STRUCTURAL IMPACT LABORATORY, NTNU

Materials Mechanics
Part II
Odd Sture Hopperstad & Tore Børvik
8/31/2017

Second version 2017

Table of contents
1 Introduction 1-1

2 Continuum mechanics for large deformations 2-1

2.1 Motion of a body 2-1
2.2 Physical field quantities and material time derivatives 2-1
2.3 Rate-of-deformation and spin 2-3
2.4 Some simple motions 2-6
2.4.1 Uniform extension 2-6
2.4.2 Simple shear 2-8
2.4.3 Rigid rotation about the origin 2-9
2.5 Corotational base vector system 2-10
2.6 Stress and equations of motion 2-12
2.7 Balance of mechanical energy 2-13
2.8 Principle of virtual power 2-14
2.9 References 2-15

3 Plasticity and viscoplasticity 3-1

3.1 General formulation 3-1
3.2 Rate-independent J2 flow theory 3-3
3.3 The modified Johnson-Cook model 3-4
3.4 Proportional loading 3-7
3.5 Plastic limit theorems 3-8
3.6 References 3-11

4 Ductile damage and failure criteria 4-1

4.1 Stress invariants 4-2
4.2 Growth of an isolated cylindrical void 4-5
4.3 Damage evolution rules 4-8
4.4 Damage coupling 4-14
4.5 References 4-18

5 Porous plasticity 5-1

5.1 General formulation 5-2
5.2 Yielding of thick-walled sphere under external hydrostatic stress 5-4
5.3 Gurson model 5-8
5.4 Nucleation, coalescence and shearing of voids 5-11
5.5 References 5-14

I
6 Hypoelastic-plasticity 6-1
6.1 Problem definition 6-1
6.2 Corotational formulation 6-3
6.3 Jaumann stress rate 6-5
6.4 Jaumann formulation for isotropic materials 6-7
6.5 Thermal effects 6-8
6.6 Corotational integrals 6-10
6.7 Numerical integration 6-12
6.8 References 6-15

7 Explicit finite element methods 7-1

7.1 Semi-discrete equations of motion 7-1
7.2 Central difference time integration scheme 7-3
7.3 Stable time step 7-4
7.4 Energy balance 7-4
7.5 Mass scaling 7-5
7.6 Strain softening, element erosion and mesh dependence 7-6
7.7 References 7-13

8 Crystal plasticity 8-1

8.1 Introduction 8-1
8.2 Single crystal plasticity 8-3
8.2.1 Lattice rotation and plastic slip 8-3
8.2.2 Yield condition and normality 8-5
8.2.3 Principle of maximum plastic power 8-8
8.2.4 Complementary minimum principle 8-9
8.2.5 Work-hardening 8-10
8.2.6 Rate-dependent behaviour 8-11
8.2.7 Lattice rotation and elastic behaviour 8-12
8.2.8 Numerical integration 8-14
8.3 Polycrystal plasticity 8-16
8.3.1 Crystal orientation and texture 8-16
8.3.2 Taylor’s approach 8-19
8.4 References 8-20

9 Localization of deformation 9-1

9.1 Introduction 9-1
9.2 Bifurcation analysis 9-2
9.3 Interlude: Material time derivatives of unit vectors 9-7

II
 9.4 Imperfection analysis 9-9
9.5 References 9-11

10 Thermodynamics 10-1
10.1 First and second law of thermodynamics 10-1
10.2 Thermo-elastic materials and heat conduction 10-3
10.3 Inelastic solid materials 10-5
10.4 Heat equation 10-8
10.5 Thermoelastic-thermoviscoplastic materials 10-10
10.6 References 10-18

Appendix A: Polar coordinates

Appendix B: Homogeneous and convex functions

III
IV
1 Introduction
The main aims of Part II of the lecture notes in Materials Mechanics are as follows: to
extend the theory of plasticity to finite deformations, to introduce models for damage
evolution and ductile fracture, and to give a brief presentation of explicit finite element
methods. Explicit finite element methods are used mainly for highly nonlinear problems
like metal forming, crashworthiness, explosions and structural impact; the main application
area being fast transient dynamics. In these applications, the strains and rotations are usually
finite, as the structural elements are subjected to large deformations, and the mathematical
formulation of elastic-plasticity and elastic-viscoplasticity from Chapters 7 and 8 of Part I
of the lecture notes has to be modified accordingly. In addition, when considering large
deformations, plastic instability, damage evolution and fracture become important
phenomena to describe.
Chapter 2 gives a brief introduction to continuum mechanics for solids undergoing
large deformations. The kinematics of large deformations with reference to the current
configuration of the solid is first described. The rate-of-deformation tensor is found to be a
suitable measure of strain rate, while the spin tensor is a measure of rotation rate. A
corotational coordinate system is defined which rotates with the spin of the material, and
will later be used to construct valid constitutive relations for large-deformation problems.
Next the Cauchy stress tensor is defined in the current configuration and the equations of
motion are established. Finally, the balance of mechanical energy and the principle of virtual
power for dynamically loaded solids are derived from the equations of motion.
In Chapter 3, the theories of plasticity and viscoplasticity are extended to large
deformation, while the formulation of the elastic behaviour is saved for Chapter 6. Thus, we
only consider plastic deformations in Chapters 3–5. The main emphasis of Chapter 3 is on
the general formulation of large-deformation plasticity and viscoplasticity, while the J2 flow
theory and the Johnson-Cook model are used as examples of plastic and thermo-viscoplastic
models, respectively. It will be shown that the extension from small to large deformation is
almost trivial as long as the elastic behaviour is excluded. Employing the principle of virtual
power, the upper-bound and lower-bound theorems of rigid perfect plasticity are derived for
materials obeying the principle of maximum plastic dissipation.
When the plastic deformations become large, the risk for plastic instability, damage
evolution and fracture increases and these phenomena have to be included in the constitutive
modelling. The modelling of ductile fracture is the topic of Chapter 4, which describes
uncoupled damage models for use in nonlinear finite element simulations of plastic and
viscoplastic materials. The word “uncoupled” means that the damage development is
assumed to have negligible influence on the plastic behaviour until fracture occurs. In
explicit finite element simulations, fracture is most often described by element erosion,
which implies that the stress tensor is set to zero when the fracture criterion is reached in an
integration point. When all or a number of integration points have failed, the element is
removed from the simulation. This way crack propagation may be described at least in a
qualitative sense. Ductile damage evolution and fracture occurs by nucleation, growth and
coalescence of voids and these processes are strongly dependent upon the stress state. Thus,
the uncoupled damage evolution rules depend on the stress invariants for isotropic materials,
but, as the failure process is ductile, the main role is played by plastic straining or plastic
dissipation. Finally, a simple approach to couple the damage evolution with the plastic and
viscoplastic constitutive equations is presented.

1-1
 Chapter 5 presents the mathematical formulation of porous plasticity with main
emphasis on the Gurson model. In these models, the void volume fraction (or porosity) is
included in the plasticity formulation as an internal variable, and processes like void
nucleation, growth and coalescence are described by evolution equations. Since the damage,
in the form of the void volume fraction, is now included in the plasticity formulation, this
is a coupled damage model, and the evolution of damage will soften the material. Indeed, if
the damage evolution exceeds the work-hardening, the material will exhibit strain softening;
i.e., the strength of the material decreases with increasing plastic straining. Strain softening
leads to concentration of the plastic deformation in certain regions of the structure and is a
precursor to failure, either in the form of plastic instability (localization of deformation) or
ductile fracture.
In Chapter 6, we describe a way to introduce elasticity into our models of large-
deformation plasticity and viscoplasticity. The hypoelastic formulation will be adopted here,
while hyperelastic formulations are covered in Part III of the lecture notes. The plastic
strains and the rotations are allowed to be finite, but the elastic strains are assumed to remain
infinitesimal. The latter is a reasonable assumption for several classes of materials, with
metals and alloys as the most important examples. The formulation is called hypoelastic
because the elastic constitutive equations are not derived from a strain energy function but
expressed directly in rate form—the result being that dissipation may occur in a closed
elastic deformation cycle. It should, however, be mentioned that the theory of hypoelastic-
plasticity described in this chapter is adopted today in most commercial finite element
codes. Our main emphasis is on the corotational formulation in which the stresses and strain
rates are defined in a corotational coordinate system. This formulation is valid for isotropic
and anisotropic materials. However, we will also derive a formulation based on the Jaumann
stress rate, which is used in most commercial codes for isotropic materials. The reason for
this is its computational efficiency. Finally, the various approaches for numerical
implementation of hypoelastic-plasticity is described.
Chapter 7 gives a brief introduction to explicit nonlinear finite element methods for
solids subjected to dynamic loading and undergoing large deformations. The main aim is to
show where the constitutive model enters into the formulation of the semi-discrete equations
of motion and then to explain the explicit time integration of these equations. As already
mentioned, explicit finite element methods are particularly suitable for rapid and highly
nonlinear problems (e.g. including fracture by element erosion), since the explicit time
integration requires exceedingly small time steps. However, by using mass scaling (or time
scaling), the range of application of explicit finite element methods is extended to highly
nonlinear quasi-static problems like metal forming and structural collapse. The central
difference scheme used to integrate the semi-discrete equations of motion in time is only
conditionally stable. An estimate on the stable time step is used in simulations, but still
numerical instabilities may occur for highly nonlinear problems. Such instabilities are most
easily detected by making a check of the energy balance, since numerical instabilities lead
to spurious generation of energy, thus violating the law of conservation of energy. As a
consequence, it is pivotal to make energy checks when running explicit finite element
simulations and thereby to ensure that the energy is conserved during the simulation.
Thermal softening due to adiabatic heating or damage softening due to nucleation, growth
and coalescence of voids tend to make finite element simulations mesh dependent. In some
circumstances this mesh dependence becomes pathological, i.e., as the mesh is refined the
plastic dissipation approaches zero—obviously a physically meaningless result for a body
loaded plastically. The result is that some kind of regularization scheme is needed to

1-2
enhance the convergence properties of the simulation. Various regularization schemes are
briefly presented and discussed.
In crystal plasticity, the behaviour of the metallic material is modelled at the level of
the single crystal, or grain, and constitutive equations valid at the slip system level are
developed. This is the topic of Chapter 8. By use of crystal plasticity, the anisotropy caused
by non-random orientation of the grains of a polycrystal—or the crystallographic texture—
is modelled in a natural way, and the evolution of the texture due to grain rotation with
plastic deformation comes as a result of the formulation. Two important applications of
crystal plasticity are to establish the anisotropic yield surface of metallic materials based on
crystallographic texture and to simulate work-hardening and texture evolution during plastic
forming operations.
Localization of the deformation of a homogeneous and homogeneously deformed solid
into narrow planar bands is considered in Chapter 9. Softening due to damage evolution or
adiabatic heating may lead to non-unique solutions of the boundary value problem for rate-
independent material behaviour, such as plasticity, which means that other solutions than
the fundamental solution are possible. Localization of the deformation within shear bands
is a frequent precursor to failure in metallic materials. The criterion for such a bifurcation
to occur is established in this chapter. Some consequences of this for finite element solutions
are also addressed through a one-dimensional example. Finally, we consider localization of
deformation in a pre-existing planar band in an otherwise homogeneous and homogeneously
deformed solid, where the constitutive equations inside and outside the band may differ.
Finally, in Chapter 10 the laws of thermodynamics are presented and applied to define
thermodynamically admissible constitutive equations for elastic and inelastic materials. One
important result is the heat equation that can be used to calculate the temperature change
during deformation of a material, accounting for heat conduction, thermoelastic effects and
inelastic dissipation. Another result is the dissipation inequality which has to be fulfilled for
any thermodynamically admissible constitutive relation. Lastly, we obtain expressions for
the free energy of thermoelastic-thermoviscoplastic solids with isotropic hardening, which
allow us to estimate the Taylor-Quinney coefficient giving the percentage of the plastic
power that is converted into heat.
Chapters and sections marked with an asterisk * are not part of the syllabus of the course
TKT 4128 Impact Mechanics. The notation used in these lecture notes follows what was
established in Part I. For completeness, a brief review of tensor notation is given below.

1-3
 Brief review of tensor notation
(from Chapter 3 of Part I)

Let v and T be first and second order tensors, respectively, and let  ei  and  eˆ i 
be two rectangular (or Cartesian) basis systems. The basis vectors e i and eˆ i , i  1, 2,3 ,
are mutually perpendicular and of unit length, i.e., they are orthonormal. The relation
between the basis vectors are given by
eˆ i  aij e j , ei  a ji eˆ j

where aij  eˆ i  e j is the orthogonal transformation matrix, i.e., aik a jk  aki akj   ij ;  ij
being the Kronecker delta.
The components of tensors v and T are denoted by vi and Tij in the basis  ei 
and by vˆi and Tˆij in the basis  eˆ i  . The transformation rules for the vector and tensor
components read as
vˆi  aij v j , vi  a ji vˆ j

Tˆij  aik a jlTkl , Tij  aki aljTˆkl

The component matrices of the two tensors are denoted by v and T in the basis  ei 
and by v̂ and T̂ in the basis  eˆ i  . In matrix notation, the coordinate transformation
rules are expressed by
vˆ  av, v  aT vˆ
Tˆ  aTa T , T  a T Ta
ˆ

where a  [aij ] is the 3×3 transformation matrix, v  [vi ] and vˆ  [vˆi ] are row vectors
(3×1), while T  [T ] and Tˆ  [Tˆ ] are 3×3 matrices.
ij ij

Note here that bold italic letters, e.g. v and T , indicate matrices of any
dimensions, whereas bold letters, e.g. v and T , are reserved for vectors and tensors.

1-4
2 Continuum mechanics for large deformations
Chapter 4 of Part I presented the continuum mechanics under the assumption that the
deformations of the solid body remained small. In this chapter, we give a primer to the
continuum mechanics for solid bodies subjected to large deformations, while Part III of
these lecture notes gives a more comprehensive presentation. We recommend the textbooks
by Irgens [1], Lubliner [2], Belytschko et al. [3] and Malvern [4] for further reading.

2.1 Motion of a body

Figure 2-1 shows the initial and current configurations of a solid body. To define the
components of vectors and tensors, we introduce a Cartesian coordinate system with base
vectors ei , i  1, 2,3 . In the initial configuration at t  0 , a material point (or particle) P has
position vector X . In the current configuration, at time t , this material point has moved
and the position vector is x . Thus, the motion of the material point P is defined by
x  x  X,t  (2-1)
This equation represents a one-to-one mapping between the sets of points in the initial and
current configurations. The displacement of the material point P from the initial to the
current configuration is given by the displacement vector
u  X, t   x  X, t   X (2-2)
The velocity and acceleration vectors are defined as
x  X, t   2 x  X, t 
v  X, t    x  X, t  , a  X, t    x  X, t  (2-3)
t t 2
The time rate of change of a physical field quantity   X,t  attached to the particle X is
called the material time derivative and defined as   X, t     X, t  / t . Here,   X,t 
may represent a scalar or the component of a tensor quantity.

Current configuration
Initial configuration

P*

Figure 2-1: Initial and current configuration of the solid body.

2-1
2.2 Physical field quantities and material time derivatives
There are two types of physical quantities: intensive and extensive. An intensive
quantity is a physical property that does not depend on the amount of material, while an
extensive quantity does and is additive for parts of the body. Temperature and density are
examples of intensive quantities, while mass and volume are examples of extensive
quantities.
The mass m of a continuous body with current volume V is per definition
m    dV (2-4)
V

where     x,t  is the density in the current configuration. According to the principle of
mass conservation, m is constant during any motion of the body. We may consider the
volume V to consist of many small volume elements dV and the mass m to be the sum of
the contributions  dV to the total mass from each of these small volume elements. Mass
conservation implies that the mass  dV of any volume element dV is constant, even
though the volume element itself may change in shape and size.
Due to the one-to-one mapping between the initial and current configuration, we may
express an intensive quantity  either as a function of the position vector X in the initial
configuration,     X,t  , or as a function of the position vector x in the current
configuration,     x,t  . The first choice is called a Lagrangian (or material) description,
while the second choice is denoted an Eulerian (or spatial) description of the field. In the
Lagrangian description, the material time derivative is simply the partial derivative with
respect to time, because X is constant. In the Eulerian description, we have
    x  X, t  , t  and the material time derivative is calculated from the chain rule:

D   x, t    x, t  x  X, t 
    (2-5)
Dt t x t
By introducing the velocity vector, we get
D  
   v (2-6)
Dt t x
The first term is local and stems from the change of the field at a fixed position; the second
term is called convective and is caused by the motion of the particle.
Now, consider the extensive quantity  defined by
    dV (2-7)
V

where the quantity  is intensive and the volume V is time dependent. Due to mass
conservation, the time rate of change of the quantity  dV is simply  dV , and the time
rate of change of  is equal to the sum of the contributions  dV from each of the small
volume elements dV . As a result, the material time derivative of the extensive field
quantity  may be calculated as
    dV (2-8)
V

2-2
2.3 Rate-of-deformation and spin
Adopting the Eulerian description, the velocity field is expressed as
vi  vi  x, t  (2-9)
The gradient of the velocity field with respect to the spatial coordinate, denoted the velocity
gradient L , is defined by
vi
Lij   Dij  Wij (2-10)
x j
The symmetric part of the velocity gradient is called the rate-of-deformation tensor D ,
whereas the skew-symmetric part is denoted the spin tensor W :
Dij  Lij   12  Lij  L ji  , Wij  Lij   12  Lij  L ji  (2-11)

In terms of the velocity vector, the rate-of-deformation and spin tensors are expressed as

1  v v j  1  v v j 
Dij   i   , Wij   i   (2-12)
2  x j xi  2  x j xi 

By comparison with Equation (4-15) in Part I, it is understood that for small deformations
(i.e., when there is no need to distinguish between the initial and current configurations),
Dij and Wij are in turn equal to the strain rate tensor  ij and the rate-of-rotation tensor ij .
We note that even if D is called the rate-of-deformation tensor, it is not the proper rate of
any tensor for large deformations and the time integral of D is not always zero for a closed
deformation path.
Consider now an infinitesimal line segment dx at time t emanating from the particle
P and terminating at a particle Q, and let dv be the relative velocity of Q with respect to P.
The situation is illustrated in Figure 2-2. From the definition of the velocity gradient in
Equation (2-10), we have
vi
dvi  dx j  Lij dx j or dv  v  dx  L  dx (2-13)
x j
It is evident that the velocity gradient defines the relative motion in the neighbourhood of a
particle in the current configuration. In view of Equation (2-3)1, we may also write [4]
D
dxi  dxi  Lij dx j (2-14)
Dt
Thus, the velocity gradient is a tensor defined with respect to the current configuration that
transforms the infinitesimal line segment dx into its material time derivative dx .

2-3
Figure 2-2: Relative motion of the material points P and Q.
We will now demonstrate that the rate-of-deformation tensor D represents the
stretching of the material element, whilst the spin tensor W defines the rotation rate. To
this end, we consider the motion of four adjacent material points P, Q, R and S in the time
interval between t and t  dt . The material line elements PQ, PR and PS are directed along
the coordinate axes x1 , x2 , x3 at time t , i.e., they are instantaneously mutually orthogonal.
The line elements are further assumed to be infinitesimal and their lengths are defined by
PQ  dx1 , PR  dx2 and PS  dx3 . At time t  dt , the lengths of the material line
elements and the angle between them have changed due to the imposed velocity field. Figure
2-3 shows the material points P, Q and R located in the x1 x2 plane and the local deformation
during the time interval. It is clear from Figure 2-3 that during the time interval the
incremental deformation of the infinitesimal neighbourhood around material point P is
defined by
d ij  Dij dt (2-15)
We will denote this tensor the incremental deformation tensor. From the figure we deduce
that the normal components d11 , d 22 , d 33 define the specific changes in length of the
particle distances PQ , PR and PS , while the shear components 2d12 , 2d 23 , 2d 31
represent the change in angle between the initial orthogonal pairs of material line elements
PQ  PR , PR  PS and PS  PQ .

An interpretation of the spin tensor W is obtained by assuming that the coordinate

axes x1 , x2 , x3 instantaneously coincides with the principal directions of the rate-of-
deformation tensor D at time t , so that all the shear components 2d12 , 2d 23 , 2d 31 vanish.
This implies that the material line elements PQ, PR and PS remain mutually orthogonal but
they may rotate rigidly during the time interval dt . This incremental rotation is defined by
the tensor
dij  Wij dt (2-16)

with non-zero components d12  d21 , d23  d32 , d31  d13 , which we will
denote the incremental rotation tensor (see also Chapter 4 of Part I). From this we infer that
the spin tensor W defines the instantaneous spin of three orthogonal material line elements
oriented along the principal axes of the rate-of-deformation tensor D [1].

2-4
Figure 2-3: Two-dimensional graphical illustration of the role of the components of the
rate-of-deformation and spin tensors [5].

In the coordinate system defined instantaneously by the principal directions of D , it is

also straightforward to calculate the rate of change of volume of the infinitesimal volume
element based on Figure 2-3. At time t , the volume is dV  dx1dx2 dx3 , while at time t  dt
the volume has change into
dV  1  D11dt  dx1 1  D22 dt  dx2 1  D33dt  dx3  1  Dkk dt  dV (2-17)
where higher-order terms were neglected to obtained the latter approximation. We may thus
define the volumetric strain rate as
 v  Dkk  D11  D22  D33 (2-18)

Since Dkk is the first invariant of D , the expression for the volumetric strain rate holds in
any coordinate system.

2.4 Some simple motions

In the following, we will address some important motions and calculate the rate-of-
deformation and spin tensors for them. The first is uniform extension which is a motion
often aimed at in materials testing. Uniaxial tension, plane-strain tension and equi-biaxial
tension are frequently used tests to determine material properties and they are all examples
of uniform extension. We may of course substitute tension with compression in these three
cases and consider compression tests instead. The spin tensor is zero in uniform extension,
while in the second and third examples presente below, namely simple shear and rigid
rotation, it is not.

2-5
2.4.1 Uniform extension

Initial
configuration

Current
configuration

Figure 2-4: Uniform extension.

Consider the motion of a material element defined by

x1  1 X1 , x2  2 X 2 , x3  3 X 3 (2-19)

where X i and xi are the components of the initial and current position vectors, respectively.
The principal stretch ratios i  t  are defined as the ratio of the current to initial dimensions
of the material element along the coordinate axes, i  li / li0 (no sum), where li0 and li are
the initial and current dimensions of the material element. This field prescribes a uniform
extension of the body as illustrated in Figure 2-4. If the stretch ratio i is less than unity,
there is a contraction along the xi axis.
If we assume an isotropic material subjected to isochoric motion (i.e., the volume is
preserved during the motion, which is a good approximation for metal plasticity up to large
deformations), we may identify the following three special cases:
 Uniaxial tension: 1  1, 2  3  11/2
 Plane strain tension: 1  1, 2  1, 3  11
 Equi-biaxial tension: 1  2  1, 3  1121  12
It is noted here that plane strain tension could also be denoted pure shear. We may also
identify similar motions in compression by letting 1 be less than unity. Isochoric motion
implies that l1l2l3  l10l20l30 , which is fulfilled provided 123  1 . By taking the material time
derivative of the latter expression, we readily find an alternative criterion for volume
preserving motion, namely
1 2 3
  0 (2-20)
1 2 3
The velocity field v  vi ei is obtained in terms of the initial coordinates as

v1  x1  1 X1 , v2  x2  2 X 2 , v3  x3  3 X 3 (2-21)

2-6
where the superimposed dot implies material time derivative. Inserting the equations for the
motion, we may instead express the velocity field in terms of the current coordinates
1  
v1  x1 , v2  2 x2 , v3  3 x3 (2-22)
1 2 3
Using the indicated coordinate system, the component matrix of the velocity gradient
tensor Lij  vi / x j reads as

 1 0 0
1 
L  0 2
2 0 (2-23)
 
0 0 3 
 3 

Since the velocity gradient is diagonal—and thus symmetric, the spin is zero for this motion,
and the component matrices of the rate-of-deformation and spin tensors are
 1 0 0
1  0 0 0
D  0 2
2 0  , W  0 0 0  (2-24)
 
0 3  0 0 0 
0
 3 

The velocity field is volume preserving on condition that the trace of the rate-of-deformation
tensor is zero, Dkk  0 . Clearly, the diagonal elements of the rate-of-deformation tensor
represent the strain rates in uniform extension, and the logarithmic principal strains are
obtained as

  /   dt  ln   no sum 
i
i   i i i (2-25)
1

Note, however, that this is a special case, as the spin is zero, and in general  D dt
ij does not
represent a physically meaningful measure of strain.

2-7
2.4.2 Simple shear

Initial Current
configuration configuration

Figure 2-5: Simple shear.

The geometry of simple shear is illustrated in Figure 2-5. The motion is described by
x1  X1   X 2 , x2  X 2 , x3  X 3 (2-26)
where   tan  and the angle  is defined in Figure 2-5. The associated velocity field
v  vi ei is then
v1   X 2   x2 , v2  0, v3  0 (2-27)
where  is denoted the shearing rate. Using the indicated coordinate system, the component
matrix of the velocity gradient tensor is
0  0 
L  0 0 0  (2-28)
0 0 0 

Thus, the component matrices of the rate-of-deformation and spin tensors read as
 0 2 0  0 
2 0
   
D   2 0 0  , W    2 0 0 (2-29)
0 0 0 0 0 0 
  
where

 (2-30)
cos 2 
It is apparent that simple shear motion involves combined shearing ( D12  D21  0 ) and
rotation ( W12  W21  0 ) of the material element. Note that simple shear is volume
preserving, since the trace of the rate-of-deformation tensor is always zero ( Dkk  0 ).

2-8
2.4.3 Rigid rotation about the origin

Initial
configuration

Current
configuration

Figure 2-6: Rigid rotation about the origin.

Rigid rotation about the origin is illustrated in Figure 2-6. The velocity field of this
motion is given by
v  n  x (2-31)
where  is the angular speed and n is the axis of rotation. By the definition of the cross
product, the velocity vector v is mutually perpendicular to the rotation axis n and the
position vector x and proportional to the angular speed  as well as the magnitude of the
position vector | x | . In our case, we have n  e3 , and thus

e1 e2 e3
v   e3  x   0 0 1   x2e1   x1e 2 (2-32)
x1 x2 x3
The component matrix of the velocity gradient is
 0  0
 
L   0 0 (2-33)
0 0 0 

Since the velocity gradient tensor is skew-symmetric, the rate-of-deformation vanishes and
the spin equals the velocity gradient. Thus, the component matrices are
0 0 0   0  0
   
D  0 0 0  , W   0 0 (2-34)
0 0 0  0 0 0 

The fact that D is zero under rigid rotation is important for the development of
hypoelasticity in Chapter 6.

2-9
2.5 Corotational base vector system

Figure 2-7: Corotational base vector system.

When formulating constitutive equations for large deformations, it is often useful to

introduce a Cartesian basis  eˆ i  which is assumed to co-rotate with the spin of the material
element, see Figure 2-7. This basis is said to be corotational. Thus, the material time
derivative of the basis vector eˆ i is defined by

eˆ i  W  eˆ i (2-35)
Since Equation (2-35) represents a rigid rotation, there must exist a rotation tensor  that
transforms the fixed basis  ei  into the corotational basis  eˆ i  , viz.

eˆ i    ei (2-36)

The components of eˆ i in the fixed basis are per definition the projections of this vector onto
the unit vectors e j :

e j  eˆ i  e j    ei   ji (2-37)
Thus
eˆ i   ji e j (2-38)

Analogously, the scalar product ei  eˆ j can be interpreted as the projection of ei onto the
unit vectors eˆ j , i.e., as the components of ei in the corotational basis. Accordingly, we have

ei  ij eˆ j (2-39)
Since both sets of base vectors are orthonormal, we have
ij  eˆ i  eˆ j  kiek lj el  kilj kl  kikj (2-40)

ij  ei  e j  ik eˆ k  jl eˆ l  ik  jl kl  ik  jk (2-41)

We conclude that the rotation tensor  is orthogonal:

ik  jk  kikj  ij (2-42)

2-10
Orthogonal tensors transform a vector into another vector with the same length but with a
different orientation in space.
Next, we need to find a differential equation for  in terms of the spin of the material
element. Using Equations (2-35) and (2-38), we get two alternative expressions for the
vector eˆ i , namely

eˆ i  W  eˆ i  W  e j  ji and eˆ i   jie j (2-43)

Projection of eˆ i onto e k gives

ek  eˆ i  ek  W  e j  ji  Wkj  ji and ek  eˆ i  ki (2-44)

Thus, by comparison and change of indices, we arrive at
ij  Wik kj (2-45)

which is the differential equation for the rotation tensor  . Obviously, we need to define
an initial value of  at time t  0 , i.e., in the initial configuration. In some cases, for
instance for isotropic materials, we may set the initial value of  equal to the identity tensor
I , which implies that the two base vector systems are initially identical. For anisotropic
materials, the coordinate axes of the corotational base vector system should be aligned with
the principal axes of symmetry in the initial configuration. It is then assumed that the
principal axes of anisotropy rotate with the spin of the material, which is a reasonable
assumption provided the deformations are moderate.
Let a be an arbitrary vector and A an arbitrary second-order tensor. The components
are obtained using the definitions established in Chapter 3 of Part I, and thus
ai  ei  a   ij eˆ j   a  ij aˆ j
(2-46)
Aij  ei  A  e j   ik eˆ k   A   eˆ l  jl   ik Aˆkl  jl

Using that  is orthogonal, we get the transformation rules

ai  ij aˆ j and aˆi   ji a j (2-47)

Aij  ik Aˆkl  jl and Aˆij  ki Akl lj (2-48)

Recall that these coordinate transformations are time dependent, since  evolves according
to Equation (2-45). The reader should compare these transformation rules with those
established in Chapter 3 of Part I.
We will use the corotational coordinate system in the following to establish physically
admissible hypoelastic constitutive equations for the material. As a final point, we note that
other definitions of the rotation tensor  are possible, see Part III for further details.

2-11
2.6 Stress and equations of motion
If we assume that the derivations in Chapter 4.2 of Part I are carried out in the current
configuration of the body they are valid also for large deformations. The stress tensor σ ,
defined with respect to the current configuration of a body subjected to large deformations,
is called the Cauchy stress tensor. Cauchy’s law defines the Cauchy stress tensor as, see
Figure 2-8
ti   ji n j (2-49)

where t is the traction force on the surface element with unit normal vector n in the current
configuration.

Figure 2-8: Components of the Cauchy stress tensor in two dimensions.

For quasi-static loading, the force and moment equilibrium equations are (cf. Chapter
4 of Part I)
 ij
 b j  0 and  ij   ji (2-50)
xi
where b is the body force vector, i.e., force per unit current volume of the body, and x is
the current position vector of a material point. The latter equation, which expresses moment
equilibrium, shows that the Cauchy stress tensor is symmetric. In the case of dynamic
loading conditions, we may apply D’Alemberts principle which states that any system of
forces is in equilibrium if the inertia forces are added to the external forces [6]. Thus, we
substitute the body force b with the effective body force b  b  a , where  a is the
inertia force per unit current volume; a is the acceleration vector and  is the density of
the material. The minus sign is due to the fact that the inertia force vector acts in the opposite
direction of the acceleration vector.
Inserting the effective body force into Equation (2-50), gives the equations of motion
(or the dynamic equilibrium equations) as
 ij
 bj   a j (2-51)
xi

2-12
As the inertia force tends to zero, the quasi-static solution is retrieved. The equations of
motion could instead have been directly derived by considering an infinitesimal material
element, as shown in Figure 2-9.

Figure 2-9: Dynamic equilibrium of infinitesimal material element in two dimensions.

