CE 7018

Computational Inelasticity

Serdar Göktepe
Middle East Technical University
Department of Civil Engineering · Division of Construction Materials

Fall 2020
Contents

I Overview

II Introduction to Tensor Algebra and Tensor Calculus

1 Fundamentals of Tensor Algebra . . . . . . . . . . . . . . . . . . . . . 7
1.1 Algebra of Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2 Algebra of Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2 Eigenvalue Problem of Tensors . . . . . . . . . . . . . . . . . . . . . . 20
3 Tensor Functions and Derivatives . . . . . . . . . . . . . . . . . . . . 23

III Thermodynamics of the 3-D Continuum

1 Kinematics at Small Strains . . . . . . . . . . . . . . . . . . . . . . . 31
2 Stress Tensor and Heat Flux . . . . . . . . . . . . . . . . . . . . . . . 33
3 Balance Laws of Continuum Thermomechanics . . . . . . . . . . . . . 34
4 Closure Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
5 Material Equations in General Internal Variable Formulation . . . . . 40
5.1 Concepts of Dissipation and Thermodynamic Consistency . . . . 40
5.2 General Internal Variable Formulation . . . . . . . . . . . . . . . 42
5.3 Coleman’s Exploitation of the 2nd Law of Thermodynamics . . . . 43
6 Initial Boundary-Value Problem (IBVP) of Thermoinelasticity . . . . 45

IV Finite Element Discretization of Non-linear Boundary-Value

Problems
1 Strong Form of a Nonlinear BVP (Mechanics) . . . . . . . . . . . . . 49
2 Weak Form of Nonlinear BVP . . . . . . . . . . . . . . . . . . . . . . 49
3 Finite Element Discretization . . . . . . . . . . . . . . . . . . . . . . 52
4 Iterative Solution Techniques for Nonlinear Problems . . . . . . . . . 58

V Linear and Nonlinear Isotropic Elasticity

1 Constitutive Formulation of Elasticity . . . . . . . . . . . . . . . . . . 61
2 Representation of Isotropic Free Energy Functions . . . . . . . . . . . 62
2.1 Representation of Stresses . . . . . . . . . . . . . . . . . . . . . . 63
2.2 Representation of Elasticity Moduli . . . . . . . . . . . . . . . . . 65
3 Linear Isotropic Elasticity . . . . . . . . . . . . . . . . . . . . . . . . 66
4 Decoupled Volumetric-Isochoric Elasticity . . . . . . . . . . . . . . . 67

VI Viscoelasticity: Formulation and Numerical Implementation

1 Motivation: One-Dimensional Representations of Linear Viscoelasticity 69
1.1 Representation A: Generalized Internal Variable Formulation . . . 69
1.2 Representation B: In Terms of Internal Forces . . . . . . . . . . . 70
1.3 Representation C: Convolution Integral Representation . . . . . . 70

2
 Contents

2 Three-Dimensional Representation of Linear Viscoelasticity . . . . . . 71

2.1 Volumetric-Isochoric Split . . . . . . . . . . . . . . . . . . . . . . 71
2.2 Representation A: Internal Variable Formulation . . . . . . . . . . 71
2.3 Representation B: In Terms of Internal Forces . . . . . . . . . . . 72
2.4 Representation C: Convolution Integral Representation . . . . . . 73
3 Stress Updates and Consistent Tangent Moduli in Linear
Viscoelasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.1 Algorithmic Representation of Evolution Di↵erential Equations . 74
3.2 Algorithm for Representation C in Terms of Convolution Integral
Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4 Representation of Nonlinear Viscoelasticity . . . . . . . . . . . . . . . 79
4.1 Viscoelasticity with Nonlinear Equilibrium Elastic Response . . . 79
4.2 Viscoelasticity with Nonlinear Evolution Equations . . . . . . . . 80

VII Rate-Independent and Rate-Dependent Elastoplasticity

1 Ideal Elastoplasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
1.1 Elastic Stress Response . . . . . . . . . . . . . . . . . . . . . . . . 82
1.2 Plastic Flow Response . . . . . . . . . . . . . . . . . . . . . . . . 83
1.3 Tangent Moduli (Prandtl-Reuss Tensors) . . . . . . . . . . . . . . 85
1.4 Specification of a Model Problem: von Mises Plasticity . . . . . . 86
1.5 Stress Update Algorithm for von Mises Plasticity: The Radial
Return Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
2 Rate-Independent Plasticity with Hardening . . . . . . . . . . . . . . 94
2.1 General Considerations . . . . . . . . . . . . . . . . . . . . . . . . 94
2.2 Combined Linear Isotropic-Kinematic Hardening in von Mises
Plasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
2.3 Stress Update Algorithm for Hardening Plasticity . . . . . . . . . 100
3 Rate-Dependent Elastoplasticity (Viscoplasticity) . . . . . . . . . . . 104
3.1 Perzyna-type Viscoplasticity for the von Mises Model . . . . . . . 104
3.2 Numerical Implementation of Perzyna-Type Viscoplasticity . . . . 107
4 Associative and Non-Associative Plasticity . . . . . . . . . . . . . . . 109

3
Chapter I

Overview

The graduate level course CE7018 - Computational Inelasticity has been o↵ered
since Spring Semester 2014 at the Department of Civil Engineering, Middle East
Technical University. CE7018 is concerned with the theoretical and computational
aspects of constitutive modeling within the three-dimensional framework. In this
course, we restrict ourselves to geometrically linear kinematics. Those who are
also interested in learning continuum mechanics treated within the geometrically
non-linear setting are referred to another graduate level course CE7026 - Contin-
uum Mechanics. The latter course is o↵ered biennially in an alternating manner
with CE7018 in fall semesters.

