 9.00 (Itasca 2023). The numerical implementation and DLL module are described. The
behavior of the model is illustrated by simulations of element loading tests covering a broad range of
conditions, including drained and undrained, cyclic, and monotonic loading under a range of initial
confining and shear stress conditions, which can then be compared to typical design relationships. The
model is shown to provide reasonable approximations of desired behaviors and to be relatively easy to
calibrate.

i
TABLE OF CONTENTS

1. INTRODUCTION ........................................................................................................................................... 1
1.1 Revisions .....................................................................................................................................3

2. MODEL FORMULATION ........................................................................................................................... 16

2.1 Basic stress and strain terms ......................................................................................................16
2.2 Critical state ...............................................................................................................................17
2.3 Bounding, dilatancy, and critical surfaces.................................................................................18
2.4 Yield surface and image back-stress ratio tensors.....................................................................19
2.5 Stress reversal and initial back-stress ratio tensors ...................................................................20
2.6 Elastic strains and moduli..........................................................................................................21
2.7 Plastic components without fabric effects .................................................................................23
2.8 Fabric effects .............................................................................................................................30
2.9 Post-shaking reconsolidation .....................................................................................................38
2.10 Summary of constitutive equations .........................................................................................39

3. MODEL IMPLEMENTATION ..................................................................................................................... 58

3.1 Aspects of FLAC's numerical approach .....................................................................................58
3.2. Implementation of PM4Sand in FLAC .....................................................................................59
3.3 Effect of time step size on element responses ............................................................................61
3.4 Effect of time step size on system responses .............................................................................62
3.5 DLL module ...............................................................................................................................62
3.6 Additional notes on use in boundary value problem simulations .............................................63

4. MODEL INPUT PARAMETERS AND RESPONSES ................................................................................ 73

4.1 Model input parameters .............................................................................................................73
4.2 Example calibration and model responses for a range of element loading conditions .............80
4.3 Example calibration with user-defined critical state undrainded shear strength .......................84

5. CONCLUDING REMARKS ....................................................................................................................... 110

ACKNOWLEDGMENTS ................................................................................................................................ 112
REFERENCES ................................................................................................................................................. 112

ii
 PM4Sand (Version 3.3):
A Sand Plasticity Model for Earthquake Engineering Applications

1. INTRODUCTION

Nonlinear deformation analyses for problems involving liquefaction are increasingly common in
earthquake engineering practice. Constitutive models for sand that have been used in practice range
from relatively simplified, uncoupled cycle-counting models to more complex plasticity models (e.g.,
Wang et al. 1990, Cubrinovski and Ishihara 1998, Dawson et al. 2001, Papadimitriou et al. 2001, Yang
et al. 2003, Byrne et al. 2004, Dafalias and Manzari 2004, Tasiopoulou and Gerolymos 2016,
Khosravifar et al. 2018, Liu et al. 2020, Yang et al. 2022). Each constitutive model has certain
advantages and limitations that can be illustrated for potential users by documents showing the
constitutive response of the model in element tests that cover a broad range of the conditions that may
be important to various applications in practice (e.g., Beaty 2009). Considerations in selecting and
calibrating constitutive models for nonlinear dynamic analyses and the documentation of results are
discussed in Boulanger and Beaty (2016), Boulanger and Ziotopoulou (2018), and Boulanger 2022).

The information available for the calibration of constitutive models in design practice most
commonly includes basic classification index tests (e.g., grain size distributions), penetration
resistances (e.g., SPT or CPT), and shear wave velocity (Vs) measurements. More detailed laboratory
tests, such as triaxial or direct simple shear tests, are almost never available due to the problems with
overcoming sampling disturbance effects and the challenge of identifying representative samples from
highly heterogeneous deposits.

Constitutive models for geotechnical earthquake engineering applications must be able to

approximate a broad mix of conditions in the field. For example, a single geotechnical structure like
the schematic earth dam shown in Figure 1.1 can have strata or zones of sand ranging from very loose
to dense under a wide range of confining stresses, initial static shear stresses (e.g., at different points
beneath the slope), drainage conditions (e.g., above and below the water table), and loading conditions
(e.g., various levels of shaking). The engineering effort is greatly reduced if the constitutive model can
reasonably approximate the predicted stress-strain behaviors under all these different conditions. If the
model cannot approximate the trends across all these conditions, then extra engineering effort is
required in deciding what behaviors should be prioritized in the calibration process, and sometimes by
the need to repeat the calibrations for the effects of different initial stress conditions within the same
geotechnical structure.

The PM4Sand (Version 3.3) plasticity model for geotechnical earthquake engineering applications
is presented herein. The PM4Sand model follows the basic framework of the stress ratio-controlled,
critical state-based, bounding surface plasticity model for sand initially presented by Manzari and
Dafalias (1997) and later extended by Dafalias and Manzari (2004). Modifications to the Dafalias-
Manzari model were developed and implemented to improve its ability to approximate engineering
design relationships that are used to estimate the stress-strain behaviors that are important to predicting
liquefaction-induced ground deformations during earthquakes. These developments are described in
the manuals (Version 1 in Boulanger 2010, Version 2 in Boulanger and Ziotopoulou 2012, Version 3
in Boulanger and Ziotopoulou 2015, Version 3.1 in Boulanger and Ziotopoulou 2017, Version 3.2 in

1
Boulanger and Ziotopoulo 2022, and Version 3.3 herein) and associated publications (Boulanger and
Ziotopoulou 2013, Ziotopoulou and Boulanger 2013a, Ziotopoulou 2014, Ziotopoulou and Boulanger
2016). The model is coded as a dynamic link library (DLL) for use with the commercial program FLAC
8.1 (Itasca 2019) and FLAC2D 9.00 (Itasca 2023).

It is unlikely that any one model can be developed or calibrated to simultaneously fit a full set of
applicable design correlations for monotonic and cyclic, drained and undrained behaviors of sand, in
part because the various design correlations are not necessarily physically consistent with each other;
e.g., they may include a mix of laboratory test-based and case history-based relationships, or they have
been empirically derived from laboratory data sets for different sands. Nonetheless, it is desirable that
a model, after calibration to the design relationship that is of primary importance to a specific project,
be able to produce behaviors that are reasonably consistent with the general magnitudes and trends in
other applicable design correlations or typical experimental data.

Stress-strain behaviors of sand that are most commonly the focus in design are listed below, along
with reference to a figure showing an example design correlation or typical experimental test result.

• The cyclic resistance ratio (CRR) against triggering of liquefaction, which is commonly estimated
based on SPT and CPT penetration resistances with case history-based liquefaction correlations
(e.g., Figure 1.2). The CRR is the cyclic stress ratio (e.g., CSR = cyc/'vc, with cyc = horizontal
cyclic shear stress, 'vc = vertical consolidation stress) that is required to trigger liquefaction in a
specified number of equivalent uniform loading cycles.
• The response under the irregular cyclic loading histories produced by earthquakes, which is
approximately represented by the relationship between CRR and number of equivalent uniform
loading cycles (e.g., Figure 1.3). This aspect of behavior also directly relates to the magnitude
scaling factors (MSF) that are used with liquefaction correlations in practice.
• The dependence of CRR on effective confining stresses and sustained static shear stresses. These
aspects of behavior are represented by the K (Figure 1.4) and K (Figure 1.5) correction factors,
respectively, which are used with liquefaction correlations in practice.
• The accumulation of shear strains after triggering of liquefaction. Evaluations of reasonable
behavior are often based on comparisons to laboratory tests results for similar soils in the literature
(e.g., Figure 1.6).
• The strength loss as a consequence of liquefaction, which may involve explicitly modeling
phenomena such as void redistribution or empirically accounting for it through case history-based
residual strength correlations (e.g., Figure 1.7).
• The small-strain shear modulus which can be obtained through in-situ shear wave velocity
measurements.
• The shear modulus reduction and equivalent damping ratio relationships prior to triggering of
liquefaction. These aspects of behavior are commonly estimated using empirical correlations
derived from laboratory test results for similar soils in the literature (e.g., Figure 1.8).
• Drained monotonic shear strengths and stress-strain behavior (e.g., Figure 1.9). Peak friction angles
may be estimated using relationships such as Bolton's (1986) relative dilatancy index, IR (Figure
1.10) or correlations to SPT and CPT penetration resistances.
• Undrained monotonic shear strengths and stress-strain behavior (e.g., Figure 1.11), which may be

2
 estimated using correlations to SPT and CPT penetration resistances.
• The volumetric strains during drained cyclic loading (Figure 1.12 and Figure 1.13) or due to
reconsolidation following triggering of liquefaction (e.g., Figure 1.14), both of which may be
estimated using empirical correlations derived from laboratory test results for similar soils in the
literature.
The constitutive model described herein was developed for earthquake engineering applications,
with specific goals being: (1) the ability to reasonably approximate empirical correlations used in
practice, and (2) an ability to be calibrated within a reasonable amount of engineering effort. In essence,
the approach taken was to calibrate the constitutive model at the equation level, such that the functional
forms for the various constitutive relationships were chosen for their ability to approximate the
important trends embodied in the extensive laboratory-based and case history-based empirical
correlations that are commonly used in practice.

The organization of this report is structured as follows:

• Section 2 of this report contains a description of the model formulation.

• Section 3 contains a description of the model's implementation as a user defined material in a
dynamic link library for use in the commercial program FLAC 8.1 (Itasca 2019) and FLAC2D
9.00 (Itasca 2023).
• Section 4 of this report contains a summary of the model input parameters, guidance on model
parameter selections, and then illustrations of the model responses to a broad range of elemental
loading conditions.
• Section 5 contains summary remarks regarding the model and its use in practice.

1.1 Revisions

Revisions to PM4Sand in Version 3, relative to Version 2, included: (1) revised dependency of

dilatancy and plastic modulus on fabric and fabric history, (2) modifications to the initial back-stress
ratio tracking logic, (3) modifications for improved modeling of post-liquefaction reconsolidation
strains after the end of strong shaking in a simulation, (4) addition of a minor cohesion term to reduce
potential hour-glassing and improve behavior of zones near a free surface, (5) a more efficient tensor
library which reduces computational time, and (6) re-calibration of the model, resulting in changes to
the default values for some secondary parameters.

Revisions to PM4Sand in Version 3.1, relative to Version 3, were minor. They included: (1)
recompiling to run with FLAC 8.0 (Itasca 2016), (2) changes to the logic for input of secondary
parameters, (3) a minor correction to the algorithm for post-liquefaction reconsolidation strains, and
(4) renaming of some tracking variables. For boundary value problem simulations not involving post-
liquefaction reconsolidation, these revisions have been found to have no effect on simulation results.

The report for Version 3.1 provided, relative to the one for Version 3, clarifications on some aspects of
the model formulation and implementation in response to questions from users, corrected typographic
errors, and described the above revisions to the input logic for secondary parameters.

Revision 1 to Version 3.1 (posted January 2018) documented that the Version 3.1 DLL module was
recompiled on November 28, 2017 with a modification to the initialization scheme that prevents a
problem that can arise with the use of the static solver during model initialization (i.e., introduction of
3
PM4Sand into a model) with certain sequences of commands. Note that it is still recommended that
PM4Sand only be used with the dynamic solver, especially for any conditions involving significant
applied loading, as explained in Section 3.6.

Revisions to PM4Sand Version 3.2 relative to Version 3.1, include a revision to the initial back-stress
ratio initialization routine, a modification to the elastic shear modulus equation, a recalibration of the
Fsedmin secondary parameter as well as a proper renaming of its input string for FLAC from F_sed to
F_sedmin. The initial back-stress ratio at the time of model initialization is now limited to have a
magnitude that is no greater than 90% of the bounding surface stress ratio (Mb); this constraint
eliminates a problem that can occur when the model is initialized with consolidation stress states that
are outside the bounding surface. The equation for the elastic shear modulus includes a CSR term that
reduces the elastic shear modulus at stress ratios close to the bounding surface. The CSR term is
normalized in Version 3.2 to produce a value of unity at the time of model initialization, which
simplifies the calibration of the model. These changes do not significantly affect the general features
of model responses but do affect responses for a given set of calibration parameters. Therefore,
calibrations using PM4Sand Version 3.1 need to be revised when switching to Version 3.2.
Furthermore, the manual was revised to reflect these changes as well as address minor typos and
inconsistencies.

Revision to PM4Sand to Version 3.3 relative to Version 3.2 was adding a restriction that CSR be less or
equal to unity. In addition, the model was compiled for use with FLA2D 9.00. The DLL module for
FLAC 8.1 modelpm4sand005_64.dll was compiled on June 12, 2023 using Microsoft Visual Studio
Community 2015 C++. The DLL module for FLAC2D 9.00 cmodelPM4Sand2D009.dll was compiled
on June 12, 2023 using Microsoft Visual Studio Community 2022 C++. The DLL was recompiled on
October 18, 2023 to correct a bug that caused errors when used with triangular elements with a single
overlay in FLAC2D. Note that the compilation date and version are included in the properties of the
DLL file. The simulations presented in this report were prepared using PM4Sand Version 3.3 in FLAC
8.1. PM4Sand Version 3.3 in FLAC2D 9.00 produced the same results.

4
 (N1 )60 = 25 (N1 )60 = 25

(N1 )60 = 15 (N1 )60 = 10

Figure 1.1. Schematic cross-section for an earth dam.

Figure 1.2. Correlations for cyclic resistance ratio (CRR) from SPT data
(after Idriss and Boulanger 2010).

5
 0.4
(a)

Cyclic Stress Ratio CSR = cyc / 'vc

Shaking Table tests conducted by
De Alba et al (1976):
Relative Density = 90%
0.3
Relative Density = 82%
Relative Density = 68%
Relative Density = 54%

Fit with
0.2
b = 0.25
= 0.21

= 0.21
0.1
= 0.23

Initial confining pressure = 8 psi (55 kPa)

0
1 10 100
Number of Cycles to cause initial liquefaction, N
2

(b)
Yoshimi et al (1984):
Cyclic Stress Ratio CSR = d / 2'o

Cyclic TX on frozen samples

1 of dense Niigata sand

0.8

0.6

Silver et al. (1976): b = 0.34

0.4 Cyclic TX on moist tamped
Monterey sand ( DR=60% )

0.2 b = 0.22

Toki et al. (1986):

0.1 Cyclic TX on air pluviated
Toyoura sand ( DR=50% )
0.08 b = 0.10
0.06

0.04
0.1 1 10 100
Number of Cycles, N, to reach 5% D.A. strain

Figure 1.3. Relationships between cyclic resistance ratio (CRR) and number of equivalent uniform
loading cycles for undrained loading of reconstituted and undisturbed samples of clean sand
(from Ziotopoulou and Boulanger 2012).

6
Figure 1.4. K factor describing the effect that effective overburden stress has on cyclic resistance
ratio of sands (from Idriss and Boulanger 2008).

Figure 1.5. K factor describing the effect that sustained static shear stress ratio (=s/'vc) has on
cyclic resistance ratio of sands (Boulanger 2003a).

7
 0.4 0.4 Sacramento river sand
B ACU cyclic triaxial
DR=57%, '3c=200 kPa
B
q/2p'c

0.2
0.2

q/2p'c
0 A
A
-0.2 0

1 A -0.2
0.75 B
-4 -3 -2 -1 0 1 2 3 4
Axial strain, a (%)
ru

0.5

0.25

0 0.4
4
B
0.2

q/2p'c
2 B
a (%)

0 0.0
A
-2 A
-0.2
-4
0 5 10 15 20 25 30 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4
Cycle number p'/p'c

Figure 1.6. Undrained cyclic triaxial test on clean sand (test from Boulanger and Truman 1996; from
Idriss and Boulanger 2008).

8
Figure 1.7. Empirical relationship for estimating residual strength of liquefied sands based on case
histories (Boulanger and Idriss 2011).

9
 1

Shear modulus ratio, G/Gmax

0.8 Depth =
500-1000 ft
250-500 ft
0.6 120-250 ft
50-120 ft
20-50 ft
0.4 0-20 ft

0.2 EPRI (1993) recommendations

for sand at different depths:

0
0.0001 0.001 0.01 0.1 1 10
Shear strain amplitude (%)
40
Equivalent damping ratio (%)

30 Depth =
500-1000 ft
250-500 ft
120-250 ft
20 50-120 ft
20-50 ft
0-20 ft

10

0
0.0001 0.001 0.01 0.1 1 10
Shear strain amplitude (%)

Figure 1.8. Shear modulus reduction and equivalent damping ratio relationship for sands, as
recommended by EPRI (1993).

10
Figure 1.9. Drained triaxial compression tests on loose and dense sand specimens under a range of
effective confining stresses (after Lee and Seed 1967; from Idriss and Boulanger 2008).

11
 16
IR = DR (Q - ln p')-R
Q=10, R=1
75 % DR = 100 %

DR = 80 %
12
'max - 'crit (deg)

'max - 'crit = 3IR

DR = 50 %
8 50 %

4
25 %

0
10 100 1000 10000 100000

Mean Effective Stress p' ( kPa )

Figure 1.10. Triaxial test data for sands with initial relative densities in the vicinity of 80% or 50%
failing at various mean effective stresses. The difference of peak friction angle from the critical
friction angle is related to the relative dilatancy index (IR) (after Bolton 1986).

12
 1.2 1.2
DR = 16% DR = 16% Ishihara (1993):
Toyoura sand,

Deviator stress, q (atm)

Deviator stress, q (atm)
1 1 Critical ICU-TC
'3c  1.0 atm
state
0.8 0.8
0.6 atm
0.6 0.6
0.2 atm

0.4 0.4

0.2 0.1 atm 0.2

0 0
0 5 10 15 20 25 30 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Axial strain, a (%) Mean principal effective stress, p' (atm)
(a)

20 DR = 38% 20 DR = 38%

Deviator stress, q (atm)

Deviator stress, q (atm)

'3c  30 atm Critical

15 15 state
20 atm

10 10
10 atm

5 1 atm 5

0 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30 35
Axial strain, a (%) Mean principal effective stress, p' (atm)
(b)

35 35 Critical
'3c  30 atm DR = 64%
state
DR = 64%
Deviator stress, q (atm)
Deviator stress, q (atm)

30 30

25 25

20 20 atm 20
10 atm 15
15
1.0 atm
10 10

5 5

0 0
0 5 10 15 20 25 30 0 10 20 30 40 50 60
Axial strain, a (%) Mean principal effective stress, p' (atm)
(c)

Figure 1.11. Undrained triaxial compression tests on very loose and loose sand specimens under a
range of effective consolidation stresses (after Ishihara 1993; from Idriss and Boulanger 2008).

13
 0.58
Youd (1972): Drained simple shear,
Ottawa sand, emax = 0.752, emin = 0.484,
0.56 'vc = 48 kPa
Cycle 1
C A
B E
0.54 D

5
0.52

10

Void ratio
0.50
20

0.48
50

0.46 100

200

0.44
500
1000
0.42 10000

0.40
-0.8 -0.4 0 0.4 0.8
Shear displacement (mm)

Figure 1.12. Drained cyclic simple shear test showing densification of a sand specimen with
successive cycles of loading (after Youd 1972; from Idriss and Boulanger 2008).

Figure 1.13. Volumetric strains in drained cyclic direct simple shear tests on clean sands (Duku et al.
2008): (a) Results from 16 sands at a relative density of about 60% with an overburden stress of 1.0
atm, and (b) Comparison of trends with earlier relationships by Silver and Seed (1971) for sands at
relative densities of 45, 60, and 80%.

14
Figure 1.14. Relationship between post-liquefaction volumetric strain and the maximum shear strain
induced during undrained cyclic loading of clean sand (after Ishihara and Yoshimine 1992; redrawn
in Idriss and Boulanger 2008).