2.7 Balance of mechanical energy

The balance of mechanical energy for a dynamic system subjected to large
deformations is derived from the equations of motion. Multiplying both sides of Equation
(2-51) with the velocity, and integrating over the volume V of the body in the current
configuration, leads to
 ij
 x
V i
v j dV   b j v j dV    a j v j dV
V V
(2-52)

Using the same procedure as in Chapter 4.3 of Part I, we get the mechanical energy balance
in the form
v j
  ij
V
xi
dV    ai vi dV   bi vi dV   ti vi dS
V V S
(2-53)

Here the two terms on the left-hand side of the equation are the power of the internal and
inertia forces, while the two terms on the right-hand side are the power of the body and
traction forces.
Using the symmetry of the Cauchy stress tensor σ and the definition of the rate-of-
deformation tensor D , the power of the internal forces may be re-written as

2-13
 Pd    ij Dij dV (2-54)
V

where Pd is the deformation power (or the stress power). The balance of mechanical energy
states that the power of the external forces equals the sum of the deformation power and the
power of the inertia forces, i.e., if the deformable body is subjected to external forces, it is
generally partly deformed and partly accelerated. If the process is quasi-static, the power of
the inertia forces is negligible.
The kinetic energy of the body is defined by
K   12  v  vdV (2-55)
V

Using Equation (2-8), we readily find that the material time derivative of K is
K    a  vdV (2-56)
V

which shows that the power of the inertia forces equals the material time derivative of the
kinetic energy of the body.

2.8 Principle of virtual power

The principle of virtual power can be derived from the equations of motion using the same
procedure that was used to establish the balance of mechanical energy. But instead of
multiplying with the real velocity vector v , we multiply both sides of the equations of
motion by a virtual velocity vector  v . The virtual velocity field is assumed to fulfil the
kinematic boundary condition
 vi  0 on Su (2-57)

where Su is the part of the surface S with prescribed velocity. The boundary has two parts:
Su where the velocity is prescribed and St where the traction is prescribed, so that
S  Su  St and Su  St   . Apart from this, the virtual quantities are arbitrary. The result
is
 ij
 x
V i
 v j dV   b j v j dV    a j v j dV
V V
(2-58)

Using once more the same procedure as in Chapter 4.3 of Part I, we obtain the principle of
virtual power as
 v j

V
ij
xi
dV    a j v j dV   b j v j dV   t j v j dV
V V S
(2-59)

The similarity with Equation (2-53) is obvious. Introducing the virtual rate-of-deformation
tensor  D as

1   vi  v j 
 Dij     (2-60)
2  x j xi 

2-14
we obtain (applying again the symmetry of the Cauchy stress tensor) the principle of virtual
power in its final form

   D dV    a  v dV   b  v dV   t  v dS
V
ij ij
V
i i
V
i i
S
i i (2-61)

Using the same procedure as in Chapter 4.4 of Part I, the dynamic equilibrium
equations (or the equations of motion) and the traction boundary conditions may be obtained
from the principle of virtual power as
 ij
 bi   ai in V , ti   ij n j on St (2-62)
x j

where ti is the prescribed traction on St .

2.9 References
[1] F. Irgens. Continuum Mechanics. Springer, 2008.
[2] J. Lubliner. Plasticity Theory. Macmillan, 1990
[3] T. Belytschko, W. K. Liu, B. Moran. Nonlinear Finite Elements for Continua and
Structures. Wiley, 2000.
[4] L.E. Malvern. Introduction to the mechanics of a continuous medium. Prentice-Hall,
1969.
[5] K. Hellan. Plastisitetsteori. Tapir, 1972.
[6] C. Lanczos. The variational principles of mechanics. Fourth edition. Dover
Publications, 1986.

2-15
2-16
3 Plasticity and viscoplasticity
In this chapter, we will describe the theories of plasticity and viscoplasticity for large
deformations, and save the formulation of the elastic behaviour for Chapter 6. We refer the
reader to Chapters 7 and 8 of Part I and references therein for more details on the theory of
plasticity and viscoplasticity. We suggest the textbooks by Belytschko et al. [1], Lubliner
[2], and Khan and Huang [3] for further reading.

3.1 General formulation

Analogously to the additive decomposition of the strain rate tensor in the theory of
plasticity for infinitesimal deformations, the rate-of-deformation tensor D decomposes
additively into elastic and plastic parts
Dij  Dije  Dijp (3-1)

where De and D p are the elastic and plastic rate-of-deformation tensors, respectively. We
note that due to large rotations the time integral  Ddt generally has no physical meaning
under finite deformations.
As derived in Chapter 2, the deformation power (or stress power) is defined by
Pd   d dV    ij Dij dV (3-2)
V V

Using Equation (3-1), the stress power per unit volume d   ij Dij is decomposed into
elastic and plastic parts as
d   ij Dij   ij Dije   ij Dijp (3-3)
The dissipation inequality is then expressed as
Dp   ij Dijp  0 (3-4)

To fulfil the dissipation inequality, the existence of a plastic potential g  g  σ  is usually

assumed, so that
g
Dijp   (3-5)
 ij

where   0 is the plastic parameter (or plastic multiplier). It follows from Equations (3-4)
and (3-5) that the function g  g  σ  must be chosen so that

g
D p   ij 0 (3-6)
 ij
How to a priori fulfil the dissipation inequality, as expressed in Equation (3-6), has been
discussed in some detail in Chapter 7 of Part I. Almost invariably g is assumed to be a non-
negative and convex function containing the origin, and quite often it is a positive
homogeneous function of order one.

3-1
 In the case of plasticity, the loading/unloading conditions define the plastic parameter
 , and in Kuhn-Tucker form they are formulated by
f  0,   0,  f  0 (3-7)
where f is the yield function
f  f  σ, R     σ    0  R  (3-8)
Based on Equation (3-7) we may deduce the consistency condition: the stress has to stay on
the yield surface during plastic loading, viz.
f 0 (3-9)
The function   0 is assumed to be convex and positive homogeneous of order one,
 eq    σ  is the equivalent stress,  0 is the initial yield stress, R  R  p  is the isotropic
hardening variable, and p is the equivalent plastic strain, which is defined by the evolution
equation
Dp
p (3-10)
 eq
With this definition, p  0 by Equation (3-4) and further the plastic dissipation (or plastic
power) reads as Dp   eq p . Hence, the equivalent stress  eq and the equivalent plastic
strain rate p are power conjugate quantities. The equivalent plastic strain is then found by
time integration
t t
Dp
p   pdt   dt (3-11)
0 0
 eq
Only isotropic hardening has been allowed for here. Kinematic hardening may be
included in a similar way as described in Chapter 7 of Part I, but care must be exercised
when defining the evolution rule for the backstress. Any of the yield functions and isotropic
hardening rules discussed in Part I may be used, but, as discussed below, care should be
taken for anisotropic yield functions. Also, the experimental identification of the material
parameters is usually more involved, since the work hardening response for large plastic
strain is required.
For viscoplastic materials, a constitutive relation explicitly defines the plastic
parameter  in the form
 0 for f 0
 (3-12)
  σ, R  for f 0

where     σ, R  is a state function that defines the rate sensitivity of the material. We
may apply the same functions  as exemplified in Chapter 8 of Part I.

3-2
3.2 Rate-independent J2 flow theory
In the J2 flow theory, the yield function is defined by the von Mises equivalent stress
and is only a function of the second principal invariant of the deviatoric stresses J 2  12  ij ij .
The von Mises equivalent stress is
 eq    σ   3
2  ij ij (3-13)

Here,  ij   ij  13  kk ij is the deviatoric part of the Cauchy stress tensor, also called the
deviatoric stress tensor. The associated flow rule is further assumed and the plastic rate-of-
deformation tensor is obtained as
 3  ij
Dijp    (3-14)
 ij 2  eq
which implies that the plastic dissipation is non-negative and given by
3  ij ij
D p   ij Dijp     eq  (3-15)
2  eq
Comparison with Equation (3-10) shows that in J2 flow theory, we have
p  Dp /  eq   (3-16)
Using the established equations, we may also derive another important expression for J2
flow theory. Multiplying Equation (3-14) with itself and then inserting Equation (3-13), we
get

 32   ij ij 3 2
2

D D 
p p 2
 2 (3-17)
ij ij
 eq2
Use of Equation (3-16), leads to

p 2
3 Dijp Dijp (3-18)

The equivalent plastic strain is finally obtained by integration

t
p 2
3 Dijp Dijp dt (3-19)
0

Note that this direct relation between the equivalent plastic strain and the plastic rate-of-
deformation tensor is only valid for J2 flow theory.

3-3
Example 3-1: Simple motions of rigid-plastic material
Let us consider a rigid-plastic material subjected to simple motions, either uniform
extension or simple shear, and find the expression for the equivalent plastic strain rate
according to J2 flow theory. These motions are described in Chapter 2.
In uniform extension, the rate-of-deformation tensor is diagonal with non-zero
components (see Chapter 2)
1  
D11p  1p  , D22p   2p  2 , D33p   3p  3
1 2 3
Using Equation (3-18), we find

p 2   p  2    p  2    p  2 
3  1 2 3 

Applying plastic incompressibility,  vp   kkp  0 , we then arrive at the following three

special cases:
 Uniaxial tension: 1p  0,  2p   3p   12 1p  p  1p  p  1p
 Plane-strain tension: 1p  0,  2p  0,  3p  1p  p  23 1p  p  23 1p
 Equi-biaxial tension: 1p   2p  0,  3p  21p  p  21p  p  21p

In simple shear, the non-zero components of the rate-of-deformation tensor are (see
Chapter 2)

D12p  D21p 
2
Inserting these values into Equation (3-18), leads to
p 1
3
  p 1
3
  1
3
tan  for   0

3.3 The modified Johnson-Cook model

As an example of a model of thermo-viscoplasticity, we consider the modified
Johnson-Cook (MJC) model [4]. This model is valid for isotropic materials and is much
used in simulations of structural impact.
The yield function is defined as
f    σ    0 T   R  p, T   (3-20)

where T is the temperature. The equivalent stress  eq , the initial yield stress  0 and the
isotropic hardening variable R are given by
 eq    σ   3
2  ij ij ,  0 T   A 1  T *m  , R  p, T   Bp n 1  T *m  (3-21)

where A , B and n are constants determining the work hardening, and m is an exponent
defining the thermal softening. The homologous temperature T * is defined as

3-4
 T  Tr
T*  (3-22)
Tm  Tr
where Tr and Tm are the reference temperature and the melting temperature of the material,
respectively. It is assumed here that T  Tr , Tm  and as a result T *   0,1 . It is useful to
introduce the current yield stress as
 y  p, T    0 T   R  p, T  (3-23)
The plastic flow is defined by the associated flow rule as
f 3  ij
Dijp    (3-24)
 ij 2  eq
where

 σ 
1/ C
 
  p  p0   1 (3-25)
  0  T   R  p, T  

The positive constants C and p0 define the rate-sensitivity of the material, and
a  max  a, 0  for any scalar a . The equivalent plastic strain is defined by p    dt . By
re-arranging the equation for the plastic parameter, the constitutive relation for thermo-
viscoplasticity reads

 ij ij   A  Bp n 1  p*  1  T *m  for

C
 eq  3
2 f 0 (3-26)

where p*  p / p0 is a dimensionless plastic strain rate. Evidently, the flow stress (or
equivalent stress) has three multiplicative contributions: strain hardening, strain-rate
hardening and thermal softening. To identify the constants of the MJC model, we need data
from tensile tests (or compression tests) over a range of strain rates and temperatures.
An alternative form of the constitutive relation is
 eq  3
2  ij ij   y  p, T    v  p, p, T  for f 0 (3-27)

where the viscous stress is given by

 v  p, p, T    y  p, T  1  p*   1
C
(3-28)
 
The viscous stress is non-negative and increases with increasing plastic strain rate. For
quasi-static loading conditions p*  0 and the viscous stress  v vanishes, implying that
the equivalent stress  eq approaches the current yield stress  y .
The original Johnson-Cook (JC) model [5] is obtained by the following substitution in
the constitutive relation:

1  p * C
 1  C ln p  *

3-5
The two versions of the model gives similar results, but in the original model the strain-rate
hardening term approaches  when p  0 . This is somewhat cumbersome in numerical
implementations, since the rate sensitivity has to be switched off for p  p0 .
In structural impact, the deformation process may often be assumed to take place under
adiabatic conditions. The plastic dissipation then leads to heating of the body and the
temperature rate is estimated as (see Chapter 10 for the derivation of this equation)
 ij Dijp
T  TQ (3-29)
 c
where  is the mass density, c the specific heat of the material and TQ is the Taylor-
Quinney coefficient. The specific heat capacity c is the amount of heat required per unit
mass to increase the temperature by one degree. The Taylor-Quinney coefficient TQ gives
the fraction of the plastic deformation power dissipated as heat (see Chapter 10 for further
details). Note that in Equation (3-4) we have assumed that TQ  1 , while in reality a part
of the plastic deformation power is stored and TQ is less than unity. Often analysts use
TQ  0.9 in numerical simulation of adiabatic processes.
In adiabatic processes, the increase of temperature may eventually lead to strain
softening, i.e., the current yield stress  y decreases with increasing plastic strain. From
Equation (3-26) it is evident that an increase in temperature will soften the material and at
some strain the thermal softening may exceed the work hardening. At this point, strain
softening occurs and the material becomes unstable. The general condition for incipient
strain softening is
 y  y
d y  dp  dT  0 (3-30)
p T
After some straightforward calculations, using the expressions for the current yield stress
and the temperature increase by adiabatic heating, Equation (3-30) leads to
1 R TQ  y
 0 (3-31)
 eq p c  T
It is evident that there is a competition between the work hardening and the thermal
softening, and strain softening commences when these to contributions balance exactly.
Note also that increasing the equivalent stress reduces the stability of the material because
it leads to a faster increase of the temperature because of the adiabatic heating.

3-6
3.4 Proportional loading
Proportional loading implies that the ratios between the components of the stress tensor
are constant during the loading process. We can express this as
 ij  k ij0 (3-32)

where  ij0 is a reference stress state and k  k  t  is a non-negative, monotonically

increasing number. The deviatoric stress tensor  ij and the equivalent stress  eq    σ 
(assuming  to be a positively homogeneous function of order one) are
 ij  k ij0 ,  eq  k eq0 (3-33)

 
where  ij0   ij0  13  kk0  ij and  eq0   σ 0 .

Assuming J2 flow theory, the associated flow rule for proportional loading becomes
3  ij
0

d  ijp  Dijp dt  dp (3-34)

2  eq0
where dp  pdt is the equivalent plastic strain increment. An incremental formulation has
been introduced instead of the rate form for convenience. Integration gives
3  ij
0

 ijp   Dijp dt  p (3-35)

2  eq0
and we find that

p 2
3  ijp ijp (3-36)

The tensor  ijp   Dijp dt defined above represents a physically meaningful plastic strain
measure, but only for proportional loading conditions. The reason for this is that the
principal directions of Dijp are equal to those of  ij0 and they are accordingly fixed during
the loading programme. According to Chapter 2, the principal values of  ijp are then the
logarithmic strains  ip  ln i , i  1, 2,3 , since the elastic deformations are assumed
negligible. Under general loading conditions the principal directions of Dijp will vary and

D
p
ij dt loses its physical meaning.

If rigid-plastic J2 flow theory is assumed, the uniform extension (or compression) and
simple shear motions described in Chapter 2 represent cases of proportional loading. In
these cases, the plastic rate-of-deformation tensor is constant and it has fixed principal
directions. According to Equation (3-34) the stress tensor has the same principal directions,
and it follows that the ratios between the components of the stress tensor are constant during
the loading programme—and thus the loading is proportional.

3-7
3.5 Plastic limit theorems
In the following, we will derive the limit theorems for rigid-perfect plasticity. We use
these theorems to calculate approximate failure loads of structures [2]. We consider a
structure loaded proportionally by traction forces ti and body forces bi . The structure is
made of a rigid-perfectly plastic material, which obeys the principle of maximum plastic
dissipation. We define a limit state (or collapse of the structure) as a state in which the
plastic deformations occur under constant external loads. Let the loads ti and bi define such
a limit state of the structure.
Assuming quasi-static loading conditions, the principle of virtual work reads (see
Chapter 4 of Part I)

  
V
ij ij dV   bi ui dV   ti ui dS
V S
(3-37)

where the stress field  ij is in static equilibrium with the external forces ti and bi . The
virtual strain field  ij and the virtual displacement field  ui satisfy kinematic
compatibility

1   ui  u j 
 ij     xi V ,  ui  0 xi  Su (3-38)
 2  x j xi 
where Su is the part of the surface S for which displacements are prescribed. In the
derivation of the plastic limit theorems, we will use kinematically admissible fields instead
of virtual fields
 ij  Dij t xi V ,  ui  vi t xi  S (3-39)

where vi and Dij are the actual velocity and rate-of-deformation fields, respectively, and
 t is a small time increment. Note that this entails that  ui is not equal to zero on Su , but
given by the kinematical boundary conditions, and the work done by the reaction forces on
the boundary must be included [2]. The principle of virtual work now takes the form


V
ij Dij dV   bi vi dV   ti vi dS
V S
(3-40)

Since rigid plasticity holds, we will not distinguish between total and plastic rates-of-
deformation. We obtain the upper and lower bound theorems as follows:
 Assume that a stress field  ij* has been established that is in static equilibrium with
 
the loads ti*   s ti and bi*   s bi , and fulfils the yield condition f  ij*  0 , where  s
is called the “static” load factor. Then by the principle of virtual work, we have

 
 Dij dV   s   bi vi dV   ti vi dS    s   ij Dij dV
*
ij (3-41)
V V S  V

3-8
 But, owing to the principle of maximum plastic dissipation (recall that Dij  Dijp for
 
the rigid-perfectly plastic material),  ij   ij* Dij  0 , so that  s  1 . Thus, the loads
ti*   s ti and bi*   s bi represent lower bounds to the actual loads.

 Assume that a velocity field vi* and strain-rate field Dij* have been determined so
that kinematic boundary conditions and kinematic compatibility are satisfied. The
corresponding loads ti*  k ti and bi*  k bi , where  k is called the “kinematic” load
factor, are defined by

 Dij* dV   bi*vi*dV   ti*vi*dS

*
ij (3-42)
V V S

 
The stress  ij* fulfils the yield condition, f  ij*  0 , where plastic deformation

 Dij* dV  0 . Using
*
takes place, and the plastic dissipation is assumed positive: ij
V
again the principle of virtual work, we find
 
 Dij* dV   k   bi vi*dV   ti vi*dS    k   ij Dij* dV
*
ij (3-43)
V V S  V

But according to the principle of maximum dissipation,  ij*   ij Dij*  0 , since 

f  ij   0 , so that k  1 . Thus, the loads ti*  k ti and bi*  k bi are upper bounds
to the actual loads.

Example 3-2: Tensile specimen with central hole

As an example on the use of the limit theorems, we will consider a tensile specimen
with a central hole. The specimen has width w , thickness h and hole diameter d , see
Figure 3-1 (a). The von Mises yield criterion is assumed with yield strength equal to  0 .
The task is to find an estimate of the limit load. Figure 3-1 (b) shows a sketch of an
admissible stress field for a lower bound solution, which gives
Fmin   0 h  w  d  (3-44)
Figure 3-1 (c) shows an admissible velocity field for an upper bound solution, where
plasticity only takes place along the slip line inclined at 45. The velocity of each part of the
specimen in the direction along the slip line is v* ; i.e., the relative velocity between the two
rigid parts is 2v* . The internal and external plastic power should be equal, and thus
0 2v*
h 2  w  d  2v  Fmax
*
(3-45)
3 2
where the shear stress at yielding along the slip line equals  0 / 3 according to the von
Mises criterion. The upper bound solution is
2 0
Fmax  hw  d  (3-46)
3

3-9
The real solution is then in the range:
2 0
 0h  w  d   F  hw  d  (3-47)
3
Note that the upper bound solution is about 16 % higher than the lower bound solution.
■

F F

(a)

Fmin 0 0 Fmin

0 0

(b)
45

Fmax v v Fmax

v v

(c)

Figure 3-1 Tensile specimen with central hole (a), admissible stress field for lower limit
solution (b) and admissible velocity field for upper bound solution (c).

3-10
3.6 References
[1] T. Belytschko, W. K. Liu, B. Moran. Nonlinear Finite Elements for Continua and
Structures. Wiley, 2000.
[2] J. Lubliner. Plasticity Theory. Macmillan, 1990
[3] A.S. Khan, S. Huang. Continuum Theory of Plasticity. Wiley-Interscience Publication
(1995).
[4] T. Børvik, O.S. Hopperstad, T. Berstad, M. Langseth. A computational model of
viscoplasticity and ductile damage for impact and penetration. Eur. J. Mech. –
A/Solids 20 (2001) 685–712.
[5] G.R. Johnson, W.H. Cook. A constitutive model and data for metals subjected to large
strains, high strain rates and high. Proceedings of the 7th International Symposium on
Ballistics (1983): 541–547.

3-11
3-12
4 Ductile damage and failure criteria
In large-scale simulations of structures, it is often convenient to use uncoupled damage
models. The damage variable then evolves as a function of stress state and plastic strain, but
has no coupling back to the elastic-plastic or elastic-viscoplastic response. Failure occurs
when the damage variable reaches a critical value, which may be set to unity without loss
of generality. The main advantage of this approach is a simpler identification of the damage
parameters. The Gurson model described in Chapter 5 is an example of a coupled damage
model, where the evolution of damage is interacting with the plastic behaviour. Then, if the
damage evolution exceeds the work-hardening, strain softening occurs, which is typically a
necessary condition for plastic instability and strain localization in three-dimensional stress
states.
The main mechanisms governing ductile fracture in metals and alloys are nucleation,
growth and coalescence of microscopic voids at various length scales [1]. The voids
nucleate at particles when the stress on the particle is sufficient to induce either particle
cracking or particle-matrix decohesion, see illustration in Figure 4-1(a). In addition,
microscopic voids are to some degree present in the material as a result of the manufacturing
process. While the influence of the stress triaxiality on ductile failure is well established,
studies showing the role played by the deviatoric stress state, as represented by the Lode
parameter, are more recent [2]. The influence of the Lode parameter is particularly important
at low stress triaxialities and it is typically found that the ductility is lower in shear-
dominated stress states than in axisymmetric stress states. Failure by void coalescence
typically occurs by localized plastic deformation and necking of the ligament between
adjacent voids. If two classes of particles of different size and spacing are present in the
material, void-sheet formation may take place and lead to shear fracture, see Figure 4-1(b)
[3].

(a) (b)
Figure 4-1 Ductile damage and fracture by a) nucleation, growth and coalescence of voids
and b) shear fracture mechanisms (or void-sheeting) (from [3]).

4-1
 In this chapter, we start by defining the stress invariants used to describe damage
evolution and failure of ductile isotropic materials. We proceed by addressing the growth
of an isolated cylindrical void in a rigid-plastic solid in order to demonstrate the strong
influence of the stress state on the void growth. Then some frequently used failure criteria
are presented, such as the Johnson-Cook, Cockcroft-Latham and modified Mohr-Coulomb
criteria. Finally, a simple way of coupling damage and plasticity is demonstrated, which
makes it possible to simulate damage softening.

4.1 Stress invariants

We will here assume isotropic materials and define the stress state in terms of stress
invariants (see also Chapters 3 and 7 of Part I). The first invariant we will make use of is
the von Mises stress, which is defined as

 VM  3J 2  3  I   H 2   II   H 2   III   H 2  (4-1)

2  
where  I   II   III are the ordered principal stresses,  H  13  I   II   III  is the
hydrostatic stress, and J 2 is the second principal invariant of σ :

J 2  12  ij ij  12  I   H    II   H    III   H  

2 2 2
(4-2)
 
The second invariant employed to describe the stress state is the stress triaxiality which is
defined by
I 
*   H (4-3)
3 3J 2  VM

where the first principal invariant of the stress tensor is defined as I  I kk  3 H . The third
invariant adopted here is the Lode angle  L which is defined as (cf. Chapter 7 of Part I)
J3 27 J 3 
cos 3 L   , 0  L  (4-4)
2  J 2 / 3
3 2 VM
3
3

where J 3 is the third principal invariant of σ :

J3  det  σ   I   H  II   H  III   H  (4-5)

In Chapter 7 of Part I, the ordered deviatoric principal stresses  I   II   III

 were
expressed in terms of the deviatoric stress invariants as
 I  23 cos  L   VM
 II  23 cos  23   L   VM (4-6)
  23 cos  23   L   VM
 III

4-2
 Major Intermediate Minor
0,8

0,6

0,4

0,2

0
0 10 20 30 40 50 60
-0,2

-0,4

-0,6

-0,8

Figure 4-2 Variation of the major, intermediate and minor principal deviatoric stresses
against the Lode angle.

Figure 4-2 illustrates the variation of the deviatoric principal stresses with Lode angle. It is
 for  L  0 ,  I   III
observed that  II   III  and  II  0 for  L  30 , and  I   II for
 L  60 . As will be discussed in some detail below, these three Lode angles represent in
turn generalized states of tension, shear and compression. Using the relation  i   i   H
with i  I , II , III , the ordered principal stresses are obtained as functions of the invariants
of the stress tensor by
 I   *  23 cos  L    VM
 II   *  23 cos  23   L    VM (4-7)
 III   *  23 cos  23   L   VM
where Equation (4-3) was used to express the hydrostatic stress in terms of the stress
invariants.
Experiments show that ductile damage depends strongly on the stress triaxiality, since
this parameter is important for the void growth, but also to various degrees on the deviatoric
stress state through the Lode angle. The trend seems to be that the strain to fracture is lower
in shear dominated than in axisymmetric stress states. Any axisymmetric stress state fulfils
one of the conditions
 I   II   III or  I   II   III (4-8)
which represent states of generalized tension and compression, respectively. It is
straightforward to show J 3   272  I   III    272  VM
3 3
for these cases, respectively, and
thus cos3 L  1 for generalized tension and cos3 L  1 for generalized compression.

4-3
Two important cases are uniaxial tension with cos 3 L  1 and  *  13 , and uniaxial
compression with cos3 L  1 and  *   13 . Any stress state defined by [4]
 I     H ,  II   H ,  III     H (4-9)
where   0 is a pure shear stress, has cos3 L  0 . This is readily seen from Equations (4-4)
and (4-5). Such stress states will be denoted generalized shear. An important case is pure
shear with cos3 L  0 and  *  0 , which implies that  I   III and  II  0 .

Evidently, the quantity cos3 L is useful for characterizing the deviatoric stress state,
since cos3 L  1 for generalized tension, cos3 L  1 for generalized compression, and
cos3 L  0 for generalized shear (i.e., all stress states being the sum of a pure shear stress
and the hydrostatic stress). The variation of cos3 L as a function of the Lode angle  L is
shown in Figure 4-3. In terms of the Lode angle we then have the following generalized
stress states:
 Generalized tension (GT):  L  0  cos3 L  1

 Generalized shear (GS): L   cos 3 L  0
6

 Generalized compression (GC): L   cos 3 L  1
3
The generalized stress states are depicted in the  -plane in Haigh-Westergaard space in
Figure 4-4 (see Chapter 7 of Part I for further details).

0,8

0,6

0,4

0,2

0
0 10 20 30 40 50 60
-0,2

-0,4

-0,6

-0,8

-1

Figure 4-3 Variation of cos3 L as a function of the Lode angle  L .

4-4
Figure 4-4: Identification of generalized tension (GT), generalized shear (GS) and
generalized compression (GC) in the  -plane in the Haigh-Westergaard space.

4.2 Growth of an isolated cylindrical void

Since the physical mechanisms governing ductile fracture involve void growth under
plastic deformation, it is interesting to study the growth of an isolated void in a rigid-
perfectly plastic J2 material. To this end, consider a rigid-perfectly plastic solid obeying the
J2 flow theory with an infinitely long circular cylindrical hole of radius a along the x3
direction, see Figure 4-5. A uniform rate-of-deformation D33p is imposed on the material,
which we will denote  for simplicity. The rate of hole-growth a as a function of the
current radius a and the remotely applied radial stress state  rr is sought for. Generalized
t
plane strain conditions are assumed: the principal logarithmic strain    D33p dt is non-zero
0

but constant along the x3 axis. Polar coordinates are adopted in the x1 x2 plane, i.e.,
x1  r cos and x2  r sin  . The derivation below is taken from Needleman et al. [5], but
was first derived by McClintock [6].
The equilibrium equation in the current configuration is obtained by considering an
infinitesimal element of the solid in polar coordinates. The resulting equilibrium equation
reads as (see Appendix A for details)
d rr  rr   
 0 (4-10)
dr r

4-5
Figure 4-5: Rigid-perfectly plastic solid with an infinitely long circular cylindrical hole of
radius a along the x3 direction.

Owing to the cylindrical symmetry of the problem, all field quantities are functions of the
radius r only, and the shear stress  r is zero. The boundary conditions on the radial stress
are given by
 rr  a   0,  rr     rr (4-11)
Hence, the radial stress is zero on the surface of the cylindrical void and takes the remote
value as the radius r goes towards infinity.
Denoting the radial velocity vr , the non-zero in-plane components of the plastic rate-
of-deformation tensor are given by [5] (see Appendix A)
dvr v
Drrp  , Dp  r (4-12)
dr r
Plastic incompressibility gives
Drrp  Dp  D33p  0 (4-13)

where D33p   was applied. Using the von Mises yield condition and the associated flow
rule, we have
3 p 3 p
Drr   rr   H  , D     H  (4-14)
2 y 2 y

where  y is the yield stress,  H  13  rr     33  is the hydrostatic stress and p is the
equivalent plastic strain rate for the von Mises material

4-6
 p 2
3  (D p 2
rr )  ( Dp )2  ( D33p )2  (4-15)

A relation between the stress and strain-rate components is obtained from Equation
(4-14), namely
 rr    2  y Drr  D
 (4-16)
r 3 r p
Combining Equations (4-10) and (4-16), we get by integration
 rr 2  D  Drr dr
 y 3 a
 (4-17)
p r
Use of Equations (4-12) and (4-13) leads to the differential equation
dvr vr
   0 (4-18)
dr r
with boundary condition
vr  a   a (4-19)
The solution is given on the form
C 
vr  r    r (4-20)
r 2
where the two terms are the homogeneous and particular solutions, respectively, and C is
a constant. Applying the boundary condition, we get the radial velocity field as
a  1 
vr  r   a 2     r (4-21)
a 2r 2
The radial and tangential components of the rate-of-deformation tensor and the equivalent
plastic strain rate are then given by
a2  a    a2  a   
D   2     , D  2    
p
rr
p
(4-22)
r a 2 2 r a 2 2
and

4  a   a4
2

p      4
2
(4-23)
3a 2 r
By inserting Equations (4-22) and (4-23) into Equation (4-17) and subsequently
introducing the substitution
a2 2  a 1 dr dx
x  2 with      (4-24)
r 3   a 2  r 2x
we get after changing the integration limits

 rr 1 dx
y
 
3 0 x2  1
(4-25)

4-7
Solving the definite integral and back-substituting the expression for  lead to the
evolution equation for the void radius as

a  3  3 rr  1 
  sinh   
   2 
(4-26)
a  2  y  
The exact solution to this problem, derived by McClintock [6], demonstrates that the void
growth rate increases exponentially with positive radial stress  rr . Integration of Equation
(4-26) for fixed remote radial stress and initial conditions a  a0 and   0 , leads to


 3  3 rr  1   
a    a0 exp  sinh       (4-27)

 2
   y  2   
t
where we recall that    D33p dt is the principal logarithmic strain along the x3 axis.
0

The evolution of the ratio a / a0 for different values of the ratio  rr /  y is shown in
Figure 4-6. The strong effect of the stress ratio  rr /  y , which is related to the stress
triaxiality, is evident. If  rr /  y is less than about ⅓, the ratio a / a0 will decrease toward
zero. Even if the shape of the void is simplified and not very realistic, it demonstrates the
very important influence of the stress state on the void growth.