This course aims to

equip students with essential knowledge about theoretical and computational
mechanics of elastic and inelastic solids that exhibit non-linear behavior,
make students assimilate the key role of elastic and inelastic material rou-
tines within the modular structure of an implicit non-linear finite element
code in the three-dimensional setting,
teach students how to employ the essential numerical and programming tech-
niques to computationally implement complex inelastic material models with
internal variables.
Taking this course, the students will
• have an essential background on the theoretical and computational mechan-
ics of non-linear elastic and inelastic materials,
• be able to critically assess the thermodynamic consistency of an inelastic
material model within the three-dimensional framework of thermodynamics
with internal variables,
• be qualified to construct algorithmic update equations for the internal vari-
ables of an inelastic model and to derive the three-dimensional tensorial
stress and consistent tangent expressions,
• be capable of constructing the algorithmic framework necessary to imple-
ment an inelastic material model,
• be in a position to implement an inelastic material model into an implicit
three-dimensional finite element code and solve (initial) boundary-value prob-
lems of continuum physics.

It is important to note that these lecture notes are based partly on the lecture
notes by late Prof. Christian Miehe of Universität Stuttgart, Germany.

5
Chapter II

Introduction to Tensor Algebra and Ten-

sor Calculus1

1. Fundamentals of Tensor Algebra

1.1 Algebra of Vectors
A vector, a first order tensor, is a directed line element in space. It is used
to denote physical quantities having both direction and magnitude, e.g. force,
velocity, acceleration. The sum of two vectors yields another vector, based on the
parallelogram law of addition, see Figure II.1.

u+v
u

v
Figure II.1.: Parallelogram law

Scalar (Dot) Product. The scalar product, aka dot or inner product, of
two vectors u and v, denoted by u · v is a scalar quantity

u · v = |u| |v| cos ✓ with 0  ✓  ⇡ (II.1)

where ✓ is the angle between

p two non-zero vectors u and v.
The quantity |u| := u · u is the length (magnitude) of the vector u, a non-
negative number.
The scalar product has the following properties

u·v =v·u
u · (↵v + w) = ↵ u · v + u · w
(II.2)
u·0=0
u·u=0 , u=0

1
Typed by H. Onur Solmaz.

7
II. Introduction to Tensor Algebra and Tensor Calculus

A vector e is called a unit vector if |e| = 1. A non-zero vector u is said to be

orthogonal to a non-zero vector v if
⇡
u · v = 0 with ✓ = (II.3)
2

e3
e2

e1
Figure II.2.: The Cartesian bases

Indicial Notation. In order to present component expressions relative to a

right-handed orthonormal system, a set of three basis vectors e1 , e2 , and e3 are
introduced, as shown in Figure II.2. These basis vectors form the Cartesian bases
with the following properties.

e1 · e2 = e2 · e3 = e1 · e3 = 0
(II.4)
e1 · e1 = e2 · e2 = e3 · e3 = 1

These vectors are unit vectors and mutually orthogonal. Any vector u in the
three-dimensional Euclidean space is represented by a linear combination of the
basis vectors, i.e.,
u = u 1 e 1 + u2 e 2 + u3 e 3 (II.5)
where the three real numbers u1 , u2 , and u3 are Cartesian components of the
vector u along the given direction e1 , e2 , and e3 , respectively. Using the indicial
notation, the expression in (II.5) can be written as
3
X
u= ui e i (II.6)
i=1

or in its short-hand form as

u = ui e i (sum over i = 1, 2, 3) (II.7)

where the Einstein’s summation convention is adapted.

The Summation Convention. Whenever an index occurs twice in a term, a
sum is implemented over that index. Any repeated index, as i in (II.7), is called
a dummy index . Observe that the actual symbol used for a repeated dummy
index is immaterial.
ai bi = a1 b1 + a2 b2 + a3 b3 = aj bj (II.8)
Any index, which is not a dummy index, is called a free index (occurs once).
For example, in the following expression,

ai = c j b j b i = c 1 b 1 b i + c 2 b 2 b i + c 3 b 3 b i (II.9)

the index i is the free index, while j is a dummy one.

8
 1. Fundamentals of Tensor Algebra

Note that in the same equation, an index can be either dummy or free. Free
indices can take of the values 1, 2, or 3, and are used to abbreviate groups of
similar equations. For instance, (II.9) is short-hand for the following equations

a1 = c j b j b 1 , a2 = cj bj b2 , a3 = cj bj b3 . (II.10)

Note that every term in an equation should have the same free indices. Moreover,
the same symbol cannot be used for a dummy and a free index. For example, the
following expressions are not permissible in indicial notation.

ai = b j , ai bj = c1 dj dj or ai bj = ci ck dk dj + dp cl cl dq (II.11)
| {z } | {z }
3 7

Kronecker’s Delta. The relations in (II.4) can be written as

(
1 if i = j
ei · ej = ij := (II.12)
0 if i 6= j

which defines the Kronecker’s Delta, ij . The following properties hold.

ii = 11 + 22 + 33 = 3
ij ui = uj
(II.13)
ij jk = ik and ij = ji (symmetry)
ij ui = 1j u1 + 2j u2 + 3j u3 = uj

where in the last equation, only the one with uj would survive.

The projection of a vector u = ui ei onto the basis vector ej yields the j th

component of u; that is,

u · ej = (ui ei ) · ej = ui (ei · ej ) = ui ij = uj (II.14)

The component representation of the scalar product then becomes

u · v = (ui ei ) · (vj ej )
= ui vj e i · e j
= ui vji
⇤ ij
= ui v i (II.15)
= uij v
⇤ j ij
= uj v j
= u 1 v 1 + u2 v 2 + u3 v 3

Similarly, the square length of u is

|u|2 = u · u = ui ui = u1 u1 + u2 u2 + u3 u3 (II.16)

9
II. Introduction to Tensor Algebra and Tensor Calculus

Cross (Vector) Product. The cross product of two vectors, denoted by u ⇥ v,

produces a vector. The cross product is not commutative.

u ⇥ v = (v ⇥ u)
u ⇥ v = 0 , u and v are linearly dependent
(↵ u) ⇥ v = u ⇥ (↵ v) = ↵ (u ⇥ v) with ↵ 2 R (II.17)
u · (v ⇥ w) = v · (w ⇥ u) = w · (u ⇥ v)
u ⇥ (v + w) = u ⇥ v + u ⇥ w

If (II.17)2 holds for any non-zero u and v, then u is parallel to v. From (II.17)4
we have
u · (u ⇥ w) = 0 (II.18)
The magnitude of the cross product is defined as

|u ⇥ v| = |u| |v| sin ✓ 0✓⇡ (II.19)

and it is equal to the area of a parallelogram spanned by u and v, as shown in

Figure II.3.
u⇥v
u⇥v
|u ⇥ v|
u u

w
v v
V = w · (u ⇥ v)
| {z }
Volume of the parallelepiped
spanned by u, v, w.