15
2. MODEL FORMULATION

The sand plasticity model presented herein follows the basic framework of the stress ratio-
controlled, critical state-based, bounding surface plasticity model for sand presented by Dafalias and
Manzari (2004). The Dafalias and Manzari (2004) model extended the previous work by Manzari and
Dafalias (1997) by adding a fabric-dilatancy related tensor quantity to account for the effect of fabric
changes during loading. The fabric-dilatancy related tensor was used to macroscopically model the
effect that microscopically-observed changes in sand fabric during plastic dilation have on the
contractive response upon reversal of loading direction. Dafalias and Manzari (2004) provide a detailed
description of the motivation for the model framework, beginning with a triaxial formulation that
simplifies its presentation and followed by a multi-axial formulation. The model described herein is
presented in its multi-axial formulation, along with the original framework of the Dafalias-Manzari
model for comparison.

2.1 Basic stress and strain terms

The basic stress and strain terms for the model are as follows. The model is based on effective
stresses, with the conventional prime symbol dropped from the stress terms for convenience because
all stresses are effective for the model. The stresses are represented by the tensor , the principal
effective stresses 1, 2, and 3, the mean effective stress p, the deviatoric stress tensor s, and the
deviatoric stress ratio tensor r. The present implementation was further simplified by casting the
various equations and relationships in terms of the in-plane stresses only. This limits the present
implementation to plane-strain applications and is not correct for general cases, but it has the advantage
of simplifying the implementation and improving computational speed by reducing the number of
operations. Expanding the implementation to include the general case should not affect the general
features of the model. Consequently, the relationships between the various stress terms can be
summarized as follows:

  xx  xy 
σ =  (1)
  xy  yy 

 xx +  yy
p= (2)
2

 sxx sxy    xx − p  xy 
s = σ − pI =  =
s yy    xy p 
(3)
 sxy  yy −

16
   xx − p  xy

s  rxx rxy   p p 

r= = =
ryy    xy
(4)
p  rxy  yy − p 
 
 p p 

Note that the deviatoric stress and deviatoric stress ratio tensors are symmetric with rxx = -ryy and
sxx = -syy (meaning a zero trace), and that I is the identity matrix.
The model strains are represented by a tensor , which can be separated into the volumetric strain
v and the deviatoric strain tensor e. The volumetric strain in plane strain is,

 v =  xx +  yy (5)

and the deviatoric strain tensor is,

 v 
  −  xy 
 xx
3
e =ε− v I =  (6)
3   v 
  yy − 
 3
xy

In incremental form, the deviatoric and volumetric strain terms are decomposed into an elastic and
a plastic part,

de = de el + de pl (7)

d  v = d  vel + d  vpl (8)

where
de el = elastic deviatoric strain increment tensor
de pl = plastic deviatoric strain increment tensor
d el
v
= elastic volumetric strain increment
d v
pl
= plastic volumetric strain increment

2.2 Critical state

Dafalias and Manzari (2004), based on findings in Li and Wang (1998), used a power relationship
to approximate the curving of the critical state line (Schofield and Wroth 1968) that occurs over a broad
range of confining stresses,

17
 m
p 
ecs = eo −   cs  (9)
 pA 
where pcs = mean stress at critical state, ecs = critical state void ratio, and eo, , and m are parameters
controlling the position and shape of the critical state line. The state of the sand was then described
using the state parameter (Been and Jefferies 1985), which is the difference between the current void
ratio (e) and the critical state void ratio (ecs) at the same mean effective stress (pcs).

The model presented herein instead uses the relative state parameter index (R) as presented in
Boulanger (2003a) and shown in Figure 2.1(a). The relative state parameter (Konrad 1988) is the state
parameter normalized by the difference between the maximum void ratio (emax) and minimum void
ratio (emin) values that are used to define relative density (DR). The relative state parameter "index" is
just the relative state parameter defined using an empirical relationship for the critical state line.
Boulanger (2003a) used Bolton's (1986) dilatancy relationship to define the empirical critical state line
and thus arrived at,

 R = DR ,cs − DR (10)

R
DR ,cs =
 p  (11)
Q − ln 100 
 pA 

where DR,cs = relative density at critical state for the current mean effective stress. The parameters Q
and R were shown by Bolton (1986) to be about 10 and 1.0, respectively, for quartzitic sands. Critical
state lines using the above expression with Q values of 9 and 10, and with R values of 1.0 and 1.5 are
shown in Figure 2.1.(b).

2.3 Bounding, dilatancy, and critical surfaces

The model incorporates bounding, dilatancy, and critical stress ratio surfaces following the form of
Dafalias and Manzari (2004). The present model simplifies the surfaces by removing the Lode angle
dependency (e.g., friction angles are the same for compression or extension loading) that was included
in the Dafalias and Manzari model, such that the bounding (Mb) and dilatancy (Md) ratios can be related
to the critical stress ratio (M) by the following simpler expressions,

M b = M  exp ( −nb R ) (12)

M d = M  exp ( nd  R ) (13)

18
where nb and nd are parameters determining the values of Mb and Md, respectively. For the present
implementation, the mean normal stress p is taken as the average of the in-plane normal stresses
(Equation 2), q is the difference in the major and minor principal in-plane stresses, and the relationship
for M is therefore reduced to,

M = 2  sin (cv ) (14)

where cv is the constant volume or critical state effective friction angle. The three surfaces can, for the
simplifying assumptions described above, be conveniently visualized as linear lines on a q-p plot
(where q = 1 - 3) as shown in Figure 2.2 or as circular surfaces on a stress ratio graph of ryy versus rxy
as shown in Figure 2.3.

As the model is sheared toward critical state (R = 0), the values of Mb and Md will both approach
the value of M. Thus, the bounding and dilatancy surfaces move together during shearing until they
coincide with the critical state surface when the soil has reached critical state.

The above functional form for the bounding stress ratio controls the relationship between peak
friction angle and relative state, which is consistent with the forms and data previously proposed by
Been and Jefferies (1985) and Konrad (1986). The data from those studies were primarily for sands
that were dense-of-critical, and the above relationship can reasonably fit those data. The few data
points for loose-of-critical sands show that the peak friction angles (presumably determined at the limit
of strains possible within the laboratory tests) were only slightly smaller than the critical state values,
such that extending the above relationships to loose-of-critical sands may tend to underestimate the
peak friction angles. Consequently, the present model allows nb and nd to be different for loose-of-
critical and dense-of-critical states for the same sand.

2.4 Yield surface and image back-stress ratio tensors

The yield surface and back-stress ratio tensor () follow those of the Dafalias and Manzari model,
although their final form is considerably simplified by the prior assumption of removing any Lode
angle dependency. The yield surface is a small cone in stress space, and is defined in stress terms by
the following expression,

f = ( s − pα ) : ( s − pα ) 2 − 1 pm = 0
1
(15)
2
The back-stress ratio tensor  defines the center of the yield surface, and the parameter m defines the
radius of the cone in terms of stress ratio. The yield function can be rewritten to emphasize the role of
stress ratio terms as follows,

f = (r − α) : (r − α) − 1 m=0
2
(16)

The yield function can then be visualized as related to the distance between the stress ratio r and the
back-stress ratio , as illustrated in Figure 2.3.

19
 The bounding surface formulation now requires that bounding and dilatancy stress ratio tensors be
defined. Dafalias and Manzari (2004) showed that it is more convenient to track back-stress ratios and
to similarly define bounding and dilatancy surfaces in terms of back-stress ratios. An image back-stress
ratio tensor for the bounding surface (b) is defined as,

α b = 1  M b − m  n (17)
2
where the tensor n is normal to the yield surface. An image back-stress ratio tensor for the dilatancy
surface (d) is similarly defined as,

α d = 1  M d − m  n (18)
2
The computation of constitutive responses can now be more conveniently expressed in terms of back-
stress ratios rather than in terms of stress ratios, as noted by Dafalias and Manzari (2004).

2.5 Stress reversal and initial back-stress ratio tensors

The bounding surface formulation, as described in Dafalias (1986) and adopted by Dafalias and
Manzari (2004), keeps track of the initial back-stress ratio (in) and uses it in the computation of the
plastic modulus Kp. This tracking of one instance in loading history is essentially a first-order method
for tracking loading history. A reversal in loading direction is then identified, following traditional
bounding surface practice, whenever,

( α − ain ) : n  0 (19)

A reversal causes the current stress ratio to become the initial stress ratio for subsequent loading. Small
cycles of load reversal can reset the initial stress ratio and cause the plastic modulus Kp to increase
accordingly, in which case the stress-strain response becomes overly stiff after a small load reversal.
This is a well-known problem in bounding surface formulations for which various approaches offer
different advantages and disadvantages.

The model presented herein tracks an initial back-stress ratio and a previous initial back-stress ratio
(inp), as illustrated in Figure 2.4a. When a reversal occurs, the previous initial back-stress ratio is
updated to the initial back-stress ratio, and the initial back-stress ratio is updated to the current back-
stress ratio.

In addition, the model tracks an apparent initial back-stress ratio tensor (αinapp) as schematically
illustrated in Figure 2.4b. The schematic in Figure 2.4b is similar to that of Figure 2.4a, except that the
most recent loading reversals correspond to a small unload-reload cycle on an otherwise positive
loading branch. The components of αinapp are taken as: (i) for positive loading directions, the minimum
value they have ever had, but no smaller than zero, and (ii) for negative loading directions, the
maximum value they have ever had, but no greater than zero. Figure 2.5 further illustrates these

20
scenarios for four different loading cases. These minimum and maximum past back-stress ratios are
stored for each component individually and for the entire loading history. The use of αinapp helps avoid
the over-stiffening of the stress-strain response following small unload-reload cycles along an
otherwise monotonically increasing branch of loading, without having to track the loading history
through many cycles of load reversals.

The computation of Kp utilizes the values of αinapp, αintrue, and αinp, as defined in Figure 2.4b and
Figure 2.5, to better approximate the stress-strain response during an unload-reload cycle. For the last
positive loading branch in this figure, the value of Kp is initially most strongly controlled (inversely)
by the distance (α - αintrue):n, such that the stiffness is initially large. As positive loading continues, the
progressive reduction in Kp becomes increasingly dependent on αinapp as well. Once the positive loading
exceeds the previous reversal point, the value of Kp becomes solely dependent on the distance
(α - αinapp):n. Thus, the computation of Kp has the following dependencies,

if ( −  inp ) : n  0  K p = f ( intrue ,  inapp )

(20)
else  K p = f ( inapp )

The equations relating Kp to these back-stress ratios are given later in Section 2.7.

The impact of the above logic for defining αin on stress-strain responses is demonstrated in
Figure 2.6 showing αxy versus shear strain  computed for two different drained DSS loading
simulations. For these two examples, the reloading stiffness of the current loading branch (green line)
is initially large because Kp is initially computed based on αin = αintrue. As the loading exceeds αinp, the
loading stiffness becomes much softer because Kp is now computed based on αin = αinapp.

The initial back-stress ratio (αin) is first established at initialization of the model or upon execution of
FirstCall (see also Section 3). The value of αin is established as being equal to the current back-stress ratio,
subject to the limitation that its corresponding stress ratio be ≤ 0.9 Mb. This constraint on the magnitude
of αin at initialization avoids a problem that can otherwise occur when the initial consolidation stress state
is above the bounding surface. In such cases, Kp = 0 since Mcur > Mb and D = 0 since α - αin = 0, which
can result in no stress changes during shearing and hence an incorrect response. Linearly scaling αin so its
corresponding stress ratio is ≤ 0.9 Mb upon initialization ensures that D > 0 at the start of shearing
whenever the initial consolidation stress state corresponds to a stress ratio > 0.9 Mb. Undrained shearing
from such an initial consolidation stress state will thus be properly accompanied by contraction and
associated strain softening.

2.6 Elastic strains and moduli

The elastic deviatoric strain and elastic volumetric strain increments are computed as,

ds
de el = (21)
2G

21
 dp
d  vel = (22)
K
where G is the elastic shear modulus and K is the elastic bulk modulus. The elastic shear modulus in
the model presented herein is dependent on the mean effective stress according to,

1
 p 2
G = Go pA   CSR (23)
 pA 
where Go is a constant, pA is the atmospheric pressure (101.3 kPa), and CSR is a factor that accounts for
stress ratio effects (described below).

Dafalias and Manzari (2004) had included dependence of G on void ratio following the form of
Richart et al. (1970). This aspect was not included in the model herein because: (1) the effects of void
ratio changes on G are small relative to those of confining stress, (2) the value of G is more strongly
affected by environmental factors such as cementation and ageing, and (3) the calibration of G to in-
situ shear wave velocity data is simplified by not including e.

Yu and Richart (1984) showed that the small-strain elastic shear modulus of sand is dependent on
the stress ratio and stress ratio history. The effect of stress ratio was shown to generally be less than
about 10% when the ratio of major to minor principal effective stresses is less than about 2.5, but to
also increase to about 20-30% at higher principal stress ratios. They also showed that stress ratio history
caused a reduction in the small-strain elastic shear modulus when the maximum previous stress-ratio
was greater than the current stress ratio. The effect of stress ratio and stress ratio history on the elastic
shear modulus was approximately accounted for in the present model by the factor CSR. The following
equation for CSR is similar in form to that used by Yu and Richart (1984) to represent stress ratio effects,
except that it uses stress ratio terms consistent with the present model. The equation for CSR was further
normalized in PM4Sand Version 3.2 to produce a value of unity at the time of model initialization, and
restricted to values less than unity thereafter, as,
m
 M 
SR

1 − CSR ,o   b 
CSR = M  1 (24)
  M  
mSR

1 − CSR ,o   b  
  M   initial

The numerator in the above equation is the same form used in Version 3.1 of the model, while the
denominator retains the value of the numerator from the time of model initialization. Thus, CSR = 1.0
at the time of initialization. The above equation approximates Yu and Richart's (1984) results for stress
ratio effects when CSR,0 = 0.3 and mSR = 2. The effects of stress ratio history would cause further
reductions, and are more complicated to represent. The calibration examples presented later in this
report worked well with CSR,0 = 0.5 and mSR = 4, which keeps the effect of stress ratio on elastic

22
modulus small at small stress ratios, but lets the effect increase to a 60% reduction when the stress ratio
is on the bounding surface.

The elastic bulk modulus is related to the shear modulus through the Poisson's ratio as,

2 (1 + v )
K= G
3 (1 − 2v )
(25)

as was done by Dafalias and Manzari (2004).

2.7 Plastic components without fabric effects

Loading index

The loading index (L) is used to compute the plastic component of the volumetric strain increment
and the plastic deviatoric strain increment tensor as,

d  vpl = L D (26)

de pl = L R' (27)

where D is the dilatancy, R is the direction of dpl, R is the deviatoric component of R, and <> are
Macaulay brackets that set negative values to zero [i.e., <L> = L if L ≥ 0, and <L> = 0 if L < 0]. The
tensor R for the assumption of no Lode angle dependency is,

1
R = n + DI (28)
3
where n is the unit normal to the yield surface (Figure 2.3). Note that the assumption of no Lode angle
dependency also means that R = n. The dilatancy D relates the incremental plastic volumetric strain to
the incremental plastic deviatoric strain,

d  vpl
D= (29)
de pl

The dilatancy D can be also related to the conventional engineering shear strain in this plane strain
approximation, as,

23
 d  vpl
D= (30)
1 d  pl
2
The loading index, as derived in Dafalias and Manzari (2004) is,

1 f 1
L= : dσ =  n : ds − n : rdp 
K p σ Kp
(31)
2Gn : de − n : rKd  v
L=
K p + 2G − KDn : r

The stress increment for an imposed strain increment can then be computed as,

d = 2Gde + Kd v I − L ( 2Gn + KDI ) (32)

Hardening and the update of the back-stress ratio

Updating of the back-stress ratio is dependent on the hardening aspects of the model. Dafalias and
Manzari (2004) updated the back-stress ratio according to bounding surface practice as,

2
dα = L   h ( α b − α ) (33)
3
where h is the hardening coefficient. The factor of 2/3 was included for convenience so that model
constants would be the same in triaxial and multi-axial derivations. They subsequently showed that the
consistency condition f=0 was satisfied when the plastic modulus Kp was related to the hardening
coefficient as,

p h ( αb − α ) : n
2
Kp = (34)
3
This expression can be rearranged so as to show that the consistency equation can be satisfied by
expressing the hardening coefficient as,

3 Kp
h=
2 p ( αb − α ) : n
(35)

24
 The relationship for the plastic modulus can subsequently take a range of forms, provided that the
hardening coefficient and updating of the back-stress ratio follow the above expressions.

Plastic modulus

The plastic modulus in the multi-axial generalized form of Dafalias and Manzari (2004), after
substituting in their expression for the hardening coefficient, can be expressed as,

2  1+ e  ( αb − α ) : n
2 (
K p = G ho  1 − Ch e )  (36)
3  ( 2.97 − e )  ( α − αin ) : n
where ho and Ch are scalar parameters and e is the void ratio. Setting aside the secondary influence of
void ratio, this form illustrates that Kp is proportional to G, proportional to the distance of the back-
stress ratio to the bounding back-stress ratio, and inversely proportional to the distance of the back-
stress ratio from the initial back-stress ratio.

The plastic modulus relationship was revised in the model presented herein to provide an improved
approximation of empirical relationships for secant shear modulus and equivalent damping ratios
during drained strain-controlled cyclic loading. The plastic modulus is computed as,

( α b − α ) : n 
0.5 (37)

K p = G ho   Crev
exp ( α − αinapp ) : n  − 1 + C 1
   

( α −α ):n
app

Crev =
in
for (α − α ) : n  0
p

(α − α ) : n
true
in
in
(38)

= 1 otherwise
The factor Crev accounts for the effect of unload-reload cycles as discussed in Section 2.5 and
illustrated in Figures 2.4 and 2.5. The constant C1 in the denominator serves to avoid division by zero
and has a slight effect on the nonlinearity and damping at small shear strains. If C1 = 0, then the value
of Kp will be infinite at the start of a loading cycle because ( - in):n will also be zero. In that case,
nonlinearity will become noticeable only after ( - in):n becomes large enough to reduce Kp closer to
the value of G (e.g., Kp/G closer to 100 or 200). Setting the value of C1 = ho/200 produces a reasonable
response as will be demonstrated later with examples of modulus reduction and equivalent damping
ratios. For stress ratios outside the bounding surface [i.e., loose-of-critical states with (b - ):n < 0],
the plastic modulus is set to zero rather than allowing for negative values. This restriction on the plastic
modulus improved numerical stability while having little effect on computed stress-strain responses.
The plastic modulus is further modified for the effects of fabric and fabric history, as described in a
later section.

25
Plastic volumetric strains – Dilation

Plastic volumetric strains are related to plastic deviatoric strains through the dilatancy D (Equations
29 and 30), which is computed in the Dafalias and Manzari (2004) model and the base component of
the model presented herein (with additional fabric effects described in a later section) as,

D = Ado  ( α d − α ) : n  (39)

Note that dilation (increasing void ratio) occurs whenever the term (d - ):n is less than zero whereas
contraction (decreasing void ratio) occurs when it is positive.

The constant Ado in this relationship can be related to the dilatancy relationship proposed by Bolton
(1986), which follows from the work of Rowe (1962), through the following sequence of steps. Bolton
showed that the difference between peak and constant volume friction angles could be approximated
as,

 pk − cv = −0.8 (40)

with
 d  pl 
 = tan  vpl
−1

 d  (41)
 
Since  ≈ tan() for  less than about 0.35 radians (20 degrees), the difference between peak and
constant volume friction angles (in radians) can be approximated as,

d  vpl 1
 pk − cv = −0.8 pl = −0.8 D
d 2 (42)

The peak friction angle is mobilized at the bounding surface, so this can be written as,

Ado  ( α d − α ) : n 
1
 pk − cv = −0.8
2
1  M d Mb   (43)
 pk − cv = −0.8 Ado   n− n  : n
2  2 2  
The term n:n is equal to unity, and the values of pk and cv (again in radians) can be replaced with
expressions in terms of Mb and M as,

26
  Mb  −1  M 
 = 0.4 Ado   M − M 
−1
 − sin 
b d
sin  (44)
 2   2 

This expression can then be rearranged to solve for Ado as,

 Mb 
−1 M  (45)
sin   − sin −1  
1  2   2 
Ado =
0.4 Mb −Md

where the angles returned by the sin-1 functions are in radians.