4.3 Damage evolution rules

There are several ways of defining the uncoupled damage evolution rule. Three
possible routes will be discussed here. In the first, a failure surface is assumed to exist,
which defines the failure strain p f for stress paths with constant values of stress triaxiality
 * and Lode angle  L :

p f  p f  * ,  L  (4-28)

The damage variable  is then usually defined as

p
dp
 (4-29)
0 p f  * ,  L 

where fracture is assumed to occur when   1 . The influence of non-proportional stress

paths is assumed to be described by the damage accumulation rule defined by Equation
(4-29).

4-8
 10

6
a/a0

0
0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1
Logarithmic strain

Figure 4-6: Evolution of void ratio a / a0 with logarithmic strain  as a function of stress
ratio  rr /  y .

An important example here is the Johnson-Cook (JC) fracture model [7], where the
influence of the Lode angle  L is neglected but the effects of strain rate p and temperature
T are included. The fracture locus of the JC fracture model reads as
p f  * , p, T    D1  D2 exp  D3 *  1  D4 ln p* 1  D5T *  (4-30)

where D1 , D2 ,..., D5 are model constants; p*  p / p0 is the dimensionless strain rate with
p0 being a user-defined reference strain rate; T *  T  Tr  / Tm  Tr  is the homologous
temperature, where Tr is the ambient (or room) temperature and Tm is the melting
temperature of the material. The JC fracture model is based on the assumptions that the
failure strain p f decreases exponentially with increasing stress triaxiality (with D3  0 )
and linearly with the increasing strain rate on a logarithmic scale (with D4  0 ), while it
increases linearly with increasing temperature (with D5  0 ). Based on experimental
evidence and the physical mechanisms responsible for ductile failure, it is expected that the
ductility decreases with increasing stress triaxiality and strain rate and increases with
increasing temperature. The strain rate tends to have the opposite effect of temperature on
the material behaviour.

4-9
 A simple way of introducing the effect of the Lode angle  L in the Johnson-Cook
fracture model is proposed here as
p f  * , L , p, T    D1  D2 exp  D3 *  1  D4 ln p* 1  D5T *  1  D6  L 
(4-31)
where the stress-dependent parameter  is defined as

  L   1  cos 2  3 L  , 0   L  (4-32)
3
This definition implies that   0,1 with   0 for all axisymmetric stress states and   1
for all stress states being the sum of a pure shear stress and the hydrostatic stress. Figure 4-7
illustrates the variation of  with the Lode angle  L . The new constant 0  D6  1 may then
be used to reduce the fracture strain for shear-dominated stress states. It should be noted
here that the parameter  was proposed by Nahshon and Hutchinson [4] in their work on
extending the Gurson model to account for shearing of voids, which we will consider in
Chapter 5. A comparable extension of the Johnson-Cook fracture model has been proposed
in [8].

0,9

0,8

0,7

0,6

0,5

0,4

0,3

0,2

0,1

0
0 10 20 30 40 50 60

Figure 4-7 Variation of   1  cos2  3L  as a function of the Lode angle  L .

4-10
 In the second approach, it is assumed that a failure criterion exists in stress space so
that failure, either by strain localization or ductile fracture, occurs when
 σ   0 (4-33)

where for isotropic materials, the function  is expressed as     I , II , III  or

equivalently in terms of the stress invariants as     * , L ,  VM  . If   σ   0 , there is no
failure and the material is intact. If we assume that the material obeys the rate-independent
J2 flow theory,  VM is a unique function of the equivalent plastic strain p via the yield
condition, f   eq   0  R  p   0 with  eq  VM  3
2  ij ij . For fixed values of stress
triaxiality  * and Lode angle  L , it is thus possible to determine either an explicit or
implicit relation for the equivalent strain at failure p f  p f  * , L  , and the damage may
again be defined by Equation (4-29).
An important representative of this class of failure criteria is the modified Mohr-
Coulomb criterion proposed by Bai and Wierzbicki [9]. In the form adopted by Gruben et
al. [10], the Mohr-Coulomb criterion reads as (see also Chen and Han [11])
  I , II , III    I  1    I   III   C  0 (4-34)

where 0    1 and C  0 are two constants. Alternatively, in terms of the invariants, we

may express the failure surface,   * , L ,  VM   0 , as
1
 VM  C  *  23 cos  L   32 1    cos  23   L   0 (4-35)

If we for simplicity assume the power law,  VM  Kp n , the failure strain p f under
proportional loading paths, as defined by fixed values of stress triaxiality  * and Lode
angle  L , reads as
1
 C n
1

p f     *  23 cos  L   32 1    cos  23   L  n (4-36)
K
To extend the flexibility of the fracture criterion, it is customary to fit the exponent n to
fracture data instead of fitting it to the stress-strain curve. The coefficient K may then be
absorbed into C and has no relevance.
In the third approach, the damage evolution is assumed to be driven by plastic straining
(or equivalently plastic dissipation) but amplified by a factor that accounts for stress state,
usually in terms of the stress triaxiality  * and the Lode angle  L . A general damage
evolution rule of this type is given by
p

     σ  dp (4-37)
0

where the function     σ  governs the influence of stress state on the damage growth.
Failure is assumed to occur when   1 . For isotropic materials, the stress state can be
represented either by the principal stresses or a suitable set of stress invariants.

4-11
 The Cockcroft-Latham criterion [12] is a particularly simple criterion of this class
p
1

WC 
0
I dp (4-38)

where  max  ,0 and  I is the major principal stress. The fracture parameter WC is
the only constant, which makes this criterion easy to calibrate from tests. Expressed in terms
of the stress triaxiality  * and Lode angle  L , the Cockcroft-Latham criterion takes the
form
p
1

WC 
0
 *  23 cos  L  VM dp (4-39)

where it was used that  VM is non-negative. A more versatile damage evolution rule of this
type, denoted the extended Cockcroft-Latham (ECL) criterion, was proposed by Gruben et
al. [10] as

1
p
     III 
   I  1     I   VM dp (4-40)
WC 0
 VM   VM 
where  0,1 and   0 are additional parameters governing the influence of the stress
state on the damage evolution. The J2 flow theory is assumed here, so that
 eq  VM  32  ij ij . For other yield functions, it is suggested to substitute  VM by
 eq    σ  in Equation (4-40). The Cockcroft-Latham criterion is obtained by setting
    1, while   0 and   1 gives an integral-based Tresca criterion
p
1

WC  
0
I   III  dp (4-41)

A fracture criterion driven by plastic dissipation only is obtained for   0 , which is often
referred to as the Freudenthal criterion. Note that we tactically assume here that 00  1 . Note
that both the integral-based Tresca criterion and the Freudenthal criterion are independent
of the stress triaxiality. The Freudenthal criterion is also independent of the Lode angle.

4-12
Example 4-1: Cockcroft-Latham criterion for proportional straining
If we assume proportional loading paths (i.e., the principal directions of the stress
tensor remain fixed), the damage evolution rule defined in Equation (4-39) for the
Cockcroft-Latham criterion can be used to establish a fracture surface in the space defined
by stress triaxiality and Lode angle. Fracture is assumed to occur as the damage variable 
equals unity. As an example, consider a rate-independent rigid-plastic material obeying the
J2 flow theory and exhibiting negligible work hardening. With good accuracy we may then
take f   VM   y  0 in the plastic domain, where  y is the yield stress. Then integration
of Equation (4-39) up to failure at   1 leads to
1 WC
p f   *  23 cos  L 
y
where p f is the fracture strain, and  *  ,  and  L  0, 3  are constant owing to the
assumption of proportional stress paths. To illustrate the variation of the fracture strain with
stress state, we take WC   y . The fracture loci for generalized tension, shear and
compression are illustrated in Figure 4-8.
■

Generalized tension Generalized shear Generalized compression

2

1,6
Fracture strain

1,2

0,8

0,4

0
-0,5 0 0,5 1 1,5 2
Stress triaxiality

Figure 4-8: Fracture strain p f versus stress triaxiality   for the generalized tension, shear
and compression according to the Cockcroft-Latham criterion.

4-13
4.4 Damage coupling
At the end of this chapter on uncoupled damage models, we will describe a simple but
physically meaningful way of coupling damage with the plastic behaviour of the material.
It is postulated that the plasticity model remains unchanged, but that the Cauchy stress σ is
substituted with an effective stress σ defined by [13]
 ij
 ij  1     ij   ij  for 0    1 (4-42)
1 
where the damage variable  is zero for the undamaged material and approaches unity for
full loss of integrity, see Figure 4-9. Evidently, plasticity can only be applied to the
undamaged matrix material and not to voids and cracks. Since parts of the load-carrying
area are damaged, the effective stress is higher than the applied stress. The damage variable
is assumed to take values between zero and C  1 , which is the critical damage at which
failure occurs. We need to set C less than unity to avoid singularity in the effective stress.
Except for using the effective stress, the constitutive equations remain unchanged. The
yield criterion and flow rule are defined by [13]
f  f  σ, R     σ    0  R  (4-43)

g  σ 
Dijp   (4-44)
 ij
where   0 and g  0 are positive homogeneous functions of order one in their argument.

Figure 4-9: Reduction of load-carrying area with damage evolution.

4-14
 Let  eq    σ  be the equivalent stress defined in terms of the effective stress tensor.
Then the equivalent plastic strain rate is defined by
 ij Dijp
p (4-45)
 eq

If the associated flow rule is adopted, so that Dijp    σ  /  ij , use of Euler’s theorem
for homogeneous functions leads to
p (4-46)

The plastic parameter  is defined by the loading/unloading conditions in the case of

plasticity
f  σ, R   0,   0,  f  σ, R   0 (4-47)
and a constitutive relation in the case of viscoplasticity
 0 for f  σ, R   0
 (4-48)
  σ, R  for f  σ, R   0

The plastic dissipation per unit current volume for the damaged (visco)plastic material
is given by
g  σ 
D p   ij Dijp  1     ij Dijp  1     ij
 ij (4-49)
 1     g  σ    g  σ   0
where the last equality follows from g being a positive homogeneous function of order one.
Damage coupling may lead to strain softening. To see this, we use that  is a positive
homogeneous function of order one and re-write the yield criterion as
 eq    σ   1    0  R  (4-50)

It is evident that the equivalent stress (defined in terms of the stress tensor σ , not the
effective one) at yielding decreases with increasing damage, and when the damage increase
exceeds the work-hardening, strain softening will take place.
The implementation of the ductile damage model described above is remarkably simple.
It turns out that if the equations are expressed in terms of the effective stress σ , the damage
elastic-(visco)plastic model is similar to an ordinary elastic-(visco)plastic model, and the
algorithms developed in Chapters 7 and 8 of Part I can be used almost without any
modifications. There are only three differences: in the beginning, we compute the effective
stress σn  σn / 1  n  at time t n , then we integrate the damage variable according to its
evolution rule, and at the end, we compute the stress tensor σn1  1  n1  σn1 at time tn 1 .
All other computations are performed with σ instead of σ in the constitutive relations.

4-15
Example 4-2: Lemaitre’s damage evolution rule*
Lemaitre [14] proposed to let damage be driven by the plastic strain and amplified by
the strain energy density release rate, which is defined by
Y  12  ijCijkl
1
 kl
1
where Cijkl is the inverse elasticity tensor (see Chapter 5 of Part I). The strain energy density
release rate measures the elastic energy stored in the damaged material which is available
for amplifying the damage evolution. For isotropic materials, Y can be expressed as [14]
 VM
2

1    3 1  2    
2
Y Rv , Rv  2
3
2E
where E and  are the elastic constants and VM  VM / 1    is the von Mises stress in
terms of the effective stress tensor σ . The damage evolution equation then takes the form

 0 for p  p0

Y / S  p for p  p0
s

where S and s are material constants and p0 is a damage threshold. It should be noted that
the way Rv depends on   makes the damage evolution invariant to the sign of the
hydrostatic stress, which is contrary to what is usually seen in experiments and predicted by
micromechanical models for ductile damage evolution.
Let us consider an isotropic material in the plastic region ( f  0 ) obeying the
constitutive relation
 VM  1     0  Kp n 

where  0 is the initial yield stress and the strain hardening is given by R  p   Kpn . In the
actual case, the damage evolution takes the form

  Kp n   2
2

1    3 1  2     
2
Y
0

2E 
3 
The damage evolution is then obtained as

  Kp n   2
2s
p s
 2
  p,  
  1    3 1  2     dp
0

 2ES   3 
s
p0

The material parameters are given the following values:  0  400 MPa ,
K  400 MPa , n  0.3 , E  210000 MPa ,   0.3 , S  4 MPa , s  2 and p0  0 . We
consider strain paths with constant stress triaxiality in the range from zero to unity. As
already mentioned, the damage evolution rule in this example is not realistic for
compressive stress states (  *  0 ). The evolution of the equivalent stress  eq and the
damage variable  with the equivalent plastic strain p is presented in Figure 4-10 and
Figure 4-11, respectively. For materials with damage there is clearly a competition between
work-hardening and damage softening, which leads to negative slope of the equivalent

4-16
stress-strain curve for large strain and high stress triaxiality. The damage variable evolves
slowly at the start and then its evolution accelerates at larger plastic strains, and the higher
the stress triaxiality, the faster the damage evolves.
■

Figure 4-10: Equivalent stress  eq versus equivalent plastic strain p at different constant
stress triaxialities   for an isotropic elastic-plastic material with ductile damage.

Figure 4-11: Evolution of the damage variable  with equivalent plastic strain p at
different constant stress triaxialities   for an isotropic elastic-plastic material with ductile
damage.

4-17
4.5 References
[1] T.L. Anderson. Fracture mechanics. Third edition. Taylor & Francis, 2005.
[2] A.A. Benzerga, J.-B. Leblond. Ductile fracture by void growth to coalescence.
Advances in Applied Mechanics 44 (2010) 169–305.
[3] D. Teirlinck, F. Zok, J.D. Embury, M.F. Ashby MF. Fracture mechanism maps in
stress space. Acta Metallurgica 36 (1988) 1213–28.
[4] K. Nahshon, J.W. Hutchinson. Modification of the Gurson Model for shear failure.
European Journal of Mechanics A/Solids 27 (2008) 1–17.
[5] A. Needleman, V. Tvergaard, J.W. Hutchinson. Void growth in plastic solids. In: A.S.
Argon (ed.), “Topics in Fracture and Fatigue”, pp. 145-178, Springer-Verlag, 1992.
[6] F.A. McClintock. A criterion for ductile fracture by the growth of holes. Journal of
Applied Mechanics 35 (1968) 363–371.
[7] G.R. Johnson, W.H. Cook. Fracture characteristics of three metals subjected to
various strains, strain rates, temperatures and pressures. Engineering Fracture
Mechanics 21 (1985) 31–48.
[8] S. Chocron, B. Erice, C.E. Anderson. A new plasticity and failure model for ballistic
application. International Journal of Impact Engineering 38 (2011) 755–764.
[9] Y. Bai, T. Wierzbicki. Application of extended Mohr-Coulomb criterion to ductile
fracture. International Journal of Fracture 161 (2010) 1–20.
[10] G. Gruben, O.S. Hopperstad, T. Børvik. Evaluation of uncoupled ductile fracture
criteria for the dual-phase steel Docol 600DL. International Journal of Mechanical
Sciences 62 (2012) 133–146.
[11] W.F.Chen, D.J. Han. Plasticity for structural engineers. Springer-Verlag, 1988.
[12] M.G. Cockcroft, D.J. Latham. Ductility and the workability of metals. Journal of the
Institute of Metals 96 (1968) 33–39.
[13] R. de Borst. Chapter 10 Damage, Material Instabilities, and Failure. In: E. Stein, R.
de Borst, T.J.R. Hughes (Editors). Encyclopedia of Computational Mechanics.
Volume 2: Solids and Structures. Pages 335–373. Wiley, 2004.
[14] J. Lemaitre. A Course on Damage Mechanics. Springer-Verlag, 1992.

4-18
5 Porous plasticity
In models of porous plasticity, a macroscopic material element is assumed to consist
of a matrix material and voids. Metals are usually assumed to be pressure-insensitive, based
on the experiments by Bridgman [1], but this is not the case for porous metals.
Ductile fracture is characterized by three stages: nucleation, growth and coalescence of
voids and micro-cracks [2][3]. Voids nucleate at second-phase particles and inclusions, and
grow under large plastic deformations. At some stage, the void volume fraction becomes so
large that neighbouring voids start to coalesce; i.e., voids join together and create macro-
cracks in the material. Void coalescence is a precursor for failure—the total loss of load-
carrying capacity. Figure 5-1 illustrates ductile failure in metallic materials by void
nucleation, growth, and coalescence.
Plastic flow of porous metals is compressible and both volumetric and deviatoric plastic
strains develop during straining. The matrix material work-hardens while the voids nucleate
and grow, and thus there is a competition between work-hardening of the matrix and
softening due to nucleation and growth of voids.

Figure 5-1: Illustration of ductile failure in metallic materials by void nucleation, growth,
and coalescence: (a) inclusions in a ductile matrix, (b) void nucleation, (c) void growth, (d)
strain localization between voids, (e) necking between voids, and (f) void coalescence and
fracture (from Anderson [2]).
5.1 General formulation
In models of porous plasticity, the yield condition is expressed quite generally as [4]
f  σ, ,  M   0 (5-1)

where  now stands for the volume fraction of the voids (or the porosity) and  M is the
flow stress of the matrix material. If  is zero, the material is without any voids, while if
 approaches unity, the density of the material goes to zero. It will be assumed here that
the material is isotropic so that f only depends on the stress tensor through the stress
invariants. In porous plasticity, there is a competition between the hardening of the matrix
material and the softening due to void growth and nucleation. There exists a critical strain
at which the damage softening becomes stronger than the strain hardening, and for larger
strains macroscopic softening occurs.
The flow rule is defined by a plastic potential
g  g  σ, ,  M  (5-2)
and thus
g
Dijp   (5-3)
 ij

Since isotropic material is presupposed, the dependence of g on the stress tensor is only via
the stress invariants. To ensure positive dissipation in the plastic domain, we will presuppose
that the plastic power per unit volume is non-negative, i.e.
g
D p   ij Dijp   ij 0 (5-4)
 ij

If associated plasticity is assumed, which is normally the case, the plastic potential g is
taken equal to the yield function f .
The equivalent plastic strain  M is defined to be conjugate in power to the flow stress
of the matrix material, i.e.,
 ij Dijp
 ij D  1     M  M
p
or  M  (5-5)
ij
1     M
It is noted that  M is non-negative by Equation (5-4). This relation for the plastic power per
unit volume acknowledges the fact that the matrix material amounts to 1    % of the
macroscopic material element, while the rest of the volume is made up by voids.
Assuming rate-independent matrix behaviour, the hardening rule is expressed as
 M   0  R  M  or  M  hM  M (5-6)

where  0 is the yield stress of the matrix material, R is the isotropic hardening variable
and hM  dR / d M is the isotropic hardening modulus. Any of the isotropic hardening rules
given in Chapter 7 of Part I may be used to define the work-hardening rule for the matrix
material.

5-2
 Let V be the volume of a representative volume element (RVE) of the material. The
size of the RVE should be taken sufficiently small to avoid smoothing of high gradients and
large enough to represent an average of the microscopic processes taking place in the
material. The volume of voids in the RVE is V and thus the volume of the matrix material
is VM  V  V . The void volume fraction (or porosity) is defined by
V V
  (5-7)
V V  VM

It is now assumed that the matrix material is incompressible, i.e., VM  0 , which is

reasonable for a metallic material in which the elastic strains are small and the plasticity is
isochoric. Using this assumption, the material time derivative of the porosity is expressed
as
V VV V  V 
    1   (5-8)
V  VM V  VM  2
V  V 

where we used that V  V  VM to obtain the last equality. Neglecting elastic straining and
recalling the plastic incompressibility of the matrix material, we have
V V
 vp  Dkkp   (5-9)
V V
where Equation (2-18) was applied. By combining Equations (5-7) to (5-9), the evolution
equation for the porosity is obtained as
  1     vp (5-10)
Thus, the void growth is a result of the volumetric plastic strain rate, which is governed by
the flow potential g :
g
 vp  Dkkp   (5-11)
 kk
Equation (5-10) represents a simple differential equation which is solved by separation of
variables, viz.
  vp
d
 1     d
p
v (5-12)
0 0

where 0 is the initial porosity. Integration gives the solution

  1  1  0  exp   vp  (5-13)

It is evident that   1 as  vp   . The minimum value of the porosity is obviously zero

and this gives the minimum value of the plastic volumetric strain. Using Equation (5-13)
with   0 , we infer that
 vp  ln 1  0  (5-14)
The loading/unloading conditions in Kuhn-Tucker form and the consistency condition
for rate-independent porous plasticity take the usual form (see Chapter 7 of Part I)
5-3
 f  0,   0,  f  0 (5-15)
and
f 0 (5-16)
With the yield function defined by Equation (5-1), the time rate of change of the yield
function is
f f f
f   ij  M   (5-17)
 ij  M 
Using in turn Equations (5-6)2, (5-5)2, (5-3), (5-10) and (5-11), we get
f
f   ij  h  (5-18)
 ij
with
 f  ij g /  ij f g 
h    hM  1     (5-19)
  M 1     M   kk 

Applying the consistency condition, the plastic parameter comes out as

1 f
  ij (5-20)
h  ij

where   0 pertains to plastic loading, and further

1  f  g
Dijp    ij  (5-21)
 
h   ij  ij
A rate-dependent version of the porous plasticity model is obtained by assuming the
flow stress of the matrix material to depend on the equivalent plastic strain rate  M , i.e.,

 M   0  R  M    v  M ,  M  (5-22)

where  v is the viscous stress. The loading/unloading conditions and the consistency
condition still hold, but in the rate-dependent case the condition f  0 defines a dynamic
yield surface in stress space which coincides with the regular yield surface only for
vanishing plastic strain rates [5].

5-4
5.2 Yielding of thick-walled sphere under external hydrostatic stress
In order to justify the yield function of the Gurson model to be presented in Section 5.3,
we will now consider the yielding of a thick-walled sphere subjected to external hydrostatic
stress. The thick-walled sphere has internal radius a and external radius b , see Figure 5-2.
The external hydrostatic stress  H at the outer surface leads to overall yielding of the sphere.
The remote stress tensor will here be denoted  ij , while  ij will be used for the local
stresses. The remote stresses can be considered as the stresses acting upon the boundaries
of a representative volume element containing a void and matrix material. The void volume
fraction, i.e., the ratio of the void volume to the total volume of the sphere, is given by
   a / b  . The spherical symmetry of the problem implies that      , where   and
3

  are the circumferential stresses in spherical polar coordinates  r , ,   .

Figure 5-2: Thick-walled sphere with void subjected to external hydrostatic tension.

By considering an infinitesimal element of the thick-walled sphere, the equilibrium

equation in spherical polar coordinates is obtained as (see Appendix A for further details)
d rr 2
     rr  (5-23)
dr r
where  rr is the radial stress component. We assume that the solid obeys the J2 flow theory
(i.e., the von Mises yield criterion and the associated flow rule) and that the behaviour is
rigid-perfectly plastic with yield stress  y . Due to the spherical symmetry, the yield
condition for the thick-walled sphere reduces to
   rr   y  0 (5-24)

5-5
where it was used that  eq   VM  3
2  ij ij and further that the non-zero stress
components are  rr and      . The hydrostatic stress is  H  13  rr  2   .

The sphere is subjected to an externally hydrostatic stress  H on the outer surface,

while the inner surface (void surface) is un-stressed. Thus, the boundary conditions are
given by
 rr  a   0,  rr  b    H (5-25)
Since the stress is zero on the inner surface, it follows that
d rr
 0 if  H  0
dr
(5-26)
d rr
 0 if  H  0
dr
Overall yielding of the sphere implies that we also have yielding locally, and according to
Equation (5-24)
f   rr     y  0   rr      y (5-27)
Thus, the equilibrium Equation (5-23) may be written as
d rr 2 2
   y  sgn   H   y (5-28)
dr r r
where Equation (5-26) was used to obtain the last equality. Integration of this differential
equation, using the appropriate boundary conditions, gives
H b
dr
 d rr  sgn   H  2 y  (5-29)
0 a
r
The result is an expression for the external hydrostatic stress at yielding in terms of the yield
strength of the matrix material and the geometry of the thick-walled sphere:
b
 H  sgn   H  2 y ln   (5-30)
a

Inserting the expression for the porosity, i.e.,    a / b  , Equation (5-30) may be re-
3

written as

 H  2sgn   H   y ln  1/3    sgn   H   y ln 

2
(5-31)
3
Since the void volume fraction is always non-zero and less than unity, the inequality
ln   0 will always be fulfilled. This clarifies the minus sign on the right-hand side which
might seem a bit strange at first glance. Solving this equation for the porosity yields

3 H  3 H 
ln       exp  
 2  
(5-32)
2 y  y 

5-6
To proceed, we need the mathematical relation
1  exp  2 x 
cosh x  (5-33)
2exp   x 
After some straightforward manipulations, the yield condition for the thick-walled sphere
under external hydrostatic stress is established by
 3 H 
f  2 cosh    1     0
2
(5-34)
2 
 y 
where cosh   x   cosh   x  was used to get rid of the absolute value sign.

If the sphere did not contain a void, i.e., in the limit   0 , the yield criterion should
reduce to the von Mises yield criterion. However, if the material is subjected purely to
hydrostatic loading, thus leaving the deviatoric stress components identically equal to zero,
the yield condition should reduce to Equation (5-34). Thus, we may postulate the yield
condition for the thick-walled sphere subjected to a remote stress ij  ij   H  ij as

eq
2
 3 
f   2 cosh  H   1     0
2
(5-35)
 2  2
y  y 
where eq  3
2 ij ij is the von Mises equivalent stress based on the remote deviatoric
stresses ij acting on the solid. We note that as   0 , Equation (5-35) reduces to

eq
2

f  1  0  eq   y  0 (5-36)
 y2
which is the von Mises yield criterion. On the contrary, if the stress deviator is zero, so that
eq  0 , we obtain Equation (5-34).

We will see in the next section that the yield criterion that we have postulated for the
thick-walled sphere has the same form as the Gurson yield criterion for porous materials
obeying the J2 flow theory.

5-7
5.3 Gurson model
A popular model of porous plasticity is the Gurson model [6] (see also [7][8]). The
Gurson model is based on micro-mechanics of voided materials, and is valid for isotropic,
porous metallic materials. The Gurson yield criterion, as modified by Tvergaard [7], is
defined by
 eq2   
f  2  21 cosh  2 kk   1   3 2  0 (5-37)
M  2 M 
where 1 ,  2 , 3 are material constants (typical values for the constants are 1  32 ,  2  1
and 3  12 ),  M is the flow stress of the matrix material, and  eq   VM  3
2  ij ij is the
equivalent macroscopic von Mises stress. Note that the original Gurson model is obtained
by setting 1  2  3  1 . This yield condition needs some discussion. If the porosity is
zero,   0 , the yield condition equals
 eq2
f  2 1  0 (5-38)
M
which is just another form of the von Mises yield criterion. Thus, the matrix material is
governed by the von Mises yield function in the Gurson model. With porosity greater than
zero, 0    1 , yielding may take place even for vanishing deviatoric stresses, σ  0 .
Since in this case the von Mises stress is also zero, the yield criterion takes the reduced form
  
f  21 cosh  2 kk   1   3  0
2
(5-39)
 2 M 
This is the yield condition for the voided material subjected to a purely hydrostatic stress
state. Recall that the yield condition for metals and alloys is usually assumed to depend only
on the deviatoric stress, but this is not the case if the material contains a significant amount
of voids.
Using Equation (5-37), the Gurson yield locus may be expressed in terms of the
dimensionless stresses  eq /  M and  H /  M as

 eq  3  
 1   3 2  21 cosh  2 H  (5-40)
M  2 M 
The yield locus is plotted in Figure 5-3 for various levels of porosity. It is clearly seen that
for fixed value of the matrix strength, the equivalent stress at yielding, which represents the
deviatoric strength of the material, decreases with increasing porosity and increasing
magnitude of the hydrostatic stress. For very high magnitudes of  H the voided material
may completely lose its load-carrying capacity. Given that 3  12 , the yield criterion is
fulfilled at zero stress when   11 , implying a total loss of load-carrying capacity of the
material, and thus the void volume fraction has to stay within the region 0    11 .

5-8
 f = 0.00 f = 0.01 f = 0.1 f = 0.3 f = 0.6
1,2

0,8

0,6

0,4

0,2

0
-4 -3 -2 -1 0 1 2 3 4

Figure 5-3: Gurson yield locus in  eq /  M vs.  H /  M space for different void volume
fractions (   0 , 0.01, 0.1, 0.3, 0.6) as predicted by the Gurson model with 1  1.5 ,
2  1.0 and 3  12 ).

Adopting the associated flow rule, the plastic rate-of-deformation tensor is obtained as

f  3      
Dijp      2  ij  1 2 sinh  2 kk   ij  (5-41)
 ij M M  2 M  
It transpires that the plastic strain rate can be divided into volumetric and deviatoric parts,
where the first term on the right-hand side is the deviatoric part and the second term is the
volumetric part. The plastic parameter  is not dimensionless for the Gurson model, since
the yield function is not a positive homogeneous function of order one. It is, however,
readily shown that the plastic dissipation is non-negative for the Gurson model with
associated flow:

f  3      
D p   ij Dijp   ij    2  ij ij  1 2 sinh  2 kk   mm   0 (5-42)
 ij M M  2 M  
where it was used that x sinh x  0 for x  0 . The volumetric plastic strain rate is
1 2   
 vp  Dkkp  3 sinh  2 kk  (5-43)
M  2 M 
It is seen that a Gurson material obeying the associated flow rule dilates plastically under
tensile hydrostatic stresses. With the Gurson model, the void growth relation, Equation
(5-10), leads to

5-9
 31 2   
   (1   ) sinh  2 kk  (5-44)
M  2 M 
It is evident from this equation that the void volume fraction  increases for  kk  0 and
decreases for  kk  0 .

Example 5-1: Stress-strain curves obtained with the Gurson model

Let us consider a material with initial void volume fraction 0 equal to 0.01 and flow
stress defined by  M   0  K Mn , where  0  100 MPa , K  200 MPa and n  0.3 . The
constants 1 ,  2 , 3 of the Gurson model are given typical values: 1  32 ,  2  1 and
3  12 . Plots of equivalent stress  eq   VM  3
2  ij ij and void volume fraction 
versus equivalent plastic strain  M for constant values of the stress triaxiality
 *   H / VM are shown in Figure 5-4 and Figure 5-5. These results are obtained by solving
the equations numerically. It is clear that the void growth depends strongly on the stress
triaxiality. The porosity decreases towards zero for negative triaxiality, while softening due
to void growth will eventually occur for large strains under positive triaxiality.

Figure 5-4: Equivalent stress versus equivalent plastic strain for different constant stress
triaxiality levels,     1, 1 .

5-10
Figure 5-5: Void volume fraction versus equivalent plastic strain for different constant
stress triaxiality levels,  *   1, 1 .

5.4 Nucleation, coalescence and shearing of voids*

If void nucleation is included in addition to void growth, the change of the volume
fraction of voids is expressed as
  n  g (5-45)

where the subscripts n and g stand for nucleation and growth, respectively. As before, the
void growth is defined by Equation (5-10). It is common to assume that the nucleation of
new voids from particles could be either stress controlled or strain controlled. Here we will
only consider the latter case, and we define [9]
n  D  M   M (5-46)
where

N  1     2 
D  M   exp    M N
  (5-47)
sN 2  2  sN  

There are three material constants in the void nucleation rule:  N is the volume fraction of
void nucleating particles,  N is the mean plastic strain for nucleation and sN is the
associated standard deviation. An example of the nucleation of voids with plastic straining
is shown in Figure 5-6. It is seen that half of  N is nucleated when  M equals  N . By
changing sN , the evolution of the nucleation process is altered. If sN is large, nucleation

5-11
takes place over a wide range of plastic strain, while if sN approaches zero, the nucleation
happens instantaneously as the equivalent plastic strain  M reaches the value  N .