Figure II.3.: The geometric interpretations of the vector product

To express the cross product in indicial notation, we introduce the so-called per-
mutation symbol ✏ijk
8
>
<+1 for even permutation of i, j, k; i.e. (123, 231, 312)
✏ijk = 1 for odd permutation of i, j, k; i.e. (132, 213, 321) (II.20)
>
:
0 if there is a repeated index

with the properties ✏ijk = ✏jki = ✏kij , ✏ijk = ✏ikj , and ✏ijk = ✏jik .

Example II.1 (Coding).

2 3
1 0 0
ij = 1 = 40 1 05 = ix(i,j) (II.21)
0 0 1

10
 1. Fundamentals of Tensor Algebra

1
✏ijk = (i j)(j k)(k i) (II.22)
2
Observe that the right-handed orthonormal basis {ei } fulfill

ei ⇥ ej = ✏ijk ek (II.23)

which implies the following identities

e1 ⇥ e2 = e3 e2 ⇥ e1 = e3
e2 ⇥ e3 = e1 e3 ⇥ e2 = e1
(II.24)
e3 ⇥ e1 = e2 e1 ⇥ e3 = e2
e1 ⇥ e1 = e2 ⇥ e2 = e3 ⇥ e3 = 0

Furthermore, we have
✏ijk ✏pqk = ip jq iq jp
✏ijk ✏pjk = 2 pi (HE) (II.25)
✏ijk ✏ijk = 6
The coordinate representation of the cross-product w = u ⇥ v is

w = u ⇥ v = (ui ei ) ⇥ (vj ej )
= ui vj (ei ⇥ ej )
(II.26)
wk ek = ui vj ✏ijk ek using (II.23)
wk = ✏ijk ui vj = ✏kij ui vj

Then
w1 e1 + w2 e2 + w3 e3 = ✏123 u1 v2 e3 + ✏132 u1 v3 e2
+ ✏231 u2 v3 e1 + ✏213 u2 v1 e3
+ ✏312 u3 v1 e2 + ✏321 u3 v2 e1 (II.27)
= u1 v2 e 3 u1 v3 e 2 + u2 v3 e 1
u2 v 1 e 3 + u3 v 1 e 2 u3 v 2 e 1
w 1 = u2 v 3 u3 v2
w 2 = u3 v 1 u1 v3 (II.28)
w 3 = u1 v 2 u2 v1
Also, note that we have
2 3
e1 e2 e3
u ⇥ v = det 4u1 u2 u3 5 = ✏ijk ui vj ek (II.29)
v1 v2 v3

and
1
ei = ✏ijk ej ⇥ ek (HE). (II.30)
2

11
II. Introduction to Tensor Algebra and Tensor Calculus

1.2 Algebra of Tensors

Second-Order Tensors. A second-order tensor A may be conceived as a linear
operator that acts on a vector u generating another vector v; i.e.

v = Au (II.31)

which defines a linear transformation between u and v. Since A is linear, we have

A(↵u + v) = ↵Au + Av (II.32)

for all vectors u, v 2 R3 and a scalar ↵ 2 R.

If A and B are two 2nd -order tensors, we have

(A + B)u = Au + Bu
(II.33)
(↵A)u = ↵Au

where u is an arbitrary vector. The second-order identity tensor 1 and the zero
tensor 0 are defined as

1u = u1 = u and 0u = u0 = 0 (II.34)

for all u 2 R3 .

Tensor Product The tensor product (or dyad) of the vectors u and v, denoted
by u ⌦ v yields a second order tensor, which linearly transforms a vector w onto
a vector having the same direction as u. That is,

(u ⌦ v) w = u (v · w) = (v · w) u (II.35)

The dyad (⌦) has the following properties

(u ⌦ v)(↵w + x) = ↵ (u ⌦ v) w + (u ⌦ v) x
(↵u + v) ⌦ w = ↵ (u ⌦ w) + (v ⌦ w)
u (v ⌦ w) = (u · v) w (II.36)
(u ⌦ v)(w ⌦ x) = (v · w)(u ⌦ x)
A (u ⌦ v) = (Au) ⌦ v

Note that the dyad is not commutative; i.e. u⌦v 6= v⌦u. The second order tensor
A may be expressed as a linear combination of dyads formed by the Cartesian
bases
A = Aij ei ⌦ ej (II.37)

Note II.1. The number of components of a tensor can be obtained using the power
n-order
rule space , where space is the number of dimensions of the space the tensor is
decomposed in, and n-order is the order of the tensor. For example, the number
of components of a second order tensor in 3-D space is 32 = 9.

The nine Cartesian components of A with respect to {ei }, denoted by Aij , form

12
 1. Fundamentals of Tensor Algebra

the entries of the matrix [A].