The parameter Ado should thus be chosen to be consistent with the nd and nb terms that control Mb,
and Md. For example, setting the parameters nb and nd equal to 0.5 and 0.1, respectively, results in Ado
varying from 1.26 for ξR = -0.1 to 1.45 for ξR = -0.7. A default value for Ado is computed based on the
above expression using the conditions at the time of model initialization in FLAC (as described in a
later section). If an alternative value for Ado is manually input as a property of the model, then the
default value will be deactivated.

Alternatively, the stress ratio terms can be replaced with friction angles (in radians) as follows,
 pk − cv = 0.4 Ado   M b − M d 
 pk − cv = 0.4 Ado   M exp(−nb R ) − M exp(n d  R ) 
(46)
 pk − cv = 0.4 Ado   2sin( pk ) − 2sin(d ) 
 pk − cv = 0.8 Ado  sin( pk ) − sin(d ) 

The sine terms can be replaced with Taylor series, which are quite accurate with just the first two
terms as,

( )
3

sin( ) =  − (47)
3!
Substituting the Taylor series in the above equation gives,

27
 
( pk )
 
( )
3 
3
  
 pk − cv = 0.8 Ado   pk −  −  d − d   (48)
 3!   3!  
   

The parameter Ado can then be solved for as,

 pk − cv
Ado =
 ( pk ) − (d )
3
3

0.8  pk − d −  (49)
 6 
 
where the friction angles in the above expression are in radians. This expression provides an alternative
view of how the parameter Ado relates to friction angles for a given set of nb and nd terms that control
pk and d, respectively. For example, consider the case with the parameters nb and nd equal to 0.5 and
0.1, respectively, and assuming cv = 33 degrees. For ξR = -0.1, we would obtain d = 32.6 degrees, pk
=  degrees, and Ado = 1.26. For ξR = -0.7, we would obtain d = 30.5 degrees, pk = 50.6 degrees,
and Ado = 1.45.

Plastic volumetric strains – Contraction

Plastic volumetric strains during contraction (i.e., whenever (d - ):n is greater than zero) are
computed in the Dafalias and Manzari (2004) model using the same expression as used for dilatancy,

D = Ado  ( α d − α ) : n  (50)

The use of this expression was found to limit the ability of the model to approximate a number of
important loading responses; e.g., it greatly overestimated the slope of the cyclic resistance ratio (CRR)
versus number of equivalent uniform loading cycles for undrained cyclic element tests (e.g., Figure
1.3).

Plastic volumetric strains during contraction for the model presented herein are computed using
the following expression,

D = Adc  ( α − α
(α d
− α) : n
) : n + Cin 
2
app

( α d − α ) : n + CD
in (51)

Ado
Adc = (52)
hp

28
 The various forms in the above relationships were developed to improve different aspects of the
calibrated model's performance. The value of D was set proportional to the square of (( - in):n + Cin)
to improve the slope of the relationship between CRR and number of uniform loading cycles. The Cin
term depends on fabric and is described in a later section along with other modifications to the above
expression for the effects of fabric and fabric history. The inclusion of the term Cin improves the stress
paths for undrained cyclic loading and the volumetric strain response during drained cyclic loading.
Inclusion of this constant enables some volumetric strain to develop early in the unloading from a point
outside the dilatancy surface (as described later). The remaining terms on the right-hand side of the
equation were chosen to be close to unity over most of the loading range, while ensuring that D
smoothly goes to zero as  approaches d; reasonable results were obtained using a CD value of 0.10.

The parameter Adc for contraction was related to the value of Ado for dilation by dividing it by a
parameter hp that can be varied during the calibration process to obtain desired cyclic resistance ratios.
The effect of confining stress on cyclic loading behavior was then conveniently incorporated by making
hp depend on R, with the following form chosen so that the model produces results consistent with the
design K relationships presented earlier in Figure 1.4.,

(
hp = hpo exp −0.7 + 7.0 ( 0.5 −  R )
2
) for  R  0.5 (53)

hp = hpo exp ( −0.7 ) for R  0.5 (54)

Thus, the scalar constant hpo provides a linear scaling of contraction rates while the functional form
of the remaining portion in Equations (53) and (54) is what controls the effect of overburden stress on
CRR. The variation of hp with R for different values of hpo is plotted in Figure 2.7. Once the other
input parameters have been selected, the constant hpo can be calibrated to arrive at a desired cyclic
resistance ratio.

An upper limit was imposed on the contraction rate, with the limiting value computed as,

D  1.5  Ado
(α d
− α) : n
( α d − α ) : n + CD
(55)

This limit prevented numerical issues that were encountered with excessively large contraction
rates. It does not appear to have limited the ability of the model to recreate realistic contraction rates as
illustrated in the calibration examples shown later.

29
2.8 Fabric effects

Dafalias and Manzari (2004) introduced a fabric-dilatancy tensor (z) that could be used to account
for the effects of prior straining. Their fabric tensor (z) evolved in response to plastic volumetric dilation
strains, according to,

dz = −cz −d  vpl ( zmax n + z ) (56)

where the parameter cz controls the rate of evolution and zmax is the maximum value that z can attain.

The fabric-dilatancy tensor was modified for the present model as,

cz −d  vpl
dz = − ( zmax n + z )
z D (57)
1 + cum − 1
2 zmax

In this expression, the tensor z evolves in response to plastic deviatoric strains that occur during dilation
only (i.e., dividing the plastic volumetric strain by the dilatancy gives plastic shear strain). In addition,
the evolution of fabric is restricted to only occur when (d – ):n < 0; this additional constraint
precludes fabric evolution during dilation above the rotated dilatancy surface (introduced later) but
below the non-rotated dilatancy surface. The parameter zcum is the cumulative value of absolute changes
in z computed according to,

dzcum = dz (58)

The rate of evolution for z therefore decreases with increasing values of zcum, which enables the
undrained cyclic stress-strain response to progressively accumulate shear strains rather than lock-up
into a repeating stress-strain loop. In addition, the greatest past peak value (scalar amplitude) for z
during its loading history is also tracked,

 z:z 
z peak = max  , z peak  (59)
 2 
The values of z, zpeak, and zcum are later used to facilitate the accumulation of shear strains under
symmetric loading through their effects on the plastic modulus and dilatancy relationships.
The evolution of the fabric tensor terms is illustrated in Figure 2.8 and Figure 2.9 showing the
response of a loose sand to undrained cyclic DSS loading without any sustained horizontal shear stress
(Figure 2.8) and with a sustained horizontal shear stress (Figure 2.9). These figures show the stress path
and stress-strain response of the sand, along with time histories for the back-stress ratios and fabric
tensor terms. Note how the fabric terms do not grow until the soil reaches the dilatancy surface, and

30
how the stress-ratios are limited by the bounding stress ratio. There is no horizontal shear stress reversal
for the case shown in Figure 2.9 and thus the back-stress ratio and fabric terms do not reverse either.

Additional memory of fabric formation history

Memory of the fabric formation history was included in the model presented herein to improve the
ability of the model to account for the effects of sustained static shear stresses and account for
differences in fabric effects for various drained versus undrained loading conditions.

The initial fabric tensor (zin) at the start of the current loading path is determined whenever a stress
ratio reversal occurs. The zin tracks the immediate history terms without any consideration of whether
an earlier loading cycle had produced greater degrees of fabric (i.e., the logic is different from that
adopted for the updating of back-stress ratio history terms). This history term is used for describing the
degree of stress rotation and its effects on plastic modulus, as described later.

Another aspect of the fabric history that is tracked is the mean stress at which the fabric is formed.
This aspect of fabric history is tracked by tracking the product of z and p, and defining pzp as the mean
stress at the time that this product achieves its greatest peak value. The pzp is used in addressing a couple
of issues, including the issue of how fabric that is formed during liquefaction may be erased during
reconsolidation. For example, a saturated sand that develops cyclic mobility behavior during undrained
cyclic loading clearly remembers its history of plastic deviatoric strains and then subsequently forgets
(to a large extent) this prior strain history when it reconsolidates back to its pre-earthquake confining
stress. As another example, the memory of prior strains during undrained cyclic loading is very different
than the memory of prior strains during drained cyclic loading. This memory conceptually could be
related to the history of plastic and total volumetric strains, but a simpler method to account for this
effect is to consider how the mean stress p relates to the value of pzp. Conceptually, it appears that prior
strain history (or fabric) is most strongly remembered when the soil is operating under mean stresses
that are smaller than those that existed when the fabric was formed (i.e., p << pzp) and then largely
forgotten when they are of the same order (i.e., p  pzp). This attribute will be used in the relationships
described later for describing the effects of fabric on dilatancy.

Effect of fabric on plastic modulus

An effect of fabric on the plastic modulus was added to the model presented herein by reducing
the plastic modulus as the fabric tensor increased in peak amplitude, as follows:

( α b − α ) : n 
0.5

K p = G  ho    Crev
 ( app
)
exp ( α − αin ) : n − 1 + C 1

(60)
CK

z 
1 + CK p  peak  ( α b − α ) : n 1 − Czpk 2
 zmax 

where,

31
 CK  f
CK = 1 + C pzp 2 Czpk1 (61)
(
1 + 2.5 ( α − aintrue ) : n )
2

z peak
Czpk1 =
z (62)
zcum + max
5

z peak
Czpk 2 =
z (63)
zcum + max
100

− − ( pzp − p )
C pzp 2 =
− − ( pzp − p ) + pmin
(64)

The above expressions produce a reduction in plastic modulus when fabric is favorable (z:n ≥ 0) and
with increasing plastic shear strains (which conceptually would break down any cementation). This
reduces both the plastic modulus and the hysteretic damping at larger shear strains (note that zpeak = 0
unless the soil has been loaded strongly enough to pass outside the dilatancy surface), improves the
volumetric strains that develop in drained cyclic loading, and improves the path in undrained cyclic
loading.

The CKα and square root of (1-Czpk2) terms both serve to increase Kp during non-reversal loading
by amounts that depend on the fabric and stress history. During reversal loading, the (1-Czpk2) term
approaches unity and Kp evolves as it previously had. The roles of each of the other terms are discussed
below.

Czpk1 and Czpk2 are terms that start from zero and grow to be unity for uni-directional growth of
fabric which is the case during non-reversing loading conditions. These two terms differ by the rate
under which they approach unity by the use of the constant zmax /5 or zmax/100 with these respective
values chosen for their ability to better approximate the engineering behaviors of interest and
correlations. For full reversal loading where the fabric alternates between positive and negative values,
these terms will both go to zero.

Cpzp2 starts initially at zero and stays equal to zero until fabric is formed. After fabric is formed, this
term quickly transitions to unity for values of mean effective stress p that are less than the value that p
had when the maximum fabric was formed (pzp). If p increases beyond the value of pzp the term will
return to zero according to the Macaulay brackets.

The values for the calibration parameters CKp and CKαf were chosen for their ability to reasonably
approximate the targeted behaviors, as discussed later. Setting CKp to a default value of 2.0 was found
to produce reasonable responses with particular emphasis on improving (reducing) the equivalent
damping ratios at shear strains of 1 to 3% in drained cyclic loading. The parameter CKαf is particularly
useful for adjusting the undrained cyclic loading response with sustained static shear stresses; a default
calibration which depends on DR is presented later.

32
 The cumulative effect of the above parameters can be understood as follows. If a soil is strongly
loaded in uni-directional loading and forms significant amount of fabric and is then unloaded, then
upon subsequent reloading the terms Cpzp2 and Czpk1 will be unity and CKα will become large. If the
loads are increased to where the soil is being sheared and forming fabric at even higher stresses (higher
values of p than fabric was previously formed at) then CKα will be unity (Cpzp2 = 0). In this way, an
element that has developed strong fabric under monotonic or cyclic loading without reversal of the total
shear stress direction (e.g., an element within a steep slope where the static shear stresses are greater
than the cyclic shear stresses) will, when unloaded and reloaded, be initially much stiffer (increased
Kp) followed by a softening (smaller Kp) if the soil is loaded into virgin territory.

Effect of fabric on plastic volumetric dilation

A rotated dilatancy surface with slope MdR which evolves with the history of the fabric tensor z was
added to the framework of the model to facilitate earlier dilation at low stress ratios under certain
loading paths (Ziotopoulou 2014, Ziotopoulou and Boulanger 2016). The rotated surface, schematically
illustrated in Figure 2.10 as a line in q-p space and Figure 2.11 as a circular surface on a stress-ratio
graph of ryy versus rxy, is equal to the original dilatancy surface scaled-down by a factor Crot1,

Md
M dR = (65)
Crot1
2  −z : n
Crot1 = 1 + (1 − Czin1 )  1 (66)
2 zmax
where Md is the slope of the unrotated dilatancy surface. Experimental results (Ziotopoulou 2014,
Ziotopoulou and Boulanger 2016) indicate that the loading history, the loading direction and the
loading pattern play important roles in the response of the soil to irregular cyclic loading. Thus the
scaling factor that defines the rotated dilatancy surface was made dependent on whether fabric is
favorable (z : n > 0) or unfavorable (z : n < 0) and on the factor Czin1 which is an indirect measure of
whether there are reversals or not,

 z :n−z:n 
Czin1 = 1 − exp  −2.0 in  (67)
 zmax 
where zin is the fabric tensor at the beginning of the current loading branch. Czin1 can take values ranging
from 0, when there are no reversals, to 1, when there are reversals. The rotated dilatancy surface is
operating only for loading with an unfavorable fabric since the factor Crot1 becomes 1 when the fabric
is favorable (i.e., −z : n = 0).

A back-stress ratio tensor for the rotated dilatancy surface (αdR) was introduced as,

 dR =
1
2
( M dR − m ) n (68)

33
Dilation occurs whenever the term (αdR − α) : n is negative whereas contraction occurs when it is
positive. The calculation of D is still treated separately during dilation and contraction.

D during dilation is now computed according to the following expressions. First, a value for D is
computed from the rotated dilatancy surface,

− z : n ( dR −  ) : n
Drot = Ad (69)
2 zmax CDR
where the CDR factor is applied to reduce the rate under which dilatancy is increasing and is discussed
further below. Second, another value for D is computed that would be obtained from the non-rotated
dilatancy surface,

(
Dnon−rot = Ad  − −(α d − α) : n ) (70)
The Macaulay brackets in the above expression ensure that Dnon-rot is equal to zero whenever (αd − α) : n
> 0 while (αdR − α) : n < 0. Lastly, the operating value of D is selected from the above two values based
on,

if Dnon − rot  Drot  D = Dnon− rot

M b − M cur (71)
else D = Dnon − rot + ( Drot − Dnon− rot ) 
M −M
b cur
+ 0.01
The above logic is illustrated in Figure 2.12 where D is plotted for a half-cycle of loading that goes
from contraction to dilation. This figure shows that Dnon-rot is used whenever it is smaller (more
negative) than Drot. For cases where Drot is smaller than Dnon-rot, the value of D is interpolated based on
the additional term on the right that multiplies the difference between Drot and Dnon-rot. This interpolation
term is close to unity for stress ratios away from the bounding surface (Mcur < Mb), such that D will be
equal to Drot as illustrated in the figure. However, this term will also go smoothly to zero as the stress
ratio gets close to the bounding surface, so that dilatancy smoothly goes to zero as a soil approaches
the critical state where M = Md = Mb. The constant of 0.01 in the denominator controls the rate under
which D goes to zero as the stress ratio nears the bounding surface and was found to provide reasonable
results in trial simulations.

The factor CDR in the denominator of the expression for Drot is applied so that the D computed based
on the rotated dilatancy surface is consistent with experimental observations. Its value, for the default
calibration described later, has been made dependent on the initial DR of the soil.

Lastly, the parameter Ad in the expressions for both Drot and Dnon-rot is expressed as,

Ado ( Czin 2 )
Ad = 3
z 2
 −z : n 
 ( C ) ( C pzp )( C pmin ) ( C zin1 ) + 1
(72)
 1 −
cum 2
 
 zmax  2  z peak 

34
 1
C pzp = 5
  (73)
1 +  2.5 p
 pzp 
1
C pmin = 2
1+  pmin 2  (74)
 p 
 z :n−z:n 
Czin1 = 1 − exp  −2.0 in  (75)
 zmax 
zcum − z peak
1 + Czin1
3zmax
Czin 2 = (76)
zcum − z peak
1 + 3Czin1
3zmax

Consider the six terms added to the denominator of the expression for Ad. The first term [zcum2/zmax]
facilitates the progressive growth of strains under symmetric loading by reducing the dilatancy that
occurs when a liquefied soil has been sheared through many cycles of loading; note that this term
progressively increases with subsequent cycles of loading. The second term facilitates strain-hardening
when the plastic shear strain reaches the prior peak value, wherein the term approaches zero (i.e., when
z:n approaches zpeak√2) and the dilation rate consequently rapidly approaches the virgin loading value
of Ado. The third term C is a calibration constant that can be used to modify the rate of plastic shear
strain accumulation. The fourth term Cpzp causes the effects of fabric on dilation to be diminished
(erased) whenever the current value of p is near the value of pzp; this term enables the model to provide
reasonable predictions of responses to large numbers of either drained or undrained loading cycles. The
fifth term Cpmin provides a minimum amount of shear resistance for a soil after it has temporarily
reached an excess pore pressure ratio of 100%; This term is almost zero when p'=0, such that the soil
will initially dilate until some minimum p' has developed, after which the term quickly approaches 1.0.
The parameter pmin2 is currently set to become equal to 5% of the value of p' at consolidation (which is
the value that exists when the flag FirstCall –see Section 3– was last set equal to 0), with the minimum
value of pmin2 being 10 times the minimum value of p' (i.e., pmin = 1/200 times the larger of pA or the
value of p' at consolidation). The sixth term Czin1 facilitates strain-hardening when stress reversals are
not causing fabric changes; i.e., when the initial and current fabric terms are close to equal, the term
Czin1 goes to zero. Lastly, the second term in the numerator, Czin2, causes the dilatancy to be decreased
by up to a factor of 3 under conditions of large strains and full stress (and fabric) reversals, which
improves the prediction of cyclic strain accumulation during undrained cyclic loading.
An additional constraint is placed on D during dilation at very low effective stresses. For p < 2pmin,
the value of D cannot be smaller in magnitude than computed by the following expression,

35
 2 pmin − p
D = −3.5 Ado M b − M d for pmin  p  2 pmin (77)
pmin

This expression ensures that the model will, for dense-of-critical soils (i.e., Mb > Md), be dilative when
p falls below 2pmin.

Effect of fabric on plastic volumetric contraction

Dafalias and Manzari (2004) used the fabric tensor to modify the dilatancy during contraction
(D > 0) as follows,

D = Ad ( α d − α ) : n  ( 1 + z : n ) (78)

This relationship enhances the volumetric contraction whenever the fabric is favorable (z:n ≥ 0), based
on the term 1+<z:n> as recommended by Dafalias and Manzari (2004).

The effect of fabric on dilatancy during contraction was modified for the present model as,

D = Adc  ( α − α
(α d
− α) : n
):n+C
2
app
 C p min 2
 ( α d − α ) : n + CD
in in (79)

Ado ( 1 + z : n )
Adc = (80)
hp Cdz

CD = 0.1 (81)
2 z : n
Cin = (82)
2 zmax
 2 z peak   zmax  1
Cdz = 1 − Crot 2   
 zmax   zmax + Crot 2  zcum  z (83)
  1 + max
2

( = 1 − Czpk 2 )
z peak
Crot 2 = 1 −
zmax (84)
zcum +
100
C p min 2 = 0 for p  2 pmin
C p min 2 = 1 for p  18 pmin
(85)
p − 2 pmin
C p min 2 = otherwise
16 pmin

36
 The factor Cin in the expression for D has been modified so it now depends on fabric; Cin is zero
for unfavorable fabric, and increases with increasing z:n for favorable fabric to enhance the contraction
rate at the start of an unloading cycle (note that D would be zero at the start of an unloading cycle if Cin
was zero).