Figure 5-6: Nucleated void volume fraction n versus equivalent plastic strain  M for
nucleation parameters: N  0.1 , sN  0.05, 0.1, 0.15 and  N  0.5 .

A simple criterion for incipient fracture can be coupled to the porous plasticity model
if it is assumed that void coalescence (i.e., neighbouring voids grow together to form a
macrocrack) occurs at a critical value of the void volume fraction
  C void coalescence (5-48)
Some authors use a modified void volume fraction to simulate rapid loss of strength during
the final stages of void coalescence, i.e., for volume fractions  greater than C , see
Reference [10]. It should be noted here that the porosity at coalescence is not constant but
depends generally on the stress state and the initial void volume fraction, e.g. [11][12].

5-12
 In order to describe damage evolution for low stress triaxialities, as for example shear-
dominated stress states, Nahshon and Hutchinson [13] proposed a phenomenological
extension of the porous plasticity model in which the damage growth is modified under low
stress triaxiality to account for shear softening due to void distortion and inter-void linking.
In the modified porous plasticity model, the void growth equation reads
  n  g  s (5-49)

The evolution of the shear damage s is defined as

 ij Dijp
s  ks  σ  (5-50)
 VM

where k s is a constant and  VM  3

2  ij ij is the von Mises stress. The stress-dependent
factor  is given as

  σ   1  cos 2  3 L  , 0   L  (5-51)
3
where 0    1 and the Lode angle  L is defined in Equation (4-4). Based on the discussion
in Chapter 4 it follows that   0 for all axisymmetric stress states and   1 for all stress
states being the sum of a pure shear stress and the hydrostatic stress. It is thus understood
that s equals zero for all axisymmetric stress states while it accelerates the damage
evolution in the material for more shear-dominated stress states. Note that the evolution of
s depends only on the deviatoric stress state. It should be further noted that in the modified
porous plasticity model  is no longer the void volume fraction but rather a damage
variable that accounts for the softening of the material due to void growth and distortion.
In the case of pure shear, we have the ordered principal stresses  I   ,  II  0 and
 III   , where  is a positive shear stress, and the von Mises equivalent stress is
 VM  3
2  ij ij  3 . The non-zero principal values of the plastic rate-of-deformation
tensor are D33p   D11p , where it is assumed that the principal directions coincide with the
coordinate axes. Recall that σ and D p have the same principal directions for isotropic
materials. The equivalent von Mises strain is defined by VM
p
 2
3 Dijp Dijp  2
3
D11p . With zero
pressure,  kk  0 , the Gurson yield criterion, Equation (5-40) takes the form

VM   M 1  21  3 2 (5-52)

With zero pressure and in absence of nucleation, the only contribution to the void growth is
from void shearing, i.e.,   s . Under these conditions, Equation (5-50) gives the porosity
evolution as
  ksVM
p
(5-53)

where it was used that   σ   1 for pure shear. Integrating this differential equation with
initial conditions   0 and VM
p
 0 at t  0 , gives

5-13
   0 exp  ksVM
p
 (5-54)

Thus, in pure shear the porosity grows exponentially with the von Mises equivalent strain.
By combining Equations (5-52) and (5-54) the model parameter k s is readily calibrated
from pure shear tests.

5.5 References
[1] P.W. Bridgman. Studies in large plastic flow and fracture with special emphasis on
the effects of hydrostatic pressure. McGraw-Hill, 1952.
[2] T.L. Anderson. Fracture mechanics. Third edition. Taylor & Francis, 2005.
[3] A.A. Benzerga, J.-B. Leblond. Ductile fracture by void growth to coalescence.
Advances in Applied Mechanics 44 (2010) 169–305.
[4] A. Needleman, V. Tvergaard, J.W. Hutchinson. Void growth in plastic solids. In: A.S.
Argon (ed.), “Topics in Fracture and Fatigue”, pp. 145-178, Springer-Verlag, 1992.
[5] M. Ristinmaa, N.S. Ottosen. Consequences of dynamic yield surface in
viscoplasticity. International Journal of Solids and Structures 37 (2000) 4601–4622.
[6] A.L. Gurson. Continuum theory of ductile rupture by void nucleation and growth: Part
I—Yield criteria and flow rule for porous ductile media. J. Eng. Mater Tech. 99 (1977)
2–15.
[7] V. Tvergaard. On localization in ductile materials containing spherical voids. Int. J.
Fracture 18 (1982) 237–252.
[8] V. Tvergaard. Material failure by void growth to coalescence. Adv. Applied
Mechanics 27 (1990) 83–151.
[9] C.C. Chu, A. Needleman. Void nucleation effects in biaxially stretched sheets. J. Eng.
Mater. Tech. 102 (1980) 249–256.
[10] V. Tvergaard, A. Needleman. Analysis of cup-cone fracture in round tensile bar. Acta
Metall. 32 (1984) 157–169.
[11] Z.L. Zhang, E. Niemi, A new failure criterion for the Gurson-Tvergaard dilational
constitutive model. International Journal of Fracture 70 (1995) 321–334.
[12] Z.L. Zhang, C. Thaulow, J. Ødegård. A complete Gurson model approach for ductile
fracture. Engineering Fracture Mechanics 67 (2000) 155–168.
[13] K. Nahshon, J.W. Hutchinson. Modification of the Gurson Model for shear failure.
European Journal of Mechanics A/Solids 27 (2008) 1–17.

5-14
6 Hypoelastic-plasticity
6.1 Problem definition
We have now completed the description of the plastic and viscoplastic behaviour of
materials for large plastic strains, and we will turn to describe the elastic response. It is clear
from the definition of D that this tensor should be equal to zero for local rigid-body rotation,
and in this case we have L  W . On the other hand, the stress tensor σ is not constant
during rigid-body rotation. This fact is illustrated in Figure 6-1 showing a rotating uniaxial
tension specimen subjected to a follower load F with constant magnitude. At points of time
t1 and t2 , the components of the stress tensor in the fixed global basis  ei  are

 FA 0 0 0 0 0
σt t1  
 0 0 0 and σt t2  0 FA 0 (6-1)
   
 0 0 0 0 0 0

where A is the cross-section area of the specimen. At time t1 , we have 11  F / A and
 22  0 , while at t2 , it is  11 that is zero and  22  F / A .

Figure 6-1: Rotating tensile specimen subjected to a follower load F with constant
magnitude (after [1]).

This simple example shows that the naïve extension of Hooke’s law on rate form
 ij  Cijkl Dkle (6-2)

is not physically admissible for large deformations. In this equation, Cijkl is the fourth order
tensor of elastic constants, as defined in Chapter 5 of Part I. If we, on the other hand, express
the stress tensor in a coordinate system with basis  eˆ i  that co-rotates with the specimen,
then the components of the stress tensor remain constant:

6-1
  FA 0 0
σˆt t1  σˆt t2   0 0 0 (6-3)
 
 0 0 0
It thus seems that working with a corotational coordinate system might be beneficial in
handling the large rotations in a correct way.
To proceed, we define a local corotational coordinate system with basis  eˆ i  , as
defined by Equation (2-36). We also need to define the initial basis  eˆ i0  at time t  0 ,
which corresponds to the initial configuration of the body. For isotropic materials, we
usually take  eˆ i0  equal to the fixed global basis  ei  , whereas for anisotropic material
 eˆ  is aligned with the principal axes of anisotropy. In the latter case,  eˆ  may vary from
0
i
0
i

one position to another in the solid body. From Equation (2-48) the transformations of the
stress tensor and the rate-of-deformation tensor between the two coordinate systems are
given by
ˆ ij  ki kl lj and  ij  ikˆ kl  jl (6-4)

Dˆ ij  ki Dkl lj and Dij  ik Dˆ kl  jl (6-5)

For short we will call ˆ ij the corotational stress and Dˆ ij the corotational rate-of-
deformation. Since local rigid body rotations have been eliminated in the corotational basis,
we should have ˆ ij  0 for all Dˆ ij  0 , i.e., the components of the stress tensor in the
corotational coordinate system are constant during local rigid body rotations. We may then
express Hooke’s law on rate form as
ˆ ij  Cˆijkl Dˆ kle (6-6)

where Cˆ ijkl represent the components of the fourth-order tensor of elastic moduli in the
corotational coordinate system. Since small elastic deformations are assumed, it is
reasonable to adopt the same elastic moduli as for small strains. This implies that the elastic
behaviour is not depending significantly on the plastic deformation, which is a good
approximation for most metals and alloys.
The reason why this theory is called hypoelastic (or linear hypoelastic, to be precise) is
that we do not work with a strain energy function from which we derive the stress tensor.
Instead we define the rate of the corotational stress ˆ ij as a linear function of the
corotational elastic rate-of-deformation Dˆ kle . For large deformations it is not ensured that
the hypoelastic formulation will conserve the energy and thus the work performed in a
closed elastic deformation cycle is not necessarily equal to zero [1], viz.

 ˆ Dˆ dt  0
e
ij ij (6-7)

By a closed elastic deformation cycle we understand that the body is taken from the initial
configuration into successive deformed configurations, all of which are obtainable by pure
elastic deformation, and finally it is taken back to the initial configuration. However, since

6-2
we have assumed that the elastic deformations are small (or even infinitesimal), the
hypoelastic dissipation is negligible compared to the plastic dissipation in most situations.

6.2 Corotational formulation

If the material is anisotropic, it is convenient to formulate all the constitutive relations
in the coordinate system defined by the principal axes of anisotropy and then to let this
system co-rotate with the local spin of the material. This approach is reasonable for
moderate plastic deformations for which the anisotropy can be assumed to remain constant.
On the other hand, if the material is isotropic, the choice of coordinate system is arbitrary.
The corotational formulation of hypoelastic-plasticity is compiled in Box 6-1.
Combining the equations in Box 6-1, the rate constitutive equations for the elastic-
plastic material are obtained as (see Chapter 7 of Part I for details on the derivation)
ˆ ij  Cˆijkl
t ˆ
Dkl (6-8)
where
Cˆ ijkl for elastic loading/unloading

Cˆ ijkl
t
 1ˆ g   f ˆ  (6-9)
ˆ
Cijkl  H  Cijpq ˆ   ˆ Cmnkl  for plastic loading
  pq   mn 
In the derivations it was assumed that
f ˆ g
H Cijkl  hR  0 (6-10)
ˆij ˆ kl

where hR  dR / dp is the plastic modulus.

The rate constitutive equations for the elastic-viscoplastic material take the form
g
ˆ ij  Cˆ ijkl Dˆ kl  Cˆ ijkl (6-11)
ˆ kl

with   0 for f  0 and    (σˆ , R ) for f  0 . It is noted here that the last term on the
right-hand side of Equation (6-11) is a function of state for viscoplastic loading, i.e., it
depends on the stress σ and the hardening R .

6-3
Box 6-1: Corotational formulation of large-deformation hypoelastic-plasticity.

Domain of validity:
 Infinitesimal elastic strains
 Finite plastic strains and rotations
 Elastic and plastic anisotropy

Main equations:
 Additive decomposition of the rate-of-deformation tensor:
Dˆ ij  Dˆ ije  Dˆ ijp

 Hypoelastic relation:
ˆ ij  Cˆijkl Dˆ kle
 Plastic dissipation:
Dp  ˆij Dˆ ijp  0

 Flow rule:
g  σˆ 
Dˆ ijp  
ˆ ij

 Yield function:
f    σˆ    0  R  ,  eq    σˆ 

 Work hardening:
Dp
R  R  p , p
 eq
 Plastic parameter:
o Plasticity:
f  0,   0, f 0
o Viscoplasticity:
 0 for f0

  σˆ, R  for f0

6-4
6.3 Jaumann stress rate
Let us take a closer look at the corotational stress rate ˆ ij . The coordinate
transformation rule gives

ˆ ij  ki kl lj  ki kl lj  ki kl lj  ki kl lj (6-12)

Using Equation (2-45), we get
ˆij  Wkmmi kl lj  ki kllj  ki klWlmmj (6-13)

The Jaumann stress rate  ijJ is now defined by transforming ˆ ij into the fixed global
coordinate system:
 pqJ   piˆijqj (6-14)

It follows by this definition that  ijJ is zero during local rigid-body rotations because this
is the case for ˆ ij . Such a rate of the stress tensor is said to be objective. It is important to
recall that  is time dependent, since it is generated from the spin of the material. The
transformation is therefore local in the sense that  varies from one material particle to the
other, because of the spatial variation of the spin tensor.
From the definition of  ijJ we get

 pqJ   piˆ ijqj

  pi Wkmmi kl lj  ki kl lj  ki klWlmmj  qj (6-15)
 Wkp kq   pq   plWlq

By re-arranging the indices and using the skew-symmetry of the spin tensor, we arrive at
[1]:
 ijJ   ij  Wki kj   ikWkj   ij  Wik kj   ikWjk (6-16)
Thus, the relations between the material time derivative of the stress tensor and the Jaumann
stress rate are
 ijJ   ij  Wik kj   ikWjk and  ij   ijJ  Wik kj   ikWjk (6-17)
The latter equation shows that the material time derivative of the stress has two
contributions: the first is defined by the Jaumann stress rate and stems from the deformation
of the material; the second is caused by the material spin. The Jaumann stress rate is said to
be a corotational derivative of the stress tensor.
Using Equations (6-14) and (6-6), the Jaumann stress rate is calculated from the elastic
rate-of-deformation as

 
 ijJ  ipˆ pq jq  ip jqCˆ pqrs Dˆ rse  ip jqkrlsCˆ pqrs Dkle (6-18)

Making the identification

Cijkl  ip jqkrlsCˆ pqrs (6-19)

6-5
the hypoelastic relation takes the form
 ij J  Cijkl Dkel (6-20)

Note that for anisotropic materials the components Cijkl in the fixed global coordinate
system vary with time for a material element, even if the coefficients Cˆ are material pqrs

constants. The reason is that the principal axes of anisotropy rotate relative to the fixed
global axes due to the spin of the material. It follows that the transformation in Equation
(6-19) has to be performed for each time increment for each material element, which is
clearly a computationally expensive operation.
The rate-constitutive equations of the elastic-plastic model are expressed in terms of
the Jaumann stress rate by combining Equations (6-14) and (6-8), i.e.,
 ijJ   ij  Wik kj   ikW jk  ikˆ kl jl  Cijkl
t
Dkl (6-21)
where
t
Cijkl  ip jqkrlsCˆ pqrs
t
(6-22)
Thus, the stress rate tensor in the fixed coordinate system is obtained as
 ij  Cijkl
t
Dkl  Wik kj   ikWjk (6-23)

Similarly, for the elastic-viscoplastic model, we get

 ijJ   ij  Wik kj   ikW jk  ikˆ kl jl  Cijkl Dkl  Qijp (6-24)

where Cijkl is given by Equation (6-19). The tensor Q p  Q p  σ, R  represents the stress
relaxation caused by viscoplasticity, and depends on the state of the material, as defined by
the stress tensor and the hardening variable. From Equation (6-11), we deduce that
0 for f 0
Qijp  ik Qˆ klplj , Qˆ klp   (6-25)
  σˆ , R  Cklmn g  σˆ  / ˆ mn
ˆ for f 0

The stress rate tensor with regards to the fixed basis reads
 ij  Cijkl Dkl  Qijp  Wik kj   ikWjk (6-26)
It is seen that the corotational formulation involves several coordinate transformations
to express the stress rate tensor in the fixed coordinate system. These are costly operations
and it is well worth establishing a more efficient formulation for isotropic materials.

6-6
6.4 Jaumann formulation for isotropic materials
If we restrict our attention to isotropic materials, the choice of coordinate system should
play no role in the formulation of the constitutive equations. The isotropic elasticity tensor
takes the same form in any coordinate system and is given as (see Chapter 5.3 of Part I)
Cijkl  e ij kl  e  ik jl   il jk  (6-27)

where e and  e are the Lamé elastic constants. Thus, the hypoelastic relation in Equation
(6-20) takes the form
 ijJ   ij  Wik kj   ikWjk  e Dkke ij  2e Dije (6-28)

The yield function f and the flow potential g should only depend on the stress
invariants for isotropic materials, and the same goes for the function  defining the plastic
parameter for viscoplastic materials. Note that the stress invariants I , J 2 , J 3 may be
calculated based on either  ij or ˆ ij ; the result will be the same. Thus, for isotropic
materials we may formulate the constitutive equations in the fixed coordinate system for
simplicity and computational speed, but we have to use the Jaumann stress rate to obtain
physically relevant results. The constitutive relations for the isotropic hypoelastic-
(visco)plastic material are compiled in Box 6-2.
In the Jaumann formulation, the rate-constitutive equations of the hypoelastic-plastic
models are expressed as
 ijJ   ij  Wik kj   ikWjk  Cijkl
t
Dkl (6-29)
where
Cijkl for elastic loading/unloading

 1 g   f 
t
Cijkl (6-30)
C   C   C  for plastic loading
H   pq    mn
ijkl ijpq mnkl
 
These two equations are obtained from Equations (6-8) and (6-9) by taking the corotational
coordinate system to instantaneously coincide with the fixed coordinate system, i.e.,
kj   kj in Equation (2-45) so that ij  Wij . The detailed derivation is left for the reader
as an exercise. Note that Equation (6-10) is still valid but without the hats. The stress rate
tensor with regards to the fixed basis is given by
 ij  Cijkl
t
Dkl  Wik kj   ikWjk (6-31)
For the hypoelastic-viscoplastic model, we have
 ijJ   ij  Wik kj   ikWjk  Cijkl Dkl  Qijp (6-32)
with
0 for f 0
Qijp   (6-33)
  σ , R  Cijkl g  σ  /  kl for f 0

6-7
Consequently, the stress rate tensor in the fixed coordinate system is expressed as
 ij  Cijkl Dkl  Qijp  Wik kj   ikWjk (6-34)
It should be noted that the Jaumann-rate formulation is usually used in commercial
finite element codes for isotropic solids. The corotational formulation is mostly employed
for anisotropic solids and for shells. The reason for using the Jaumann-rate formulation for
isotropic solids is that it is more computationally efficient than the corotational formulation
in this case. It should be noted that the Jaumann formulation can also be established for
anisotropic materials, but in this case it seems to have no significant advantages over the
corotational formulation and the details are therefore left out here.

6.5 Thermal effects

In some applications it is important to include the influence of temperature on the
material behaviour at large strains. There are thermal effects on both the hypoelastic and
viscoplastic behaviours.
The thermo-hypoelastic relation is defined by (cf. Chapter 5 of Part I)
ˆij  Cˆijkl Dˆ kle  ˆijT or  ijJ  Cijkl Dkle  ijT (6-35)

where De  D  D p , as before, and β is the second-order, symmetric tensor of thermal

stress coefficients. In general, the thermoelastic tensors C and β depend on temperature,
but they are here assumed to be temperature independent. Thus, only small temperature
variations are allowed for. In the case of isotropy, the thermo-hypoelastic relation reads
 ijJ  e Dkke ij  2e Dije  eTij (6-36)

where  e is the thermal stress coefficient. The more general case with temperature-
sensitive thermoelastic constants is addressed in Chapter 10.
The constitutive equations of viscoplasticity remain essentially the same as before, but
the yield function f , the flow potential g and the plastic parameter  all depend on the
temperature T . An important example is the Johnson-Cook model discussed previously. It
is observed experimentally that the work-hardening and rate sensitivity of materials are
strongly dependent on temperature. The work-hardening tends to decrease with increasing
temperature whereas the rate sensitivity increases.

6-8
Box 6-2: Jaumann-rate formulation of large-deformation isotropic hypoelastic-plasticity.

Domain of validity:
 Infinitesimal elastic strains
 Finite plastic strains and rotations
 Isotropic materials (  , g and  are isotropic functions, i.e., they depend
only on the stress invariants)

Main equations:
 Additive decomposition of the rate-of-deformation tensor:
Dij  Dije  Dijp

 Hypoelastic relation:
 ijJ   ij  Wik kj   ikWjk  e Dkke ij  2e Dije
 Plastic dissipation:
Dp   ij Dijp  0

 Flow rule:
g  σ 
Dijp  
 ij

 Yield function:
f    σ    0  R  ,  eq    σ 

 Work hardening:
Dp
R  R  p , p
 eq
 Plastic parameter:
o Plasticity:
f  0,   0, f 0
o Viscoplasticity:
 0 for f0

  σ , R  for f0

6-9
6.6 Corotational integrals*
As the perceptive reader has already noted, we have not defined any measure of strain
in the theory of hypoelastic-(visco)plasticity. The reason for this is that we have taken the
rate-of-deformation tensor as our measure of the stretching of the material, and as already
mentioned a couple of times, the rate-of-deformation tensor is not the proper rate of a strain
tensor. However, by using corotational integrals it is possible to establish deformation
measures, even if they are not proper strain measures. We will first define the corotational
integral for an arbitrary second-order tensor and then use this approach to define a suitable
deformation measure for hypoelastic-(visco)plasticity.
The corotational integral of a second-order tensor T is defined as [2][3]
t 
ij  T   ik  t    Tˆkl  t   dt   jl  t  (6-37)
0 
where Φ  T  is the symbol for the corotational integral of T , t is the current time (and the
limit of integration) and t the integration variable running from 0 to t . The integrand is
Tˆkl  t  mk  t Tmn  t nl  t (6-38)
The procedure to obtain the corotational integral of T is to integrate the tensor components
in the corotational basis over the time history and then at the current time transform the
integral (which is of course also a second-order tensor) to the fixed global basis.
Let us take the corotational integral of the corotational (or Jaumann) derivative of the
second-order tensor T . We get
t  t ˆ 
 ij  T   ik   mk Tmn nl dt   jl  ik   Tkl dt   jl  ik Tˆkl jl  Tij (6-39)
J J

0  0 
where we have assumed that the rotation tensor  is generated by Equation (2-45). It
transpires that corotational integration is the inverse operation to corotational derivative,
where the corotational derivative of T is defined by [2]

TijJ  ik Tˆkl  jl  Tij  Wik Tkj  TikW jk (6-40)

where Equations (6-38) and (2-45) were used to obtain the last equality.
We are now ready to consider useful deformation measures in the hypoelastic-
(visco)plastic theory. The corotational deformation tensor ε  is defined as
 ij  ij  D  ikˆkl jl (6-41)

where
t t t
ˆkl   ˆkldt   Dˆ kl dt   mk Dmnnl dt  (6-42)
0 0 0

Equation (6-42) shows the relation between the corotational deformation tensor ε  and the
corotational integral of D . From Equation (6-41), we get by material time differentiation

6-10
  ij  ik ˆkl jl  ik ˆkl jl  ikˆkl jl
(6-43)
 ik Dˆ kl  jl  Wimmk  ˆkl jl  ik ˆkl W jmml 

where the identity ˆkl  Dˆ kl and Equation (2-45) were used. Applying Equations (6-5) and
(6-41), we finally arrive at
 ij  Dij  Wik kj  ikWjk (6-44)

Thus, the material time derivative of  ij is seen to have the same form as the material time
derivative of the Cauchy stress  ij , cf. Equation (6-17)2. In Equation (6-44), Dij represents
the rate of  ij due to stretching of the material, whereas the remaining terms are caused by
spin.
Hypoelastic-(visco)plasticity is used in several commercial finite element codes to
describe the behaviour of metals and alloys at large deformation, since the assumption of
small elastic deformation is reasonable. In these models, the corotational deformation tensor
 ij is sometimes used as a deformation measure, and it should be noted that it is
occasionally denoted “strain”. This is not appropriate since  ij is generally path dependent.

6.7 Numerical integration*

The integration algorithms developed for elastic-(visco)plasticity in Chapters 7 and 8
of Part I can be used with minor modifications for hypoelastic-(visco)plasticity. The
challenge is to treat the large rotations correctly, as will become apparent in the following.
It is assumed that the state variables σ n and Rn are known at time tn and that the rate-
of-deformation Dn1/2 and spin Wn1/2 at time tn1/2  12  tn  tn1  are given. The reason for
this becomes clear in Chapter 9 where the central difference scheme for time integration of
the equations of motion is described. By definition Dn1/2 and Wn1/2 are, respectively, the
symmetric and skew-symmetric parts of the velocity gradient tensor L n1/2 , viz.

vi  x n 1/2 
Lij ,n 1/2  (6-45)
x j ,n 1/2
where
x j ,n 1/2  1
2 x j ,n  x j ,n 1  (6-46)

The task is to obtain the state variables σ n1 and Rn 1 at time tn 1  tn  tn 1 , where
tn1 is the time step. The incremental deformation tensor ε n1 is defined as
ε n1  tn1Dn1/2 (6-47)

It is understood that ε n1 plays the role of the strain increment in the large deformation
formulation.

6-11
Corotational formulation
We start by considering the corotational formulation that is valid for isotropic as well
as anisotropic materials. In the corotational formulation, the integration of the constitutive
equations takes place in the local, corotational basis, which is defined by the rotation tensor
 and for anisotropic materials coincides with the principal axes of anisotropy. The
procedure is to transform ε n1 into the corotational basis, then to integrate the rate
constitutive equations using the return mapping algorithms presented in Chapters 7 and 8 of
Part I, and finally to transform the stress tensor σ n1 into the global basis.
The difficult task is to integrate accurately the differential equation for the rotation
tensor in time. In matrix form, the differential equation takes the form (see Chapter 2)
  W (6-48)
A first-order update of  is simply obtained as [4]
1
 n1   n   Wn1/2 n (6-49)
tn 1
and thus
n1   I  tn1Wn1/2  n (6-50)
This update is not very accurate and should only be used if the incremental rotations within
the time step are very small. A second-order update is obtained by using the central
difference scheme to Equation (6-48) [4]
1 1
 n1   n   Wn1/2  n   n1  (6-51)
tn 1 2

Solving for n1 yields

 n 1   I  12 tn 1Wn 1/2   I  12 tn1Wn1/2  n

1
(6-52)
Since this update is not exact, the rotations within each time step should be limited. It was
shown by Hughes and Winget [5] that Equation (6-52) preserves the orthogonality of the
rotation tensor  and further that the update should be limited to rotation increments less
than 180° per time step (which in practice is no real limitation).
In the corotational formulation it is convenient to store the stress matrix σˆ n in the local
coordinate system [4]. With the rotation tensor defined, the incremental deformation tensor
ε n1 is transformed into the local coordinate system

εˆn 1   nT1εn 1 n 1 (6-53)

The integration of the stress and the isotropic hardening variable due to the deformation
increment over the time step is symbolically expressed as [4]
σˆn1, Rn1    σˆn , Rn  (6-54)

where any of the return mapping algorithms described in Chapters 7 and 8 of Part I can be
adopted. The stress in the global coordinate system is then obtained by
σn 1   n 1σˆ n 1 nT1 (6-55)

6-12
The accuracy of this algorithm has been evaluated by Flanagan and Taylor [6] for a different
definition of the rotation tensor.

Jaumann-rate formulation
Next we consider the Jaumann-rate formulation of hypoelastic-(visco)plasticity which
is often used for isotropic materials. The idea is to first update the stress to account for local
rigid body rotations and then to update the resulting stress for the incremental deformation
during the time step.
A first-order update of the stress σ n at time tn to account for the rotations during the
time step tn 1 is defined by [4]

σn  σn  tn 1 Wn 1/2σn  σnWnT1/2  (6-56)

The state variables σ n and Rn are used with the deformation increment ε n1 as input to
the integration algorithms described in Chapters 7 and 8 of Part I, and the result is the new
state σ n1 and Rn 1 at time tn 1 . We write this symbolically as

σn1, Rn1    σn , Rn  (6-57)

A second-order, time-centred update for rotations is obtained as follows [4]
tn 1
σn  σn 
2
Wn 1/2σ n  σ nWnT1/2  (6-58)

σn1, Rn1    σn , Rn  (6-59)

tn 1
σn 1  σn 1 
2
Wn1/2σn1  σn1WnT1/2  (6-60)

This scheme is somewhat more time consuming than the first-order update, but more
accurate.
The third possible update was proposed by Hughes and Winget [5]. Let us consider the
configuration of the body at tn and assume that we attach a local, corotational coordinate
system to the material point which equals the global system at tn . In this corotational
system, the components of σ n are unchanged by the rotations imposed during the time step
tn1 . The incremental rotation tensor  n1 between the configurations at tn and tn 1 is
defined by Equation (6-52) with  n  Ι , and thus

 n 1   I  12 tn 1Wn 1/2   I  12 tn1Wn1/2 

1
(6-61)

The transformation of the components of σ n from the corotational system to the global
system at time tn 1 then reads (c.f. Equation (2-48)1)

σn   n 1σn  nT1 (6-62)

The integration of the rate constitutive equations is then symbolically expressed by Equation
(6-57).

6-13
 The reader is referred to Benson [4] (from which this presentation was motivated) and
the textbooks by Belytschko et al. [1] and Simo and Hughes [7] for more details on the
temporal integration of constitutive equations.

6.8 References
[1] T. Belytschko, W. K. Liu, B. Moran. Nonlinear Finite Elements for Continua and
Structures. Wiley, 2000.
[2] Y.F. Dafalias. Corotational integrals in constitutive formulations. Journal of
Engineering Mechanics ASCE 113 (1987) 1967–1973.
[3] A.S. Khan, S. Huang. Continuum Theory of Plasticity. Wiley-Interscience Publication
(1995).
[4] D.J. Benson. Explicit finite element methods for large deformation problems in solid
mechanics. Encyclopaedia of Computational Mechanics. Edited by E. Stein, R. De
Borst, T.J.R. Hughes. Vol. 2: Solids and Structures. Chapter 25. Wiley, 2004.
[5] T.J.R Hughes, J. Winget. Finite rotation effects in numerical integration of rate
constitutive equations arising in large-deformation analysis. Int. J. Num. Meth. Eng.
15 (1980) 1862-1867.
[6] D.P. Flanagan, L.M. Taylor. An accurate numerical algorithm for stress integration
with finite rotation. Comp. Meth. Appl. Mech. Eng. 62 (1987) 305–320.
[7] J.C. Simo, T.J.R Hughes. Computational Inelasticity. Springer, 2000.

6-14
7 Explicit finite element methods
Explicit finite element methods (FEM) are particularly suited for transient dynamics
problems, like structural impacts and explosions, but also for highly nonlinear quasi-static
problems, like plastic forming. In the following, we will extend the results we derived in
Chapter 4.5 of Part I first to dynamic loading conditions, i.e., by including the inertia forces,
and then to large deformations. The result is the semi-discrete equations of motion of the
solid in the current configuration. These equations are solved in time by employing the
central difference scheme which is succinctly described. Since the central difference scheme
is only conditionally stable, the issue of setting a stable time step is discussed. Instabilities
in the solution of the dynamic problem typically induces spurious energy growth and it is
therefore of utmost importance to check the energy balance when performing explicit finite
element simulations. An explanation of how the different energy contributions are
calculated is provided as a basis for the energy check. To make possible solutions of quasi-
static problems, mass scaling (or time scaling) is frequently used in explicit FEM, and this
“numerical trick” is discussed at the end of the chapter. For further reading we suggest
References [1] and [2].