2 3
A11 A12 A13
[A] = 4A21 A22 A23 5 (II.38)
A31 A32 A33

which is the matrix representation of A. Analogous to (II.25)2 , we have

Aij⇤ jk
⇤
= Aik (II.39)

The Cartesian components of the unit tensor 1 form the Kronecker’s delta

1= ij ei ⌦ ej = ei ⌦ ei = ej ⌦ ej (II.40)

The projection of a 2nd -order tensor A onto the orthonormal bases ei is given by

Aij = ei · Aej
= ei · (Akl ek ⌦ el ) ej
(II.41)
= ei · (Akl ek lj )
= Akl lj ik = Aij

The linear transformation v = Au in (II.31) can be written as

vi = Aij uj (II.42)

Proof.
v = Au
= (Aij ei ⌦ ej )(uk ek )
= Aij uk (ei ⌦ ej ) ek
(II.43)
= Aij uk jk ei
vi ei = Aik uk ei
= Aij uj ei
Then
v1 = A1j uj , v2 = A2j uj , v3 = A3j uj (II.44)

The second order tensors A and B are equal if and only if

Au = Bu for all u 2 R3 and u 6= 0 . (II.45)

Note II.2 (Programming linear transformations).

v = Au ) vi = Aij uj (II.46)

13
II. Introduction to Tensor Algebra and Tensor Calculus
2 3 2 32 3
[ !] "
44 5 = 4[ 4 !]5 4 | 5 (II.47)
⇤ [ ⇤ !] #
| {z } | {z } |{z}
v A u

Pseudo-code for v = Au

1 for i = 1 ,2 ,3
2 v(i) = 0
3 for j = 1 ,2 ,3
4 v ( i ) = v ( i ) + A (i , j ) * u ( j )
5 end
6 end

Product of Second-Order Tensors (Single Contraction). The product of

A and B, denoted by AB, is again a second-order tensor. It fulfills the following
identity
(AB)u = A(Bu) (II.48)
The components of AB can be obtained as

(AB)ij = ei · (AB)ej = ei · A(Bej )

= ei · (A(Bkn ek ⌦ en )ej )
= (ei · Aek )Bkn nj (II.49)
= Aik Bkn nj
= Aik Bkj

Furthermore,
(AB)C = A(BC)
A2 = AA (II.50)
(A + B)C = AC + BC
Note that in general AB 6= BA and Au 6= uA. Moreover AB = 0 and Au = 0
do not always imply that A, B, or u is zero.
Transpose of a Tensor. The unique transpose of A, denoted AT , is governed
by
v · AT u = u · Av = (Av) · u (II.51)
for all u, v 2 R3 and A 2 R3⇥3 . Every tensor A has a transpose

AT = Aji ei ⌦ ej (II.52)

and the following hold

(AT )T = A
(↵A + B)T = ↵AT + B T
(II.53)
(AB)T = B T AT
(u ⌦ v)T = v ⌦ u

14
 1. Fundamentals of Tensor Algebra

Furthermore, from (II.53), we obtain

(AT )ij = Aji = (ei · AT ej ) (II.54)

Trace and Contraction. The trace is a linear operation that assigns a scalar
tr(A) to each tensor A through

tr(u ⌦ v) = u · v (II.55)

for any u, v 2 R3 . Linearity is the requirement that

tr(↵A + B) = ↵ tr(A) + tr(B) (II.56)

Then,
tr(A) = tr(Aij ei ⌦ ej )
= Aij tr(ei ⌦ ej )
(II.57)
= Aij ei · ej
= Aij ij = Aii
where we have used (II.55). Therefore

tr(A) = Aii = A11 + A22 + A33 (II.58)

The trace operator fulfills the following

tr(Au ⌦ v) = tr(v ⌦ Au)

tr(AT ) = tr(A)
tr(AB) = tr(BA)
(II.59)
tr(A + B) = tr(A) + tr(B)
tr(↵A) = ↵ tr(A)
tr(1) = 3

Double Contraction of Two Tensors. The double contraction of A and B,

denoted by A : B, yields a scalar and is defined by

A : B = tr(AT B) = tr(B T A)
(II.60)
= tr(AB T ) = tr(BAT ) = B : A

In indicial notation, knowing

AT B = ATim Bmj ei ⌦ ej
(II.61)
= Ami Bmj ei ⌦ ej

A : B = tr(AT B)
= tr(Ami Bmj ei ⌦ ej )
= Ami Bmj tr(ei ⌦ ej ) (II.62)
= Ami Bmj ij
= Ami Bmi

15
II. Introduction to Tensor Algebra and Tensor Calculus

Thus,
A : B = Aij Bij (II.63)
and the following hold

1 : A = tr(A) = A : 1
A:B=B:A
A : (BC) = (B T A) : C = (AC T ) : B (II.64)
A : (u ⌦ v) = u · Av
(u ⌦ v) : (w ⌦ x) = (u · w)(v · x)

Note that, in general A : B = C : B does not imply A = C.

The norm of a tensor A is denoted by kAk. This non-negative quantity is defined
by
1 1
kAk = (A : A) 2 = (Aij Aij ) 2 0. (II.65)
Determinant and Inverse of a Tensor. The coordinate-free definition of the
determinant is given through the volume, i.e.

(Au) · (Av ⇥ Aw) = det(A) u · (v ⇥ w) (II.66)

for u, v, w 2 R3 and A 2 R3⇥3 . Recall that u · (v ⇥ w) is none other than the

volume V of a parallelepiped spanned by u, v, and w as shown in Figure II.3.

V = u · (v ⇥ w) = v · (w ⇥ u) = w · (u ⇥ v) (II.67)

From (II.66) and (II.67), det(A) can be considered the ratio of the volume of the
parallelepiped spanned by Au, Av, and Aw to that of the parallelepiped spanned
by u, v, and w. It can be shown that
2 3
A11 A12 A13
det(A) = det 4A21 A22 A23 5 (II.68)
A31 A32 A33

1
det(A) = ✏ijk ✏mno Aim Ajn Ako (II.69)
6
det(AT ) = det(A)
det(AB) = det(BA) = det(A) det(B) (II.70)
det(↵A) = ↵3 det(A)
for ↵ 2 R and A, B 2 R3⇥3 . A tensor A is said to be singular if det(A) = 0 . If
A is non-singular, then there exists a unique inverse A 1 of A such that
1
AA = 1 = A 1A (II.71)

16
 1. Fundamentals of Tensor Algebra

For invertible tensors A and B, we have

(AB) 1
= B 1A 1

1 1
(↵A) = ↵ 1A
(A 1 ) 1
=A
(II.72)
(A 1 )T = (AT ) 1

2
A = A 1A 1

det(A 1 ) = (det(A)) 1

T
In what follows, we use A := (A 1 )T .