The term Cdz in the denominator of the expression for Adc serves to increase the rate of contraction
as zpeak nears zmax or as a large amount of cumulative fabric formation/destruction has taken place. This
term was developed for improved modeling of the cyclic strength of denser sands, for which the value
of hp can be on the order of 100 (Figure 2.7). The degrading of the denominator as zpeak or zcum increases
enables the generation of high excess pore pressures at higher loading levels, and controls the slope of
the CRR versus number of uniform loading cycles relationship obtained for undrained element loading.
Note that the denominator degrades whether fabric is favorable or not, but that the overall rate of
contraction is more enhanced if the fabric is favorable (z:n ≥ 0). The factor Crot2 was introduced into
the factor Cdz to provide better control over the rate of contraction as zpeak nears zmax or as a large amount
of cumulative fabric formation/destruction has taken place. The factor Crot2 takes values that range from
1 for loading with zero fabric or cyclic loading that causes reversals of fabric (since zcum will become
much larger than zpeak), to 0 for loading that causes fabric to grow monotonically in one direction such
as in non-reversal cyclic loading (since zcum will equal zpeak ). Lastly, the limit on the minimum value
of Cdz is required for avoiding division by zero and to avoid over-estimating contraction rates (i.e.,
small values of hp and large values of zpeak or zcum).

The term Cpmin2 slows the rate of contraction when p is approaching its minimum allowable value,
and stops further contraction when p is less than twice the minimum allowable p.

Effect of fabric on the elastic modulus

The elastic shear modulus and elastic bulk modulus may degrade with increasing values of
cumulative plastic deviator strain term, zcum. This component of the model was added to account for
the progressive destruction, with increasing plastic shear strains, of any minor cementation bonds or
other ageing- or strain history-related phenomena that produced an increase in small-strain shear
modulus. The destruction of minor cementation by plastic shear strains is evidenced in the field by
measurements of shear wave velocities in sand that are lower after earthquake shaking than before
earthquake shaking (e.g., Arai 2006). The degradation of the elastic shear modulus is computed as,

 zcum 
1
 1 + 
 p  2 zmax
G = Go p A   C SR
  (86)
 pA   1 + zcum C 
 GD 
 zmax 
where CGD is the factor by which the shear modulus is degraded (divided) at very large values of zcum.
This change in the elastic shear modulus G causes the bulk modulus K to progressively decrease with
increasing zcum. The change in K improves the model's ability to track the stress-strain response of
liquefying sand. In particular, decreasing K with increasing zcum reduces the rate of strain-hardening
after phase transformation at larger shear strain levels, and improves the ability to approximate the
hysteretic stress-strain response of a soil as it liquefies.

37
Effect of fabric on peak mobilized friction angles in drained and undrained loading

Kutter and Chen (1997) showed that plastic dilation rates are different in drained and undrained
loading of the same clean sand, with the consequence being that the peak mobilized friction angles are
also different for drained and undrained loading. This aspect of behavior would appear to be
contradictory to having a bounding surface that is only dependent on the relative state of the sand (i.e.,
through the parameter nb) if the mobilized friction angles for drained and undrained loading paths are
both controlled by the bounding surface. The model presented herein produces the same peak mobilized
friction angles for drained and undrained loading because both conditions become limited by the same
bounding surface. This aspect of behavior deserves closer examination in future efforts.

2.9 Post-shaking reconsolidation

Volumetric strains that develop during post-liquefaction reconsolidation of sand are difficult to
numerically model using the conventional constitutive separation of strains into elastic and plastic
components since a large portion of the post-liquefaction reconsolidation strains are due to
sedimentation effects which are not easily incorporated into either the elastic or plastic components of
behavior. Single element simulations using various constitutive models show that they generally predict
post-liquefaction reconsolidation strains that are an order of magnitude smaller than observed in various
experimental studies (e.g., Ziotopoulou and Boulanger 2013b, Howell et al. 2014).

The present model was modified to provide more realistic estimates of reconsolidation strains
during the post-shaking portion of a numerical simulation. The modification involved the pragmatic
approach of reducing the post-shaking elastic shear modulus G (and hence elastic bulk modulus K)
which increases reconsolidation strains, thereby compensating for the sedimentation strains which are
not explicitly modeled. The user may activate this feature after the end of strong shaking, such that
post-liquefaction reconsolidation strains are better approximated in the remainder of the simulation.
This feature should not be activated for the strong shaking portion of a simulation.

The post-shaking elastic moduli are determined by multiplying the conventional elastic moduli
(computed using the expressions described earlier) by a reduction factor Fsed as,

G post −shaking = Fsed G (87)

K post −shaking = Fsed K (88)

The Fsed value is computed as,

2
 p 
Fsed = Fsed ,min + (1 − Fsed ,min )   1
 
(89)
 20 psed

38
 0.25
 zcum  M cur
 = psed
psed  o  1− d (90)
z +
 cum max 
z M

Fsed,min = 0.04 (91)

Patm
 o =−
psed (92)
5
where the constant value Fsed,min represents the smallest value that Fsed can attain, and the parameter
p'sed,o is the mean effective stress up to which reconsolidation strains are enhanced. The value of Fsed
progressively reduces from unity toward the value of Fsed,min as zcum progressively increases and
provided that Mcur is less than Md. Setting Fsed,min to 0.04 was found to produce reasonable responses
as shown later. The user can select other values for p'sed,o and Fsed,min.

2.10 Summary of constitutive equations

The constitutive equations for the model presented herein are summarized in Table 2.1 along with
the equations for the Dafalias and Manzari (2004) model.

39
Table 2.1. Comparison of constitutive equations
Dafalias-Manzari (2004) model Present model
Critical state line Critical state line

p  R =
R
− DR
ec = eo −   cs   p 
 pA  Q − ln 100 
 pA 
Elastic deviatoric strain increment Elastic deviatoric strain increment
ds ds
de el = de el =
2G 2G
( 2.97 − e )
2
 p 
1
 z 
1 + cum 
2

1
G = Go p A    p  2
zmax
1+ e  pA  G = Go p A   CSR  
 A
p  z 
 1 + z CGD 
cum

 max 
mSR
 M 
1 − CSR ,o   b 
CSR = M  1
  M 
mSR

1 − CSR ,o   b  
  M   initial
CSR ,o = 0.5
mSR = 4
Elastic volumetric strain increment Elastic volumetric strain increment
dp dp
d vel = d vel =
K K
2 (1 + v ) 2 (1 + v )
K= G K= G
3 ( 1 − 2v ) 3 ( 1 − 2v )
Yield surface Yield surface

f = ( s − pα ) : ( s − pα )  f = ( s − pα ) : ( s − pα ) 
1 1
2
− 2 pm = 0 2
− 1 pm = 0
3 2

40
Plastic deviatoric strain increment Plastic deviatoric strain increment
de pl = L R de pl = L R
 1  1
R = Bn − C  n2 − I  + DI 1 1
R = R + DI = n + DI
 3  3
3 3
3 1− c M b = M  exp ( −nb R )
B = 1+ g(  ,c)cos( 3 )
2 c M = 2  sin (cv )
3 1− c
C =3 g ( ,c ) αb = 1  M b − m n
2 c 2
Qext
c=
( α b − a ) : n 
0.5
Qcompr  
K p = G  ho Crev
2c exp ( ( α − ain ) : n ) − 1 + C 1
g ( ,c ) =
(1 + c ) − (1 − c ) cos ( 3 ) CK

z 
M b = M  exp ( −nb R ) 1 + CK p  peak  ( α b − a ) : n 1 − Czpk 2
 zmax 
αb = 2  g ( , c ) M b − m  n (α − α ) : n
app
3 Crev =
in
for (α − α ) : n  0
p

2 ( αb − α ) : n (α − α ) : n
true
in
in

K p = G  ho 
3 ( α − αin ) : n = 1 otherwise
CK f
CK = 1 +  C pzp 2  Czpk1
(
1 + 2.5  ( α − a ):n )
2
true
in

z peak
Czpk1 =
z
zcum + max
5
z peak
Czpk 2 =
z
zcum + max
100

41
 − − ( pzp − p )
C pzp 2 =
− − ( pzp − p ) + pmin
ho
C 1 =
200
CKp =2

Plastic volumetric strain increment Plastic volumetric strain increment

d  vpl = L D d  vpl = L D

M d = M  exp ( n d  R ) M d = M  exp ( n d  R )

αd = 2  g ( ,c) M d − m  n Md
3 M dR =
Crot1
2  −z : n
Crot1 = 1 +  (1 − Czin1 )  1
2 zmax
D = Ad  ( αd − α ) : n 
 z :n−z:n 
Czin1 = 1 − exp  −2.0 in 
Ad = Ao ( 1 + z : n )  zmax 
 ( M d − m) n
1
d =
2
 ( M dR − m ) n
1
 dR =
2

If dilating (D<0):

Dnon −rot = Ad  ( α d − α ) : n 

− z : n ( dR −  ) : n
Drot = Ad  
2 zmax CDR

42
 if Dnon − rot  Drot  D = Dnon− rot
M b − M cur
else D = Dnon − rot + ( Drot − Dnon− rot ) 
M b − M cur + 0.01
Ado ( Czin2 )
Ad = 3
 zcum
2
 −z : n 
−  ( C ) ( C pzp )( C p min ) ( Czin1 ) + 1
2
  1
 zmax 2  z peak 
 −1  M b  −1  M 

 sin   − sin  
1   2   2 
Ado = 
0.4 Mb −Md
1
C pzp = 5
 2.5 p 
1+ 
 pzp 
1
C p min = 2
1+   pmin 
 p 
 z :n−z:n 
Czin1 = 1 − exp  −2.0 in 
 zmax 
zcum − z peak
1 + Czin1
3zmax
C zin2 =
z −z
1 + 3C zin1 cum peak
3zmax

43
 If contracting (D≥ 0)

D = Adc  ( α − α
(α d
− α) : n (α d
− α) : n
):n+C
2
app
 C p min 2  1.5  Ado
in in  (α d
− α ) : n + CD (α d
− α ) : n + CD

Ado ( 1 + z : n )
Adc =
hp Cdz
2 z:n
Cin =
2 zmax
 2 z peak   zmax  1
Cdz = 1 − Crot 2   
 zmax   zmax + Crot2 zcum  z
  1 + max
2

( = 1 − Czpk 2 )
z peak
Crot2 = 1 −
z
zcum + max
100
CD = 0.1

C p min 2 = 0 for p  2 pmin

C p min 2 = 1 for p  18 pmin
p − 2 pmin
C p min 2 = otherwise
16 pmin

(
h p = hpo exp −0.7 + 7.0 ( 0.5 −  R )
2
) for  R  0.5
h p = hpo exp ( −0.7 ) for  R  0.5

44
Fabric-dilatancy tensor update
( )
Fabric-dilatancy tensor update if α − α : n < 0
d

dz = −cz −d  pl
( zmax n + z )
−d  vpl
v
cz
dz = − ( zmax n + z )
zcum D
1+ −1
2zmax

dzcum = dz
Stress increment Stress increment
2Gn : de − n : rKd v 2Gn : de − n : rKd  v
L= L=
K p + 2G ( B − C  trn3 ) − KDn : r K p + 2G − KDn : r

dσ = 2Gde + Kd  v I dσ = 2Gde + Kd v I − L ( 2Gn + KDI )

   1  
− L  2G  Bn − C  n 2 −  I  + KDI 
   3  
Post-shaking reconsolidation
G post −shaking = Fsed G
K post −shaking = Fsed K
 p 
Fsed = Fsed,min + (1 − Fsed,min )   1
 
 20 psed
0.25
 zcum  M cur
 
psed = psedo   1− d
z +
 cum max 
z M
Fsed,min = 0.04
Patm
 o =−
psed
5

45
 R = DR,CS − DR

R
DR,CS =
100 (1 + 2Ko ) vc
Q − ln
3 pA

Figure 2.1. Definition of the relative state parameter index, R (Boulanger 2003a) and the effects of
varying Q and R.

46
 Figure 2.2. Schematic of yield, critical, dilatancy, and bounding lines in q-p space
(after Dafalias & Manzari 2004). Relative location of dilatancy and bounding lines corresponds to
dense-of-critical states of stress.

47
Figure 2.3. Schematic of the bounding, dilatancy, and yield surfaces on the ryy-rxy stress-ratio plane
with the yield surface, normal tensor, dilatancy back-stress ratio, and bounding back-stress ratio.
Relative locations of the surfaces differ from those of Figure 2.2.

48
Figure 2.4. Schematic showing definitions of back-stress ratio tensors on the yy-xy plane for: (a) a loading history with reversals in
the sign of the shear stress ratios, and (b) a loading history with a recent loading reversal that does not involve reversal of the sign of
the shear stress ratios.

49
Figure 2.5. Example scenarios of back-stress ratio tracking: (a) positive loading direction with
𝑎𝑝𝑝
minimum value of back-stress ratio (𝜶𝑚𝑖𝑛 ) being less than zero such that 𝜶𝑖𝑛 = 𝜶𝑡𝑟𝑢𝑒 𝑖𝑛 , (b)
positive loading direction with minimum value of back-stress ratio (𝜶𝑚𝑖𝑛 ) being greater than zero
𝑎𝑝𝑝
such that 𝜶𝑖𝑛 = 𝜶𝑚𝑖𝑛 , (c) negative loading direction with maximum value of back-stress ratio
𝑎𝑝𝑝
(𝜶𝑚𝑎𝑥 ) being greater than zero such that 𝜶𝑖𝑛 = 𝜶𝑡𝑟𝑢𝑒
𝑖𝑛 , (d) negative loading direction with
𝑎𝑝𝑝
minimum value of back-stress ratio (𝜶𝑚𝑎𝑥 ) being less than zero such that 𝜶𝑖𝑛 = 𝜶𝑚𝑎𝑥 .

50
Figure 2.6. Drained DSS simulations showing αxy versus γ with the points corresponding to the
current back-stress ratio α, the apparent initial back-stress ratio αinapp, the true initial back-stress
ratio αintrue, and the previous initial back-stress ratio αinp for: (a) monotonic shearing with one
intermediate unload-reload cycle, and (b) a more general sequence of cyclic loading.

51
Figure 2.7. Variation of contraction rate function hp with relative state parameter index R and the
contraction rate parameter hpo.

52
Figure 2.8. Undrained cyclic DSS loading response for DR = 35% with an initial static shear stress
ratio of α=0.0, showing the variation in stresses, stress ratios, and fabric tensor terms.

53
Figure 2.9. Undrained cyclic DSS loading response for DR = 35% with an initial static shear stress
ratio of α=0.20, showing the variation in stresses, stress ratios, and fabric tensor terms.

54
Figure 2.10. Schematic of the rotated dilatancy line added to PM4Sand Version 3, along with the yield,
critical, dilatancy, and bounding lines in q-p space. Relative location of dilatancy and bounding lines
corresponds to dense-of-critical states of stress.

55
Figure 2.11. Schematic of the rotated dilatancy line, along with the bounding, dilatancy, and yield
surfaces on the ryy-rxy stress-ratio plane with the yield surface, normal tensor, dilatancy back-stress
ratio, and bounding back-stress ratio. Relative locations of the surfaces differ from those of
Figure 2.10.

56
 Figure 2.12. Schematic of the dilatancy D calculation based on the stress state with regards to the
rotated dilatancy (MdR), dilatancy (Md) and bounding (Mb) surfaces during a half-cycle of loading that
goes from contraction to dilation.

57
3. MODEL IMPLEMENTATION

The model has been implemented as a user defined material (udm) for use with the commercial
finite difference program, FLAC 8.1 (Itasca 2019). This section includes a brief description of the
mixed discretization scheme used in FLAC, the numerical implementation scheme used for PM4Sand,
some additional comments on alternative implementation schemes, and information regarding the
dynamic link library (DLL) for PM4Sand.

3.1 Aspects of FLAC's numerical approach

Explicit integration

FLAC is an explicit finite difference program which uses time steps equal to or smaller than the
minimum time required for waves to travel between any pair of nodes. This approach ensures that
physical information does not propagate faster than numerical information. FLAC computes a default
time step based on the properties of the model (e.g., element size, material stiffness, permeability,
damping). Users may specify a time step that is smaller than the default value.

Obtaining numerically convergent solutions to nonlinear problems using FLAC requires that:

1) integration of the constitutive models be convergent, and

2) the explicit global solution is convergent.

The default time step computed by FLAC does not necessarily ensure a numerically convergent
solution, especially for FLAC models that are subjected to very high loading rates. Convergence of the
constitutive model's integration depends more strongly on the strain increment size, which is dependent
on both the loading rate and time step size. Convergence of the explicit global solution depends more
strongly on the sizes of the stress increments generated in the materials, which again are only indirectly
controlled by the default time step size. For this reason, the user needs to evaluate the sensitivity of the
solution to the time step size and not automatically assume that the default time step size ensures a
convergent solution.

Mixed discretization scheme

FLAC uses a mixed discretization technique in which each quadrilateral zone (analogous to an
element) is subdivided internally by its diagonals into two overlaid sets of constant-strain triangles. The
term “mixed” stems from the fact that different discretizations are used for the isotropic and deviatoric
parts of the strain and stress tensor (Marti and Cundall 1982). Isotropic stress and strain components
are taken to be constant over the whole quadrilateral zone, while the deviatoric components are treated
separately for each triangular sub-zone. Essentially, the shear strains are computed and maintained for
each individual triangle, while the volumetric strains are computed for each quadrilateral as a weighted
average of the volumetric strains within the juxtaposed pairs of triangles. Hour-glass modes are resisted
by shear stresses generated in the triangles and the scheme accurately predicts plastic collapse loads
because constant volume deformations are possible within the quadrilaterals. Note that discretization
using triangles or four-node quadrilaterals alone would result in meshes that are over-constrained (too
stiff) and which would tend to over-predict plastic collapse loads. Since each quadrilateral can be

58
divided by two possible diagonals, a symmetric response of this discretization can only be obtained by
running two complete meshes in parallel, each representing one half of the overall stiffness. At the end,
the procedure reduces the number of constraints on plastic flow and, at the same time, reduces unwanted
hour-glassing by ensuring that hourglass modes produce non-zero stresses.

The implementation of a complex constitutive model in FLAC requires special attention to the way
stresses and strains are handled under FLAC's mixed discretization scheme. During each time step,
FLAC calls the constitutive model once per triangular subzone (four times per zone). The isotropic
components of the stress outputs from the four subzones are then averaged internally by FLAC
according to the Mixed Discretization scheme. A consequence of this averaging of stresses is that the
final stress state for any subzone is unlikely to satisfy the consistency condition of elasto-plastic models,
meaning that the newly calculated stress states will not necessarily lie on the yield surface.
Andrianopoulos (2006) addressed this problem by adopting a vanished elastic region in their elasto-
plastic model.

3.2. Implementation of PM4Sand in FLAC

The implementation scheme for PM4Sand and how it relates to the challenges posed by FLAC's
mixed discretization scheme (Section 3.1) are described here. Recall that each zone (consisting of four
subzones) will start off at the beginning of each time step with a stress state and will be loaded by a
strain increment whose volumetric components are the same in all four subzones while their deviatoric
components are different (due to the mixed discretization scheme). The constitutive model will be
called once per subzone (four times per zone) to obtain stresses from strains according to the following
equation where Cijkl denotes the constitutive law:

σij(i +1) = σij(i ) + dσij(dt ) =  ij(i ) + Cijkl dεij(dt ) (93)

At the end of the step, each subzone will have its own stress state, which will be handled by FLAC
independent of the constitutive model, and its own internal parameters, which FLAC will be unaware
of. The subzones can therefore all have different stress states at the beginning of the next loading
increment, and as such would need to maintain their own sets of internal parameters.