7.1 Semi-discrete equations of motion

In Chapter 4.5 of Part I, we derived the semi-discrete equilibrium equations for a solid
body subjected to small deformations and quasi-static loading conditions
Rext  Rint  r   0 (7-1)
where r is the global nodal displacement vector, Rint is the vector of internal nodal forces
and Rext is the vector of external nodal forces. These vectors all have dimensions nSD nN 1 ,
where nSD is the number of space dimensions and nN is the number of nodes in the
discretized body. The nodal force vectors were defined as
Rint  r    BT σdV , Rext   N T bdV   N T tdS (7-2)
V V St

where σ =  11  22  33  23  31  12  is the column vector containing the components of the

T

stress tensor, b = b1 b2 b3  and t =  t1 t2 t3  are the body force and traction vectors,
T T

respectively, N  x  is the global shape function matrix and B  x  =  N  x  is the gradient

operator (or strain-displacement) matrix,  being the gradient operator and x =  x1 x2 x3 
T

the position vector. The dimension of N is nSD  nSD nN , while B has dimension
n  nSD nN with n being the number of independent stress components in σ (equal to 6
for nSD  3 ).
In explicit finite element methods, dynamic behaviour is considered. Applying
Newton’s 2nd law of motion, the sum of the internal and external nodal forces should be
equal to the nodal mass times its acceleration. We may thus write the semi-discrete
equations of motion as
Mr  Rext  Rint  r  (7-3)

7-1
where M is the mass matrix and r is the vector of nodal accelerations. The dimension of
M is nSD nN  nSD nN . A more stringent derivation of the semi-discrete equations of motion
will be given in the next paragraph. In explicit FEM, a diagonal lumped mass matrix is
generally used, since this makes the inversion trivial, it works well and the nodal
accelerations are found for one node at the time. There are several ways of defining a lumped
mass matrix and this is done on the element level [2].
The extension to large deformation is rather straightforward and we will soon see that
Equations (7-2) and (7-3) remain valid. The kinematic fields are interpolated in the current
configuration by
u  Nr , v  Nr , a  Nr (7-4)
D  v   Nr = Br (7-5)

where u  u1 u2 u3  , v   v1 v2 v3  and a   a1 a2 a3  are the displacement, velocity and

T T T

acceleration vectors, respectively, D   D11 D22 D33 2 D23 2 D31 2 D12  is the column vector
T

with the components of the rate-of-deformation tensor as its entries, and  is the gradient
operator on the current configuration:
 x1 0 0 
 

0 x2 0 
  
0 0 x3 
   
(7-6)
0
 x3 x2 
 0  
 x3 x1

 x 
0 
 2 x1

The virtual quantities needed in the principle of virtual power are interpolated in the same
manner
 v  N  r ,  D  B r (7-7)
The principle of virtual power, see Chapter 2, may then be written on matrix form as

  D σdV    v  adV    v T bdV    v T tdS

T T
(7-8)
V V V St

where we have also used that  v  0 on S u and that t  t on St . Inserting the interpolated
kinematic fields, we get
 
 r T   BT σdV    N T NdV r   N T bdV   N T tdS   0 (7-9)
V V V St 
Except for nodes I located on Su , the entries of  r are arbitrary, and we arrive at the semi-
discrete form of the equations of motion

 N NdV r   N T bdV   N T tdS   BT σdV I  Su

T
(7-10)
V V St V

7-2
Introducing the definitions of the external and internal nodal force vectors in Equation (7-2),
it is evident that Equation (7-10) is equivalent to Equation (7-3), provided the mass matrix
M is defined as
M    N T NdV (7-11)
V

Using this definition, M is called the consistent mass matrix, but as already mentioned, a
lumped, diagonal mass matrix is used in explicit FEM. Using the “sum of row method”, we
may symbolically define the diagonal elements of the lumped mass matrix as
nSD nN
M PP  M PQ , P  1, nSD nN  (7-12)
Q 1

while all the off-diagonal terms are zero.

7.2 Central difference time integration scheme

The semi-discrete equations of motion are solved by using the explicit central
difference integration scheme. Let tn 1 , t n and tn 1 denote time increments n  1 , n and
n  1, respectively, with t0  0 as the initial value of time. The time increments between
these points of time are denoted tn  tn  tn1 and tn1  tn1  tn , respectively. In the
following, subscripts n and n  1 indicate quantities at times t n and tn 1 , respectively,
whereas subscripts n  12 and n  12 in turn designate quantities evaluated at the mid-
increment times tn  1 
2
1
2  tn  tn1  and tn  1 
2
1
2  tn1  tn  . It is assumed that all quantities at
time steps prior to and including t n have been found using the central difference scheme.
The global nodal velocities and displacement vectors are updated according to [2]
tn 1  tn
rn  1  rn  1  rn (7-13)
2 2
2
rn1  rn  tn1rn 1 (7-14)
2

In these equations, quantities at tn 1 and t n are known from the previous increment, and
2

thus the update scheme is explicit. The central difference scheme for time integration is not
self-starting, and at time t  0 , it is necessary to prescribe a value of the nodal velocity
vector r 1 . In [3] it is assumed that
2

t1
r 1  r0  r0 (7-15)
2
2
and Equation (7-13) gives

t0
r 1  r0  r0 (7-16)
2
2
The initial nodal velocities r0 and displacements r0 are set to zero if they are not specified
by the user.

7-3
 The nodal accelerations rn are defined by the semi-discrete equations of motion, cf.
Equation (7-3), and are given by
rn  M 1  Rnext  Rnint  (7-17)

7.3 Stable time step

The central difference scheme is only conditionally stable and very small time steps
are required to obtain an accurate solution. The stable time step is estimated by [1][2]
h 
t   min  e  , e  1, nel  (7-18)
c
where he is the characteristic dimension of element e ; c  E /  is the current wave
speed in the material; E being Young’s modulus;  is the Courant number and accounts
for destabilizing effects of the nonlinearity of the problem; nel is the total number of
elements. According to [1], typical values of  is between 0.8 and 0.98. However, for very
nonlinear problems it might be necessary to reduce  to achieve a stable solution. It is
notable that he / c is the time it takes for an elastic wave to propagate across the element,
and the criterion given by Equation (7-18) ensures that the distance travelled by an elastic
wave during t is less than the characteristic dimension of the smallest element.

To get a feeling of a typical stable time step, we can consider a steel structure meshed
with solid elements of uniform characteristic size he  1 mm . For steel materials, we have
E  210 109 N/m2 and   7850 kg/m3 , and by taking   0.9 , the time step t becomes
1.74 107 s , which explains why the main usage of explicit FEM is fast transient dynamics.
Using double precision in the computations, the total number of time steps in a simulation
should not be more than, say, about 107 to avoid round-off errors (caused by the limited
machine precision) to accumulate and completely dominate the numerical solution. At first
sight, it thus seems that only phenomena with duration less than about a second can be
analysed with explicit FEM, which would mean that metal forming could not be simulated
with these methods. However, a workaround exists, namely mass (or time) scaling, which
we will describe in Chapter 7.5.

7.4 Energy balance

Use of a time step that fulfils the criterion defined by Equation (7-18) does not imply
that instabilities in the solution of nonlinear problems are precluded, since the criterion is
based on stability analysis of the central difference time integration scheme for the linear
equations of motion [1]. To detect possible instabilities in the solution of a nonlinear
problem that lead to spurious energy generation—and thus violates the law of conservation
of energy—it is good practice to always check the energy balance when performing explicit
FEM simulations. To do this, we need to compute the work done by the internal and external
forces and the kinetic energy of the finite element model. The work of the internal and
external global nodal forces is determined using the trapezoidal rule for time integration,
viz.
tn 1 T
Wnint1  Wnint  rn  1  Rnint  Rnint1  (7-19)
2 2

7-4
 tn 1 T
1  Wn
Wnext ext
 r 1  Rnext  Rnext1  (7-20)
2 n 2
The kinetic energy at tn 1 is computed as
1 T
1 
Wnkin rn 1 Mrn 1 (7-21)
2
where the nodal velocities at tn 1 are calculated as
tn 1
rn 1  rn  1  rn 1 (7-22)
2
2
Energy conservation implies that [1]

1  Wn 1  Wn 1   max Wn 1 , Wn 1 , Wn 1 
Wnkin int ext kin int ext
(7-23)

where  is a tolerance, typically on the order of 102 . It is proposed in [1] that for large
systems the energy balance should be checked for subdomains of the model. If the
conservation of energy is violated, the results of the simulation should be discarded and the
reason for the instability detected and resolved.

7.5 Mass scaling

It was indicated above that explicit FEM needs exceedingly small time steps to provide
stable solutions, which makes the method mostly useful for fast transient dynamic problems.
However, explicit FEM may be used also for quasi-static problems by fictitiously increasing
the density of the materials to increase the stable time step. According to Equation (7-18),
the stable time step is increased by a factor 10 if the density is increased by a factor 100,
everything else being unchanged. Such a fictitious increase of the density is called mass
scaling and can be used to increase the stable time step for quasi-static problems in which
the inertia forces have negligible influence on the solution. Typical examples of such
problems are various types of metal forming. There are two important versions of mass
scaling. If the density is scaled uniformly for all elements, the mass scaling is said to be
fixed. On the contrary, if the density is scaled for a selection of elements to keep the time
step above a user-defined lower limit, the mass scaling is said to be variable. In the latter
case, the density of an element is scaled so that


 s   max 1,  tmin / te 
2
 (7-24)

where  s is the scaled density, tmin is a user-specified minimum time step, and
te   he / c is the stable time step for element e , according to Equation (7-18). Note that
only elements for which the stable time step, te , is less than the minimum time step, tmin ,
are scaled with variable mass scaling. The total mass gained due to variable mass scaling
should be checked to evaluate possible effects on the solution.
If mass scaling, either fixed or variable, is adopted, it should always be checked that
the ratio of the kinetic energy to the internal energy, Wnkin int
1 / Wn 1 , remains small, say less

than 0.01, throughout the simulation. Too extensive mass scaling is sometimes seen as
spurious oscillations superimposed on the quasi-static solution. To obtain best possible
results with mass scaling, it is important to apply traction and displacement (or velocity)
7-5
boundary conditions smoothly in order to avoid excitation of high frequency modes.
Typically, the loading is ramped up during the first 10% of the process time. Mass scaling
should always be used with utmost caution, but if used properly, it is a very useful way of
solving highly nonlinear quasi-static problems. It is noted that mass scaling should never be
used for dynamic problems where the inertia forces play a significant role.
Finally, it should be mentioned that an alternative to mass scaling for quasi-static
problems is time scaling, which implies that the problem is run during a shorter time interval
than the duration of the physical process that is modelled. Time scaling increases the strain
rates in the structure and should only be used for rate-insensitive constitutive models, and
in this case the two approaches are equivalent. As for mass scaling, care should be taken
that the ratio of the kinetic energy to the internal energy stays small.

7.6 Strain softening, element erosion and mesh dependence

Strain softening because of damage evolution (i.e., nucleation, growth and coalescence
of voids) or adiabatic heating in transient dynamic problems may lead to non-unique
solutions of the boundary value problem for rate-independent plasticity [4]. When
uniqueness is lost, general bifurcation modes, e.g. local necking and shear banding, become
possible. Indeed, experiments show that strain localization in a narrow band is a frequent
precursor to failure. As long as the elastic-plastic material work-hardens, the finite element
solution converges when the characteristic element size he decreases towards zero,
provided an appropriate element formulation is applied. When strain softening occurs, the
strains tend to localize and the convergence will be slower—the solution becomes mesh-
size dependent. This mesh dependence may become pathological for strain-softening
materials, meaning that the deformation localizes in narrow bands with dimension governed
by the characteristic element size, i.e., the plastic dissipation goes to zero when he  0 . Of
course, this is not a physically meaningful result and some kind of regularization technique
is called for to improve the convergence characteristics of the numerical solution of the
discretized problem. Somehow we need to introduce a physical length scale into the problem
that determines the size of the localization bands.
Strain softening means that the elastic domain, as defined by the yield surface,
decreases during plastic flow. In transient dynamic problems the strain softening may be
induced by adiabatic heating. Recall that under adiabatic conditions the temperature
increases as a result of rapid plastic straining, i.e., there is not enough time for heat
conduction to take place. If the strength and work-hardening of the material depend on
temperature, like in the Johnson-Cook constitutive relation, strain softening commences as
the rate of thermal softening exceeds the work-hardening rate. In porous plasticity models,
like the Gurson model, strain softening starts when the rate of damage softening, due to an
increasing porosity with plastic straining under hydrostatic tension, surpasses the work-
hardening rate. In both cases, the strain softening tends to localize the plastic deformations
and the propensity for strain localization into narrow bands increases.

7-6
Example 7-1: Softening and mesh sensitivity

(1) (2) (m)

Figure 7-1: Bar loaded in uniaxial tension and discretized into m elements.

In this example, we will demonstrate the influence of softening on the growth of

imperfections in a bar subjected to uniaxial tension. The stress state is assumed to be one-
dimensional. Figure 7-1 shows the bar discretized into m elements. The constitutive
behaviour of the bar is defined by

 E
 for 0     y   y / E
 
 y  Et     y  for    y
(7-25)

where
y
Et   (7-26)
 f y
Here,  is the uniaxial stress,  is the strain,  y is the yield stress,  y   y / E and  f
are the yield and failure strains, E is the elastic modulus and Et is the tangent modulus in
the plastic domain. Since Et is negative, strain softening occurs. The constitutive behaviour
of the bar is sketched in Figure 7-2.
It is assumed that one of the elements have a marginally lower yield stress than the
other elements, i.e., the bar is not perfectly homogeneous and the stress is not perfectly
uniform along the bar. As the element with the lower yield stress starts to deform plastically,
this element softens due to the constitutive behaviour shown in Figure 7-2. The consequence
is that, since we require equilibrium in the entire structure and thus between the elements,
the other elements start to unload elastically. These elements will never reach their yield
limit, and will only experience elastic deformations. The weaker element will eventually
fail, while the remaining elements continue to unload elastically.

7-7
Figure 7-2: Constitutive behaviour of the bar up to failure.

Equilibrium requires that F1  F2  ...  Fm , where m is the total number of bar elements.
Assuming that all bar elements have the same cross-sectional area, we find that the
equilibrium condition is also given by
1   2  ...   m   (7-27)
Compatibility of the bar means that the total increase in length of the bar is equal to the
summed increase in length of all the elements, i.e. L  i 1 Li . Using the small strain
m

assumption, we may then find that

m
L   L    i Li (7-28)
i 1

where  is the average strain of the bar.

In order to proceed, we further assume that all the bar elements have equal length Le
in the initial configuration. Then we may write Equation (7-28) in the form
m
1 m
 mLe  Le   i    i (7-29)
i 1 m i 1
Now, we need to find the strain of each bar element. For the m 1 elements that remain
in the elastic range, the strains are simply

i  , i  2,..., m (7-30)
E
For the single plastic element, labelled i  1 , we have
  y   y y
1  y   (7-31)
H H E

7-8
Now, Equation (7-29) yields
1    y  y  1     y  y  
   m  1     m    (7-32)
m E H E  m E H E 
which may be written
  E  H    y
    (7-33)
E  EH  m
The stress-average strain curves for various numbers of elements are plotted in Figure
7-3. It is readily seen from this figure that for a single element ( m  1 ) we obtain the
constitutive behaviour from Figure 7-2. As we increase the number of elements, the average
strain decreases because more elements unload elastically, i.e., a greater part of the bar is
elastically unloaded and contracts. When the number of elements reaches the value
m   f /  y , the average strain is constant after incipient yielding, and the bar has constant
length even though the stress decreases. As the number of elements increases beyond this
value, the elastic unloading dominates the problem and the response of the bar approaches
the elastic loading curve. This is clearly indicated by the curve for m  100 .
This example illustrates the problem of pathological mesh sensitivity that occurs for
elastic-plastic materials exhibiting strain softening and material instability, cf. Chapter 9.
When simulating the behaviour of elastic-plastic materials with strain-softening behaviour,
utmost care should be taken and some kind of regularization of the numerical solution
should be employed.
■

Figure 7-3: Stress-average strain curves for different total element numbers m.

While strain softening localizes the plastic deformation into narrow bands, work-
hardening tends to distribute plastic flow. Another mechanism for distributing plastic flow
is rate-sensitivity. In an elastic-viscoplastic material, the viscous stress contributes more to
the flow stress in regions of high plastic strain rate. The result is that the plastic flow has a
tendency to distribute over a wider region than in a rate-insensitive material, everything else
being the same. If strain softening takes place, the strain rate in the regions of strain
localization will increase, because the elastic domain shrinks with plastic straining. The
7-9
outcome is that the viscous stress increases and thus works against the strain softening.
Hence, changing from plasticity to viscoplasticity is a simple and physically appealing way
to regularize the finite element solution in the case of strain-softening materials, in particular
in transient dynamics problems where the range of strain rates is wide. Obviously, the
method is less effective if the material at hand exhibits low rate sensitivity. It should be
noted that introducing viscoplasticity for strain-softening materials does not remove the
mesh dependence [1], but the mesh convergence should improve compared with rate-
independent plasticity.
Another possible route to improve the convergence characteristics of finite element
solutions for strain-softening materials is to use a non-local approach. This way a
characteristic length scale , other than the characteristic element size he , is introduced into
the problem. The length scale should be somehow related to the microstructure of the
material and the physical mechanisms governing the modelled behaviour. For instance, it
could be related to the grain size of the material if ductile damage is the source of strain
softening. Let  denote an internal variable, for instance the equivalent plastic strain p or
the void volume fraction  , and let  be the non-local average of this variable over a
neighbourhood domain defined by the length scale . In the non-local approach, the
evolution of  is then defined by (e.g. [5][6] and references therein)
1
  x    y  w  x, y,  dV
W  x  V
(7-34)

where x is the position vector of the actual material particle, y is the position vector to a
neighbouring material particle, and V is the volume of the neighbourhood domain. The
weight function w is typically defined by a bell-shaped function in the form (e.g. [5])
q
 1 
w  x, y,    p
1   x  y /   (7-35)

where  q, p  are parameters and x  y is the magnitude of the vector x  y . The function
W  x  has to be defined such that    when  is spatially uniform and  0 . Thus,
proper normalization requires that
W  x    w  x, y,  dV (7-36)
V

It is further noted that the local formulation is obtained for  0 . The weight function
w  x, y,  is illustrated in Figure 7-4 for some realistic values of the parameters  q, p  . The
non-local approach is an efficient way of improving the convergence properties of finite
element solutions for strain-softening materials, but it is often not available in commercial
finite element codes. If the reason for softening is adiabatic heating, it is suitable to take the
equivalent plastic strain as the non-local variable. If the softening is caused by damage, it
seems more appropriate to take the porosity as the non-local variable.

7-10
 p=1, q=1 p=2, q=1 p=1, q=2 p=2, q=2 p=8, q=2
1

0,8

0,6
w

0,4

0,2

0
0 1 2 3 4 5 6
|x-y|/l

Figure 7-4: Illustration of bell-shaped weight function w for some combinations of the
constants p and q .

While the two regularization methods described above have some physical relevance,
more pragmatic approaches also exist. Here we will briefly discuss the computational cell
approach and the use of he -dependent damage parameters.
In the computational cell approach [7], the problem of mesh convergence is
circumvented altogether by selecting a characteristic element size he which is sufficiently
small to describe the physical mechanisms governing the mechanical behaviour of the
material (e.g. plasticity and damage) and sufficiently large to make possible simulations of
the actual problem (e.g. a structural component or a complete structure). The appropriate
mesh size can be obtained from previous knowledge of the problem, e.g. at which scale the
damage mechanisms develop, or by performing initial mesh sensitivity studies. When the
mesh size is selected, finite element simulations of material tests, employing the selected
characteristic mesh size, are carried out to determine the material parameters governing
plastic flow and damage evolution. The same characteristic mesh size is used in the finite
element simulations of the actual problem, e.g. the crash of a car against a barrier. If the
Gurson model is used, the mesh size is set by the so-called process zone, which is the zone
in front of a propagating crack within which damage processes take place. Typically this
zone only contains some few grains and the characteristic mesh size ends up being only
fractions of a millimetre. This limits the application of the Gurson model for large-scale
structures and the uncoupled damage evolution rules (discussed in Chapter 4) are usually
more appropriate for such applications. The computational cell approach could be useful
also for uncoupled damage models.
If an uncoupled damage model is adopted, there is no damage softening and the solution
should converge as the element size is decreased, provided thermal softening is avoided.
But damage evolution and failure typically occurs in regions with high gradients in the
plastic strain field, and a prohibitively small mesh size may be required to obtain a
converged solution. Thus, to reduce the mesh sensitivity it is often useful to apply a simple
regularization technique also in combination with an uncoupled damage model. The
regularization approach based on introducing he -dependent damage parameters (see also

7-11
[1]) is perhaps best suited for the uncoupled damage models. If the Cockcroft-Latham
criterion is adopted, the failure parameter WC is taken to depend on the characteristic mesh
size, WC  WC  he  . The method could be used for any of the uncoupled damage models
presented in Chapter 4, but the calibration may become intractable if several parameters are
involved. The calibration of the failure parameters involves running finite element
simulations of one or several materials tests using a range of characteristic mesh sizes. For
each mesh size values are determined for the failure parameters that best fit the experimental
data. Typically, the failure strain will increase with decreasing mesh size, as high gradients
in the stress and plastic strain fields are progressively better resolved.
The fracture propagation is usually modelled by element erosion, i.e., as the damage in
an element reaches the critical value, the element is removed from the finite element model.
If the element has several integration points, it is typically removed when a user-defined
number of integration points have failed. However, in some commercial finite element
codes the element is eroded only when all integration points have failed. Element erosion
could be considered as an extremely rapid softening process, since removing the element is
equivalent to abruptly reducing the stress to zero. It follows that modelling of ductile failure
and crack propagation by element erosion is inherently mesh-dependent, and regularization
of the solution may be required to improve the convergence properties.

7-12
7.7 References
[1] T. Belytschko, W. K. Liu, B. Moran. Nonlinear Finite Elements for Continua and
Structures. Wiley, 2000.
[2] D.J. Benson. Explicit finite element methods for large deformation problems in solid
mechanics. Encyclopaedia of Computational Mechanics. Edited by E. Stein, R. de
Borst, T.J.R. Hughes. Vol. 2: Solids and Structures. Chapter 25. Wiley, 2004.
[3] Abaqus 6.13 Theory Guide. Dassault Systèmes SIMULIA, 2011.
[4] N. Saabye Ottosen, M. Ristinmaa. The Mechanics of Constitutive Modeling. Elsevier,
2005.
[5] V. Tvergaard, A. Needleman. Effects of nonlocal damage in porous plastic solids.
International Journal of Solids and Structures 32 (1995) 1063–1077.
[6] A. Needleman, V. Tvergaard. Dynamic crack growth in a progressively cavitating
solid. European Journal of Mechanics A/Solids 17 (1998) 421–438.
[7] C. Ruggieri, T.L. Panontin, R.H. Dodds. Numerical modeling of ductile crack growth
in 3-D using computational cell elements. International Journal of Fracture 82 (1996)
67–95.

7-13
7-14
8 Crystal plasticity*
A brief account of crystal plasticity is given in this chapter. We will assume small
elastic deformations and isothermal conditions to simplify the presentation. However, the
plastic deformations and the rotations may be finite. The textbook of Khan and Huang [1],
the book edited by Kocks et al. [2] and the scientific papers [3]–[5] provide more in-depth
treatises of crystal plasticity.

8.1 Introduction
The microstructure of metals and alloys consists of grains having different orientations
compared to the global frame. Each grain has a crystalline structure defined by a three-
dimensional array of atoms in a regular pattern or lattice, see Figure 8-1. The atoms are held
together by electromagnetic forces between the electrons of neighbouring atoms. There are
three major classes of lattice systems, namely the face centred cubic (fcc), body centred
cubic (bcc) and hexagonal close packed (hcp) systems. If the grains are randomly oriented,
the material—or polycrystal—is expected to exhibit isotropic behaviour at the macroscopic
scale, even if the grains themselves have anisotropic properties. If, on the other hand, the
grains have preferred directions, denoted crystallographic texture, the material is anticipated
to display anisotropic behaviour.
While elastic deformations can be explained by the structure of the perfect crystal,
plastic deformations can only be described by the presence of crystal defects that disturb the
regular pattern of the lattice. There are three main classes of crystal defects. Point defects
consist of atoms inserted or substituted in solid solution and of vacancies which create a
local distortion of the lattice. Surface defects are the surfaces of separation between crystals
or parts of a crystal where the orientations or characters of the phases are different, e.g. grain
boundaries and interfaces between two phases. Line defects or dislocations are the defects
that are chiefly responsible for plastic deformation. There are different types of dislocations:
edge dislocations, screw dislocations and dislocation loops. Dislocations are not static or
permanent defects, but are being nucleated and annihilated and can move through the lattice
under the action of a shear stress. Their movements cause plastic deformations which again
increases the density of dislocations. Dislocations act as obstacles for the movement of other
dislocations and thus contribute to the work-hardening of the material.
Elastic deformations occur at the atomic level. They are caused by variations in the
atomic spacing required to balance external loads (Figure 8-1). Elastic deformations are
reversible, i.e., the initial configuration of the atoms is restored when the load is removed.
Plastic deformations, on the other hand, occur at the grain level. They are caused by relative
displacement of atoms which remains after unloading—plastic deformation is thus
irreversible. Two mechanisms of plastic deformation are slip and twinning. Only slip will
be discussed here, since this is the single most important mechanism. Slip means that parts
of a crystal glide past one another by an integer number of interatomic spacing (Figure 8-1).
Slip follows certain crystallographic planes, denoted slip planes, which are most likely the
planes of closest packing. The slip direction is usually parallel to the crystallographic
direction of least interatomic spacing.

8-1
 Elastic
deformation

Un-deformed
lattice

Plastic
deformation

Figure 8-1: Un-deformed lattice, lattice after elastic deformation and lattice after plastic
deformation by slip.

A slip system is a combination of a slip plane and a slip direction. If the slip plane is
denoted by its unit normal n( ) and m( ) is the unit vector parallel to the slip direction, then
the slip system may be identified by the pair of vectors ( m( ) , n( ) ) , where  refers to one
of several possible slip systems in a crystal (e.g. 12 in fcc crystals and up to 48 in bcc
crystals), see also Figure 8-2.
The magnitude of the resolved shear stress determines whether or not slip occurs on a
slip system. The resolved shear stress is the shear component of the stress tensor on the slip
plane in the slip direction. Hence, slip occurs on a slip system when the resolved shear stress
reaches a critical value, which is called the critical resolved shear stress – this is Schmid’s
law. The kinematics of slip leads to deformation-induced rotation of the crystal lattice. This
phenomenon is of key importance for the anisotropic response of polycrystalline materials:
an initially isotropic polycrystalline material (i.e., having randomly oriented grains)
develops anisotropy (texture) during sustained plastic deformation. Slip is assisted by
dislocation motion across the lattice under the action of a shear stress. This mechanism
requires breaking of bonds only in the vicinity of the dislocation line and successively from
one atom to the next, which strongly reduces the resistance compared with the perfect
lattice.
In the following, we will describe the theory of plasticity for single crystals and then
the extension to a polycrystalline aggregate is briefly explained. It is assumed that plastic
deformation occurs by slip on a set of slip systems. The theory is developed for small elastic
deformations and isothermal conditions. One important application of this theory is to
establish the yield surface used in the phenomenological theory of plasticity based on
measurements of crystallographic texture.
8-2
8.2 Single crystal plasticity
8.2.1 Lattice rotation and plastic slip
We start by assuming that the velocity gradient can be split additively into elastic and
plastic parts
Lij = Leij + Lijp (8-1)

It follows immediately that the rate-of-deformation tensor Dij  L( ij ) and the spin tensor
Wij  L[ij ] are decomposed in the same way, namely

Dij = Dije + Dijp , Wij = Wije + Wijp (8-2)

It is assumed that the elastic deformations are small, while plastic deformations and
rotations are finite. The crystal lattice is assumed to rotate as defined by the elastic spin Wije ,
which is often denoted the substructure spin Wij* , e.g. [5].

Plastic deformations occur by slip on given planes in the crystal lattice and in given
lattice directions. The combination of a slip plane and a slip direction is called a slip system.
Let n( ) and m( ) represent the orthonormal vectors defining the normal to the slip plane
and the slip direction, respectively, for slip system  , where  = 1,2,..., ns and ns is the
number of slip systems. In general, there are many slip systems available for plastic slip in
a material (e.g. 12 slip systems in fcc materials). The plastic slip is assisted by dislocations
and represents irreversible changes to the material. Even so, the lattice is left unchanged by
crystal slip.

Figure 8-2: Single slip on a slip system where the slip normal n( ) is parallel to the x2 axis,
while the slip direction m( ) is parallel to the x1 axis.
We first consider that the body is deformed by plastic slip on a single slip system. As
indicated in Figure 8-2, showing slip on a system defined by n ( ) = e 2 and m( ) = e1 , the
plastic velocity field is then defined by
vip ( x, t ) =  ( ) ( xk nk( ) ) mi( ) (8-3)

8-3
where  ( ) is the plastic shear strain rate of the actual slip system. The plastic velocity
gradient is obtained by
vip 
Lijp  =  ( ) ( xk nk( ) ) mi( )  =  ( ) nk( ) mi( ) jk =  ( ) mi( ) n(j ) (8-4)
x j x j  

It is convenient to introduce the Schmid tensor Sij( ) as

Sij( ) = mi( )n(j ) (8-5)

Then, the plastic velocity gradient takes the simple form

vip
L 
p
=  ( ) Sij( ) (8-6)
x j
ij

In the elastic-viscoplastic approach, which we will present here, the plastic shear strain rate
 ( ) is determined from a constitutive relation and assumed to depend on the magnitude of
the resolved shear stress on the slip system, the strength of the slip system, and the rate
sensitivity of the material.
We find the plastic rate-of-deformation and the plastic spin as the symmetric and anti-
symmetric parts of the velocity gradient
Dijp =  ( ) S((ij)) , Wijp =  ( ) S[(ij]) (8-7)

where S((ij)) and S[(ij]) denote the symmetric and anti-symmetric parts of the Schmid tensor

S((ij)) =
2
( mi n j + m(j ) ni( ) ) , S[(ij]) = ( mi( ) n (j ) − m(j ) ni( ) )
1 ( ) ( ) 1
2
(8-8)

If the material experiences slip on several systems, the plastic velocity field is obtained
as the sum of the contributions from these slip systems
ns
vip (x, t ) =  ( ) ( xk nk( ) ) mi( ) (8-9)
 =1

and the plastic rate-of-deformation and plastic spin take the form
ns ns
D = 
p
ij
( )
S( )
(ij ) , W =  ( ) S[(ij])
ij
p
(8-10)
 =1  =1

Thus, when the orientations of all slip systems within the grain are known, the plastic rate-
of-deformation and plastic spin are uniquely defined by the plastic shear strain rates  ( )
on each slip system.
The plastic deformation of metals is assumed to preserve the volume, i.e., plastic
incompressibility is obeyed. The plastic volumetric strain rate is defined by the trace (i.e.,
the sum of the diagonal components) of the plastic rate-of-deformation tensor
 vp  Dkkp = D11p + D22p + D33p (8-11)

Plastic incompressibility implies that  vp is equal to zero. This is indeed the case as seen by
combining in turn Equations (8-11), (8-10)1 and (8-8)1

8-4
 ns ns
  D = 
v
p p
kk
( ) ( )
S
( kk ) =  ( )mk( ) nk( ) = 0 (8-12)
 =1  =1

The latter equality stems from the fact that the vectors mi( ) and ni( ) defining the slip
direction and slip plane for slip system  are orthonormal per definition. Since the plastic
strain-rate tensor has zero trace, it is a deviatoric tensor.

8.2.2 Yield condition and normality

The next question we have to consider is: under what conditions will plastic slip occur
on a given slip system? It is usually postulated that slip will commence on a slip system
when the magnitude of the resolved shear stress on this slip plane and in this slip direction
reaches a critical value, independent of the other stress components on the lattice plane. This
is a statement of Schmid’s law.

Figure 8-3: Slip on the slip system with slip normal n( ) and slip direction m( ) caused by
the traction force t ( ) and the resulting resolved shear stress  ( ) . The normal stress  ( ) is
assumed to have no influence on the slip activity.
The resolved shear stress is the projection of the traction vector t ( ) = ti( )ei onto the
slip plane along the slip direction, i.e.,
 ( ) = t (j )m(j ) = ni( ) ij m(j ) =  ij S((ij)) (8-13)

where the last equality follows from Equation (8-5) and the symmetry of the stress tensor.
The calculation of the resolved shear stress based on the traction force t ( ) on the slip plane
n( ) is illustrated in Figure 8-3. It is noted that the normal stress  ( ) acting on the slip
plane is assumed to have no influence on the plastic slip. Schmid’s law is then expressed as:
slip occurs on slip system  as soon as the magnitude of  ( ) reaches the critical value
 c( ) . Since the resistance to plastic slip depends on the previous plastic straining of the
crystal, i.e., the crystal work-hardens when deformed plastically because the dislocation
density increases,  c( ) is assumed to depend on the plastic strain history and to vary between
slip systems.