Orthogonal Tensors. A tensor Q is orthogonal if

Qu · Qv = u · v (II.73)

for all u, v 2 R3 . Setting v = u in (II.73), we have

Qu · Qu = u · u = |u|2 ! |Qu| = |u| (II.74)

Therefore, an orthogonal tensor Q operating on u leaves the length of u unaltered.

Similarly, the angle between two vectors u and u remains unchanged when both
transformed by Q; that is,
(
(Qu) · (Qv) = |Qu| |Qv| cos ✓
⌘ (II.75)
u · v = |u| |v| cos ✓

The basic algebraic property of an orthogonal tensor is

QT = Q 1
(II.76)

which is a necessary and sufficient condition for being orthogonal. As a conse-

quence,
QT Q = QQT = 1 (II.77)
From (II.71), we can conclude that

det(QT Q) = det(QT ) det(Q) = (det(Q))2 = 1

(II.78)
! det Q = ±1

An orthogonal tensor Q is a rotation if det Q = +1, a reflection if det Q = 1.

Symmetric and Skew-Symmetric Tensors. A tensor A is symmetric if

A = AT (II.79)

and skew-symmetric if
A= AT (II.80)

17
II. Introduction to Tensor Algebra and Tensor Calculus

We define
1
sym(A) := (A + AT )
2
1 (II.81)
skw(A) := (A AT )
2

as the symmetric and skew-symmetric parts of A respectively. Based on the defi-

nition (II.81), we have skw(sym(A)) = 0 and sym(skw(A)) = 0.

Apparently, we also have

A = sym(A) + skw(A)
1 1 (II.82)
= (A + AT ) + (A AT )
2 2
Moreover, we have

(sym(A)) : B = sym(A) : B T
(skw(A)) : B = skw(A) : B T (II.83)
sym(A) : skw(B) = 0

Higher-Order Tensors. Any tensor of order/rank n may be expressed as

A = Ai1 i2 i3 ···in ei1 ⌦ ei2 ⌦ ei3 ⌦ · · · ⌦ ein (II.84)

where (i1 , i2 , i3 , . . . , in = 1, 2, 3). As mentioned above, a tensor of order n in the

3-D space has 3n components,
Zeroth-order (scalar) n=0 30 = 1 (·)
23
·
First-order (vector) n=1 31 = 3 4· 5
·
2 3
· · ·
Second-order (tensor) n=2 32 = 9 4· · ·5
· · ·

Tensors of 3rd Order. Typical example is the permutation symbol

✏ = ✏ijk ei ⌦ ej ⌦ ek (II.85)

where ✏ijk = (ei ⇥ ej ) · ek are 33 = 27 components of ✏.

(u ⌦ v) ⌦ w = u ⌦ v ⌦ w
(u ⌦ v ⌦ w)x = (w · x)(u ⌦ v)
(II.86)
(u ⌦ v ⌦ w) : (x ⌦ y) = (v · x)(w · y)u (HE)
(u ⌦ v ⌦ w) : 1 = (v · w)u (HE)

18
 1. Fundamentals of Tensor Algebra

Also, it can be shown that

✏ : (u ⌦ v) = u ⇥ v (II.87)

Proof.

(✏ijk ei ⌦ ej ⌦ ek ) : (um em ⌦ vn en ) = ✏ijk um vn jm kn ei

= ✏ijk uj vk ei = u ⇥ v (II.88)
| {z }
(u⇥v)i

Tensors of 4th Order. Any fourth-order tensor A has 34 = 81 components. A

typical 4th -order tensor is the elasticity tensor (moduli) of a solid.

= ijkl e i ⌦ ej ⌦ e k ⌦ e l (II.89)

We have
(u ⌦ v) ⌦ (w ⌦ x) = u ⌦ v ⌦ w ⌦ x
(II.90)
(u ⌦ v ⌦ w ⌦ x) : (y ⌦ z) = (w · y)(x · z)(u ⌦ v)
The components of can be obtained as

ijkl = (ei ⌦ ej ) : : (ek ⌦ el ) (II.91)

Note that the double contraction of with a second-order tensor A is a second-

order tensor
: A = ( ijkl ei ⌦ ej ⌦ ek ⌦ el ) : (Amn em ⌦ en )
= ijkl Amn km ln ei ⌦ ej (II.92)
= ijkl Akl ei ⌦ ej
We also have
(A ⌦ B) : C = A(B : C) = (B : C)A
= Aij Bkl Ckl ei ⌦ ej
(II.93)
A : (B ⌦ C) = (A : B)C
(A ⌦ B) : (C ⌦ D) = (B : C)(A ⌦ D)

The 4th -order identity tensor is defined by

A: =A (II.94)

for any second-order tensor A 2 R3⇥3 . The 4th -order identity can be represented
as
= ik jl ei ⌦ ej ⌦ ek ⌦ el if A 6= AT
1 (II.95)
= ( ik jl + il jk ) ei ⌦ ej ⌦ ek ⌦ el if A = AT
2

19
II. Introduction to Tensor Algebra and Tensor Calculus

2. Eigenvalue Problem of Tensors

A generalized eigenvalue problem of a 2nd -tensor A is defined by

An = n (II.96)

The scalar denotes an eigenvalue (principal) value of the tensor A and n is the
corresponding eigenvector. In 3-D, a second-order tensor A has three eigenvalues
and three eigenvectors. Thus, the equality (II.96) holds for any of the three pairs.