The current implementation scheme for PM4Sand is illustrated in Figure 3.1 and described by the
pseudo-code listed in Table 3.1. At the end of each time step, the stress and internal variables are
averaged over the four subzones. A drift correction is applied to ensure that the averaged stresses and
internal variables satisfy the consistency condition; the correction involves projecting the back-stress
ratio in the direction of the zone-averaged stress ratio. Another correction is applied if the zone-
averaged stress ratio lies outside the bounding surface; the correction involves projecting the zone
averaged stress ratio back along a normal to the bounding surface. The zone-averaged stresses are then
used to compute a new dilatancy D and plastic modulus Kp that are consistent with the average response
of the zone over this step. These values for D and Kp are then used by all four subzones in the next time
step (i.e., the values of D and Kp lag one step behind the time step for which they were determined);
note that this approach is used by other elasto-plastic models available in FLAC. Consequently, the
four subzones will use a common D and Kp during each time step. Most other internal parameters are
also computed and retained at the zone level, as described by the pseudo-code in Table 3.1.

59
 Two other implementation schemes for PM4Sand were explored for comparison purposes and
found to have problems. The first of these alternative implementations was that used in Version 1 of
PM4Sand. In this implementation, each subzone had its own D and Kp and developed its own internal
variables (e.g., fabric, back-stress ratios, history terms) for each loading increment or step. At the end
of the step, the stress and internal variables were averaged at the zone level and the drift and bounding
surface corrections were applied. The four subzones therefore started each loading step with a common
set of stresses and internal variables, but each could have greatly differing values for D and Kp
depending on the loading direction imposed on each subzone. In highly nonlinear loading steps, it was
possible for one or two of the overlapping subzones to be strongly contractive (e.g., perhaps because
of a reversal in loading direction) while the other subzones were strongly dilative, such that the
incremental changes in stresses between the four subzones had competing effects on the zone's average
behavior. This implementation was found to sometimes lead to unusual deformation modes in zones
that were connected to piles by FLAC's interface springs. The unusual deformation modes are believed
to be due to strong differences in loading directions and conditions between the subzones of zones
being loaded by interface springs. This problem was effectively eliminated by the current
implementation described in Table 3.1. The second of these alternative implementations increased the
independence of the subzones, just to explore how it would affect behavior. In this implementation,
each subzone (triangle) retained its own memory and history of stresses and internal parameters. This
approach led to nonsensical results between the overlapping triangular subzones, especially when the
loading conditions were highly nonlinear. For example, the external stresses sometimes could be
carried by only two of the overlapping triangles (each having twice the correct stresses) while the
stresses in the other two overlapping triangles went to zero. The experiences with these two alternative
implementation schemes illustrate how FLAC's mixed discretization scheme requires special
considerations when implementing highly nonlinear constitutive models.

The current implementation also includes a scheme to reduce hour-glassing modes which developed
in liquefied zones in some cases because the four subzones no longer have independent states of stress.
The four subzones have, in parallel to the PM4Sand constitutive model, an elastic-plastic resistance to
shear stresses which acts independently in each of the subzones. The properties of this parallel elastic-
plastic model are set at the instance when PM4Sand is initialized; the elastic moduli of the parallel
elastic-plastic model are set equal to 0.01 times those for PM4Sand, and its plastic shear strength (chg)
is set as the product of a strength ratio (crhg) times the mean effective stress in the zone. If the user
specifies values for both chg and crhg, then chg is taken as the greater of the specified chg value and the
value computed using the specified crhg. The default value for crhg is 0.005. The parallel elastic-plastic
model only responds to deviatoric strains (producing shear stresses) and not to volumetric strains
(producing no mean stress). This nominal amount of independent shearing resistance in the subzones
was found to adequately control hour-glassing modes for the range of problems examined to date.

Implementation of PM4Sand uses explicit integration and thus the user should routinely check that
the solutions are not sensitive to time step size. The addition of substepping could improve the
constitutive model's integration but would not eliminate the need to evaluate the effect of time step size
on the global solution. In the developers’ experiences, the default time steps of FLAC in dynamic
analyses of liquefaction problems have been small enough to ensure that numerical solutions are not
significantly affected by time step size, and thus the additional computational cost of including
substepping at the constitutive level was not considered necessary. Examples of the effects of time step
size are presented in Section 3.3.

60
 Numerical stability of the implemented model has been evaluated for a wide range of simulations
of both element responses and system responses using the default range of parameters which are also
summarized in the next section. Numerical stability problems may, however, develop when using input
parameters which fall outside the ranges explored during model development, calibration, and
implementation. Some initial bounds have therefore been placed on certain parameters whenever
parametric analyses identified the potential for such problems; e.g., the minimum value of mean stress
is limited to 0.5 kPa or 0.005 times the initial consolidation stress; the relative density was limited to
values less than 1.2. The user must be aware that other limits may be identified as additional analyses
explore a broader range of the possible input parameters.

3.3 Effect of time step size on element responses

Numerical convergence of the current implementation of PM4Sand was evaluated by running

numerous problems using a range of dynamic time steps (dydt), beginning with the default (maximum)
time step computed by FLAC and then trying smaller and smaller values. These comparisons have
shown that the solutions are not sensitive to the time step size for the range of problems and loading
rates examined. The user must always check the sensitivity of boundary value problems to the time step
size, however, as the accuracy of the explicit integration is strongly dependent on the size of the strain
increments which are only partly controlled by the time step size.

For example, the effect of time step size (or strain increment size) on integration of the PM4Sand
model is shown in Figure 3.3a for a single element simulation of a cyclic drained DSS test for sand at
DR = 55% at σ'vo = 100kPa. The element was subjected to two cycles of strain-controlled loading with
a single-amplitude shear strain of 1%. The strain rate was constant, with each cycle having a total
duration of 1 sec (i.e., average loading frequency was 1 Hz). The default time step was 1.038e-4 s and
the strain rate was 4 %/s which gives a step size of Δγ = 4.15e-6 %/step. To evaluate different Δγ’s, the
time step was reduced by factors of ½, ¼, and 1/8. The simulated stress-strain responses showed
minimal differences, indicating that the integration was sufficiently accurate for practical purposes.

A second example of the effect of time step size is shown in Figure 3.3b for a stress-controlled
cyclic undrained DSS test for sand at DR = 55% at σ'vo = 100kPa. The default time step was 3.604e-5 s
and the strain rate was again 4 %/s which gives a step size of Δγ = 1.44e-6 %/step. The default time
step is smaller for the undrained element test because of the higher wave speed in the pore water. To
evaluate different Δγ, the time step was again reduced by factors of ½, ¼, and 1/8. The simulated stress-
strain responses again showed minimal differences, indicating that the integration was sufficiently
accurate for practical purposes.

Figure 3.3a and b presents the same examples as Figure 3.2a and b but for a very high strain rate of
12%/s. The corresponding step sizes for the drained (a) and undrained (b) cases are 12.45e-6%/step and
4.32e-6 %/step, respectively. To evaluate different Δγ’s, the time step was again reduced by factors of
1/2, 1/4 and 1/8. The simulated stress-strain responses for the drained case showed minimal differences,
whereas the undrained case showed some slight differences. The differences for the undrained case are
attributed to the very high strain rate of 12%/sec, which was only used to examine the limits of
behaviors.

61
 Note that comparisons of solutions at different step sizes Δγ generally cannot be made by just
varying the strain rate or cyclic loading frequency. FLAC is always solving the dynamic equation of
equilibrium so changing the strain rate by changing any loading rate parameter also changes the
dynamic excitation for the system, which can cause a change in the dynamic response of the element.
In that case, any changes in the stress-strain response caused by changes in loading rate parameters
may be a realistic simulation result that reflects the change in the dynamic excitation of the element.

3.4 Effect of time step size on system responses

The effect of time step size on the solution of full boundary value problems can similarly be
examined by repeating simulations with successively smaller dynamic time step sizes. As an example,
the effect of time step size on the seismic response of an embankment dam with PM4Sand used for the
shells [(N1)60cs = 25] and foundation alluvium [(N1)60cs = 15] is illustrated in Figure 3.4. Time histories
of crest settlement and horizontal displacements at points on the upstream and downstream faces are
shown for the default time step size and for time step sizes that are ½ and ¼ the default size. The
differences in the displacements at these three points are small enough to be virtually indistinguishable
in the figure. The effect of time step size has been observed to be more significant in some other
boundary value problem simulations (e.g., 5% differences), but they are generally small enough for
practical applications.

The sensitivity of simulation results to the dynamic time step size should always be evaluated as part
of the sensitivity studies. As previously discussed in Section 3.1, the effects of changing time step size
may result from a combination of the effects on the constitutive model integration and the explicit
global solution. Implementation of substepping in the constitutive model may reduce its effect, but will
not remove the need to check the global solution's sensitivity to the step size. Since the sensitivity to
step size should always be checked, the additional computational costs of including substepping at the
constitutive level was not considered warranted at this time.

3.5 DLL module

The PM4Sand model was coded in C++ and compiled as a User Defined Model (UDM) dynamic
link library (DLL) in Microsoft Visual Studio 2015 for FLAC 8.1 and in Microsoft Visual Studio 2022
for FLAC2D 9.00. The steps required for using a DLL are described in the respective FLAC/FLAC2D
manuals.

Automatic loading of the DLL file in FLAC8.1

(1) Load the DLL file in the /Exe64/plugins/models subdirectory of the folder where FLAC has
been installed.

(2) Open the FLAC 8.1 executable file or the FLAC graphical user interface. If the DLL is properly
located, then the model should be automatically loaded. In order to verify that it has been loaded,
the user can type “print model” in the console. If the model has been loaded then it should
appear as “pm4sand” under the list of “Currently loaded CPP models”.

62
 (3) Before constitutive model plug-ins can be assigned to zones, the model must be configured for
their use by giving the config cppudm command. Otherwise, the user will get a “model will
not cycle” error message.

Automatic loading of the DLL file in FLAC2D 9.00

(1) Load the DLL file in the /Exe64/plugins/cmodels subdirectory of the folder where FLAC2D has
been installed.

(2) Open the FLAC2D 9.00 executable file or the FLAC2D graphical user interface. If the DLL is
properly located, then the model should be automatically loaded and the console will indicate
“flac2d>program load cmodel "plugins/cmodel/cmodelPM4Sand2D009.dll --
- cmodel plugin pm4sand2d loaded.”. In order to also verify that it has been loaded,
the user can type “zone cmodel list” in the console. If the model has been loaded then it
should appear as “pm4sand2d” under the list of “Currently loaded CPP models”.

(3) Before constitutive model plug-ins can be assigned to zones, the model must be configured for
their use by giving the model configure plugins command. Otherwise, the user will
get a “model will not cycle” error message.

In order to assign the model to the preferred zones the following command should be given in FLAC8.1:

model pm4sand <...>

or in FLAC2D 9.00:

zone cmodel assign pm4sand2d range <...>

3.6 Additional notes on use in boundary value problem simulations

FLAC includes both "static" and "dynamic" solution procedures. PM4Sand has been extensively
validated for use with the dynamic procedure. The use of PM4Sand with FLAC's static solution
procedure requires special attention to the loading and solution procedures. The static solution
procedure uses extremely high damping values which can carry significant shear and normal stresses,
which can cause problems with the response of a highly nonlinear, stress-dependent material. For
example, if the user imposes a large strain rate (e.g., high rate of loading on a foundation) in a problem
involving drained loading of a contractive soil with the static solution procedure, the drained volumetric
contraction of the soil can result in normal stresses being transferred to the damping component which
causes an artificial reduction in normal effective stress in the soil. For this reason, the use of PM4Sand
with FLAC's static solution procedure requires a higher degree of scrutiny and evaluation to ensure that
such problems do not develop.

A nominal amount of Rayleigh damping should be included with PM4Sand zones to control
numerical noise during dynamic solutions. A damping ratio of 0.005 has been found sufficient for most
applications.

63
 Zones at the ground surface, particularly within slopes and above the water table, are susceptible to
developing large deformations at strong shaking levels (i.e., when the frictional shear resistance is
exceeded). Excessive distortion of surface zones can lead to premature stoppage of a simulation,
particularly for soils that liquefy or cyclically soften. Some analysts will use Mohr Coulomb materials
in lieu of complex sand models for surface zones, for which they can then include a nominal amount
of cohesion to reduce the potential for surficial shear failures. In the current version of PM4Sand, a
similar effect can be achieved by increasing the nominal shear resistance chg above the default value
used to control hour-glassing in liquefied zones.

Loading conditions that cause a progressive increase in the mean effective stresses in PM4Sand, or
any other pressure-dependent material, require special consideration during the solution process. The
elastic moduli will increase with increasing mean effective stress, such that the time step required for
a stable solution will decrease as the loading progresses. FLAC only determines the required time step
at certain instances, like when the step or solve commands are executed. For this reason, the loading
should be applied in small increments with the solve command periodically repeated so that the required
time step is updated as appropriate during the applied loading.

Initial stresses in a boundary value problem are sometimes established using simpler constitutive
models, like a Mohr Coulomb or elastic model, prior to switching the materials to a more complex
model like PM4Sand. Problems can develop if the initial states of stress fall outside the greater of the
bounding and dilatancy surface lines for the PM4Sand model. This can happen in zones where the
initial state of stress was computed for a Mohr Coulomb material with a nonzero cohesion or for an
elastic material. For this reason, it is helpful to first ensure that the initial states of stress in all zones
correspond to a stress ratio that is less than some reasonable limit prior to switching the material model
to PM4Sand.

The ability to use the DLL with FLAC's "free-field" lateral boundary conditions option or compliant
base option has not been configured at this time. Thus, the user should not have PM4Sand in the outer
column of elements against which the free-field lateral boundary condition will be applied. Instead, the
outer columns can be replaced with elastic materials having a secant modulus compatible with the
adjacent PM4Sand zones.

64
Table 3.1: Simplified pseudo-code of PM4Sand (Version 3.3)

Operations within one subzone:

1. Initialize the model parameters; this only happens when the model is first assigned or when FirstCall is set to zero
at some point during the analysis. For detailed information on what parameters are initialized (or reset) see Table
3.2.
2. Obtain the strain increment from FLAC 𝑑𝜺.
3. Decompose the strain increment into volumetric and deviatoric components, 𝑑𝜀𝑝 and 𝑑𝜀𝑠 .
4. Calculate the trial elastic stress increment and trial elastic stress:
σtr = σ0 + dσtr = σ0 + 2Gde + Kd v I
5. Calculate the trial stress ratio ,rtr the distance from the yield surface 𝑑𝑖𝑠𝑡, the unit normal to the yield
surface n and the inner product of the change in back-stress ratio tensor with unit normal vector daxn.
σ − ptr I
rtr = tr
ptr
dist = ( rtr − α0 ) : ( rtr − α0 )

n=
( rtr − α0 )
dist
daxn = ( α0 − αin ) : n
6. Check for yield:
a. If elastic then commit the trial stresses. Go to step 8.
1
dist  m
2
σ0 = σtr
b. If inelastic:
i. Calculate loading index L:
2Gn : de − n : rKd  v
L=
K p + 2G − KDn : r
ii. Calculate trial stress increment and trial stress:
σtr = σ0 + dσtr = σ0 + 2Gdεs + Kd  p I − L 2Gn + KDI 
iii. Apply penalties to stress ratios and back-stress ratios to meet the consistency condition and to remain
within the greater of the bounding and dilatancy surfaces.
iv. Calculate image back-stress ratios and inner products:
α b = 1  M b − m  n
2
αd = 1 M d − m n
2 
α dR = 1  M dR − m  n
2 
v. Commit the trial stresses (back-stress ratios, stress ratio, mean stress, stress)
7. Return all stress tensor components to FLAC (at this point FLAC takes over and will average them according to
the mixed discretization scheme)

Operations referring to the whole zone:

8. After the calculation has completed the 4th subzone, the following additional calculations are performed for the
overall zone. Recall the following parameters for all 4 subzones and compute area-weighted average values for:
• Volumetric strain d  p
• Strain increment dε

65
 • Mean stress p
• Stress tensor (committed one) σ0
• Back-stress ratio tensor α0
• Unit normal to yield surface vector n
9. Apply penalties to the averaged zone parameters to meet the consistency condition and maintain the yield surface
inside the greater of the bounding and dilatancy surfaces.
10. Calculate image back-stress ratios and inner products for the averaged zone parameters.
11. Calculate daxn for the averaged zone parameters and determine whether a loading reversal has occurred.
12. Compute Dilatancy D and Plastic Modulus Kp for the past average step in the zone.
13. Compute plastic volumetric strain for use in fabric terms.
( )
14. If α − α : n < 0, update the fabric tensor for the zone and if exceeding its former value, update the cumulative
d

fabric term.
cz d  vpl
z=z− ( zmax n + z )
 z  D
max  1, cum 
 2zmax 

15. Update the relative state parameter, the bounding and dilatancy stress ratios, the elastic shear modulus (depends on
fabric) and the elastic bulk modulus for the next step.
16. Update the initial and previous initial back-stress values and the strain increment accumulators.
17. Update initial back-stress ratios upon reversal.
18. Commit zone stress tensor, zone mean stress, zone back-stress ratio tensor, zone stress ratio tensor to memory.

66
Table 3.2: Initialization function of PM4Sand (called during the first application of the model and
whenever First_Call=0)

1. Obtain stresses from FLAC and create stress tensor (these will be the committed stresses from which the calculation
will start):
𝒊𝒋
𝝈𝝄
2. Check stresses and calculate mean effective stress:
a. If stresses compressive (following FLAC’s sign convention that tensile stresses and strains are positive):
1
𝜎𝜊11 < 0 → 𝑝𝑜 = 𝑚𝑖𝑛 (𝑝𝑚𝑖𝑛 , 𝜎𝜊𝑖𝑖 )
2
𝑝
𝑝𝑚𝑖𝑛 = 𝑚𝑖𝑛 (𝑝𝑚𝑖𝑛 , )
200
𝑝
𝑝𝑚𝑖𝑛2 = 𝑚𝑖𝑛 (𝑝𝑚𝑖𝑛2 , )
20
b. If stresses tensile:
𝑃𝑎𝑡𝑚
𝜎𝜊11 > 0 → 𝑝𝑜 = −
20
𝒊𝒋
𝝈𝝄 = 𝑝𝑜 ∙ [𝐼]

3. Calculate relative state parameter and subsequently calculate the bounding and dilatancy stress ratios and Ado (from
input property DR and secondary parameters R, Q, nb and nd – see Chapter 4):

𝑅
𝜉𝑅 = 𝑝 − 𝐷𝑅
𝑄 − 𝑙𝑛 (−100 )
𝑃𝑎𝑡𝑚
a. If dense-of-critical (𝜉𝑅 < 0):
𝑀𝑏 = 𝑀𝑒𝑥𝑝(−𝑛𝑏 𝜉𝑅 ) , 𝑀𝑑 = 𝑀𝑒𝑥𝑝(𝑛𝑑 𝜉𝑅 )
𝑀𝑏
𝑠𝑖𝑛−1 ( ) − 𝜑𝑐𝑣
2
𝐴𝑑𝑜 = 2.5 [ ]
𝑀𝑏 − 𝑀 𝑑

b. If loose-of-critical (𝜉𝑅 > 0):

𝑛𝑏
𝑀𝑏 = 𝑀 ∙ 𝑒𝑥𝑝 (− 𝜉 ) , 𝑀𝑑 = 𝑀 ∙ 𝑒𝑥𝑝(4𝑛𝑑 𝜉𝑅 ), 𝐴𝑑𝑜 = 1.24
4 𝑅
4. Check that initial stresses are inside the greater of the bounding and dilatancy surfaces and compute the committed
back-stress and stress ratio tensors from the stress tensor:
2 1 𝒊𝒋 𝒊𝒋
𝑀𝑐𝑢𝑡 = 𝑚𝑎𝑥(𝑀𝑏 , 𝑀𝑑 ) , 𝑀 𝑓𝑖𝑛 = − ∙ √ (𝝈𝝄 − 𝑝𝑜 [𝐼]): (𝝈𝝄 − 𝑝𝑜 [𝐼])
𝑝0 2
a. If 𝑀 𝑓𝑖𝑛 > 𝑀𝑐𝑢𝑡 𝑤ℎ𝑒𝑟𝑒 𝑀𝑐𝑢𝑡 𝑖𝑠 𝑡ℎ𝑒 𝑔𝑟𝑒𝑎𝑡𝑒𝑟 𝑜𝑓 𝑀𝑏 𝑎𝑛𝑑 𝑀𝑑 :
𝒊𝒋
𝒊𝒋 𝝈𝝄 − 𝑝𝑜 [𝐼] 𝑀𝑐𝑢𝑡
𝒓𝝄 = ( ) ( 𝑓𝑖𝑛 )
𝑝𝑜 𝑀