8-5
 If we for now assume that the plastic behaviour of the crystal is rate independent, the
yield condition for the single crystal may be written as
f ( ) =  ( ) −  c( ) = 0,  = 1, 2,..., ns (8-14)

or explicitly in terms of the stress components

f ( ) =  ij S((ij)) −  c( ) = 0,  = 1,2,..., ns (8-15)

This is a set of ns equations and slip occurs on systems for which Equation (8-14) is fulfilled.
If f ( )  0 for a given slip system  , there is no slip activity on this system, while f ( )  0
is inadmissible. The latter condition implies that for rate-independent behaviour the
magnitude of the resolved shear stress is bounded from above by its critical value. Equation
(8-15) defines a yield surface in the space of the stress components, which might be
considered as a multi-dimensional vector space. The elastic region bounded by the ns
hyper-planes is a convex domain in this space. When single slip occurs, i.e., there is plastic
deformation on one slip system only, the stress vector is located on the active slip plane. For
multiple slip, several slip planes are active and the only possibility is that the stress vector
is positioned in the vertex made by the intersecting hyper-planes (see Figure 8-4).
The gradients to the ns hyper-planes are obtained as the partial derivatives of f ( )
with respect to the stress  ij and thus

f ( )
f ij( )  = S((ij)) sgn( ( ) ) (8-16)
 ij

Here we have, in addition to Equation (8-15), used that |  ( ) | =  ( ) sgn ( ( ) ) , where

sgn ( •) denotes the sign of the variable ( • ) . Comparing Equation (8-16) with Equation
(8-10)1, we find that
ns
Dijp =   ( )fij( ) where  ( ) = sgn( ( ) ) ( ) (8-17)
 =1

In this equation,  ( ) are proportionality constants known as the plastic parameters. Since
the gradient is normal to the surface in the actual point, Equation (8-17) implies that the
plastic rate-of-deformation tensor is normal to the yield surface. This is the normality rule
in metal plasticity.

8-6
Figure 8-4: Yield surface with vertex and single slip (left) and double slip (right).

The cases of single slip and double slip are illustrated in Figure 8-4, where the hyper-
planes representing yielding on slip systems 1 and  2 are included. In the case of single
slip on 1 , the stress is located on the hyper-plane defined by f (1 ) = 0 and the plastic strain
increment d ijp = Dijpdt is normal to this hyper-plane. For slip on both systems 1 and  2 ,
the stress has to be located in the vertex defined by the two hyper-planes f ( 1) = 0 and
f (2 ) = 0 . The plastic strain increment has then a contribution from each of the two active
p ( 1 )
slip systems and is calculated as d  ijp = d  ij + d  ijp (2 ) where d  ijp ( ) =  ( )fij( )dt . It is
seen that in double slip the direction of the plastic strain increment is within the hyper-cone
defined by the normal vectors of the intersecting hyper-planes.
In the following, we will establish the principle of maximum plastic power, which is of
great importance in phenomenological theories of plasticity, and the complementary
minimum principle. The latter principle is used in finding kinematically as well as
physically admissible sets of active slip systems for a given deformation rate in the rate-
independent theory of single crystal plasticity.

8-7
8.2.3 Principle of maximum plastic power
The plastic dissipation produced in the crystal during plastic flow reads
ns ns
Dp =  ij D =   S
p
ij
( ) ( )
ij (ij ) =  ( ) ( )  0 (8-18)
 =1  =1

By assuming plasticity to be a fully dissipative process, we obtain the reasonable result that
the resolved shear stress  ( ) and the plastic shear rate  ( ) on system  are expected to
have the same sign.
Let  ij be the stress tensor that produces the plastic deformations in the crystal and let
 be any other stress that does not violate the yield condition. Then from Equation (8-18)
*
ij

we obtain
ns

( ij −  ij ) Dij =  (
* p ( )
−  *( ) )  ( ) (8-19)
 =1

where  *( )  ni( ) ij*m(j ) has been introduced. Consider the slip system  . Let  ( ) and
 ( ) be positive for slip in the direction of mi( ) and negative in the direction of − mi( ) .
Then for  ( )  0 , we have
 ( ) −  *( ) =  c( ) −  *( )  0 (8-20)

When  ( )  0 , we have
 ( ) −  *( ) = − c( ) −  *( )  0 (8-21)

since per assumption − c( )   *( )   c( ) . We have then proved that the product
( ( ) −  *( ) )  ( ) is positive or zero for all slip systems  , and we obtain the important
result of Bishop and Hill [6]:

( ij −  ij* ) Dijp  0 (8-22)

This is the principle of maximum plastic power which in word reads: Among all admissible
stress states  ij* in the elastic range (with the points on the yield surface included), the actual
stress state  ij (at which the plastic rate-of-deformation is created) produces the maximum
plastic power:  ij Dijp   ij* Dijp .

Recall that we have already used this principle in the phenomenological theory of
plasticity presented in Chapter 7 of Part I to establish the normality rule and the convexity
of the yield surface.

8-8
8.2.4 Complementary minimum principle
As shown by Bishop and Hill [6], it is possible to obtain a complementary minimum
principle to the principle of maximum plastic power. To this end, let Dijp be the prescribed
plastic rate-of-deformation and  ij the stress that produces this plastic rate-of-deformation
by a set of slip rates  ( ) . Let  *( ) be any other set of slip rates that are kinematically
equivalent to the prescribed plastic rate-of-deformation, but are not necessarily achievable
by any stress fulfilling the yield condition. Since the two sets of slip rates are kinematically
equivalent, the principle of virtual velocities requires that
ns ns
 ij Dijp =  ( ) ( ) =  ( ) *( ) (8-23)
 =1  =1

For the set of slip systems producing the plastic rate-of-deformation for the given stress
state, the resolved shear stresses must fulfil the yield condition: |  ( ) | =  c( ) . On the
contrary, for the kinematically equivalent set of slip rates, the resolved shear stresses should
be admissible, i.e., they should not violate the yield condition: |  ( ) |   c( ) . Thus, we obtain
the result
ns ns ns ns

 ( ) ( ) =  c( ) |  ( ) | and
 =1  =1
 ( ) *( )   c( ) |  *( ) |
 =1  =1
(8-24)

The complementary minimum principle follows from combination of Equations (8-23) and
(8-24), and reads
ns ns

 c( ) |  ( ) |   c( ) |  *( ) |

 =1  =1
(8-25)

Hence, the plastic power expended by the physically possible slip rates producing a
given plastic rate-of-deformation is less than that for a set of slip rates which is only
kinematically possible. In particular, if we assume that the critical resolved shear stress takes
the same value for all slip systems, we get
ns ns



=1
| ( )
|   |  *( ) |
 =1
(8-26)

In words: the sum of the absolute values of the physically possible slip rates producing
a given plastic rate-of-deformation is less than that for a set of slip rates which is only
kinematically equivalent. In the case that more than one set of slip rates is physically
possible the sums of their absolute values are equal.
Owing to the symmetry and deviatoric nature of the plastic strain-rate tensor, it has
only 5 independent components. In general, since ns is greater than 5, there may be several
sets of slip rates that are kinematically as well as physically admissible. These sets are
determined from Equation (8-10)1 using Equation (8-26) as a linear constraint. The fact that
there may be several solutions to the problem of finding the active slip systems is called the
Taylor ambiguity. There are various ways to circumvent this problem. Perhaps the most
obvious one is to assume that the critical resolved shear stress has a statistical distribution
and thus varies between the different slip systems (see e.g. [7]). This will in most cases lead
to unique solutions. Another way out is to adopt a rate-dependent formulation, where
Schmid’s law is replaced by a power-law relationship between the resolved shear stress and
8-9
the slip rate on each slip system. This also leads to unique stress-strain relationships. The
rate-dependent formulation is discussed in Section 8.2.6

8.2.5 Work-hardening
Based on experimental observations, the following two conclusions are drawn. First,
slip systems are hardened by slip on other systems, whether they themselves are active or
not. This mechanism is called latent hardening. Second, the latent hardening rate is at least
comparable in magnitude to the self-hardening rate, i.e., the hardening of the active system
by its own plastic slip.
Work-hardening is introduced by making the critical resolved shear stress  c( ) a
function of the plastic strain. A generic relation for work-hardening reads [1][2]
ns
 c( ) =  h  (  ) (8-27)
 =1

where h are the hardening moduli, assumed to be non-negative numbers. In particular,
h (no sum on  ) is the self-hardening rate on system  , whereas h (   ) is the
latent-hardening rate of system  caused by slip on system  . The hardening moduli h
depend somehow on the structure of the material, and internal variables may be introduced
to describe this dependence.
An example of a very simple work-hardening rule is given by [8]
n
 c( ) =  (  )  q  (  ) (8-28)
 =1

where h =  ( ) q and the latent hardening matrix q is defined as

 q for   
q =  (8-29)
1 for  = 
Typically, the latent hardening coefficient q is set equal to 1.4, which means that latent
hardening is usually stronger than the self-hardening. The accumulated plastic strain  is
defined as
t ns
 =   |  ( ) | dt (8-30)
0  =1

The work-hardening modulus  =  (  ) is the derivative with respect to  of a master

hardening curve defined by the extended Voce hardening rule, namely
nV
  k 
 c (  ) =  0 +  k  1 − exp  −  (8-31)
k =1   k 
Here  0 is the initial critical resolved shear stress, whereas k and  k are the initial
hardening rate and saturated value of hardening term k = 1,2,..., nV . Thus, the work-
hardening modulus is obtained as

8-10
 d c (  ) nV   
 ( ) = = k exp  − k   (8-32)
d k =1  k 
The initial hardening modulus is  ( 0) = kV=1k , while the saturation value of the slip
n

resistance is  c ( ) =  0 + kV=1 k .
n

8.2.6 Rate-dependent behaviour

Above we have assumed that the material behaviour is rate independent, i.e., the
strength, flow and hardening of the material is independent upon the plastic strain rate at
which the plastic deformation is produced. Most metals are rate dependent to lesser or
greater extent, and the rate dependence is often increasing with increasing temperature. Thus,
it is useful to develop a viscoplastic formulation of crystal plasticity.
In viscoplasticity it is assumed that the slip rate on a given slip system depends uniquely
on the state of the material defined by the stress state and the structural parameters
determining e.g. the work-hardening. A simple viscoplastic model is obtained by taking the
shearing rate on slip system  to be defined by the constitutive relation [1][2]
1
 ( ) m
 ( ) =  0 ( ) sgn ( ( ) ) (8-33)
c
where  0 is a reference shearing rate, m is the instantaneous strain rate sensitivity of the
material and  c( ) represents the strength of slip system  , as defined by the work-
hardening law. It is noted that slip will occur for all resolved shear stresses different from
zero, and, as such, no yield condition exists. However, for realistic rate sensitivities the
shearing rate will be negligible when |  ( ) |   c( ) , since m often takes a relatively low
value for metals.
It is important to check that the plastic dissipation is non-negative for rate-dependent
material. Combining Equation (8-18) with Equation (8-33), we find
1
ns ns
 ( ) m
D p =  ( ) ( ) =   0 |  ( ) | ( ) 0 (8-34)
 =1  =1 c

where |  ( ) | = sgn ( ( ) ) ( ) was used. Hence, it has been shown that the plastic dissipation
is non-negative for the viscoplastic single crystal model. A percentage of the plastic work
will be stored in the material due to storage of new dislocations – the dislocation density
increases as the material is work-hardened and since the dislocation distorts the crystal
lattice in its neighbourhood, some elastic energy is stored in the material. This topic is,
however, outside the scope of these lecture notes.

8-11
8.2.7 Lattice rotation and elastic behaviour
In crystal plasticity, the lattice is assumed to rotate according to the elastic spin Wije ,
which includes spin from elastic deformations and rigid body rotation. Recall that the plastic
spin is uniquely defined by plastic slip. The slip direction and slip normal are thus updated
by
mi( ) = Wijem(j ) , ni( ) = Wijen(j ) (8-35)

Since both the slip direction and slip normal rotates in the same way, they remain
orthonormal throughout the deformation. The rotation of the individual grains leads to an
evolution of the texture of the polycrystal.
Referring to Chapter 2, a corotational coordinate system is defined by the rotation
tensor  with evolution equation
ij = Wikekj (8-36)

The fixed basis ( ei ) and corotational basis ( eˆ i ) are then related by

eˆ i =  ji e j , ei = ij eˆ j (8-37)

The transformation rules between the two basis systems are defined in Chapter 2. Expressed
in the corotational coordinate system, the components mˆ i( ) and nˆi( ) of the unit vectors
m( ) and n( ) , respectively, are invariant to the lattice rotations. It follows that the
components of the Schmid tensor S ( ) in the corotational system, Sˆij( ) = mˆ i( )nˆ (j ) , are also
constant. We will take advantage of this below to develop an efficient integration algorithm
for single crystal plasticity. We also have the useful relations (see Chapter 2)
mi( ) = ij m
ˆ (j ) , ni( ) = ij nˆ (j ) (8-38)

The corotational stress and rate-of-deformation tensors by

ˆij = ki kllj , Dˆ ij = ki Dkllj (8-39)

which are invariant to rigid body motion as they are referred to a coordinate system that
rotates with the rigid body spin. Then, the rate form of generalized Hooke’s law reads as
ˆ ij = Cˆijkl Dˆ kle (8-40)

where Cˆ ijkl is the fourth-order elasticity tensor. When defined with regards to the
corotational system, the components of the elasticity tensor C are constant. The single
crystal has certain material symmetry depending on the lattice system, but the elastic
relation is generally anisotropic [1][2].

8-12
Box 8-1 Rate-dependent single crystal plasticity

Domain of validity:
• Infinitesimal elastic strains
• Finite plastic strains and rotations
• Elastic and plastic anisotropy

Main equations:

• Additive decomposition:
Dij = Dije + Dijp , Wij = Wije + Wijp

• Generalized Hooke’s law:

 ijJ = Cijkl Dkle
 ijJ   ij − Wike kj −  ikW jke
Cijkl = im jnkplq Cˆ mnpq
• Resolved shear stress:
 ( ) =  ij S((ij))
• Constitutive relation:
1
 ( ) m
 ( )
=  0 ( ) sgn ( ( ) )
c
• Plastic rate-of-deformation and spin tensor:
ns ns
D = 
p
ij
( ) ( )
S
(ij ) , W =  ( ) S[(ij])
ij
p

 =1  =1

• Work-hardening rule:
ns
 ( )
c =  h |  (  ) |
 =1

• Plastic work rate per unit volume:

ns
D p =  ij Dijp =  ( ) ( )  0
 =1

• Evolution of slip system:

mi( ) = Wijem(j ) , ni( ) = Wijen(j )

8-13
Combining Equation (8-40) with Equation (8-39), we get after some straightforward tensor
manipulations (see also Chapter 6)
 ijJ = Cijkl Dkle with  ijJ   ij − Wike kj −  ikWjke (8-41)

where  ijJ is now the Jaumann stress rate based on the elastic spin, and Cijkl is the elasticity
tensor referred to the fixed base system
Cijkl = im jnkplqCˆ mnpq (8-42)

It is noted that Cijkl depends on the rotation of the lattice because of the anisotropic elastic
properties of the crystal.
The constitutive equations of the rate-dependent single crystal plasticity theory are
compiled in Box 8-1.

8.2.8 Numerical integration

A simple forward-Euler algorithm for the temporal integration of the rate-dependent
single crystal plasticity model presented in Box 8-1 is established in the following, using a
corotational formulation. The algorithm is only conditionally stable and very small time
steps are required to obtained accurate and stable results.
We start by defining the incremental strain and rotation tensors in the fixed, global
coordinate system as  ij  Dij t and ij  Wij t . Both tensors are additively
decomposed into elastic and plastic parts. Then we assume that the variables  ij , n ,  n ,  c(, n)
and ij ,n at time tn have been determined, and that the incremental strain and rotation
tensors  ij , n +1  Dij ,n +1/2 tn +1 and ij ,n +1  Wij ,n +1/2 tn +1 during the time step tn+1 = tn+1 − tn
is given. The slip systems are defined by unit vectors n( ) and m( ) with constant
components mˆ i( ) and nˆi( ) in the corotational coordinate system. These vectors are
predefined for a given crystal structure. Also the components Sˆ ( ) of the Schmid tensor
ij
( )
S in the corotational system are predefined and independent of the deformation.
The stress-update algorithm has the following steps (see also Zhang et al. [9] for a
similar forward-Euler algorithm):
1. Transform the incremental strain tensor into the corotational coordinate system
using the rotation tensor
tn : at
ˆij ,n +1 = ki ,n  kl ,n +1lj ,n (8-43)

2. Calculate the resolved shear stresses at tn :

 n( ) = ˆij , n Sˆ((ij)) (8-44)

3. Estimate the shear strain increments during the time step tn +1 by use of known
quantities at tn :

8-14
 1

 ( ) m
 n(+1)  tn +1 0 n
sgn ( n( ) ) (8-45)
 ( )
c, n

4. Update accumulated plastic strain:

ns
n+1 = n +   n(+1) (8-46)
 =1

5. Calculate the incremental plastic strains and rotations during the time step tn +1 in
the corotational coordinate system:
ns ns
ˆijp, n +1 =   n(+1) Sˆ((ij)) , ˆijp, n +1 =   n(+1) Sˆ(ij) (8-47)
 =1  =1

6. Calculate the Cauchy stress tensor at tn +1 in the corotational coordinate system:

ˆ ij , n +1 = ˆ ij , n + Cˆ ijkl ( ˆkl , n +1 − ˆklp , n +1 ) (8-48)

7. Update the critical resolved shear stresses:

ns
 c(, n)+1 =  c(, n) +  h ( n )  n( +1) (8-49)
 =1

8. Calculate the incremental elastic rotation during the time step tn +1 in the fixed,
global coordinate system:
ije, n+1 = ij ,n+1 − ik ,n ˆ klp, n+1 jl ,n (8-50)

9. Update the rotation tensor using a 2nd order update (cf. Chapter 6):

ij ,n +1 = ( ik − 12 ike ,n +1 ) ( + 12 kle ,n+1 ) lj ,n

−1
kl (8-51)

10. Update the orientations of the slip systems:

mi(,n+) 1 = ij ,n+1mˆ (j ) , ni(,n+) 1 = ij ,n+1nˆ (j ) (8-52)

11. Transform stress tensor into the fixed, global coordinate system:
 ij ,n +1 = ik ,n +1ˆ kl ,n +1 jl ,n +1 (8-53)

As already mentioned, the forward-Euler scheme is only conditionally stable and very
small time steps should be used. One possibility here is to use sub-stepping to limit the strain
increment. There are, however, several other explicit as well as implicit integration
algorithms available for single crystal plasticity in the literature, see e.g. Zhang et al. [9].

8-15
8.3 Polycrystal plasticity
8.3.1 Crystal orientation and texture
In polycrystal plasticity, we consider an aggregate of single crystals (or grains). Each
grain within the aggregate has its own orientation with respect to the orientation of the
aggregate. It is important to establish a procedure for representing the grain orientations in
a convenient way for computations of the plastic behaviour of the crystal.
Let e ic denote the basis vectors of a Cartesian coordinate system defining the
orientation of a given crystal. The relation between the basis vectors of the global system
e i and the local system of the crystal eic is then given by

eic = aicj e j (8-54)

where aijc  eic  e j are the cosines of the angles between the unit vectors e ic and e j . The
components of the transformation matrix aijc may be determined as a product of three simple
rotations by means of the Euler angles 0  1  2 , 0     and 0  2  2 (see Engler
and Randle [10] for details).
To this end, consider an orthonormal basis e0j initially aligned with e j . The
transformation matrix aijc taking a tensor from e j into e ic is then obtained via the following
procedure (see also Figure 8-5):
• Rotate e0j by an angle 1 about the e 03 axis to get ei , so that e3 is aligned with e 03 :

 cos 1 sin 1 0
ei = a e , aij  =  − sin 1 cos 1 0
0
(8-55)
ij j  
 0 0 1

• Rotate ej an angle  about the e1 axis to get ei , so that e1 is aligned with e1 :

1 0 0 
ei = aijej , aij  = 0 cos  sin  
 (8-56)
 
0 − sin  cos  

• Rotate ej an angle  2 about the e3 axis to obtain ei which is aligned with the local
basis vectors e ic :

 cos  2 sin  2 0
ei = aijej ,  aij =  − sin  2 cos  2 0 (8-57)
 
 0 0 1

8-16
The compound transformation from e j to e ic is then obtained as

eic = aikakl alj e j = aicj e j (8-58)

where matrix multiplication gives [10]

 cos 1 cos  2 − sin 1 sin  2 cos  sin 1 cos 2 + cos1 sin  2 cos  sin  2 sin  
a  = − cos 1 sin  2 − sin 1 cos 2 cos  − sin 1 sin  2 + cos1 cos 2 cos  cos 2 sin   (8-59)
c 
ij  
 sin 1 sin  − cos 1 sin  cos  

The definition of the Euler angles is illustrated in Figure 8-5. It is important to note that
other definitions of the Euler angles have been proposed in the literature. The polycrystal
with the global coordinate system ( x1 , x2 , x3 ) and some of the individual grains with their
crystallographic coordinate systems ( x1c , x2c , x3c ) are illustrated in Figure 8-6.

Figure 8-5: Definition of the Euler angles.

The texture of a material is defined by the distribution of the crystallographic

orientations of a polycrystalline aggregate of a representative size, where the
crystallographic orientation, i.e., the orientation of the crystal with respect to the global
frame, is conveniently defined by the set of Euler angles (1, ,2 ) . The texture of the
polycrystal is then the statistical distribution of the grain orientations. The representation of
crystallographic texture is most frequently given by the orientation distribution function
(ODF). Let dV ( g ) / V be the volume fraction of crystals having their orientation within
dg around g , where g denotes a given crystal orientation, e.g. defined by Euler angles [2].
The ODF, denoted f ( g ) , is then defined by [2]

dV ( g )
= f ( g ) dg (8-60)
V
If f ( g )  1 the grains have random orientation and we say that the polycrystal has random
texture. If this is not the case, the polycrystal is textured and the mechanical behaviour is
expected to exhibit anisotropy.

8-17
 During plastic deformation of the polycrystalline aggregate, the grains rotate (i.e., the
crystal orientation evolves) and thus the texture of the aggregate will change. The texture
evolution is uniquely defined from the single crystal plasticity model. The evolution of the
crystal orientation is then given by
eic =   e 0c i (61)

where e 0c i denotes the orthonormal basis defining the grain orientation in the initial
configuration and the rotation tensor  is generated by the elastic spin of the grain.

x2

Polycrystal

x1
x3

x2c x2c
x2c x1c
x3c

x1c
x3c
x1c
x3c
Individual grains

Figure 8-6: Polycrystal with global coordinate system ( x1, x2 , x3 ) and individual crystals (or
grains) having their own crystallographic coordinate systems ( x1c , x2c , x3c ) (after [11]).

8-18
8.3.2 Taylor’s approach
When the texture of a crystal aggregate is determined, the question arises how to
establish the properties of the aggregate given the behaviour of each and every single crystal.
Homogenization techniques have thus to be used. There are basically two ways, either
analytically or computationally. One example of an analytical homogenization technique is
the Taylor approach, which will be briefly described below. The crystal plasticity finite
element method (CP-FEM) is an example of a computational homogenization procedure in
which each grain in the aggregate is discretized by finite elements [8]. In both cases, it is
important to select an aggregate consisting of a representative number of grains describing
the measured texture of the material (typically around 1000 grains).
In the Taylor approach, originally developed by Taylor in 1938 [12], all grains are
assumed to be subjected to the same homogeneous deformation as the polycrystalline
aggregate, i.e.,
DijI = Dij , I = 1,2,..., ng (8-62)

where Dij and DijI are the rate-of-deformation tensors of the polycrystalline aggregate and
grain I , respectively, and n g is the number of grains in the aggregate. For every grain I ,
the rate-dependent single crystal plasticity theory in Box 8-1 is adopted. The volume-
averaged Cauchy stress of the polycrystal  ij is thus obtained as [2]
n
1 g
 ij = V I ijI (8-63)
V I =1
where  ijI and V I are the Cauchy stress and the volume of grain I , respectively, and V is
the volume of the aggregate of all the grains.
The Taylor approach can be considered as a homogenization technique where every
material point of the solid is assumed to be composed of a finite number n g of grains to
which a texture is associated. It is important to note that the Taylor approach does not ensure
stress equilibrium at the grain boundaries, and further the stress and strain gradients present
within the grains are not accounted for. However, by combining the Taylor approach with
the rate-dependent single crystal plasticity model, the macroscopic stress-strain response
(e.g. work-hardening curves for various straining conditions and the anisotropic yield
surface) of an aggregate of many grains can be computed when the texture of the aggregate
and work-hardening at the grain level have been determined.

8-19
8.4 References
[1] A.S. Khan, S. Huang. Continuum Theory of Plasticity. Wiley-Interscience Publication
(1995).
[2] U.F. Kocks, C.N. Tomé and H.R. Wenk. Texture and Anisotropy. Cambridge
University Press (1998).
[3] K.S. Havner. The theoretical behaviour of a polycrystalline solid as related to certain
general concepts of continuum plasticity. Int. J. Solids Structures 5:215-226 (1969).
[4] J.R. Rice. Inelastic constitutive relations for solids: an internal variable theory and its
application to metal plasticity. J. Mech. Phys. Solids 19:433-455 (1971).
[5] R.J. Asaro. Micromechanics of crystals and polycrystals. Adv. Appl. Mech. 23:1-115
(1983).
[6] J.F.W. Bishop, R. Hill. A theory of the plastic distortion of a polycrystalline aggregate
under combined stresses. Phil. Mag. (Ser. 7) 42:414-427 (1951).
[7] P. Van Houtte, S. Li, M. Seefeldt, L. Delannay. Deformation texture prediction: from
the Taylor model to the advanced Lamel model. Int. J. Plasticity 21 (2005) 589–624.
[8] A. Saai, S. Dumoulin, O.S. Hopperstad, O.G. Lademo. Simulation of yield surfaces
for aluminium sheets with rolling and recrystallization textures. Computational
Materials Science 67 (2013) 424–433.
[9] K. Zhang, O.S. Hopperstad, B. Holmedal, S. Dumoulin. A robust and efficient
substepping scheme for the explicit numerical integration of a rate-dependent crystal
plasticity model. International Journal for Numerical Methods in Engineering 99
(2014) 239–262.
[10] O. Engler, V. Randle. Introduction to texture analysis. Macrotexture, microstexture,
and orientation mapping. Second edition. CRC Press (2010)
[11] O.S. Hopperstad, S. Dumoulin, A. Saai. A brief introduction to crystal plasticity.
SIMLab Technical Report, NTNU, 2010.
[12] G.I. Taylor. Plastic strain in metals. J. Inst. Metals 62:307-24 (1938).

8-20
9 Localization of deformation*
9.1 Introduction
Softening due to damage evolution or adiabatic heating may lead to non-unique
solutions of the boundary value problem for elastic-plastic solids, which means that other
than the fundamental solution are possible. Indeed, experiments show that localization of
the strain in a narrow band (typically a shear band) is a frequent precursor to failure. Figure
9-1 shows examples of shear bands in Weldox steel plates struck normally by a blunt
impactor. In the following, we will consider the criterion for localization of deformation
into a planar band. We suggest References [1]–[5] and references therein for further reading.

Figure 9-1: Light microscopy images of shear bands in 12 mm thick plates of Weldox 460
E, Weldox 700 E and Weldox 900 E struck at high velocity by a blunt impactor [6].

We will address localization in both rate-independent and rate-dependent elastic-plastic

solids, and we will find that rate dependence plays an important role. The constitutive
equations of the rate-independent material are defined by (see Chapter 6)
 ijJ   ij  Wik kj   ikWjk  Cijkl
t
Dkl (9-1)
t t
where Cijkl is the tangent modulus tensor. In the bifurcation analysis in Section 9.2, Cijkl is
assumed to be a state function. The material behaviour is then incrementally linear, and the
material behaviour is said to be that of a linear comparison solid. These equations are
assumed to represent the behaviour of elastic-plastic materials with strain softening due to
damage evolution. It should be noted here that an elastic-plastic material is not
incrementally linear due to the loading/unloading conditions. However, the linear
comparison solid gives good estimates on loss of uniqueness and localization in most cases
[7]. In the imperfection analysis in Section 9.4, we consider instead the ordinary elastic-
plastic material behaviour, including the loading/unloading conditions. The rate-dependent
material is defined by (see Chapter 6)
 ijJ   ij  Wik kj   ikWjk  Cijkl Dkl  Qijp (9-2)

where Cijkl is the elastic modulus tensor. The tensor Qijp    σ, R  Cijkl g  σ  /  kl
represents the stress relaxation caused by viscoplasticity. It depends on the state of the
material, as defined by the stress tensor and the internal variables, but not on the rate
quantities. Evidently Qijp vanishes in the elastic domain. These equations represent the

9-1
behaviour of elastic-viscoplastic materials with strain softening due to damage or other
possible phenomena leading to strain softening (such as adiabatic heating).

9.2 Bifurcation analysis

L, σ
n
L, σ


L, σ

Figure 9-2: Body with discontinuous velocity gradient field across a planar band  with
unit normal vector n .

We consider a homogeneous and homogeneously deformed solid which is subjected to

a uniform stress field  ij and investigate the possibility of a bifurcation. Specifically, we
search for conditions “for which continued deformation may result in an incipient non-
uniform field in which deformation rates vary with position across a planar band but remain
uniform outside the band” [2]. The planar band is denoted  and the unit normal to the band
is n , see Figure 9-2. We will adopt the notation  and  for the same physical quantity
outside and inside the band  , respectively, while      represents the difference of
the local field  at a point x inside the band and the uniform field  outside the band.
The homogeneous velocity gradient field Lij leads to the homogeneous stress field  ij ,
which is assumed to fulfil the quasi-static equilibrium equations with zero body force. In
the absence of body forces, the equilibrium equations read as
 ij
0 (9-3)
xi
Continuing stress equilibrium at a fixed position x in space in the current configuration
implies that
   ij     ij 
    ij  vk 0 (9-4)
xi  t  xi  xk 
where the definition of the material time derivative was used to obtained the first equality,
see Chapter 2. But as the stress field is assumed to be uniform at incipient bifurcation,
 ij / xk  0 and the equations for continuing stress equilibrium reduces to

9-2
  ij
0 (9-5)
xi
The velocity field v is assumed to remain continuous, at least initially, and the partial
derivatives of v in the directions parallel to the band remain uniform [2]. The velocity field
inside the band then takes the form
vi  x, t   vi  x, t   vi  , t  (9-6)

where   x  n  xk nk is the coordinate along n and vi  , t  is the non-uniformity of the

velocity field inside the band. To obtain a homogeneous velocity gradient field Lij , the
velocity field outside the band has to vary linearly with position, viz.
vi  x, t   Lij  t   x j  x 0j   vi0  t  (9-7)

where the reference point x 0 moves with the translational velocity v 0 . The partial
derivatives of the velocity field inside the band are
vi vi vi vi vi  vi vi
      nj (9-8)
x j x j x j x j  x j x j 
The velocity gradient field follows as
Lij  Lij  gi n j (9-9)

where g  v /  is a function of the coordinate  across the band and vanishes outside
the band. Let m be the unit vector parallel to the vector g defining the type of non-
uniformity. If m is orthogonal to n , the band is a shear band; if m is parallel with n , the
band is said to be a dilatation band. In-between these two extreme cases, the band represents
a mixed mode of localization. The non-uniformities in the rate-of-deformation and spin
tensors inside  are then
Dij  1
2 g n
i j  g j ni  , Wij  1
2 g n
i j  g j ni  (9-10)

From the kinematics it follows that the stress rates are uniform outside the band and
depend on the coordinate  inside the band, viz.

 ij  , t    ij  t    ij  , t  (9-11)
Clearly, continuing stress equilibrium is fulfilled outside the band (the stress rates are
uniform), while inside the band Equation (9-5) gives
 ij  ij   ij   ni  ij 
  ni  0 (9-12)
xi  xi  
Using the definition of the non-uniformity of the stress rate field, we get [2][5]
ni ij  ni ij (9-13)

which expresses continuing equilibrium of the tractions across the band, as illustrated in
Figure 9-3.