An↵ = ↵ n↵ (↵ = 1, 2, 3; no sum) (II.97)

The expression (II.97) can be recast into the following form

(A 1) n = 0 . (II.98)

For non-zero eigenvectors (n 6= 0), the latter is satisfied for vanishing determinant.

det(A 1) = 0 (II.99)

which can be expressed as

3 2
I1 + I2 I3 = 0 (II.100)

This latter is referred to as the Characteristic Equation of A. The roots of

(II.100) give the eigenvalues ↵=1,2,3 of A. The coefficients of (II.100) are called
the Principal Invariants of A and defined as

I1 (A) := tr(A) = Aii = A11 + A22 + A33

1 1
I2 (A) := [(tr A)2 tr(A2 )] = (Aii Ajj Aij Aji ) (II.101)
2 2
I3 (A) := det(A) = ✏ijk A1i A2j A3k (HE)

The principal invariants can also be expressed in terms of the eigenvalues of A;

that is,
I1 (A) := 1 + 2 + 3
I2 (A) := 1 2 + 2 3 + 1 3 (II.102)
I3 (A) := 1 2 3
Note that the eigenvalues of a symmetric and positive-definite second-order tensors
are real and positive. Moreover, the eigenvectors of a symmetric tensor constitute
an orthonormal basis system.
Proof.
T
A
| =
{zA } u · Au = A : (u ⌦ u) > 0 8u 2 R3 (II.103)
| {z }
symmetry positive-definiteness

20
 2. Eigenvalue Problem of Tensors

n · (An↵ = ↵ n↵ )
n↵ · (An = n )

(II.104)
n · An↵ n↵ · An = ↵n · n↵ n↵ · n
n↵ · (AT A) n = ( ↵ ) n↵ · n
0=( ↵ ) n↵ · n
for ↵ 6= . Therefore,
n↵ ?n ↵ 6= . (II.105)

It can be shown that every tensor satisfies its own characteristic equation.

A3 I 1 A2 + I 2 A I3 1 = 0 (II.106)

which is referred to as the Cayley-Hamilton Theorem.

Proof.
3 2
I1 + I2 I3 = 0 (II.107)
which is factorized and multiplied with the eigenvector n
2
n I1 n + I2 n I3 n = 0 (II.108)

Substituting n = An, we obtain

2
An I1 An + I2 An I3 n = 0 (II.109)

Repeating this procedure twice more, we eventually obtain (II.106).

Spectral Decomposition of a Symmetric Tensor. Since the eigenvectors of
a symmetric tensor form orthonormal bases n1 , n2 , and n3 , these can be related
to the Cartesian bases e1 , e2 , and e3 through a rotation tensor Q

n↵ = Qe↵ for all ↵ = 1, 2, 3 (II.110)

The components of Q are the so-called direction cosines and can be expressed

e3
n2

n1
e2 Q

e1
n3
Figure II.4.: The transformation of Cartesian bases

in terms of the components of the eigenvectors n↵ .

21
II. Introduction to Tensor Algebra and Tensor Calculus

From (II.110) we can deduce that Q = nk ⌦ ek . The components Qij of Q with

respect to the Cartesian bases can be obtained as

Qij = ei · Qej = ei · (nk ⌦ ek ) ej = ei · nj (II.111)

This leads us to the following matrix representation of Q with respect to {ei }

2 3 2 3
" " " n1,1 n2,1 n3,1
[Q] = [Qij ] = 4n1 n2 n3 5 = 4n1,2 n2,2 n3,2 5 (II.112)
# # # n1,3 n2,3 n3,3

Since ei ⌦ ei = 1, we have

Q1QT = QQT = 1 = Q(ei ⌦ ei )QT

= (Qei ) ⌦ (Qei ) (II.113)
1 = ni ⌦ ni

Multiplying (II.97) with n↵ through the dyadic product from right and summing
the resulting expression over ↵ = 1, 2, 3, we get
3
X 3
X
A n↵ ⌦ n↵ = ↵ n↵ ⌦ n↵ (II.114)
↵=1 ↵=1

Factorizing the left-hand side, we get

3
X 3
X
A n↵ ⌦ n↵ = ↵ n↵ ⌦ n↵ (II.115)
↵=1 ↵=1
| {z }
1

and thus
3
X
A= ↵ n↵ ⌦ n↵ (II.116)
↵=1

which is called the spectral representation of A. The latter implies that

A = Aij ei ⌦ ej = 1 (n1 ⌦ n1 ) + 2 (n2 ⌦ n2 ) + 3 (n3 ⌦ n3 ) (II.117)

That is, the components of A with respect to {n↵ } is a diagonal tensor. Observe
that the components of A with respect to ei can be obtained through
3 3
!
X X
A= ↵ (n↵ ⌦ n↵ ) = Q ↵ e↵ ⌦ e ↵ QT (II.118)
↵=1 ↵=1
| {z }
⇤

or in a shorter form
A = Q⇤QT (II.119)
where ⇤ is a diagonal tensor in {e↵ }. This equality can be demonstrated in matrix

22
 3. Tensor Functions and Derivatives

form as
2 3 2 32 32 3
A11 A12 A13 " " " 1 n1 !
4A21 A22 A23 5 = 4n1 n2 n3 5 4 2
54 n 2 !5 (II.120)
A31 A32 A33 # # # 3 n3 !