𝒊𝒋 𝒊𝒋
𝝈𝝄 = 𝑝𝑜 [𝐼] + 𝒓𝝄 𝑝𝑜

𝒊𝒋 𝒊𝒋 𝑀𝑐𝑢𝑡 − 𝑚
𝜶𝝄 = 𝒓𝝄 ∙
𝑀𝑐𝑢𝑡

b. If 𝑀𝑐𝑢𝑡 > 𝑀 𝑓𝑖𝑛 :

𝒊𝒋
𝒊𝒋 𝝈𝝄 − 𝑝𝑜 [𝐼]
𝒓𝝄 = ( )
𝑝𝑜

67
 𝒊𝒋 𝒊𝒋
𝜶𝝄 = 𝒓𝝄
5. Create/Initialize the initial back-stress ratio, initial previous back-stress ratio, minimum initial back-stress ratio and
maximum initial back-stress ratio tensors (see also Section 2.5 on Stress Reversal):

a. If 𝑀 𝑓𝑖𝑛 < 0.9𝑀𝑏 :

𝒊𝒋 𝒊𝒋
𝜶𝒊𝒏 = 𝜶𝝄
𝑓𝑖𝑛 𝑏
b. If 𝑀 > 0.9𝑀 :
𝒊𝒋 0.9𝑀𝑏 𝒊𝒋
𝜶𝒊𝒏 = 𝜶𝝄 ( )
𝑀 𝑓𝑖𝑛
Note that 𝑀 𝑓𝑖𝑛 in the above expression would have been updated if step 4 had required adjusting the stresses. The
other back-stress ratio history terms are then set as:
𝒊𝒋 𝒊𝒋 𝒊𝒋 𝒊𝒋
𝜶𝒊𝒏𝑷 = 𝜶𝒊𝒏𝑴𝒂𝒙 = 𝜶𝒊𝒏𝑴𝒊𝒏 = 𝜶𝒊𝒏

6. Calculate initial values of elastic shear modulus, elastic bulk modulus, plastic modulus, dilatancy:

−𝑝𝑜
𝐺 = 𝐺𝑜 𝑃𝑎𝑡𝑚 √
𝑃𝑎𝑡𝑚

2(1 + 𝑣)
𝐾=𝐺
3(1 − 2𝑣)

𝐾𝑝 = 100𝐺

𝐷 = 0

7. Initialize fabric related terms (see Section 2.8) – note that these terms will be referring to the whole zone:

𝑝𝑜
𝑝𝑧𝑝 =
100
𝑧𝑚𝑎𝑥
𝑧𝑝𝑒𝑎𝑘 =
100000

𝑧𝑥𝑝 = 𝒛: 𝑝 = 0
𝑝𝑜
𝑧𝑥𝑝𝑃𝑘 = −𝑧𝑚𝑎𝑥
50
𝒊𝒋 𝒊𝒋
𝒛𝒊𝒋 = 𝒛𝒊𝒏 = 𝒛𝜶 = 𝑧𝑐𝑢𝑚 = 0

68
 Figure 3.1. Schematic illustration of the averaging procedure followed in the implementation of
PM4Sand: zone-averaged values are computed for some internal variables of the model, denoted as
“m”, at the end of each step, after which other internal parameters, denoted as “q”, are computed
based on the zone-averaged parameters.

69
Figure 3.2. Effect of dynamic time step on the results obtained from (a) drained and (b) undrained
cyclic DSS element test simulations (DR=55%, σ'vo=1atm) loaded at a shear strain rate of 12%/s.
The black line in each case denotes the response obtained with FLAC’s default dynamic time
step.

70
Figure 3.3. Effect of dynamic time step on the results obtained from (a) drained and (b) undrained
cyclic DSS element test simulations (DR=55%, σ'vo=1atm) loaded at a shear strain rate of 4%/s.
The black line in each case denotes the response obtained with FLAC’s default dynamic time
step.

71
Figure 3.4. Effect of dynamic time step on seismic analysis of an embankment dam: cross-section,
materials, and displacement monitoring points (top figures) and time histories of crest settlement and
horizontal displacements for points on the upstream and downstream faces (lower figures).

72
4. MODEL INPUT PARAMETERS AND RESPONSES

4.1 Model input parameters

The model parameters are grouped into two categories; a primary set of six parameters (three
properties, two flags, and atmospheric pressure) that are most important for model calibration, and a
secondary set of parameters that may be modified from their default values in special circumstances.

Primary input parameters

The three primary input properties are the sand’s apparent relative density DR, the shear modulus
coefficient Go, and the contraction rate parameter hpo. These three parameters are discussed below and
summarized in Table 4.1.

Relative density (DR) can be estimated in practice by correlation to penetration resistances. For
example, a common form for SPT correlations is,

( N1 )60
DR = (94)
Cd

where DR is expressed as a ratio rather than a percentage. Idriss and Boulanger (2008) reviewed
published data and past relationships, and then adopted a value of Cd = 46 in the development of their
liquefaction triggering correlations. For the CPT, they similarly reviewed available relationships and
arrived at the following expression,

0.264
q 
DR = 0.465  c1N  − 1.063 (95)
C
 dq 
for which they adopted Cdq = 0.9. For the example loading responses shown later, DR values of 35%,
55%, and 75% were used, which would be correlated to SPT (N1)60 values of 6, 14, and 26 by the above
correlations.

The input value of DR is best considered an "apparent relative density," rather than a strict measure
of relative density following conventional laboratory tests. The input value of DR influences the
response of the model and thus it is just another input parameter that the user can adjust as part of the
calibration process. The above correlations are provided for the purpose of obtaining a reasonable
estimate for the apparent DR so that the resulting model behaviors are also reasonable. There are
situations, however, where the user may choose to adjust the input DR, up or down relative to the above
relationships, to improve its calibration to some other relationship or data.

73
 The second primary input parameter is the constant Go which controls the elastic (or small-strain)
shear modulus. The equation for the elastic shear modulus (Equation 23) at the time of model
initialization can be simplified to,
1
 p  2
G = Go p A   (96)
 pA 
The full equation for the elastic shear modulus includes adjustments for the effects of stress-ratio
(Equation 24) and fabric (Equation 86), but these are both unity at the time of model initialization. The
elastic shear modulus can be calibrated to fit in-situ Vs measurements, according to,

G =   (Vs )
2
(97)

or alternatively fit to values of Vs that may be estimated by correlation to penetration resistances. For
the examples shown herein, the correlation by Andrus and Stokoe (2000) in Figure 4.1 was used, with
a slight modification that constrains the extrapolation to very small (N1)60 values, as shown in the figure,

Vs1 = 85 ( N1 )60 + 2.5 

0.25
(98)

The above relationships in combination with the default values for the maximum and minimum
void ratios (described later) produce Vs1 values of 145, 171, and 196 m/s and corresponding Go values
of 477, 677, and 906 for the DR values of 35%, 55%, and 75%, respectively.

Calibration of G to fit in-situ Vs measurements requires an estimate for the in-situ coefficient of earth
pressure at rest (Ko) for computing p (i.e., the mean of the vertical and horizontal stresses for the present
2D implementation; Equation 2). Maintaining consistency between the calibration procedure and the
boundary value problem solution requires that the initial Ko conditions in the boundary value problem
reasonably match the value assumed during calibration.

Alternatively, the above expressions were combined together with a range of typical densities to
arrive at the following simpler expression for estimating Go,

Go = 167 ( N1 )60 + 2.5 (99)

This expression produces Go values of 487, 678, and 892 for the DR values of 35%, 55%, and 75%,
respectively.

The third primary input parameter is the constant hpo which is used to modify the contractiveness
and hence enable calibration of the model to specific values of cyclic resistance ratio (CRR). For the
examples presented herein, the target CRR values were based on the liquefaction triggering correlation
by Idriss and Boulanger (2008) in Figure 1.2. This relationship produces target CRR values for an
effective overburden stress of 1 atm and an earthquake magnitude of M=7.5 of 0.090, 0.147, and 0.312

74
for the corresponding SPT (N1)60 values of 6, 14, and 26, respectively. The corresponding values of hpo
to achieve these CRR values, given the already set values for DR and Go, are 0.52, 0.40 and 0.62,
respectively.

The value of atmospheric pressure, pA, should also be specified in the unit set being used for the
analysis. If not specified, it will default to 101,300 Pascal.

The flag FirstCall is used to re-set the back-stress ratio history terms equal to the current stress ratio,
and to erase all fabric terms. The first time the model is called, the flag should be unspecified or have
a value of 0. The model will then initiate the back-stress ratios and all pertinent history terms using the
current state of stress. The flag is then set equal to 1.0 internally. If FirstCall is later set equal to 0.0
using the property command in FLAC, this will cause the material to re-initiate all internal terms,
thereby re-setting the back-stress and stress ratio history terms and erasing all fabric terms. FirstCall
should usually be set to 0.0 just before initiating dynamic earthquake loading. Otherwise, the model
will retain memory of the loading during the static initiation of the model, which may or may not be
desired.

The flag PostShake is used during the post-shaking portion of a simulation to improve the modeling
of post-liquefaction reconsolidation strains. The flag is set to 0 internally and remains 0 unless the user
specifies otherwise. If the flag is set to 1.0, the elastic moduli will be reduced according to the
expressions presented previously. PostShake should only be set to 1.0 at the end of strong shaking, as
the reductions in elastic moduli were not calibrated for cyclic loading behavior.

75
Table 4.1 – Primary input parameters (parameter names in square brackets correspond to the input
name to be used within FLAC)

Parameter
[FLAC property Comments
name]
DR Apparent relative density: Primary variable controlling dilatancy
[D_r] and stress-strain response characteristics. Input as a fraction, not as
a percentage.

Commonly estimated based on CPT or SPT penetration resistances,

such as the following relationships used by Idriss and Boulanger
(2008):
( N1 )60
DR =
Cd
with Cd = 46, and
0.264
q 
DR = 0.465  c1N  − 1.063
C
 dq 
with Cdq = 0.9.
Go Shear modulus coefficient: Primary variable controlling the small
[G_o] strain shear modulus, Gmax. Should be chosen to match estimated or
measured shear wave velocities according to Gmax =  Vs2.

A value for Go can be estimated based on the modified correlation

between SPT (N1)60 and Vs1 values shown in Figure 4.1. The value
of Go can thus be computed as,

Go = 167 ( N1 )60 + 2.5

hpo Contraction rate parameter: Primary variable that adjusts
[h_po] contraction rates and hence can be adjusted to obtain a target cyclic
resistance ratio, as commonly estimated based on CPT or SPT
penetration resistances and liquefaction correlations.

Calibration of this parameter should be performed last because its

value can depend on the values assigned to other parameters.
pA Atmospheric pressure in the unit set being used. Defaults to
[P_atm] 101,300 Pascals if not specified.
FirstCall Flag used to re-set the back-stress ratio history terms equal to the
[First_Call] current stress ratio, and to erase all fabric terms. FirstCall should
usually be set to 0.0 at model initiation and/or just before initiating
dynamic earthquake loading. Otherwise, the model will retain
memory of the loading during the static initiation of the model,
which may or may not be desired.

76
 PostShake Flag used during post-shaking portion of a simulation to improve
[Post_Shake] modeling of post-liquefaction reconsolidation strains. PostShake
should only be set to 1.0 after the end of strong shaking.

Secondary input parameters

Secondary input parameters are those parameters for which default values have been developed that
will generally produce reasonable agreement with the trends in typical design correlations. The user
must, however, still confirm through element loading calibrations that the default parameters are
appropriate for their particular conditions. The secondary input parameters (21 in total) are listed in
Table 4.2, along with commentary on the recommended default values.

The recommended default values for all secondary parameters have been embedded within the
initialization section of the code and thus these parameters will take their default values unless the user
specifies otherwise. In addition, the input logic is structured such that secondary parameters will take
their default value if the user inputs a value of zero for that parameter.

Table 4.2 – Secondary input parameters

Parameter
Comments
[FLAC name]
ho Variable that adjusts the ratio of plastic modulus to elastic modulus. The
[h_o] default value of ho=(0.25+DR)/2, with a minimum value of 0.30, was chosen
to provide reasonable G/Gmax and damping relationships for the default
value of Go. This variable may require adjustment in combination with any
adjustments to Go.
emax and emin The maximum and minimum void ratios affect the computation of density,
[e_max] and affect how volumetric strains translate into changes in relative state.
[e_min] Default values of 0.8 and 0.5, respectively, were adopted. Refinements in
these parameters for a practical problem may not be necessary, as the
calibration of other parameters will have a stronger effect on monotonic or
cyclic strengths.
nb Default value is 0.50. Controls dilatancy and thus also the peak effective
[n_b] friction angles. Note that Mb for loose-of-critical states is computed using
nb/4.
nd Default value is 0.10. Controls the stress-ratio at which contraction
[n_d] transitions to dilation, which is often referred to as phase transformation. A
value of 0.10 produces a phase transformation angle slightly smaller than
cv, which is consistent with experimental data. Note that Md for loose-of-
critical states is computed using 4nd.
Ado Default value is computed based on Bolton's dilatancy relationship at the
[A_do] time of initialization; typical values will be between 1.2 and 1.5.

77
zmax Default value is computed at the time of initialization as,
[z_max] zmax = 0.70  exp ( −6.1 Ro )  20
This returns 0.7 if R is initially 0.0, and increases to its maximum value of
20 with increasing dense-of-critical states. May require varying if the
relationship between DR and cyclic strength is significantly different from
that implied by the liquefaction correlations of Idriss and Boulanger (2008).
Cz Default value is 250. Controls strain levels at which fabric affects become
[c_z] important.
C Default value varies with DR. The value is 0.5 for DR less than 55%, and
[c_e] linearly decreases to its minimum value of 0.2 at DR = 75%. Can be used
to adjust the rate of strain accumulation in undrained cyclic loading.
'cv Default value is 33 degrees.
[phi_cv]
 Default value is 0.30. For 1-D consolidation of an elastic material, the
[pois] value of Ko would correspond to,

Ko =
1−
The default value for  results in a Ko value of 0.43 in 1-D consolidation.
CGD Default value is 2.0. The small-strain elastic modulus degrades with
[G_degr] increasing cumulative plastic deviator strains (zcum). The maximum
degradation approaches a factor of 1/CGD.
CDR Default value is computed at the time of initialization as,
[C_DR] CDR = 5+ 25 (DR - 0.35) ≤ 10
Controls the rotated dilatancy surface and is applied to reduce the rate under
which dilatancy is increasing.
CKf Default value varies with DR. as,
CK f = 5 + 220  ( DRo − 0.26 )   4.0; 35.0 
3
[C_kaf]
The value is 4.0 for DR less than 10%, and increases to its maximum value
of 35.0 at DR = 77%. This variable controls the effect that sustained static
shear stresses have on plastic modulus.
Q Default value is 10. Default value is for quartzitic sands per
[Q_bolt] recommendations of Bolton (1986).
R Default value is 1.5. Default value for quartzitic sands would be 1.0 per
[R_bolt] recommendations of Bolton (1986); a slight increase in R is used to lower
the critical state line to better approximate typical results for direct simple
shear loading.
m Default value is 0.01. Default value provides reasonable modeling and
[m_par] numerical stability.
Fsed,min Default value is set at time of initialization and equal to 0.04. Controls the
[F_sedmin] minimum value the reduction factor of the elastic moduli can attain during
reconsolidation (used when Post_Shake=1.0).
p'sed,o Default value is –Patm/5. It is the mean effective stress up to which
[p_sedo] reconsolidation strains are enhanced when Post_Shake=1.0

78
 crhg Nominal plastic shear strength ratio used to compute chg at the time of
[MC_ratio] initialization or when FirstCall is set equal to 0. Default value is 0.005.
chg Nominal plastic shear strength assigned at initialization or when FirstCall
[MC_c] is set equal to 0. It is computed as the greater of: (1) crhg times p, and (2)
the user-specified value for chg. Thus, the user-specified value for chg is the
minimum value it will be assigned.

Tracking variables

Many of the parameters internal to PM4Sand may be tracked for debugging purposes. The table
below lists six internal parameters which may be of interest. Other internal parameters which can be
tracked include: max_G, max_K, pmin, pmin2, MM, alfa_11, alfa_12, r_11, r_12, aIn_11, aIn_12,
aInP_11, aInP_12, z_11, z_12, zcum, zpeak, zxpPk, pzp, zxp, Cka, eqsum, evsum, LoadInd, Dilat, Kp,
zabs, evol, eq_11, eq_22, eq_12, epsIncr and daxn. Note that Cka tracks the product of the Crev and Ck
terms. Other internal parameters are visible through the FLAC interface, and can be phonetically
mapped to different terms in constitutive equations in this manual.

Table 4.3 – Internal parameters available for tracking

Parameter [FLAC Name] Comments

Mb
Bounding surface stress ratio
[Mb]
Md
Dilatancy surface stress ratio
[Md]
Mcur
Current stress ratio
[Mcur]
G
Elastic shear modulus
[shearG]
K
Elastic bulk modulus
[bulkK]
ξR
Relative state parameter
[rsp]
Relative density, which evolves in response to
DR volumetric strains. Note that the input
[Dr] parameter D_r is an initial parameter and does
not evolve during an analysis.

79
4.2 Example calibration and model responses for a range of element loading conditions

The response of the model is illustrated in this section by presenting simulation results for a set of
input parameters that were calibrated to emphasize realistic modeling of liquefaction behavior. Results
are presented for sands having initial apparent relative densities of 35%, 55%, and 75% with
corresponding SPT (N1)60 values of approximately 6, 14, and 26, respectively, based on the correlations
presented previously. All secondary input parameters were assigned the default values summarized
previously in Table 4.2. Values for Go were obtained using the previously presented correlation
between SPT (N1)60 values and overburden-corrected shear wave velocity VS1 (Figure 4.1). Values for
hpo were obtained by matching the CRR values from direct simple shear (DSS) simulations with the
CRRM=7.5 values that were computed using the SPT-based liquefaction triggering correlation by Idriss
and Boulanger (2006, 2008); an SPT-based estimate of CRR for an M=7.5 earthquake and effective
overburden stress of 1 atm was assumed to be approximately equal to the CRR corresponding to 15
uniform loading cycles causing a peak shear strain of 3% in direct simple shear loading. The model
input parameters for the examples presented in this section are summarized in Table 4.4 below.

Table 4.4. Input parameters for example element responses

Scenario field condition Model input parameters (a)

VS1 using CRRM=7.5
DR (N1)60 Andrus & using Idriss & DR Go hpo
Stokoe (2000) Boulanger (2008)
0.35 6 145 0.090 0.35 477 0.52

0.55 14 171 0.147 0.55 677 0.40

0.75 26 196 0.312 0.75 906 0.62

(a) All secondary input parameters were assigned the default values listed in Table 4.2.

Undrained cyclic loading

The undrained cyclic loading responses for the calibrated models are illustrated in Figures 4.2-4.4.
These figures show the stress-strain and stress-path responses for undrained uniform cyclic loading in
DSS with a vertical consolidation stress of 1 atm and initial static shear stress ratios () of 0.0, 0.1, and
0.2. Results for initial DR of 35%, 55%, and 75% are presented in Figures 4.2, 4.3, and 4.4, respectively.
Close up views of the stress-strain responses for DR=35%, 55%, and 75% with an initial static shear
stress ratio  = 0.0 are presented in Figure 4.5.