9-3
Figure 9-3: Continuing stress equilibrium across the planar band, where the coordinate axes
are defined as xn  x  n and xm  x  m with m embedded in the band so that m  n  0 .
Note that  equals the coordinate xn across the planar band.

To proceed, we need to introduce the constitutive relations. Assuming a rate-

independent material, as defined by Equation (9-1), the non-uniformity in the stress is
expressed as
 ij  Cijkl
t
Dkl  Wik kj   ik Wjk (9-14)
where it was used that the stress is uniform at incipient bifurcation and that the tangent
modulus tensor is assumed to be a function of state (i.e., the linear comparison solid is
assumed). Inserting Equation (9-10), we get
 ij  Cijkl
t
g k nl  12  gi nk kj  g k ni kj  g j nk ik  g k n j ik  (9-15)

where the minor symmetry of the tangent modulus tensor was employed. Combination of
Equations (9-13) and (9-15), gives after some straightforward manipulations the bifurcation
condition as
Atjk  n  gk  0 (9-16)

Here, the acoustic tensor At  n  is introduced as

Atjk  n   niCijkl
t
nl  R jk (9-17)

with the tensor R defined by

2 R jk  n j  ni ik    nl lj  nk   ni il nl   jk   kj (9-18)

Localization is assumed to occur when Equation (9-16) has a non-trivial solution for the
first time during the deformation process. A non-trivial solution exists for the actual state

9-4
when the acoustic tensor At  n  has at least one zero eigenvalue, i.e., when its determinant
equals zero
det At  n   0 (9-19)
We have thus arrived at the condition for localized bifurcation of the rate-independent
hypoelastic-plastic material.
At a given state of stress and internal variables, a computational procedure may be used
to check for localization. The problem is to find the unit vector n that minimizes the
function h  n   det At  n  . To this end, it is convenient to express n in terms of the
spherical angles, namely
n  cos  sin e1  sin  sin e2  cos e3 (9-20)

where the azimuthal angle   0, 2  and the polar angle  0, 2  are defined in Figure
9-4. The search for the critical orientation of the band is then done by sweeping through the
range of variation 0, 2   0, 2  of the spherical angles  ,   at suitable increments. One
way to perform the search for the critical orientation of the band is proposed in [8]. If
localization is established for a given unit normal n , the unit vector m , characterizing the
mode of failure, is found as the zero eigenvector of the acoustic tensor At  n  .
Analytical solutions for various constitutive relations, such as the J2 flow theory,
pressure-sensitive dilatant plasticity models and the Gurson model (approximate) are given
in References [2]−[5],[9][10]. These solutions show that strain localization is usually not
possible for associated plasticity with positive work-hardening modulus, while non-
associated plasticity makes the solid more prone to localization.

Figure 9-4: Definition of the unit normal n to the planar band in terms of spherical angles.

The situation is different for a hypoelastic-viscoplastic solid with stress-strain

behaviour defined by Equation (9-2). It turns out that bifurcations are precluded for this
class of solids, except for very special (and unlikely occurring) cases. By combining
Equation (9-2) with Equations (9-10) and (9-13), we get

9-5
 Ajk g k  0 (9-21)

where Ajk  ni Cijkl nl  R jk is the elastic acoustic tensor. It was used here that Q p depends
on the state only, and thus has the same value outside and inside the band at inception of
localization. The localization condition for the hypoelastic-viscoplastic solid is then given
by
det A  0 (9-22)
Note that in this case the localization condition is completely independent of plasticity
altogether and is the same as for a hypoelastic solid.
Since the components of the elasticity tensor C is of the order of the elastic modulus
E , which under normal circumstances is much larger than the components of the Cauchy
stress tensor σ , we may usually neglect the contribution from R to the elastic acoustic
tensor. Thus, to a first approximation, the elastic acoustic tensor takes the form
Ajk  ni Cijkl nl . This simplification is equivalent to neglecting the rotational effects on the
stress rate and thus using σ instead of σ J , which is what we would do for small
deformations. For small deformations, we assumed the existence of a positive definite strain
energy function, which implies that U 0  12  ij Cijkl  kl  0  ij  0 , where ε is the
infinitesimal strain tensor, cf. Chapter 5 of Part I. The implication is that C is a positive
definite 4th order tensor. Assuming this to be the case also for hypoelasticity and neglecting
terms of the order of stress divided by the elastic modulus, we have
g j Ajk g k  g j  niCijkl nl  g k  Dij Cijkl Dkl  0  Dij  0 (9-23)

Thus, the elastic acoustic tensor A is positive definite, which implies that det A  0 , and
localization is precluded for the rate-dependent material. We may conclude that for all,
except for very special circumstances where the stress components are of the same order as
the elastic modulus, localization of deformation into a planar band does not take place for a
homogeneous and homogeneously deformed hypoelastic-viscoplastic solid subjected to a
uniform stress field. If the solid is heterogeneous or has initial imperfections, localization
of deformation may take place by accelerated growth of the non-uniformity.

9-6
9.3 Interlude: Material time derivatives of unit vectors
In the next section, we need the expression for the material time derivative of the unit
normal vector to a planar surface. We will now derive the expression for the material time
derivative of the unit vector along a line segment and the unit vector normal to a plane. To
this end, we consider the unit normal vector n to a planar surface in the current
configuration of a deformed solid and the unit vector m along a line segment embedded in
this plane. The two unit vectors m and n are then orthogonal, so that m  n  0 .
We now assume that the solid is subjected to a uniform deformation defined by the
velocity gradient field L . According to our derivations in Chapter 2, the material time
derivative of a line segment s  sm oriented along m is then given by s  L  s . The scalar
s has a fixed value.

Figure 9-5: Rate of change of vectors s  sm and ν   n instantaneously on the unit sphere.

The vector s may be decomposed into components s and s  that are respectively
parallel and perpendicular to the unit vector m , viz.
s  s  s (9-24)
where
s  s  m  m  s m  L  m  m
(9-25)
s   s  s  s L   m  L  m  I   m

The component s changes the length of s but conserves the direction, whereas the
component s  rotates s but conserves the length, see Figure 9-5. It then follows by setting
s  1 that the material time derivative of the unit vector m is given by
m  L   m  D  m  I   m (9-26)

where it was used that the skew-symmetric spin tensor W has diagonal components all
equal to zero. It is evident that m is orthogonal to m .
Let ν   n be a vector directed along the normal vector n , so that ν  s  0 . The scalar
 is fixed. Taking the material time derivative of the scalar product ν  s , we obtain
0  Dt    
D ν  s  ν  s  ν  s  ν  ν  L  s  ν   ν  L  LT  ν (9-27)

9-7
The vector ν has components ν and ν  that are parallel and perpendicular to n , and we
have
ν  ν  ν (9-28)
where
ν   ν  n  n    n  L  n  n
(9-29)
ν   ν  ν    n  L  n  I  LT   n

While the component ν changes the length of ν but conserves the direction, the
component ν  rotates ν but conserves the length, see Figure 9-5. By taking   1 the
material time derivative of the unit normal vector n is obtained by
n   n  D  n  I  LT   n (9-30)

where n  W  n  0 was utilized. It is readily checked that n and n are orthogonal and
further that n  ndt is orthogonal to m  mdt as long as dt is infinitesimal.

9-8
9.4 Imperfection analysis
Obviously strains localize also in rate-dependent materials, even if a bifurcation is
excluded. To analyse this problem, we consider a solid body that is homogeneous and
homogeneously deformed, except that there exists a thin planar band, in which the stress
and strain rates are allowed to take values different from their values outside the band. The
unit normal to the planar band is denoted n . Thus, the situation is as depicted in Figure 9-2
but the band is now assumed to be present in the un-deformed solid. We will assume that
some of the material properties differ slightly inside the band compared with the
surrounding material. As before, let quantities outside and inside the band be denoted ( )
and ( ) , respectively.
We assume as in Section 9.2 that non-uniformities in the velocity gradient field may
occur across the planar band while the velocity field remains continuous. Compatibility then
requires that (cf. Equation (9-9))
Lij  Lij  gi n j (9-31)

where g has the same meaning as in the homogeneous case. Thus

Dij  Dij  12  gi n j  g j ni  , Wij  Wij  12  gi n j  g j ni  (9-32)

Static equilibrium across the band implies that

ni ij  ni ij or ni  ij   ij   0 (9-33)

Taking the material time derivative of the latter equation, we get the condition for continuing
equilibrium in the form
ni  ij   ij   ni  ij   ij   0 (9-34)

According to Equation (9-30), the material time derivative of the unit normal n to the band
reads as
ni   nk Dkl nl  ni  nk Lki (9-35)
Then combination of Equations (9-34) and (9-35) gives
ni  ij   ij   ni Lik  kj   kj   0 (9-36)

or alternatively
ni   ij  Lik  kj   0 (9-37)

To arrive at this relation we used that the band is in static equilibrium, as expressed by
Equation (9-33).
Addressing first the rate-independent elastic-plastic solid, the stress rates outside and
inside the band are given by Equation (9-1) as
 ij  Cijkl
t
Dkl  Wik kj   ikWjk (9-38)

 ij  Cijkl
t
Dkl  Wik kj   ikWjk (9-39)

9-9
Using Equation (9-32), we get after some manipulations
 ij  Cijkl
t
Dkl  Wik kj   ikW jk
(9-40)
 Cijkl
t
gk nl  12  gi nk kj  gk ni kj  g j nk ik  gk n j ik 

Inserting Equations (9-38) and (9-40) into Equation (9-36), we get [9]
Atjk g k  ni  Cijkl
t
 Cijkl
t
 Dkl  nW
i ik  kj   kj   ni Lik  kj   kj  (9-41)

After simplification, the equations for the unknown vector g are obtained as [9]

Atjk gk  ni  Cijkl

t
 Cijkl
t
  ik  lj   lj  Dkl (9-42)

The acoustic tensor At inside the band is given by

Atjk  n   niCijkl
t
nl  R jk (9-43)
with
2 R jk  n j  ni ik    nl lj  nk   ni il nl   jk   kj (9-44)

Equation (9-41) provides a set of three equations for the unknown vector g while the rate-
of-deformation field outside the band is assumed to be prescribed. It is seen that localization
occurs when the acoustic tensor At becomes singular for the first time. Owing to the
imperfections inside the band, this localization may occur at an earlier stage than for the
homogeneous solid.
In the case of the rate-dependent elastic-plastic solid, combination of Equations (9-2),
(9-31) and (9-33) give the equations for the unknown components g k as

Ajk g k  ni  Cijkl

t
 Cijkl
t
  ik  lj   lj  Dkl  ni Qijp  Qijp  (9-45)

where Cijkl is the elastic modulus tensor and Ajk  niCijkl nl  R jk is the elastic acoustic
tensor inside the band. Since Cijkl is assumed to be positive definite, it follows that
det A  0 , except in special cases in which the stress components are of the same order as
the elastic moduli. Localization is then defined as || D || / || D ||   , where || D ||  Dij Dij
is the Euclidean norm of the rate-of-deformation tensor. The interpretation of this is that at
localization all deformation is taking place inside the band while the deformation reaches
saturation outside the band.
The imperfection analysis can be used in the following way. A realistic imperfection is
assumed inside the band, e.g. based on measurements of variations in the porosity of the
material at hand. A strain path is then prescribed outside the band and the constitutive
relations are used to determine the stress and internal variables. Simultaneously, Equation
(9-41) or (9-45) is used to compute the strain inside the band and the associated stress and
internal variables are computed using the constitutive relations. Equation (9-35) is used to
update the band orientation with the deformation. Note that the constitutive relations may
be different inside and outside the band. The strain outside the band is increased until the
localization condition is reached. The calculation has to be repeated with different band
orientations n to find the critical orientation of the band in the same manner as for the

9-10
homogeneous case. By changing the strain path and stress state outside the band, their
influence on the strain to localization may be investigated numerically for given constitutive
behaviours.

9.5 References
[1] N. Saabye Ottosen, M. Ristinmaa. The Mechanics of Constitutive Modeling. Elsevier,
2005.
[2] J.W. Rudnicki, J.R. Rice. Conditions for the localization of deformation in pressure-
sensitive dilatant materials. Journal of the Physics and Mechanics of Solids 23 (1975)
371–394.
[3] J.R. Rice. The localization of plastic deformation. In: Theoretical and Applied
Mechanics (Proceedings of the 14th International Congress on Theoretical and
Applied Mechanics, Delft, 1976, edited by W.T Koiter) Vol. 1, pp. 207–220. North-
Holland Publishing co., 1976.
[4] A. Needleman, J.R. Rice. Limits to ductility set by plastic flow localization. In:
Mechanics of Sheet Metal Forming (Edited by D.P. Koistinen and N-M. Wang) pp.
237–267. Plenum Press, New York, 1978
[5] J.R. Rice, J.W. Rudnicki. A note on some features of the theory of localization of
deformation. International Journal of Solids and Structures 16 (1980) 597–605.
[6] J.K. Solberg, J.R. Leinum, J.D. Embury, S. Dey, T. Børvik, O.S. Hopperstad.
Localised shear banding in Weldox steel plates impacted by projectiles. Mechanics of
Materials 39 (2007) 865–880.
[7] T. Belytschko, W. K. Liu, B. Moran. Nonlinear Finite Elements for Continua and
Structures. Wiley, 2000.
[8] M. Ortiz, Y. Leroy, A. Needleman. A finite element method for failure analysis.
Computer Methods in Applied Mechanics and Engineering 61 (1987) 189−214.
[9] H. Yamamoto. Conditions for shear localization in the ductile fracture of void-
containing materials. International Journal of Fracture 14 (1978) 347−365.
[10] A. Benallal, C. Comi. Localization analysis via a geometrical method. International
Journal of Solids and Structures 33 (1996) 99−119.

9-11
9-12
10 Thermodynamics*
The laws of thermodynamics are presented in this chapter together with a framework
for establishing thermodynamically admissible constitutive relations for solid materials. The
first law of thermodynamics expresses the balance of energy and asserts the equivalence of
mechanical and non-mechanical energy [1]. Non-mechanical energy is related to the
behaviour at the micro-scale (i.e., at the molecular or atomistic level), and here only heat
energy is considered, which represents the kinetic energy at the micro-scale [2]. The second
law of thermodynamics, or the principle of irreversibility, states that the entropy production
is never negative [1]. It is an experimental fact that physical processes tend to develop in a
certain direction: heat flows spontaneously from a hot body to a cold body, the molecules
of a gas will spread out to fill the whole volume available to them, etc. Accordingly, the
energy of a system tends to disperse, and entropy change can be considered “as a measure
of the dispersal of energy from localized to spread out” [3]. The second law of
thermodynamics leads to the dissipation inequality, which implies that the dissipation, i.e.,
the transformation of mechanical energy into non-mechanical energy in the form of heat, is
either zero or positive.
When developing a new constitutive model, it is important to ensure that it is
thermodynamically admissible, i.e., that the model fulfils the dissipation inequality [4].
Different ways of ensuring thermodynamic admissibility are discussed in this chapter on a
rather general basis, and examples are presented for thermoelastic and thermo-viscoplastic
materials. The laws of thermodynamics also make it possible to calculate the temperature
change during a thermo-mechanical process by use of the heat equation, provided a
constitutive model has been determined for the material [5]. Since field measuring
techniques, such as digital image correlation and digital infrared thermography, allow us to
acquire the displacement, strain and temperature fields on the surface of a test specimen or
a structural component during a laboratory test, the heat equation may be used to improve
and validate constitutive equations.
The presentation of thermodynamic offered in this chapter is rather brief, and the reader
should consult the reference list for more comprehensive presentations. In particular,
Chapters 21–23 of the textbook of Saabye Ottosen and Ristinmaa [4] are recommended for
further reading.

10.1 First and second law of thermodynamics

The first law of thermodynamics is here stated as [6]
D
Dt V
 udV  Pd  Q (10-1)

where u is the internal energy per unit mass of the body, Pd    ij Dij dV is the
V
deformation power and the density  is the mass per unit volume of the material. The heat
flow Q , i.e., the rate at which energy in the form of heat is transferred to the body, is given
as
Q   rdV   qi ni dS (10-2)
V S

10-1
where r is the heat sources per unit volume (e.g. induction heating) and qi is the heat flux
vector per unit area (heat conduction). The outward unit normal ni to the infinitesimal
surface element dS is directed out of the body; thus the minus sign of the heat flux term.
Equation (10-1) expresses the equivalence between mechanical work and heat, and further
the energy balance: the rate of change of the internal energy equals the sum of the
deformation power and the heat flow.
Using Gauss’ theorem, we get
 q 
Q   rdV   qi ni dS    r  i  dV (10-3)
V S V
xi 
Assuming that Equation (10-1) is valid for any volume V of the body, we get the local form
of the first law
qi
 u   ij Dij  r  (10-4)
xi
To obtain the expression on the left-hand side, we used that  dV is constant due to mass
conservation. Equation (10-4) is a partial differential equation expressing energy
conservation.
The second law of thermodynamics is here stated as [4]
D r q

Dt V
 sdV   dV   i ni dS
V
T S
T
(10-5)

where s is the entropy per unit mass of the body, and T  0 is the absolute temperature.
The right-hand side of the inequality represents the entropy supply. Inequality (10-5) states
that the increase of the body’s entropy is always greater or equal to the entropy supply from
its surroundings; i.e., intrinsic entropy production due to internal processes is never negative.
This is a statement of the principle of irreversibility and it is the global form of the Clausius–
Duhem inequality.
Using again Gauss’ theorem, we get
r 1 qi qi T
s    0 (10-6)
T T xi T 2 xi
This is the local form of the Clausius–Duhem inequality, which is assumed to be satisfied
for every point in the body.
Combining Equations (10-4) and (10-6), we obtain the so-called dissipation inequality
qi T
D  T  s   u   ij Dij  0 (10-7)
T xi
where D represents the energy dissipation per unit volume of the body in the current
configuration. A process is said to be reversible if D  0 and irreversible if D  0 . It is
convenient here to introduce the Helmholtz free energy per unit mass of the body, which is
defined as
  u  Ts (10-8)
Combination of Equations (10-7) and (10-8) gives
10-2
 qi T
D      sT    ij Dij  0 (10-9)
T xi
This inequality should be fulfilled for thermodynamic admissible constitutive
equations. Different ways to ensure this inequality are described below. If we integrate
Equation (10-9) over the volume of the body, we arrive at (e.g. [4][5])
D  qi T 
 D dV  Pd 
V
Dt V
 dV  
V
  sT   dV  0
T xi 
(10-10)

In cases where thermal energy production can be neglected (i.e., the last term in the above
equation), the deformation power is equal to the rate of the free energy plus the dissipation,
where the latter is non-negative.

10.2 Thermo-elastic materials and heat conduction

A thermo-elastic material is defined by a free energy function which depends on the
strain tensor ε and temperature T [6]. Infinitesimal strains but finite rotations are assumed
here, and we adopt a corotational formulation of elasticity. Thus, tensor quantities are
expressed in a corotational coordinate system. The free energy function is then defined in
the form
    ˆij , T  (10-11)

where the corotational (infinitesimal) strain tensor is given by

t
ˆij   Dˆ ij dt (10-12)
0

It then follows that

 ˆ 
 Dij  T (10-13)
ˆij T
Inserting Equation (10-13) into the dissipation inequality (10-9), we obtain

      qˆ T
D    s   T   ˆ ij    Dˆ ij  i 0 (10-14)
 T   ˆij  T xˆi

This inequality should be fulfilled for all thermoelastic processes.

Consider an elastic process at constant ( T  0 ) and uniform ( T / xˆi  0 ) temperature
[4]. Since the dissipation inequality should hold for arbitrary values of the corotational rate-
of-deformation tensor Dˆ ij , it follows that


ˆ ij   (10-15)
ˆij
Assuming this equality to hold for all thermoelastic processes, we now consider a
thermoelastic process with uniform temperature ( T / xˆi  0 ). The dissipation inequality is
assumed to be valid for arbitrary values of the temperature rate T , and we therefore have

10-3
 
s (10-16)
T
This shows that the stress tensor and the entropy are derived from the free energy function
by partial differentiation with respect to the strain tensor and the temperature, respectively,
which means that  works as a potential function for ˆ ij and s .
Equations (10-15) and (10-16) are denoted state relations [4]. Inserting the state
relations into Equation (10-14), leads to
qˆi T
D  0 (10-17)
T xˆi
This implies that the dissipation of energy is solely due to heat conduction for a
thermoelastic material, i.e., if the temperature field is uniform ( T / xˆi  0 ), the dissipation
vanishes. It is important to note that inequality (10-17) expresses the experimental fact that
“heat flows from hot to cold”.
Linear thermo-elasticity is obtained by using a quadratic free energy function in the
strain tensor and the temperature change (e.g. [4]–[6]), viz.
  ˆij , T   12 Cˆijkl T  ˆij ˆkl  ˆij T  ˆij T  T0    T T  (10-18)

where Cˆ ijkl T  is the 4th order tensor of elastic coefficients and ˆij T  is the 2nd order
tensor of thermal stress coefficients. Further,  T T  is the part of the free energy
depending only upon temperature (and thus contributing to the entropy only) and is here
defined as [4]
 T T    0  c1 T T  T0   c2 T T  T0 
2
(10-19)

where  0 is the free energy at T  T0 and ˆij  0 , i.e., at the reference state, and c1 and c2
are temperature dependent coefficients. Note that small strains and small perturbations in
temperature are assumed in linear thermo-elasticity, and the density  is assumed constant.
The stress tensor is then obtained from Equation (10-15) as

ˆ ij    Cˆijkl T  ˆkl  ˆij T T  T0  (10-20)
ˆij
which is the same form as given in Section 5.4 of Part I. The entropy is obtained by Equation
(10-16) when required.
We also can make use of Equation (10-17) to establish a constitutive relation for heat
conduction which is thermodynamically consistent, i.e., a constitutive relation that is
formulated according to the Clausius-Duhem inequality. To obtain this, we may define the
heat flux vector as
T
qˆi  kˆij (10-21)
xˆ j

where kˆij is the symmetric 2nd order conductivity tensor. Then Equation (10-17) becomes

10-4
 qˆi T kˆij T T
D   0 (10-22)
T xˆi T xˆi xˆ j

Since T  0 , this inequality is satisfied if kˆij is positive definite. For isotropic materials we
have kˆ  k , and thus
ij ij

T
qˆi   k (10-23)
xˆi
where k  0 follows from the Clausius-Duhem inequality; k being the conductivity. This
is the most usual form of Fourier’s law for heat conduction, and states that the heat flux
vector has the opposite direction of the temperature gradient.

10.3 Inelastic solid materials

To define the behaviour of inelastic materials, it is necessary to introduce internal
variables. The internal variables will be denoted 1 ,  2 ,...,  N and represent scalar variables.
The corotational formulation is used in the following. The elastic behaviour is defined by
the elastic strain tensor defined as
t
ˆije   Dˆ ije dt (10-24)
0

The elastic strains are assumed to be infinitesimal and the rate-of-deformation tensor is
additively decomposed into elastic and plastic parts
Dˆ ij  Dˆ ije  Dˆ ijp (10-25)
The free energy of inelastic materials is now defined in the form
    ˆije , T ,   (10-26)

Then
 ˆ e  N


ˆij
e
Dij 
T
T  
 1 
 (10-27)

Using Equation (10-27) in combination with the dissipation inequality (10-9), we find

      ˆ e N
 qˆ T
 T   ˆ ij   ˆ e  Dij  ˆ ij Dij   
D    s  ˆp   i  0 (10-28)
 T    ij   1  T xˆi

It is assumed that Equations (10-15) and (10-16) remain valid also for inelastic materials
(cf. References [4][6] for detailed discussions on this issue), but now with ˆije instead of ˆij
in the expression for the stress:

ˆ ij   (10-29)
ˆije
The dissipation inequality is then reduced to

10-5
 N
 qˆ T
D  ˆ ij Dˆ ijp      i 0 (10-30)
 1  T xˆi
Using the experimental observation that heat flows from hot to cold (see also [6]), we
get instead
N
 qˆ T
Din  ˆ ij Dˆ ijp      0, Dth   i 0 (10-31)
 1  T xˆi
where Din is the intrinsic dissipation due to the evolution of the internal variables and the
associated change of the free energy and Dth is the thermal dissipation due to heat
conduction [4]. The inequality Din  0 is denoted the Kelvin inequality and is rewritten as
[6]
N

Din  ˆ ij Dˆ ijp   p   0, p    (10-32)
 1 
where p is the thermodynamic force conjugate with the internal variable  . Note that the
Kelvin inequality implies that internal processes, as represented by the evolution of the
internal variables, lead to a decrease of the free energy.
As discussed by Saabye Ottosen and Ristinmaa [4], there are several ways of fulfilling
the dissipation inequality in constitutive modelling. In the direct approach, the constitutive
equations are first postulated and then the dissipation inequality is checked a posteriori.
It is more usual to adopt the potential approach, in which the evolution equations of the
internal variables are defined as
g g
Dˆ ijp   ,    , 0 (10-33)
ˆ ij p

where  is an inelastic parameter and the function g is denoted the dissipation potential.
The intrinsic dissipation then takes the form
N  g N
g 
Din  ˆ ij Dˆ ijp   p     ˆ ij   p 0 (10-34)
 ˆ ij  1 p 
 1 
The dissipation potential function depends on the thermodynamic forces and possibly on
other variables. Here we will simply assume that g  g ˆ ij , p  if nothing else is explicitly
stated.
To ensure that inequality (10-34) is fulfilled, the function g can be assumed to be
positive homogeneous of degree n , i.e.,
g  aˆ ij , ap   a n g ˆ ij , p  , a  0 (10-35)

where a is any positive number. Then by Euler’s theorem for homogeneous functions (cf.
Appendix B) we get
N  g N
g 
Din  ˆ ij Dˆ ijp   p     ˆ ij   p   n g  0 (10-36)
 ˆ ij  1 p 
 1 

10-6
which means that the intrinsic dissipation is non-negative provided g  0 .
Another possibility is to assume that g is a convex function in the stress tensor ˆ ij and
the thermodynamic forces p . Let the function g be sufficiently smooth. Then g is a
convex function if and only if (cf. Appendix B)
g g
g ˆ ij , p   g ˆ ij , p    ˆ ij  ˆ ij     p  p 
N
(10-37)
ˆ ij  1 p

for any two points ˆ ij , p  and ˆ 

ij , p  in the space defined by stresses and
thermodynamic forces, where the partial derivative is evaluated at the point ˆ ij , p  [4].
A convex function of a single variable is illustrated in Figure 10-1. We further assume that
g is non-negative, g  0 , and contains the origin, g  0, 0   0 [5]. Then by selecting
ˆ ij  0 and p  0 , we get

g g
p  g ˆ ij , p   0
N
ˆ ij   (10-38)
ˆ ij  1 p

It immediately follows from Equations (10-34) and (10-38) that the dissipation inequality
holds true for such a convex dissipation potential.

Figure 10-1: Example of a non-negative, convex function of a single, scalar variable

containing the origin.

10-7
10.4 Heat equation
Since it is today possible during materials testing to measure the displacement field by
digital image correlation and the temperature field by digital infrared thermography, it is
interesting to establish the equation defining the temperature change of the body during a
thermo-mechanical process. This can be obtained by combining the two laws of
thermodynamics formulated for the inelastic material.
Using the definition of the free energy for inelastic materials, Equation (10-27), we get
 ˆ e  N
 N
   e Dij   T     ij Dij   sT   p 
ˆ ˆ e
(10-39)
ˆij T  1   1

where Equations (10-29), (10-16) and (10-32)2 were used. From Equation (10-16) we get
 2 ˆ e  2 N
 2
s
T ˆije
Dij 
T 2
T  
 1 T 

(10-40)
1 ˆ ij ˆ e s N
1 p
 Dij  T  
 T T  1  T

where we have used that     ˆije , T ,   . The specific heat c (per unit mass) is defined
by
s  2
c  T  T (10-41)
T T 2
where the deformation is fixed. An alternative definition is
u    sT   s s
c     s T T (10-42)
T T T T T
where Equation (10-16) was used to obtain the last equality. As a result, c represents the
energy required to produce a unit increase in the temperature of a unit mass of the body with
the deformation kept fixed. With the specific heat thus defined, Equations (10-40) and
(10-41) give
 ˆ ij ˆ e N p 
Ts   c T  T  Dij     (10-43)
 T  1 T 
The internal energy is u    Ts and thus
 u    Ts   sT
 ˆ N
p  (10-44)
 ˆ ij Dˆ ij  Din   c T  T  ij Dˆ ije     
 T  1 T 
where Equations (10-39), (10-32), (10-25), and (10-43) were used. Invoking the local
energy balance expressed by Equation (10-4), we obtain the heat equation
qˆi  ˆ ij ˆ e N p 
 c T  r   Din  T  Dij     (10-45)
xˆi   T  1 T 

10-8
The two first terms on the right-hand side represent the heat supply from the surroundings,
the third term is the intrinsic dissipation, whereas the terms in the parenthesis represent the
thermo-mechanical coupling. According to Lemaitre and Chaboche [5], the thermo-
mechanical coupling terms may be neglected in most applications.
In classical heat conduction, the thermo-mechanical coupling terms and the intrinsic
dissipation all vanish, and the heat equation takes the reduced form
qˆi
 c T  r  (10-46)
xˆi
Another reduced form of the heat equation is obtained for adiabatic heating, i.e., there is no
external heat input:
 ˆ ij ˆ e N p 
 c T  Din  T  Dij     (10-47)
 T  1 T 
If the thermo-mechanical coupling terms are negligible under adiabatic conditions, the
temperature increase is only due to intrinsic dissipation. The Taylor-Quinney coefficient
TQ is defined as the part of the specific plastic power dissipated as heat under adiabatic
conditions, namely
 c T  TQˆ ij Dˆ ijp (10-48)
Combining the last two equations, we may find the value of the Taylor-Quinney coefficient
as a function of the inelastic deformation. In particular, if the thermo-mechanical coupling
terms are negligible, we arrive at
Din  TQˆ ij Dˆ ijp (10-49)
We note that if there are no structural rearrangements contributing to the free energy of the
material (   0 ), the Taylor-Quinney coefficient is equal to unity, provided the
thermoelastic effect is negligible.