3. Tensor Functions and Derivatives

A tensor function is a function of one or more tensors and maps its arguments to
scalars, vectors, or tensors. For instance, the functions
( )
R3⇥3 ! R
(B) = a scalar-valued tensor function
B 7! (B)
( )
R3⇥3 ! R3
u(B) = a vector-valued tensor function (II.121)
B 7! u(B)
( )
R3⇥3 ! R3⇥3
A(B) = a tensor-valued tensor function
B 7! A(B)

Likewise, (u), v(u), and A(u) are vector functions of u with the value of a
scalar, vector, and tensor, respectively.
( )
R3 ! R
(u) =
u 7! (u)
( )
R3 ! R3
v(u) = (II.122)
u 7! v(u)
( )
R3 ! R3⇥3
A(u) =
u 7! A(u)

Let (t), u = u(t) = ui (t) ei and A = A(t) = Aij (t) ei ⌦ ej be scalar-, vector- and
tensor-valued scalar functions with fixed Cartesian bases {ei }. First derivatives of
u and A with respect to t, which are denoted by u̇ = du dt
and Ȧ = dA
dt
, are given
by
u̇ = u̇i (t) ei and Ȧ = Ȧij (t) ei ⌦ ej (II.123)
Note that dei = 0 and ˙ = d . Furthermore, the product rule applies as follows
dt dt

(t)u(t) = ˙ (t) u(t) + (t) u̇(t)

•
(II.124)
a ⌦ b = ȧ ⌦ b + a ⌦ ḃ

and •

Au = Ȧu + Au̇
• (II.125)
AB = ȦB + AḂ
•

1
Example II.2. Show that A = A 1 ȦA 1

23
II. Introduction to Tensor Algebra and Tensor Calculus

•
•
1
Since AA = 1 = 0,
• •

1 1 1
ȦA + AA =0 ) A = A 1 ȦA 1
(II.126)

Frechet and Gateaux Derivatives. The Frechet derivatives of scalar-, vector-,

and tensor-valued functions are
d (t)
D (t) = =: @t (t) ! scalar
dt
d (u)
D (u) = =: @u (u) ! vector (II.127)
du
d (A)
D (A) = =: @A (A) ! 2nd -order tensor
dA
The Gateaux derivative is non-other than a directional derivative in the specified
directions given by the increments of the corresponding arguments t, u, A,

d
D (t) t = (t + ✏ t) 8 t2R
d✏ ✏=0
d
D (u) · u = (u + ✏ u) 8 u 2 R3 (II.128)
d✏ ✏=0
d
D (A) : A = (A + ✏ A) 8 A 2 R3⇥3
d✏ ✏=0

The Gateaux derivative is a simple mechanism to determine derivatives with re-

spect to vectors or tensors, through the derivatives with respect to a scalar ✏.

Example II.3 (Derivative of the first principal invariant of a second-order tensor).

Recall that
(A) = I1 (A) = tr(A) = A : 1 (II.129)
We want to obtain
@I1 (A)
DI1 (A) = =? (II.130)
@A
In order to calculate this, we first start with the Gateaux derivative
d
D (A) : A= [tr(A + ✏ A)]✏=0
d✏
d
= [(A + ✏ A) : 1]✏=0
d✏ (II.131)
d
= [(A : 1 + ✏ A : 1]✏=0
d✏
= A:1=1: A

Thus, we can conclude that

D (A) = DI1 (A) = 1 (II.132)

24
 3. Tensor Functions and Derivatives

Component-Based Derivative of a Tensor Function. Assume (A) is a

scalar-valued tensor function and recall that

A = Aij ei ⌦ ej = A11 e1 ⌦ e1 +A12 e1 ⌦ e2 +A13 e1 ⌦ e3 +· · ·+A33 e3 ⌦ e3 (II.133)

The Frechet derivative becomes

@ @
D (A) = = ei ⌦ ej = @Aij ei ⌦ ej
@A @Aij
@ @ @ @
= e1 ⌦ e1 + e1 ⌦ e 2 + e1 ⌦ e3 + · · · + e3 ⌦ e 3
@A11 @A12 @A13 @A33
(II.134)

Example II.4 (Derivative of I1 (A) with respect to A). We know that I1 =

tr(A) = Akk . The Frechet derivative then reads

@I1 (A) @Akk

= ei ⌦ ej (II.135)
@A @Aij

where we have the following logical expression

(
@Akk 1 if (i = k) ^ (j = k)
= (II.136)
@Aij 0 otherwise

One of the quantities that demonstrates this property is Kronecker’s delta. Hence
we can write
@I1 (A)
= ik jk ei ⌦ ej
@A (II.137)
= ij ei ⌦ ej = 1

Remark II.1 (Derivative of a tensor with respect to itself).

@A @Aij
= ei ⌦ e j ⌦ ek ⌦ el
@A @Akl
(II.138)
= ik jl ei ⌦ ej ⌦ ek ⌦ el
=

if A 6= AT . Otherwise, we impose A = 12 (A + AT ) and


@A @ 1
= (A + AT )
@A @A 2
@ 12 (Aij + Aji ) (II.139)
= ei ⌦ e j ⌦ e k ⌦ e l
@Akl
1
= [ ik jl + jk il ] ei ⌦ ej ⌦ ek ⌦ el
2

25
II. Introduction to Tensor Algebra and Tensor Calculus

In literature, we often find

[ ]ijkl = ik jl for A 6= AT
1 (II.140)
[ ]ijkl = ( ik jl + il jk ) for A = AT
2
Example II.5.
@tr(A2 )
=? for A 6= AT (II.141)
@A
1. Component-Based

tr(A2 ) = A2 : 1 = AA : 1 = A : AT
(II.142)
= Aik ATik = Aik Aki

Then,
@tr(A2 ) @(Aik Aki )
= em ⌦ en (II.143)
@A @Amn

@tr(A2 ) @Aik @Aki
= Aki + Aik
@A mn @Amn @Amn
(II.144)
= im kn Aki + Aik km in
= 2Anm = 2ATmn
Hence
@tr(A2 )
= 2ATmn em ⌦ en = 2AT (II.145)
@A
2. Gateaux Derivative
d
D (A) : A= (A + ✏ A)
d✏ ✏=0
d ⇥ ⇥ ⇤⇤
= tr (A + ✏ A)2
d✏ ✏=0
d ⇥ ⇤
= tr(A2 + ✏A A + ✏ AA + ✏2 A2 )
d✏ ✏=0
d ⇥ 2 ⇤
= A : 1 + ✏(A A) : 1 + ✏( AA) : 1 + ✏2 A2 : 1
d✏ ✏=0
⇥ ⇤
= (A A) : 1 + ( AA) : 1 + 2✏ A2 : 1
✏=0
= (A A) : 1 + ( AA) : 1
(II.146)
2
Here, we aim to obtain an expression that has the form @A tr(A ) : A.
Continuing with the derivation,