The stress-strain responses for  = 0.0 illustrate the model’s ability to progressively reach larger
and larger shear strains with continued cyclic loading, rather than locking up in a repeating loop as
many plasticity models do. The ability to simulate the progressive accumulation of shear strains reflects
the inclusion of the cumulative fabric terms, as described previously. The progressive increases in peak
shear strain after the soil has reached a peak excess pore pressure ratio (ru) greater than 98% are realistic
in magnitude.

80
 The stress-strain responses with nonzero initial static shear stresses show a progressive
accumulation of shear strains in the direction of the initial static shear stress, with the rate and nature
of the stress-strain response also being reasonably realistic.

CRR versus number of loading cycles – Effect of DR and failure criterion

The CRR obtained for the calibrated models are summarized in Figure 4.6 showing the cyclic stress
ratio (CSR) required to cause an excess pore pressure ratio (ru) of 98% or single-amplitude shear strains
of 1% and 3% versus number of uniform loading cycles. These results are for DSS loading with a
vertical consolidation stress of 1 atm, an initial Ko of 0.5, and zero initial static shear stress ratio (=).
The simulation results in this figure were fitted with a power law, for which the exponent “b” is labeled
beside each curve.

The slopes of these CRR versus number of loading cycles are in good agreement with typical values
obtained in laboratory testing studies. The exponent b is generally between 0.24 and 0.27 for these
simulations. For the experimental data in Figure 1.3 the exponent b ranges from a low of 0.1 for one
study on to a high of 0.34 for another. The ability of the model to produce reasonable slopes for these
curves is attributed primarily to the changes in the plastic modulus and dilatancy relationships
(Ziotopoulou and Boulanger 2012).

The slopes of the CRR versus number of loading cycle curves can be slightly adjusted by the
parameters nb and nd. For example, repeating the same simulations for DR=75% with nb=0.8 (versus
the value of 0.5 used herein) and with all other factors the same, increases the exponent b to values of
0.28 to 0.33. Ziotopoulou and Boulanger (2012) performed the same simulations using an earlier set of
calibration parameters and got values for the exponent b ranging from 0.26 to 0.36. Note, however, that
a greater value for nb also results in greater peak friction angles and changes other responses as well,
so such adjustments cannot be made independent of other features of behavior.

CRR versus number of loading cycles – Effect of confining stress

The effect of overburden stress on CRR for the calibrated models is illustrated in Figure 4.7 showing
the CSR required to cause a single-amplitude shear strain of 3% versus number of uniform loading
cycles for different confining stresses. These results are for DSS loading with initial Ko=0.5, initial
static shear stress ratio () of 0.0, and vertical consolidation stresses of 1, 4, and 8 atm. The cyclic
strengths for DR=35% are the least affected by confining stress, while the cyclic strengths for DR=75%
are the most affected (reduced).

The equivalent K values from these simulations, with the CRR values compared at 15 uniform
loading cycles, are compared in Figure 4.8 to the relationships recommended by Boulanger and Idriss
(2004) based on the framework presented in Boulanger (2003b). The simulated effects of confining
stress are in good agreement, as expected since the expression for hpo was calibrated to this relationship.

81
CRR versus number of loading cycles – Effect of sustained shear stress

Summary plots of the CSR required to cause a single-amplitude shear strain of 3% versus number
of uniform loading cycles are presented in Figure 4.9 for different values of initial static shear stress
ratio. Results are presented for sand at DR=35%, 55%, and 75% loaded in DSS with an initial Ko=0.5,
a vertical consolidation stress of 1atm, and with initial static shear stress ratios () of 0.0, 0.1, 0.2, and
0.3. The model was initialized at the stress conditions corresponding to the end of 1D vertical
consolidation, thereby setting the back-stress ratio terms prior to application of the initial horizontal
static shear stress. The simulation results are reasonable in predicting that the presence of an initial
static shear stress ratio results in lower cyclic strengths for loose sands (e.g., the DR = 35% results) and
greater cyclic strengths for denser sands (e.g., the DR=75% results).

The effect of vertical effective stress on these behaviors is illustrated in Figures 4.10a and 4.10b
showing the cyclic resistance ratio (3% peak shear strain in 15 uniform cycles of DSS loading) against
initial static shear stress ratio for DR of 35, 55, and 75% with vertical effective consolidation stresses
of 100 and 400 kPa. Figures 4.10c and 4.10d show the same effects in terms of the static shear stress
ratio strength correction factor Kα for the same overburden stresses. Increasing the vertical effective
consolidation stress makes the material relatively more contractive, such that the effect of an initial
static shear stress ratio is relatively more detrimental at 400 kPa than at 100 kPa. This general pattern
is in agreement with experimental observations (Figure 1.5).

CRR versus number of loading cycles – Effect of Ko

Summary plots of the CSR required to cause a single-amplitude shear strain of 3% versus number
of uniform loading cycles are presented in Figure 4.11 for different values of the lateral earth pressure
coefficient at rest Ko (i.e., the ratio of horizontal to vertical effective stresses at the time of
consolidation). Results are presented for sand at DR=35%, 55%, and 75% loaded in DSS, a vertical
consolidation stress of 1atm, and with zero initial static shear stress ratio.

Drained monotonic loading

The response for drained monotonic loading in direct simple shear (DSS) and plane-strain
compression (PSC) for sand at DR of 35%, 55%, and 75% under vertical confining stresses of ¼, 1, 4,
and 16 atm is shown in Figures 4.12 and 4.13. The responses reasonably approximate the effects of
relative density and confining stress on both the stress-strain and volumetric strain responses. The plots
show the response up to shear strains of 10%, while the simulations tend to reach critical state
conditions at shear strains of 40-60%. The post-peak rate of strain-softening is dictated by the dilation
rate, which is constrained to approximate Bolton’s (1986) stress-dilatancy relationship. The simulated
post-peak softening is slower than often observed in experimental results (e.g., Figure 1.9) because
drained laboratory experiments are often affected by strain localizations in dilating sands (e.g., Desrues
et al. 1996, Sadrekarimi and Olson 2010); The rate of strain-softening in a dilating zone is much lower
than represented by global measurements of stress and strain.

The peak effective friction angles from simulations of drained monotonic loading in DSS and plane
strain compression (PSC) are shown versus vertical consolidation stress in Figure 4.14 where they are
also compared to Bolton’s (1986) relationship for plane strain conditions for Q=10 and R=1.5 (which

82
are the default values of Q and R that have been selected for the model). The peak friction angles are
lower in DSS than in PSC because of the difference in how the friction angles are computed for this
plot. For PSC, the peak friction angle is computed based on the peak stress ratio within the element,
without any predetermined assumptions regarding the orientation of the plane on which it will occur.
For DSS loading, the peak friction angle was computed as the inverse tangent of the peak stress ratio
on the horizontal plane, following the same convention commonly used in practice for interpreting such
tests. In the DSS simulation, however, the horizontal plane was not the plane of maximum stress
obliquity, and therefore the interpreted peak friction angle is slightly lower than the value obtained in
PSC. Computationally, both the DSS and PSC mobilize similar peak friction angles if the comparison
is made only for the plane of maximum stress obliquity in both simulations; In the DSS, however, the
stress ratio on the horizontal plane in the DSS simulations is often closer to sin() as opposed to tan(),
which results in the apparent differences shown in Figure 4.14. Despite these differences, the peak
friction angles are reasonable and consistent with typical design correlations (e.g., Kulhawy and Mayne
1990).

Undrained monotonic loading

The undrained monotonic loading in direct simple shear (DSS) for sand at DR of 35%, 55%, and
75% under vertical consolidation stresses of ¼, 1, 4, and 16 atm are shown in Figure 4.15, while the
same responses are shown with normalization by the vertical consolidation stress in Figure 4.16. The
stress-strain responses show strain hardening behavior at lower relative states than would be expected
based on the experimental results for reconstituted sands, such as presented by Yoshimine et al. (1999).
Experiments on loose reconstituted sands often show strain softening to some minimum shear stress
ratio (e.g., quasi-steady state condition) before beginning to strain harden, and that minimum stress
ratio is often in the range of 0.1 to 0.3. For the present calibration, the CRR for DR = 35% sands under
1 atm of confining stress was targeted to be 0.090 based on a field-based liquefaction correlation, and
it was not possible to calibrate the model to match both the target CRR values and the monotonic
undrained strengths presented in Yoshimine et al. (1999). If the monotonic behavior was more
important than the CRR values, then a different calibration would be required.

Drained, strain-controlled, cyclic loading from small to large strains

Drained strain-controlled cyclic loading in DSS for sand at DR of 35%, 55%, and 75% under vertical
consolidation stresses of 1, 4, and 16 atm with Ko=1.0 are shown in Figures 4.17 to 4.19, with results
also shown for the equivalent modulus reduction (G/Gmax) and equivalent damping ratio (ξ) versus
cyclic shear strain amplitude (γ). Also shown on these figures are the modulus reduction and equivalent
damping ratio curves recommended for sands at different depths by EPRI (1993). The simulated
modulus reduction and equivalent damping ratio curves depend on the effective confining stress in a
pattern and magnitude that is consistent with empirical design correlations, such as the ones by EPRI.
The simulated modulus reduction curves for this calibration tend to be slightly higher than the empirical
curves, whereas the simulated equivalent damping ratios are in reasonable agreement with the empirical
curves over a fairly broad range of shear strain amplitudes. The model response for this calibration
avoids the problem common to many plasticity models of producing excessively high equivalent
damping ratios as shear strain amplitudes approach about one percent.

83
Drained, strain-controlled cyclic loading: Densification under large numbers of cycles

Drained strain-controlled cyclic loading in DSS for sand at DR of 35%, 55%, and 75% subjected to
20 cycles at 1% shear strain under a vertical effective stress of 1 atm are presented in Figure 4.20 to
illustrate the accumulation of volumetric strains with increasing number of constant-amplitude strain
cycles. The model response with this calibration produces volumetric strains that are about twice the
values expected based on the empirical data presented in Figure 1.12, although the general pattern of
stress-strain behavior and its dependency on DR and confining stress are reasonably consistent with the
empirical data. Alternative model calibrations can produce better agreement with these behaviors, but
they were generally found to require compromising the fit to the CRR correlations.

Post-liquefaction reconsolidation strains

The volumetric strains that develop during post-liquefaction reconsolidation of sand are difficult to
model using the conventional separation of strains into elastic and plastic components because a large
portion of the post-liquefaction reconsolidation strains are due to sedimentation effects (i.e., volume
reductions while the effective stresses remain close to zero) which are not easily incorporated into either
the elastic or plastic components of behavior. For example, it is common for many plasticity-based
constitutive models to predict reconsolidation volumetric strains from a condition of ru=100% that are
only a fraction of one percent (Ziotopoulou and Boulanger 2013b), whereas experimental data show
values ranging from one to four percent for most relative densities (e.g., Figure 1.12).

Volumetric strains due to post-cyclic reconsolidation, with and without the PostShake
[Post_Shake = 1.0] option, are plotted in Figure 4.21 versus the maximum shear strain induced
during undrained cyclic loading. Results are shown for sand at DR = 35%, 55%, and 75% loaded in
DSS with an initial Ko=0.5, a vertical consolidation stress of 1 atm, and zero initial static shear stress
ratio. After cyclic loading to different maximum shear strains, the horizontal shear stress was reduced
to zero such that the excess pore pressure was near its maximum possible value (e.g., ru was
approximately 98% or larger for cases with maximum shear strains of 3% or greater). The computed
volumetric strains were less than about 0.25% with PostShake = 0 (default value) and are much smaller
than expected based on common experimental data. The computed volumetric strains with
Post_Shake = 1.0 (imposed at the end of cyclic loading) are in much closer agreement with
experimental data (Figure 1.14).

4.3 Example calibration with user-defined critical state undrained shear strength

Calibration of the model to a specified undrained critical state shear strength (su,cs) is illustrated in
this section. The su,cs for any set of calibration parameters can be computed from the critical state line
(Figure 2.1) as,

 R 
M PA  Q − DR 
su ,cs = e (100)
2 100

84
The critical state line parameters Q and R can be adjusted to fit an experimentally determined critical
state line (e.g., Kamai and Boulanger 2012) or fit an su,cs estimated from a case history based correlation.
Boulanger and Ziotopoulou (2018) illustrated how the model can be calibrated to a specific target su,cs
by setting R according to,

  2 s 100  
R = DR Q − ln  u ,cs  (101)
  M PA  

The logic of this approach is illustrated in Figure 4.22 wherein the target su,cs is used to compute pcs
given M, which is then used to compute R given DR and Q. Alternatively, the above expression can be
rearranged to obtain the target su,cs by adjusting Q instead of R.

The calibration of R or Q to obtain a target su,cs is based on the su,cs corresponding to the current DR.
If a numerical simulation procedure imposes undrained conditions (no fluid flow) such that the DR does
not change significantly during the simulation, then the model will produce the target su,cs at large shear
strains. If a numerical simulation allows fluid flow, then the DR may increase or decrease throughout
the simulation depending on the pore pressure diffusion pattern. In this latter case, the resulting changes
in su,cs during the simulation (per the above equations) needs to be recognized by the user.

Consider the example calibration for DR = 35% in Section 4.2. This example calibration, referred
to as Calibration 1 below, used Go = 477 and hpo = 0.52 with all other parameters retaining their default
values. Recall that Q and R have default values of 10 and 1.5, respectively. That calibration produces
a su,cs value of,

 1.5 
1.089 101.3 kPa 10− 0.35 
su ,cs = e = 167 kPa
2 100

For a zone with vc = 100 kPa, this corresponds to an initially dense-of-critical-state condition that
produces su,cs/ vc = 1.67.

A revised calibration, referred to as Calibration 2, is now presented for a case where the calibration
includes targeting an su,cs/ vc = 0.07 for a zone with vc = 100 kPa. The target su,cs is therefore only 7
kPa, which is achieved by setting R to,

  2 ( 7 kPa ) 100  
R = 0.35 10 − ln    = 2.611
  1.089 101.3 kPa  

The target Vs1 and CRRM=7.5 values remain unchanged, which means Go remains unchanged whereas
hpo had to be recalibrated to 2.2.

85
 The responses of the above two calibrations to monotonic undrained DSS loading with vc = 100
kPa are shown in Figure 4.23. Calibration 1 results in a strongly dilative response whereas Calibration
2 produces the intended contractive response.

The responses of the above two calibrations to cyclic undrained DSS loading with vc = 100 kPa
and an imposed CSR = 0.090 are shown in Figure 4.24. Calibration 1 produces a cyclic mobility
response wherein cyclic shear strains progressively accumulate during cyclic loading. The specimen
never develops flow liquefaction because it is always dense of critical state. Calibration 2 exhibits flow
liquefaction, which is triggered by the generation of excess pore pressures during the cyclic loading.
The specimen develops flow liquefaction because the peak shear stresses imposed during cyclic loading
are greater than the su,cs.

The representation of post-liquefaction residual shear strengths in practice is complicated by various

challenges and limitations as discussed in Boulanger et al. (2014, 2015). The su,cs mobilized in the field,
and the timing at which it is mobilized, will depend on pore pressure diffusion and associated void
redistribution processes, which depend on the soil properties (e.g., initial relative density, cyclic
strength, permeability), subsurface stratigraphy (e.g., layer thicknesses, slope angles, continuity of
interfaces), and ground motion characteristics (e.g., shaking intensity, shaking duration, shaking
history). Simulation of void redistribution processes and their timing are highly uncertain, such that
pragmatic approximations for representing post-triggering residual shear strengths during and after
strong shaking are often used in practice (see Boulanger et al. 2015 for additional discussion).

86
 300
Clean sand data in Andrus & Stokoe (2000)
Andrus & Stokoe (2000): Vs1=93.2(N1)0.231
Modified fit: Vs1=85(N1+2.5)0.25
260

Shear wave velocity, VS1 (m/s)

220

180

140

100
0 10 20 30 40 50
SPT (N1)60

Figure 4.1. Correlation between overburden-corrected shear wave velocity and

SPT penetration resistances in clean sands (after Andrus and Stokoe 2000).

87
Figure 4.2. Undrained cyclic DSS loading response for DR = 35% with vertical effective
consolidation stress of 100 kPa and with initial static shear stress ratios of 0.0, 0.1,
and 0.2.

88
Figure 4.3. Undrained cyclic DSS loading response for DR = 55% with vertical effective
consolidation stress of 100 kPa and with initial static shear stress ratios of 0.0, 0.1,
and 0.2.

89
Figure 4.4. Undrained cyclic DSS loading response for DR = 75% with vertical effective
consolidation stress of 100 kPa and with initial static shear stress ratios of 0.0, 0.1,
and 0.2.

90
Figure 4.5. Undrained cyclic DSS loading responses for DR = 35, 55, and 75% with a vertical
effective consolidation stress of 100 kPa and without any initial static shear stress.

91
Figure 4.6. Cyclic stress ratios versus number of equivalent uniform loading cycles in undrained
DSS loading to cause ru=98% or single-amplitude shear strains of 1% or 3% for DR =
35, 55, and 75% with a vertical effective consolidation stress of 100 kPa. Each set of
CSR-N simulations was fit with a power relationship and the exponent b is labeled
beside each curve.

92
Figure 4.7. Cyclic stress ratios versus number of equivalent uniform loading cycles in undrained
DSS loading to cause single-amplitude shear strain of 3% for DR = 35, 55, and 75%
with vertical effective consolidation stresses of 1, 4, and 8 atm.

93
Figure 4.8. Comparison of K factors, determined at 15 uniform loading cycles to cause 3% single-
amplitude shear strain, from simulations versus relationships recommended by
Boulanger and Idriss (2004).

94
Figure 4.9. Cyclic stress ratios versus number of equivalent uniform loading cycles in undrained
DSS loading to cause single-amplitude shear strain of 3% for DR = 35, 55, and 75%
with vertical effective consolidation stresses of 100 kPa and initial static shear stress
ratios of 0.0, 0.1, 0.2, and 0.3.

95
Figure 4.10. Effect of vertical effective consolidation stresses of 100 and 400 kPa on the variation
of cyclic resistance ratio (a and b) (peak shear strain of 3% in 15 uniform cycles in
DSS loading) and on the Kα factor (c and d) for sand at DR of 35, 55, and 75% with
initial static shear stress ratios of 0.0, 0.1, 0.2, and 0.3.

96
Figure 4.11. Cyclic stress ratios versus number of equivalent uniform loading cycles in undrained
DSS loading to cause single-amplitude shear strain of 3% for DR = 35, 55, and 75% with
vertical effective consolidation stresses of 100 kPa and initial Ko values of 0.3, 0.5, 0.8,
and 1.2.

97
Figure 4.12. Drained monotonic DSS loading responses for DR = 35, 55, and 75% with vertical effective confining stresses of ¼, 1, 4, 16,
and 64 atm and Ko=0.5.

98
Figure 4.13. Drained monotonic PSC (plane strain compression) loading responses for DR = 35, 55, and 75% with initial isotropic
confining stresses of ¼, 1, 4, 16, and 64 atm.

99
Figure 4.14. Peak friction angles from drained monotonic PSC and DSS loading responses for DR
= 35, 55, and 75% under effective confining stresses of ¼, 1, 4, 16, and 64 atm. For
DSS loading, the friction angle is presented using the conventional interpretations that
the horizontal plane is the failure plane (the actual plane of peak stress ratio is not
horizontal in these simulations).

100
Figure 4.15. Undrained monotonic DSS loading responses for DR = 35, 55, and 75% under vertical
effective consolidation stresses of ¼, 1, 4, and 16 atm.

101
Figure 4.16. Normalized responses to undrained monotonic DSS loading for DR = 35, 55, and 75%
under vertical effective consolidation stresses of ¼, 1, 4, and 16 atm.

102
Figure 4.17. Drained strain-controlled cyclic DSS loading responses for DR = 35% under vertical
effective consolidation stresses of 1, 4, and 16 atm.

103
Figure 4.18. Drained strain-controlled cyclic DSS loading responses for DR = 55% under vertical
effective consolidation stresses of 1, 4, and 16 atm.