10.5 Thermoelastic-thermoviscoplastic materials

The constitutive relations for thermoelastic-thermoviscoplastic materials will now be
developed within the established thermodynamic framework. Since the effects of
temperature and strain rate are strongly coupled, we will only deal with rate-dependent
materials. As before, elastic strains are assumed to be infinitesimal, while rotations and
plastic strains may be finite. All tensors are referred to a corotational coordinate system in
order to neutralize spin effects. We will also assume isochoric plastic flow, so that any
changes in the density  can be neglected.
We will consider a non-associated plasticity model with nonlinear isotropic hardening
including multiple hardening variables. Non-associated in this context implies that the
dissipation potential is not defined by the yield function alone. The free energy function is
defined by
    ˆije , r , T    e ˆije , T   p  r , T  (10-50)

10-9
where  e and  p are the thermoelastic and thermoplastic parts of the free energy,
respectively. The scalar internal variables r ,   1, 2,..., N , govern the nonlinear isotropic
hardening. (Do not confuse the internal variables r with the heat sources per unit current
volume r .) The function  e is defined by Equation (10-18) but inserting ˆije instead of ˆij
to account for the partition into elastic and plastic deformations. The thermoplastic part of
the free energy function is assumed to be quadratic in the internal variables, i.e.,
N
 p  r , T    12  T  r2 (10-51)
 1

where  are temperature-dependent hardening moduli. The stresses and hardening

variables are then obtained in terms of the partial derivatives of  with respect to free
variables ˆije and r , namely

 
ˆ ij    Cˆijkl T  ˆkle  ˆij T T  T0  , R     T  r (10-52)
ˆij
e
r

where R are the isotropic hardening variables. In a similar manner, the entropy may be
derived by taking the partial derivative of  with respect to T , see Equation (10-16). We
may now make the identifications:   r ,   1, 2,..., N , are the internal variables and
p   R are the associated thermodynamic forces.
The yield function is defined in the form

f   ˆ ij    0 T    R  r , T 
N
(10-53)
 1

where  is the convex and first-order positive homogeneous yield function, so that the
equivalent stress is defined by  eq   ˆ ij  , and R   1 R is the work-hardening. The
N

initial yield stress  0 and the hardening variables R are assumed to depend on temperature.
The current yield stress, setting the size of the elastic domain, is given by
N
 y  r , T    0 T    R  r , T  (10-54)
 1

As usual, the equivalent plastic strain p is defined as conjugate in plastic power to the
equivalent stress  eq , viz.

ˆ ij Dˆ ijp
ˆ ij Dˆ ijp   eq p  p 0 (10-55)
 eq
With this definition, we need to ensure that the specific plastic power is non-negative, so
that the equivalent plastic strain is monotonically increasing. From Equation (10-32), we
conclude that this will always be the case in the absence of internal variables 
contributing to the free energy, since the intrinsic dissipation must remain non-negative as
a result of the 2nd law of thermodynamics. However, if there are internal variables, as is the

10-10
case here, negative specific plastic power could be thermodynamically admissible if it is
balanced by a reduction of the free energy due to changes of the internal variables.
From Equation (10-32), we find that the intrinsic dissipation is given by
N
Din  ˆ ij Dˆ ijp   R r  0 (10-56)
 1

To ensure Din  0 , we introduce a dissipation function

g  g ˆ ij , R , T   0 (10-57)

so that
g g g
Dˆ ijp   , r     (10-58)
ˆ ij    R  R

where   0 is the plastic parameter. The temperature enters as a parameter in the

dissipation function. In viscoplasticity, the plastic parameter is defined by a constitutive
relation (see Chapter 8 of Part I), viz.

 0 for f ˆ ij , R , T   0
 (10-59)
 ˆ ij , R , T  for f ˆ ij , R , T   0

where the function   0 governs the rate sensitivity of the material and depends on the
temperature. As an example, consider the following definition of the function  (see
Chapter 8 of Part I)
1
 f C
  p0   (10-60)
  v0 
where  v 0 , C and p0 are material parameters that may depend on the temperature. Using
Equation (10-53) and solving for the equivalent stress  eq   ˆ ij  , we find the constitutive
relation in the plastic domain
C
N
 p
 eq   0   R   v 0   (10-61)
 1  p0 
with the viscous stress defined by
C
 p
 v  p    v0   (10-62)
 p0 
We have here assumed that  equals the equivalent plastic strain rate p for f  0 , which
is the case for associated flow. For a more general description of rate sensitivity, it is
possible to let the viscous stress  v depend also on the work-hardening as defined by the
thermodynamic forces R .
With the adopted definition of the dissipation potential, the dissipation inequality takes
the form
10-11
  g N
g 
Din    ˆ ij   R 0 (10-63)
 ˆ ij  1 R 

Ways to select g to ensure non-negative dissipation have been discussed in Section 10.3.
One possibility is to assume g to be additively decomposed into N  1 parts as

g ˆ ij , R , T   g 0 ˆ ij    g  R , T 
N
(10-64)
 1

where the functions g and g 0 fulfil the conditions

g N
g g
ˆ ij   R  0, ˆ ij 0  0 (10-65)
ˆ ij  1 R ˆ ij
It then immediately follows from Equation (10-63) that the intrinsic dissipation is always
non-negative. This decomposition also ensures the non-negativity of the equivalent plastic
strain rate defined by Equation (10-55). Associated plastic flow is obtained if g 0   , and
in this case we get by use of Equations (10-55) and (10-58)1
ˆ ij Dˆ ijp  
p  ˆ ij  (10-66)
 eq  eq ˆ ij
Recall that  was assumed to be a positive first-order homogeneous function. Another
possibility is to define the dissipation function as

g ˆ ij , R , T   f ˆ ij , R , T    g  R , T 
N
(10-67)
 1

where the functions g have to be non-negative. Recall that f  0 during plastic flow in
plasticity and f  0 during plastic flow in viscoplasticity. Thus, with g  0 it is ensured
that g  0 during plastic flow.
As an important case, we adopt Equation (10-67) with the following definitions of the
functions g :
1 R2
g  R , T   2
,   1, 2,..., N (10-68)
Q T 

where Q  0 are temperature-dependent hardening parameters. Evidently g are non-

negative as required. Using Equation (10-58)2, we obtain

g  f  N
  R 
r        g    1  
 p (10-69)
R  R R  1   Q T

where Equations (10-67), (10-53), (10-68) and (10-66) were applied. By introducing
Equation (10-52)2, we arrive at an alternative form of the evolution equations
 r 
r  1    p,   1, 2,..., N (10-70)
 q T  

10-12
with q T   Q T  /  T  . It is evident that r saturates with plastic straining and the
saturation value of r is obtained by

r  0  r  q T  (10-71)

which in view of Equation (10-52)2 implies that the saturation value of R is governed by
the parameter Q . Under isothermal conditions, solution of the differential equation in
Equation (10-70) leads to
   
R  p    r  p   Q 1  exp    p   ,   1, 2,..., N (10-72)
  Q  
which is the multi-component Voce hardening rule. The initial conditions p  0 and
R  r  0 were assumed in the integration. Under non-isothermal conditions the rate
equations for the hardening variables R are obtained from Equations (10-52)2 and (10-70)
as
 R   T 
R   T  1   p  R T ,   1, 2,..., N (10-73)
 Q T    T  

where  T  is the derivative of  T  . While the evolution of the strain-like internal
variables r depends on temperature, these variables do not change instantaneously with a
change in temperature. On the contrary, the stress-like hardening variables R do change
instantaneously with a change in temperature.
Finally, the rate constitutive equations are obtained from Equation (10-52)1 as
 g   ˆ Cˆijkl T  e ˆij T  
ˆ ij  Cˆijkl T   Dˆ kl      ij T   ˆij  T  T0  T (10-74)
 ˆ kl   T T 
where   0 for f  0 and    ˆ ij , R , T  for f  0 . It is instructive to compare this
expression with Equation (6-35)1, which is valid for constant thermoelastic tensors C and
β.
Using Equation (10-56) and (10-69), the intrinsic dissipation is expressed by
N N
 R 
Din  ˆ ij Dˆ ijp   R r  ˆ ij Dˆ ijp   R 1  p0 (10-75)
 1  1  Q 

It is left to the reader to prove that the intrinsic dissipation is non-negative. As Q is the
saturation value of R , the intrinsic dissipation will approach the specific plastic power
asymptotically with increasing plastic strain. Recall that the Taylor-Quinney coefficient
TQ is defined as the part of the specific plastic power dissipated as heat under adiabatic
conditions. If we neglect the thermoelastic effect, we may approximate the Taylor-Quinney
coefficient TQ for the adopted thermoelastic-thermoviscoplastic model by combining
Equations (10-49), (10-75) and (10-55)1, and the result comes out as

10-13
 1 N
 R 
TQ  1 
 eq
 R 1    (10-76)
1  Q  
where R are given by Equation (10-72). It is evident that TQ equals unity when R are
all equal to zero, i.e., for a virgin material without hardening, or when all R have reached
saturation. For intermediate plastic strains, TQ will be less than unity. As an example,
consider a material strained under adiabatic conditions with yield stress  0  100MPa and
hardening parameters 1  5000MPa , Q1  100MPa and N  1 (only one hardening term).
For simplicity, rate and temperature effects on the material behaviour are neglected, and the
constitutive relation in the plastic domain is reduced to  eq   0  R1 . Using Equation
(10-76), we may then compute the Taylor-Quinney coefficient as a function of the
equivalent plastic strain. The result is plotted in Figure 10-2. It is seen that for the actual
material the minimum of TQ occurs around 10% plastic strain. Note that in practical
applications TQ is often assumed to be constant and equal to 0.9.
The thermoelastic-thermoviscoplastic model with isotropic hardening is compiled in
Table Box 10-1 and Box 10-2, where the state equations are given in the former and the
evolution equations in the latter. As before, infinitesimal elastic strains and isochoric
plasticity are assumed.

Figure 10-2: An example of the evolution of the Taylor-Quinney coefficient TQ for an
elastic-plastic material with  0  100MPa , 1  5000MPa , Q1  100MPa and N  1 .

10-14
Box 10-1: Thermoelastic-thermoviscoplasticity with isotropic hardening (Part I)

Domain of validity:
 Infinitesimal elastic strains
 Isochoric plasticity
 Finite plastic strains and rotations
 Elastic and plastic anisotropy

Main equations:
 Additive decomposition of the rate-of-deformation tensor:
Dˆ ij  Dˆ ije  Dˆ ijp

 Infinitesimal elastic strain tensor:

t
ˆ   Dˆ ije dt
e
ij
0

 Free energy function:

  ˆije , r , T    e ˆije , T    p  r , T 
with
 e  ˆije , T   12 Cˆijkl T  ˆijeˆkle
 ˆij T  ˆije T  T0    T T 
N
 p  r , T    12  T  r2
 1

 State equations:

ˆ ij    Cˆ ijkl T  ˆkle  ˆij T T  T0 
ˆije

R     T  r
r

10-15
Box 10-2: Thermoelastic-thermoviscoplasticity with isotropic hardening (Part II)

Main equations (cont.):

 Yield function:

f ˆ ij , R , T    ˆ ij    0 T    R  r , T 
N

 1

 Dissipation potential:
R2
g ˆ ij , R , T   f ˆ ij , R , T   
N 1
2

 1 Q  T 

 Flow rule:

g f g  R 
Dˆ ijp    , r    1  p
ˆ ij ˆ ij R  Q T  

 Equivalent stress and plastic strain:

ˆ ij Dˆ ijp
 eq   ˆ ij  , p 
 eq
 Intrinsic (plastic) dissipation:
N
Din  ˆ ij Dˆ ijp   R r  0
 1

 Plastic parameter:

 0 for f ˆ ij , R , T   0

 ˆ ij , R , T  for f ˆ ij , R , T   0

 Adiabatic heating:
 ˆ ij ˆ e N R 
 c T  Din  T  Dij   r 
 T  1 T 

10-16
10.6 References
[1] P. Haupt. Continuum Mechanics and Theory of Materials. 2nd edition. Springer, 2002.
[2] H. Ziegler. An Introduction to Thermomechanics. 2nd edition. North-Holland, 1983.
[3] F.L. Lambert. Entropy is simple, qualitatively. Journal of Chemical Education 79
(2002) 1241–1246.
[4] N. Saabye Ottosen, M. Ristinmaa. The Mechanics of Constitutive Modeling. Elsevier,
2005.
[5] J. Lemaitre, J.-L. Chaboche. Mechanics of Solid Materials. Cambridge University
Press, 1990.
[6] J. Lubliner. Plasticity Theory. Macmillan, 1990.
[7] G.A. Holzapfel. Nonlinear Solid Mechanics. A Continuum Approach for Engineering.
Wiley, 2001.
[8] M.E. Gurtin, E. Fried, L. Anand. The Mechanics and Thermodynamics of Continua.
Cambridge, 2010.
[9] J. Lubliner. On the thermodynamic foundations of non-linear solids mechanics.
International Journal of Non-linear Mechanics 7 (1972) 237–254.

10-17
10-18
Appendix A: Polar coordinates
In these lecture notes, we have done most derivations in Cartesian coordinates. In some
instances, however, it is more convenient to work with other coordinate systems. In the
following, we will derive some useful relations in cylindrical and spherical coordinates,
with emphasis on relations important in materials mechanics.
We have seen that the infinitesimal strain tensor and the rate-of-deformation and spin
tensors are obtained from the gradient of the displacement and velocity fields, respectively,
whereas the equilibrium equations contain the divergence of the stress tensor. It is therefore
useful to establish the expressions for the del-operator  , the gradient of scalar and vector
fields, and the divergence of vector and tensor fields in the cylindrical and spherical
coordinate systems. The textbooks of Irgens [1] and Lubliner [2] are recommended for
further reading. The summary given below is mostly based on the latter reference.

Dyadic notation
In the derivation of the differential operators in polar coordinates, we will use the
dyadic notation that was introduced in Chapter 3 of Part I. For convenience, we repeat the
brief introduction to the dyadic notation given there in the following.
The tensor product u  v of two vectors u and v is called a dyad, whereas the tensor
products of several vectors are called polyads. The tensor product is defined to obey all the
rules of ordinary algebra, with one exception: the tensor product is not commutative, i.e.,
u  v  v  u . For instance, the tensor product has the following properties
 u   v   w   u  w   v  w
u   v  w   u  v  w (A-1)
u  v   w   v  w  u
where  and  are real numbers and u , v and w are vectors. The linear combination of
dyads is called a dyadic. Using that u  ui ei and v  vi ei , we find that u  v  ui v j ei  e j ,
where the summation convention applies. In general, we may use the dyadic notation to
express a 2nd order tensor L by
L  Lkl ek  el (A-2)
As before, the components of L is obtained as
ei  L  e j  ei   Lkl e k  el   e j  Lkl  ei  e k   el  e j   Lkl  ik  lj  Lij (A-3)

The transformation rule of a second-order tensor is now readily established by finding the
components of L in the starred basis ei  aij e j as

L*ij  e*i  L  e*j  e*i   Lkl ek  el   e*j  Lkl  e*i  ek  el  e*j   aik a jl Lkl (A-4)

In particular, in the basis defined by the eigenvectors ni , i  1, 2,3 , of the tensor L , the
dyadic notation gives

A-1
 3
L   Li ni  ni (A-5)
i 1

where Li are the eigenvalues of L . As a rule, we will use index notation instead of the
dyadic notation, except in cases for which the dyadic notation is particularly favourable.

Cylindrical coordinates

Figure A-1: Cylindrical coordinates.

According to Figure A-1, the Cartesian coordinates  x1 , x2 , x3  are related to the

cylindrical coordinates  r, , z  by

x1  r cos  x2  r sin  x3  z
(A-6)
r  x12  x22   tan 1  x2 / x1  z  x3

The base vectors  er , e , e z  of the cylindrical coordinate system are defined as unit tangent
vectors to the coordinate lines
x / r x /  x / z
er  , e  , ez  (A-7)
x / r x /  x / z
which gives
er  e1 cos  e2 sin  , e  e1 sin   e2 cos , e z  e3 (A-8)
The inverse transformation is readily found as
e1  er cos  e sin  , e2  er sin   e cos , e3  e z (A-9)
These relations can also be derived directly from Figure A-1. Clearly, the unit base vectors
er , e , ez  are mutually orthogonal, i.e., they constitute an orthonormal basis. While e z is
constant, the base vectors e r and e depend on the angle  , so that

A-2
 d d d
er  e , e  e r , ez  0 (A-10)
d d d
Further, we need the relations between the partial derivatives in the two coordinate
systems. The chain rule for partial derivatives gives
  xi   xi   xi
 ,  ,  (A-11)
r xi r  xi  z xi z
and from Equation (A-6) we get
       
 cos   sin  ,   sin   cos  ,  (A-12)
r x1 x2 r  x1 x2 z x3
The inverse transformation is
       
 cos   sin  ,  sin   cos  ,  (A-13)
x1 r r x2 r r  x3 z
The del-operator  was defined in Chapter 3 of Part I with reference to the Cartesian
coordinate system. Using Equations (A-13) and (A-9), we find the del-operator  referred
to the cylindrical coordinate system as
  1  
  ei  er  e  ez (A-14)
xi r r  z

Thus, in cylindrical coordinates, the gradient of the scalar field     r, , z  is

 1  
  er  e  ez (A-15)
r r  z
While the gradient of a scalar field is a vector, the gradient of a vector field
v  v  r, , z  is a 2nd order tensor

v     v 
T
(A-16)

where v  vr er  v e  vz e z in the cylindrical coordinate system. Combining these

equations, remembering that the base vectors of the cylindrical coordinate system vary with
the angle  , we get [2]
  1   
  v   er  e  e z    vr e r  v e  vz e z 
 r r  z 
 v v v 
 e r   e r r  e   e z z 
 r r r 
(A-17)
1  v v v 
 e   e r r  e vr  e   e r v  e z z 
r     
 v v v 
 e z   e r r  e   e z z 
 z z z 

A-3
The gradient v is then the transpose of dyadic product   v . The divergence
div v   v of the vector field v  v  r, , z  is calculated by applying the contraction
operation on the dyadic product   v , and we find [2]
vr 1  v  v
v    vr     z (A-18)
r r    z
Using dyadic notation, a symmetric, 2nd order tensor L may be expressed in cylindrical
coordinates by
L  Lrr e r  e r  Lr  e r  e  e  e r   L e  e
(A-19)
 Lrz  er  e z  e z  er   L z  e  e z  e z  e   Lzz e z  e z
Then, according to Equations (A-14) and (A-19), the divergence of the tensor field
L  r, , z  reads as [2]

 L L  L 1 Lr Lrz 

  L  e r  rr  rr   
 r r r  z 
 L 2L 1 L L z 
 e  r  r    (A-20)
 r r r  z 
 L L 1 L z Lzz 
 e z  rz  rz   
 z r r  z 

A-4
Spherical coordinates

Figure A-2: Spherical coordinates.

The relationships between the Cartesian coordinates  x1 , x2 , x3  and the spherical

coordinates  r, ,   are obtained from Figure A-2:

x1  r cos  sin  x2  r sin  sin  x3  r cos 

r  x12  x22  x32 
  tan 1  x2 / x1    cot x3 / x12  x22  (A-21)

The base vectors  er , e , e  of the spherical coordinate system are defined as unit
tangent vectors to the coordinate lines, viz.
x / r x /  x / 
er  , e  , e  (A-22)
x / r x /  x / 
Thus
e r  e1 cos  sin   e 2 sin  sin   e3 cos 
e  e1 sin   e 2 cos  (A-23)
e  e1 cos  cos   e 2 sin  cos   e3 sin 
Alternatively, these relations could have been established directly from Figure A-2. The unit
base vectors  er , e , e  are mutually orthogonal and they constitute an orthonormal basis.
The transformation matrix aij  ei  e j between two sets of orthonormal base vectors has to
be orthogonal and we have the transformation rules ei  aij e j and ei  a ji ej . For spherical
coordinates, ei , i  1, 2,3 , represent the base vectors e r , e and e , respectively. From
Equation (A-23), it transpires that the transformation matrix is

A-5
  cos  sin  sin  sin  cos  
a =   sin  cos  0  (A-24)
cos  cos  sin  cos   sin  
The inverse transformation is then readily obtained as
e1  e r cos  sin   e sin   e cos  cos 
e2  e r sin  sin   e cos   e sin  cos  (A-25)
e3  e r cos   e sin 

The partial derivatives of the base vectors  er , e , e  are calculated from Equation (A-23)
as
er er er
 0,  e sin  ,  e
r  
e e e
 0,  er sin   e cos  , 0 (A-26)
r  
e e e
 0,  e cos  ,  er
r  
To determine the del-operator in spherical coordinates, we need the relationships
between the partial derivatives in the two systems. To this end, the chain rule for partial
derivatives is used
  xi   xi   xi
 ,  ,  (A-27)
r xi r  xi   xi 
and combined with Equation (A-21) it gives
   
 cos  sin   sin  sin   cos 
r x1 x2 x3
  
  sin   cos  (A-28)
r  x1 x2
   
 cos  cos   sin  cos   sin 
r  x1 x2 x3
The inverse transformation is immediately established by noting that the partial derivatives
in Equation (A-28) are related by the transformation matrix a and thus the inverse
transformation follows from a T :
   
 cos  sin   sin   cos  cos 
x1 r r r
   
 sin  sin   cos   sin  cos  (A-29)
x2 r r r
  
 cos   sin 
x3 r r

A-6
 Combining Equations (A-25) and (A-29), we find the del-operator  referred to the
spherical coordinate system as
  1  1 
  ei  er  e  e (A-30)
xi r r sin   r 

In spherical coordinates, the gradient of the scalar field     r, , z  is

 1  1 
  er  e  e (A-31)
r r sin   r 
For a vector field v  v  r, ,   , we get [2]

   1  
e    vr e r  v e  v e 
1
  v   er  e 
 r r sin   r  
 v v v 
 e r   e r r  e   e  
 r r r 
  1 vr v   1 v vr cot  
 e  e r     e    v  (A-32)
  r sin   r   r sin   r r 
 1 v cot   
 e   v  
 r sin   r 
  1 vr v  1 v  1 v vr  
 e  e r     e  e   
  r  r  r   r  r  
where v  vr er  v e  v e and it should be kept in mind that the base vectors of the
spherical coordinate system depend on the angles  and  , cf. Equation (A-26). The
gradient v of the vector field is given by Equation (A-16). Applying the contraction
operator on the dyadic product   v , the divergence div v   v of the vector field
v  v  r, ,   is expressed as

vr v 1 v 1 v cot 

v  2 r    v (A-33)
r r r sin   r  r
In spherical coordinates, a symmetric, 2nd order tensor L may be written in the form
L  Lrr er  er  Lr  er  e  e  er   L e  e
(A-34)
 Lr  er  e  e  er   L  e  e  e  e   L e  e

The divergence of the tensor field L  r , ,   is established by employing Equations (A-30),

(A-34) and (A-26), and the results comes out as

A-7
  L 2 L  L  L  Lr cot  1 Lr 1 Lr 
  L  e r  rr  rr  
 r r r sin   r  
 L 3L  2 L cot  1 L 1 L 
 e  r  r  
r  
(A-35)
 r r r sin  
 L 1 L 1 L L cot   L cot   3Lr 
 e  r    
 r r  r sin   r 

References
[1] F. Irgens. Continuum Mechanics. Springer, 2008.
[2] J. Lubliner. Plasticity Theory. Macmillan, 1990.

A-8
Appendix B: Homogeneous and convex functions
When defining yield functions and flow potentials, we have adopted convex functions
and with the important exception of the Gurson model, these functions have also been
positive homogeneous of order one. At several instances, we have used Euler’s theorem for
homogeneous functions but we have not yet proved the theorem. In this appendix, we will
consider in more detail homogeneous functions, convex functions and some related
theorems, and finally norms, which are both convex and positive homogeneous functions
of order one.
Matrix notation will be employed with scalars denoted by lowercase letters ( a, b, c,... ),
column or row vectors by boldface lowercase letters ( a, b, c,... ) and matrices by boldface
uppercase letters ( A, B, C ,... ). The scalar product between two vectors u and v is defined
by uT v  ui vi , where summation extends over the range of the dummy index. The
transformation of a vector u into another vector v by a square matrix M is denoted
v  Mu and defined by vi  M ij u j . We will use and to signify the sets of all integers
and real numbers, respectively, while  is the set of all positive integers. The notation
v  n implies that the vector v has n real-valued components. Analogously, the notation
M  n  n implies a square matrix with dimension n  n and n 2 real-valued
components. We will only consider symmetric matrices in this appendix.

Homogeneous functions
Let x  n
be an n -dimensional vector and f  x  a scalar-valued function from n


to . The function f is said to be homogeneous of order k  if

f  ax   a k f  x  (B-1)

for all non-zero numbers a  . The function f is said to be positive homogeneous if

Equation (B-1) holds only for a  0 .
Assume that the function f is continuously differentiable, then f is positive
homogeneous of order k if and only if [1]

f  x  x  kf  x  (B-2)

where f  x  is the gradient vector of f  x  and defined here as a row vector. This is
Euler’s theorem for homogeneous functions.
The proof is as follows [1]. Assume first that f fulfils Equation (B-1) and introduce
the vector y  ax . Then the chain rule gives
y
ka k 1 f  x   f  y   f  y  x (B-3)
a
By taking a  1 , Equation (B-2) follows. If we instead assumed that f fulfils Equation
(B-2), integration leads to Equation (B-1). Let g  a   f  ax   f  y  , then

B-1
 k k
g   a   f  y  x  f  y  g a (B-4)
a a
k
which means that g   a   g  a   0 . The solution of this first-order differential equation
a
is g  a   g 1 a k , and Equation (B-1) follows from the definition of g  a  .

Convex functions
The function f  x  is convex if for any two vectors x and y in n
, we have

f 1    x   y   1    f  x    f  y (B-5)

where   is within the range 0    1 . Figure A-1 illustrates a convex function of a

single variable. The following results for convex functions are readily shown:

 If f and g are two convex functions, then the function f  g is also convex.
 If f is a convex function and a is a non-negative number, then the function af is
also convex.
 Any linear function f  x   aT x  b is convex, where a  n and b  are
arbitrary constants.
If f is continuously differentiable, an alternative definition of a convex function is
given by

f  y   f  x   f  x  y  x  (B-6)

This definition is also illustrated in Figure A-1. We first prove that Equation (B-5) leads to
Equation (B-6). To this end, we rewrite Equation (B-5) in the form

f  x  x   f  x 
 f  x  x   f  x  (B-7)

where x  y  x . A Taylor expansion of the left-hand side of the inequality gives

f  x  x   f  x   f  x  x  O  2  (B-8)

where O  2  is the order of the local truncation error. Letting   0 , combination of

Equations (B-7) and (B-8) leads to Equation (B-6). To prove that Equation (B-6) implies
Equation (B-5), we consider two points x1 , x2 and the number  between zero and unity.
We then take x   x1  1    x2 and y equal to either x1 or x2 . Then

f  x1   f  x   f  x  x1  x 
(B-9)
f  x2   f  x   f  x  x2  x 

Multiplying the first inequality with  and the second by 1   , and adding them together,
we get

B-2
  f  x1   1    f  x2   f  x   f  x   x1  1    x2  x  (B-10)

If we now introduce our definition of x , Equation (B-5) follows but in terms of x1 , x2

instead of x, y .

|
Figure A-1: Convex function of single variable.

The Hessian matrix H  n

 n
of the function f  x  is defined by

H  x   2 f  x  (B-11)

with components Hij   2 f / xi x j . Evidently, H is symmetric by definition. The twice

continuously differentiable function f  x  is convex if and only if the Hessian matrix H
is positive semi-definite on the domain  of x , i.e.,
xT Hx  0 x  n
(B-12)

B-3
To prove this, we make use of Taylor’s theorem [2]:

 y  x  H  x    y  x   y  x 
1
f  y   f  x   f  x  y  x  
T
(B-13)
2
for some   between zero and unity. If H is everywhere positive definite, Equation
(B-6) follows immediately, verifying the convexity of f  x  . Conversely, if convexity is
assumed, use of Equations (B-6) and (B-13), leads to

 y  x H  x    y  x   y  x   0
1 T
(B-14)
2
Since y  x is an arbitrary vector quantity, we conclude that the Hessian matrix H has to
be positive semi-definite.
Evidently, a quadratic function f  x   12 xT Ax  aT x  b with A n
 n

(symmetric), a  n and b  is convex provided the Hessian matrix, which for the
quadratic function equals A , is positive semi-definite. Another important result is obtained
as follows. Let f  y  be convex, and consider the function g  x   f  Ax  b  with
y = Ax  b , A  n
 n
and b  n
. Repeated use of the chain rule gives
g  x   AT f  y 
(B-15)
 2 g  x   AT  2 f  y  A

It immediately follows that the Hessian matrix  2 g  x  of the function g  x  is positive

semidefinite whenever  2 f  y  is and as a consequence the function g  x   f  Ax  b  is
convex. This is a result used to prove the convexity of linear transformation-based yield
functions, see Chapter 7 of Part I.
There is a more general version of this result, which is important for our purpose and
thus stated as a theorem [3]:
n
If f is a scalar-valued, convex function on and g is a convex
function on which is non-decreasing, then h  x   g  f  x   is convex
n
on .
The proof is as follows: With x, y  n
and 0    1 , we have by convexity

f  1    x +  y   1    f  x    f  y  (B-16)

Applying the function g on both sides, gives

h  1    x +  y   g  1    f  x    f  y    1    h  x    h  y  (B-17)

which shows that h is convex. Note that the first inequality assumes that g is non-
decreasing.
Another interesting question is: under what conditions is a positive homogeneous
scalar-valued function convex? We will prove in the following that a positive homogeneous
function of order one is convex if and only if [4]

B-4
 f  x  y   f  x   f  y  x, y  n
(B-18)

where we recall that positive homogeneity of order one implies that f  ax   af  x  a  0 .

Assume that f  x  is convex and positive homogeneous of order one, then by Equation
(B-5) with   12 , we have
1
2 f  x + y  f  12 x  12 y   12 f  x   12 f  y  (B-19)

which implies Equation (B-18). If we assume that Equation (B-18) holds, then for 0    1 ,
we have
f  x  1    y   f  x   f 1    y    f  x   1    f  y (B-20)

which according to Equation (B-5) implies that f  x  is convex.

Norms
n n 
A norm on is a function x from to with the following properties [5]

ax  | a | x absolute homogeneity
x+ y  x  y triangle inequality (B-21)
x 0  x0 separate points

where a  is any number. If only the two first relations hold, x is said to be a semi-
norm. Note that the first relation gives  x  x , which by the triangle inequality implies
that x  0 , i.e., the norm is non-negative. Thus, a norm is a function that ascribes a strictly
positive length or magnitude to each vector in a vector space (with the exception of the zero
vector).
Some important examples of norms are:

x 2  x12  x22   xN2 Euclidean norm

 
1
n
 p 1
p p
x p i 1
xi , p-norm (B-22)

x 
 max  x1 , x2 ,..., xN  maximum norm

where the Euclidean norm is obtained from the p-norm for p  2 and the maximum norm
is obtained by letting p approach infinity.
It is evident that the first and last relations in Equation (B-21) are fulfilled for the p-
norm, while the triangle inequality is not at all obvious. The proof that this is indeed the
case is taken from Reference [6]. In order to show that the p-norm satisfies the triangle

n p
inequality, we need first to show that i 1
xi is a convex function. It is sufficient to show
that the absolute value is convex because the function x p is convex and non-decreasing for

B-5
 p  1 and further, a sum of convex functions is convex. The convexity of the absolute value
follows from
1    x   y  1    x   y  1    x   y (B-23)

where 0    1 and the inequality is a result of the triangle inequality which is evidently

n p
valid for the absolute value. Thus, we have proved that i 1
xi is convex. Then,

1    x   y   i 1 1    xi   yi
p p p
p
(B-24)
  i 1 1    xi   yi  1    x p   y
p p p p p
p

for x, y  n
and 0    1 . This shows that 1    x   y p
 1 when x p
 y p
1 .
For x, y non-zero, we then have

x y x x y p y
p
 p
 1 (B-25)
x p
 y p
x p
 y p
x p x p y p
y p
p

where x / x p
and y / y p
are unit vectors with coefficients equal to 1   and  ,
respectively. In consequence, the triangle inequality is proved for the p-norm. Note that the
inequality

x+ y p
 x p y p
(B-26)

is termed the Minkowski inequality.

B-6
References
[1] https://en.wikipedia.org/wiki/Homogeneous_function [cited: 27.06.16].
[2] D.G. Luenberger. Linear and Nonlinear Programming. Second edition. Addison-
Wesley, 1989.
[3] R.T. Rockafellar. Convex analysis. Princeton University Press, 1970.
[4] G. Ewald. Combinatorial Convexity and Algebraic Geometry. Springer, 1996.
[5] https://en.wikipedia.org/wiki/Norm_(mathematics) [cited: 27.06.16].
[6] http://www. math.bard.edu/belk/math461/Inequalities.pdf [cited: 27.06.16].

B-7
B-8