104
Figure 4.19. Drained strain-controlled cyclic DSS loading responses for DR = 75% under vertical
effective consolidation stresses of 1, 4, and 16 atm.

105
Figure 4.20. Volumetric strains during drained strain-controlled cyclic DSS loading for DR = 35, 55, and 75% with a vertical effective
consolidation stress of 100 kPa.

106
Figure 4.21. Volumetric strain due to post-cyclic reconsolidation versus the maximum shear strain
induced during undrained cyclic DSS loading.

107
Figure 4.22. Calibrating PM4Sand to a specified undrained critical state shear strength by adjusting R
(Boulanger and Ziotopoulou 2018)

Figure 4.23. Monotonic undrained DSS responses for DR = 35% two calibrations: Calibration 1 with
defaults for all secondary parameters, and Calibration 2 with R based on su,cs = 7 kPa.

108
Figure 4.24. Cyclic undrained DSS response for DR = 35% at CSR = 0.09 for (1) Calibration 1 with default
values for all secondary parameters, and (2) Calibration 2 with R based on su,cs = 7 kPa.

109
5. CONCLUDING REMARKS

The PM4Sand (Version 3.3) plasticity model presented herein is built upon the basic
framework of the stress ratio-controlled, critical state-based, bounding surface plasticity model for
sand presented by Dafalias and Manzari (2004). A series of modifications and additions to the
model were incorporated by Boulanger (2010; Version 1), Boulanger and Ziotopoulou (2012
Version 2; 2015 Version 3; 2017 Version 3.1, 2023 Version 3.2) and further herein (Version 3.3)
to improve its ability to approximate the stress-strain responses important to geotechnical
earthquake engineering practice; in essence, the model was calibrated at the equation level to
provide for better approximation of the trends observed in empirical correlations commonly used
in practice. These constitutive modifications included:

• revising the fabric formation/destruction to depend on plastic shear rather than plastic
volumetric strains;
• adding fabric history and cumulative fabric formation terms;
• modifying the plastic modulus relationship and making it dependent on fabric;
• modifying the dilatancy relationships to include dependence on fabric and fabric history,
and provide more distinct control of volumetric contraction versus expansion behavior;
• providing a constraint on the dilatancy during volumetric expansion so that it is consistent
with Bolton’s (1986) dilatancy relationship;
• modifying the elastic modulus relationship to include dependence on stress ratio and fabric
history;
• modifying the logic for tracking the initial back-stress ratio history;
• recasting the critical state framework to be in terms of a relative state parameter index;
• simplifying the formulation by restraining it to plane strain without Lode angle dependency
for the bounding and dilatancy surfaces;
• incorporating a methodology for improved modeling of post-liquefaction reconsolidation
strains; and
• providing default values for all but three primary input parameters.
The model (Version 3.3) was implemented as a user defined material in DLLs for use with the
commercial programs FLAC 8.1 (Itasca 2019) and FLAC2D 9.00 (Itasca 2023).

The three primary model input properties are: an apparent DR which affects the peak drained
and undrained strengths and the rate of strain accumulation during cyclic loading; the shear
modulus coefficient, Go, which should be calibrated to the estimated or measured in-situ shear
wave velocity; and the contraction rate parameter, hpo which is used to calibrate to the estimated
in-situ cyclic resistance ratio after all other properties have been set.

The behavior of the model was illustrated by simulations of element loading tests covering a
broad range of conditions, including drained and undrained, cyclic and monotonic loading under a
range of initial relative densities, confining stresses, and initial shear stress conditions. The current
formulation is limited to plane strain applications. Simulations presented in this report were
completed using the dynamic link library (DLL) version modelpm4sand005_64.dll compiled on
June 12, 2023 compatible with FLAC 8.1. The use of the DLL module cmodelPM4Sand2D009.dll
compiled on June 12, 2023 in FLAC2D 9.00 showed that the results remain unaffected. The model

110
was shown to provide reasonable approximations of desired behaviors and to be relatively easy to
calibrate.

111
ACKNOWLEDGMENTS

The authors are indebted to Professors Bruce Kutter and Yannis Dafalias for their discussions
and insights regarding the formulation and implementation of constitutive models. Dr. Robbie
Jaeger was instrumental to the development of the dynamic link library. The beta versions of the
model were extensively utilized by Professors Ronnie Kamai, Jack Montgomery, and Noriaki
Sento, whose feedback resulted in improvements to the model and this manual. Comments and
results from trial applications by Lelio Mejia, Erik Newman, Richie Armstrong, Jian Hu, and Faiz
Makdisi were extremely helpful. Feedback from Professor Pedro Arduino and Drs. Long Chen and
Alborz Ghofrani in their detailed examination of the model led to many of the clarifications in the
current revision of the manual. Drs. Panagiota Tasiopoulou’s and Michael H. Beaty’s detailed
examination of model results under different conditions led to improvements in the initialization
scheme. Dr. Ashraf Zekri’s study of the model and manual resulted in improvements in the text as
well as the creation of Figure 2.6. The authors greatly appreciate all the above assistance and
feedback. Feedback from Dr. Yasser Soltanpour helped identify the bug causing errors with
triangular elements having a single overlay in FLAC2D, leading to a revision of Version 3.3.
Portions of the work leading to the current version of the model were supported by the
California Division of Safety of Dams under Contract 4600009523 and by an International
Fulbright Science and Technology Award to the second author. Any opinions, findings,
conclusions, or recommendations expressed herein are those of the authors and do not necessarily
represent the views of these organizations.

REFERENCES
Andrianopoulos, K.I., Papadimitriou, A.G., and Bouckovalas, G.D. (2009). “Explicit integration
of bounding surface model for the analysis of earthquake soil liquefaction.” International J. of
Numerical and Analytical Methods in Geomechanics, DOI 10.1002/nag.875.

Andrus, R.D., and Stokoe, K.H. (2000). "Liquefaction resistance of soils from shear-wave
velocity." J. Geotechnical and Geoenvironmental Eng., ASCE 126(11), 1015–025.

Arai, H. (2006). "Detection of subsurface Vs recovery process using microtremor and weak ground
motion records in Ojiya, Japan." Third International Conference on Urban Earthquake
Engineering, Tokyo Institute of Technology, Tokyo, Japan.

Beaty, M. (2009). "Summary of UBCSand Constitutive Model: Versions 904a and 904aR." Short
Course on Seismic Deformation Analyses of Embankment Dams Considering Liquefaction
Effects, September 22-24, 2009, Davis, California.

Been, K., and Jefferies, M. G. (1985). "A state parameter for sands." Géotechnique 35(2), 99–112.

Bolton, M. D. (1986). "The strength and dilatancy of sands." Géotechnique 36(1), 65–78.

Boulanger, R. W. (2022). "Nonlinear dynamic analyses involving soil liquefaction or cyclic

softening." Proceedings, 20th ICSMGE State of the Art and Invited Lectures, Rahman and Jaksa
(eds), Australian Geomechanics Society, Sydney, Australia, ISBN 978-0-9946261-6-5, 351-379.

112
Boulanger, R. W. (2010). "A sand plasticity model for earthquake engineering applications."
Report No. UCD/CGM-10-01, Center for Geotechnical Modeling, Department of Civil and
Environmental Engineering, University of California, Davis, CA, 77 pp.

Boulanger, R. W., (2003a). "Relating K to relative state parameter index." J. Geotechnical and
Geoenvironmental Eng., ASCE 129(8), 770–73.

Boulanger, R. W. (2003b). "High overburden stress effects in liquefaction analyses." J.

Geotechnical and Geoenvironmental Eng., ASCE 129(12), 1071–082.

Boulanger, R.W., and Beaty, M.H. (2016). Seismic deformation analyses of embankment dams: A
reviewer's checklist. Proceedings, Celebrating the Value of Dams and Levees – Yesterday, Today
and Tomorrow, 36th USSD Annual Meeting and Conference, United States Society on Dams,
Denver, CO, 535-546.

Boulanger, R. W., and Idriss, I. M. (2004). "State normalization of penetration resistances and the
effect of overburden stress on liquefaction resistance." Proc., 11th Intl. Conf. on Soil Dynamics
and Earthquake Engineering, and 3rd Intl. Conf. on Earthquake Geotechnical Engineering, Doolin
et al., eds, Stallion Press, Vol. 2, pp. 484-491.

Boulanger, R. W., and Idriss, I. M. (2011). "Cyclic failure and liquefaction: Current issues." 5th
International Conference on Earthquake Geotechnical Engineering, Santiago, Chile, Jan 10-13.

Boulanger, R. W., Kamai, R., and Ziotopoulou, K. (2011). "Numerical modeling of liquefaction
effects." Proc., Effects of Surface Geology on Seismic Motion, 4th IASPEI / IAEE International
Symposium, August 23-26, University of California, Santa Barbara, CA.

Boulanger, R. W., Kamai, R., and Ziotopoulou, K. (2014). “Liquefaction induced strength loss and
deformation: Simulation and design.” Bulletin of Earthquake Engineering, Springer, 12: 1107-
1128, DOI 10.1007/s10518-013-9549-x.

Boulanger, R. W., Montgomery, J., and Ziotopoulou, K. (2015). “Nonlinear deformation analyses
of liquefaction effects on embankment dams.” Perspectives on Earthquake Geotechnical
Engineering, A. Ansal and M. Sakr, eds., Geotechnical, Geological and Earthquake Engineering
37, 247-283, Springer, DOI 10.1007/978-3-319-10786-8_10.

Boulanger, R. W., and Ziotopoulou, K. (2013). "Formulation of a sand plasticity plane-strain

model for earthquake engineering applications." Journal of Soil Dynamics and Earthquake
Engineering, Elsevier, 53, 254-267, 10.1016/j.soildyn.2013.07.006.

Boulanger, R.W., and Ziotopoulou, K. (2018). "On NDA practices for evaluating liquefaction
effects." Proc., Geotechnical Earthquake Engineering and Soil Dynamics V, Geotechnical Special
Publication 290, S. J. Brandenberg and M. T. Manzari, eds., ASCE, 1-20.

Byrne, P. M., Park, S. S., Beaty, M., Sharp, M., Gonzalez, L., Abdoun, T. 2004. "Numerical
modeling of liquefaction and comparison with centrifuge tests." Canadian Geotechnical Journal.
41 (2): 193-211.

113
Cubrinovski, M., and Ishihara, K. (1998). "State concept and modified elastoplasticity for sand
modelling." Soils and Foundations; 38(4): 213-225.

Dafalias, Y. F. (1986). "Bounding surface plasticity. I: Mathematical foundation and

hypoplasticity." J. Engineering Mechanics, 112(9), 966-987.

Dafalias, Y. F., and Manzari, M. T. (2004). "Simple plasticity sand model accounting for fabric
change effects." Journal of Engineering Mechanics, ASCE, 130(6), 622-634.

Dawson, E. M., W. H. Roth, S. Nesarajah, G. Bureau, and C. A. Davis. (2001). "A practice oriented
pore pressure generation model." Proceedings, 2nd FLAC Symposium on Numerical Modeling in
Geomechanics. Oct. 29-31, Lyon, France.

Desrues, J., Chambon, R., Mokni, M., and Mazerolle, F. (1996). "Void ratio evolution inside shear
bands in triaxial sand specimens studied by computed tomography.” Géotechnique, 46(3), 529–
546.

Duku, P. M., Stewart, J. P., and Whang, D. H. (2008). "Volumetric strains of clean sands subject
to cyclic loads." J. Geotechnical and Geoenvironmental Eng., ASCE, 134(8), 1073-1085.

Electric Power Research Institute (1993). Guidelines for site specific ground motions, Rept. TR-
102293, 1-5, Pala Alto, California.

Howell, R., Rathje, E. M., and Boulanger, R. W. (2014). "Evaluation of simulation models of
lateral spread sites treated with prefabricated vertical drains." Journal of Geotechnical and
Geoenvironmental Engineering, ASCE, 04014076, 10.1061/(ASCE)GT.1943-5606.0001185.

Idriss, I. M., and Boulanger, R. W. (2006). "Semi-empirical procedures for evaluating liquefaction
potential during earthquakes." J. Soil Dynamics and Earthquake Eng. 26, 115–30.

Idriss, I. M., and Boulanger, R. W. (2008). Soil liquefaction during earthquakes. Monograph
MNO-12, Earthquake Engineering Research Institute, Oakland, CA, 261 pp.

Idriss, I. M., and Boulanger, R. W. (2010). "SPT-based liquefaction triggering procedures." Report
UCD/CGM-10/02, Department of Civil and Environmental Engineering, University of California,
Davis, CA, 259 pp.

Ishihara, K. (1993). "Liquefaction and flow failure during earthquakes." Géotechnique 43(3), 351–
415.

Ishihara, K., and Yoshimine, M. (1992). "Evaluation of settlements in sand deposits following
liquefaction during earthquakes." Soils and Foundations 32(1), 173–88.

Itasca (2019). FLAC – Fast Lagrangian Analysis of Continua, Version 8.1, Itasca Consulting
Group, Inc., Minneapolis, Minnesota.

Itasca (2023). FLAC2D – Fast Lagrangian Analysis of Continua, Version 9.00, Itasca Consulting
Group, Inc., Minneapolis, Minnesota.

114
Kamai, R. (2011). "Liquefaction-induced shear strain localization processes in layered soil
profiles". PhD dissertation, University of California, Davis, 259pp.

Kamai, R. and Boulanger, R. W. (2012). "Single-element simulations of partial-drainage effects

under monotonic and cyclic loading." Journal of Soil Dynamics and Earthquake Engineering, 35,
29-40.

Khosravifar, A., Elgamal, A., Lu, J., and Li, J. (2018). "A 3D model for earthquake-induced
liquefaction triggering and post-liquefaction response." Soil Dynamics and Earthquake
Engineering, 110: 43-52.

Konrad, J.-M. (1988). "Interpretation of flat plate dilatometer tests in sands in terms of the state
parameter." Géotechnique 38(2), 263–77.

Kulhawy, F. H., and Mayne, P. W., 1990. Manual on Estimating Soil Properties for Foundation
Design, Report EPRI EL-6800, Electric Power Research Institute, Palo Alto, CA.

Kutter, B. L., and Chen, Y.-R. (1997). “Constant p’ and constant volume friction angles are
different.” ASTM Geotechnical Testing Journal, ASTM, 20(3), 304-316.

Lee, K. L., and Seed, H. B. (1967). "Drained strength characteristics of sands." J. Soil Mechanics
and Foundations Div., ASCE 93(SM6), 117–41.

Li, X. S., and Wang, Y. (1998). "Linear representation of steady-state line for sand." J.
Geotechnical and Geoenvironmental Engineering, 124(12), 1215-1217.

Liu, H.Y., Abell, J.A., Diambra, A., and Pisano, F. (2020). "Memory-enhanced plasticity modeling
of sand behavior under undrained cyclic loading." Journal of Geotechnical and Geoenvironmental
Engineering, 146(11): 04020122, 10.1061/(ASCE)GT.1943-5606.0002362.

Manzari, M. T., and Dafalias, Y. F. (1997). "A critical state two-surface plasticity model for sand."
Géotechnique, 47(2), 255-272.

Marti, J. and Cundall, P.A. (1982). "Mixed discretization procedure for accurate solution of
plasticity problems." Int. J. Num. Methods and Anal. Methods in Geomechanics, 6, 129-139.

Matsuya, A., Kazama, M., Sento, N. and Uzuoka, R. (2004). "Effect of initial shear stress on
reconsolidation characteristics of sand subjected undrained cyclic shear." The Japanese
Geotechnical Engineering Society, Proceedings of the 39th Japan National Conference on
Geotechnical Engineering, pp. 497-498, July (in Japanese)

Papadimitriou, A. G., Bouckovalas, G. D., and Dafalias, Y. F. (2001). "Plasticity model for sand
under small and large cyclic strains." J. Geotechnical and Geoenvironmental Engineering, 127(11),
973-983.

Richart, F. E., Jr., Hall, J. R., and Woods, R. D. (1970). "Vibration of soils and foundations."
International series in theoretical and applied mechanics, Prentice-Hall, Englewood Cliffs, N.J.

115
Rowe, P. W. (1962). "The stress-dilatancy relation for static equilibrium of an assembly of particles
in contact." Proc. R. Soc. London, Ser. A, 269, 500-527.

Sadrekamiri, A, and Olson, S. A. (2010). "Shear band formation observed in ring shear tests on
sandy soils." J. Geotechnical and Geoenvironmental Engineering, ASCE, 366-375.

Schofield, A. N., and Wroth, C. P. (1968). Critical state soil mechanics, McGraw-Hill, New York.

Silver, M. L., and Seed, H. B. (1971). "Volume changes in sand during cyclic loading." J. Soil
Mechanics and Foundations Div., ASCE 97(SM9), 1171–182.

Tasiopoulou, P., and Gerolymos, N. (2016). "Constitutive modeling of sand: Formulation of a new
plasticity approach." Soil Dynamics and Earthquake Engineering, 82:205-221.

Wang, Z. L., Dafalias, Y. F., and Shen, C. K. (1990). "Bounding surface hypoplasticity model for
sand." J. Engineering Mechanics, ASCE, 116(5), 983-1001.

Yang, M., Taiebat, M., Dafalias, Y. (2022). "SANISAND-MSf: a memory surface and
semifluidized state enhanced sand plasticity model for undrained cyclic shearing." Géotechnique,
10.1680/jgeot.19.P.363

Yang, Z., A. Elgamal and E. Parra, (2003). "Computational model for cyclic mobility and
associated shear deformation." J. Geotechnical and Geoenvironmental Engineering, ASCE,
129(12), 1119-1127.

Yoshimine, M., Robertson, P. K., and Wride, C. E. (1999). “Undrained shear strength of clean
sands to trigger flow liquefaction.” Canadian Geotechnical Journal, 36, 891-906.

Youd, T. L. (1972). "Compaction of sands by repeated straining." J. Soil Mechanics and

Foundations Div., ASCE 98(SM7), 709–25.

Yu, P., and Richart, F. E., Jr. (1984). "Stress ratio effects on shear modulus of dry sands." J.
Geotechnical Engineering, ASCE, 110(3), 331-345.

Ziotopoulou, K. (2010). "Evaluating model uncertainty against strong motion records at liquefiable
sites", Masters thesis, University of California, Davis, 294pp.

Ziotopoulou, K., Boulanger, R. W., and Kramer, S. L. (2012). "Site response analyses of liquefying
sites." Geo-Congress 2012: State of the Art and Practice in Geotechnical Engineering, R. D.
Hryciw, A. Athanasopoulos-Zekkos, and N. Yesiller, eds., Geotechnical Special Publication No.
225, ASCE Geo-Institute, 1799-1808.

Ziotopoulou, K., and Boulanger, R. W. (2012). "Constitutive modeling of duration and overburden
effects in liquefaction evaluations." Proc., Second International Conference on Performance-based
Design in Earthquake Geotechnical Engineering, Taormina, Italy, May 28-30, paper 03.10.

116
Ziotopoulou, K., and Boulanger, R. W. (2013a). "Calibration and implementation of a sand
plasticity plane-strain model for earthquake engineering applications." Journal of Soil Dynamics
and Earthquake Engineering, 53, 268-280, 10.1016/j.soildyn.2013.07.009.

Ziotopoulou, K., and Boulanger, R. W. (2013b). "Numerical modeling issues in predicting post-
liquefaction reconsolidation strains and settlements." 10th International Conference on Urban
Earthquake Engineering, March 1-2, Tokyo Institute of Technology, Tokyo, Japan, 469-475.

Ziotopoulou, K., and Boulanger, R. W. (2016). "Plasticity modeling of liquefaction effects under
sloping ground and irregular cyclic loading conditions." Soil Dynamics and Earthquake
Engineering, 84 (2016), 269-283, 10.1016/j.soildyn.2016.02.013.

Ziotopoulou, K. (2014). "A sand plasticity model for earthquake engineering applications." PhD
Dissertation, University of California, Davis.

117