= AT AT : 1 + (AT AT ) : 1
= AT : A + AT : A (II.147)
= 2AT : A

26
 3. Tensor Functions and Derivatives

Then
@ (A) @tr(A2 )
= = 2AT (II.148)
@A @A
Example II.6. If the free energy (strain energy) of linear isotropic elasticity is
given by
1
W (") = tr2 (") + µ" : " (II.149)
2
Determine the stress tensor
@W
= (II.150)
@"
and the elasticity tensor
@ @ 2W
= = (II.151)
@" @"2
Note that, by definition, the strain tensor " is symmetric, i.e. " = "T .

d
DW (") : "= : "= W (" + ✏ ") (II.152)
d✏ ✏=0

d 1
= [(" + ✏ ") : 1]2 + µ(" : " + ✏( " : " + " : ") + ✏2 " : ")
d✏ 2 ✏=0

= [ [(" + ✏ ") : 1] [ " : 1] + µ [ " : " + " : " + 2✏ " : "]]

✏=0
= (" : 1)(1 : ") + 2µ(" : ")
= [ tr(")1 + 2µ"] : "
(II.153)
Hence
= tr(")1 + 2µ" (II.154)
The moduli are also derived as
d
ˆ = D ˆ (") : "= [ ˆ (" + ✏ "]
| {z } d✏ ✏=0
=@" ˆ
d
= [ 1 tr(" + ✏ ") + 2µ(" + ✏ ")]
d✏ ✏=0 (II.155)
d
= [ 1(" + ✏ ") : 1 + 2µ(" + ✏ ")]
d✏ ✏=0
= 1(1 : ") + 2µ "
= [ 1 ⌦ 1 + 2µ sym. ] : "

i.e.
sym.
= 1 ⌦ 1 + 2µ (II.156)
Once one becomes acquainted with such formulations, one can follow an easier
path for this derivation. We know that

@tr(") @tr("2 ) @"

= 1, = 2"T , = sym.
(⇤)
@" @" @"

27
II. Introduction to Tensor Algebra and Tensor Calculus

and we have the free energy

W (") = tr2 (") + µ(" : ") = tr2 (") + µ tr("2 ) (II.157)

2 2
Then we can use (⇤) in the derivation

@W @tr(") @tr("2 )
= = tr(") +µ
@" @" @" (II.158)
= tr(")1 + 2µ"

Similarly,
@ @tr(") @"
= = 1⌦ + 2µ
@" @" @" (II.159)
= 1 ⌦ 1 + 2µ

Di↵erential Operators: Gradient and Divergence. In continuum mechan-

ics, there are four important di↵erential operators: gradient, divergence, curl, and
Laplacian. These di↵erential operations involve derivatives of a function with re-
spect to its vector argument, which is often the position vector. The problems
tackled in this course involve mostly the gradient and divergence operations.

Gradient. The gradient operation can be defined through the following particular
Gateaux derivative

d
D (x) · x= (x + ✏ x) =: r · x 8 x 2 R3
d✏ ✏=0
(II.160)
d 3
Dv(x) · x = v(x + ✏ x) =: rv x 8 x2R
d✏ ✏=0
For example, the gradient of a vector v and a second-order tensor A can be
expressed as follows

@v @(vi ei ) @vi
rv = ⌦ ej = ⌦ ej = ei ⌦ ej
@xj @xj @xj
(II.161)
@A @(Aij ei ⌦ ej ) @Aij
rA = ⌦ ek = ⌦ ek = ei ⌦ ej ⌦ ek
@xk @xk @xk
From its definition and the latter example, we note that the gradient operator
increases the rank of the tensor by one.
Divergence. The divergence of a tensor field A(x) of order n 1 is a tensor of
rank n 1; that is, as opposed to the gradient operation, the divergence operation
reduces the order of the tensor by one.

The divergence of a vector v and a second-order tensor A are defined as

28
 3. Tensor Functions and Derivatives

@v @(vi ei ) @vi @vi @vi

div v := · ej = · ej = ei · ej = ij =
@xj @xj @xj @xj @xi
@A @(Aij ei ⌦ ej ) @Aij @Aij @Aij
div A := ek = ek = (ei ⌦ ej )ek = jk ei = ei
@xk @xk @xk @xk @xj
(II.162)

29
Chapter III

Thermodynamics of the 3-D Continuum

1. Kinematics at Small Strains

Let B ⇢ R3 a continuous body and I ⇢ R+ be a time interval, and u(x, t) be the
displacement field describing the the position change of a material point P 2 B,
initially situated at point x, at time t 2 I, see Figure III.1. That is;
(
B ⇥ I ! R3
u(x, t) : (III.1)
(x, t) 7! u(x, t)

R3 e3
Deformed body at time t
e2
e1

P u
B x2B P0

Figure III.1.: Schematic representation of the displacement map u(x, t) : B ⇥ I !

R3 within the 3-D Euclidean space

The velocity and acceleration fields are then defined as

@u(x, t)
v(x, t) := = u̇(x, t)
@t (III.2)
@v(x, t)
a(x, t) := = ü(x, t)
@t
The linearized strain tensor at x 2 B is defined by the symmetric part of the
displacement displacement gradient ru
1
"(x, t) := sym[ru] = [ru + rT u] . (III.3)
2
@ui @ui @uj
Recall that ru := @xj
ei ⌦ ej , which yields "ij = 12 ( @x j
+ @xi
).

31
III. Thermodynamics of the 3-D Continuum

The components "ij = ei · "ej of the strain tensor " are called the normal strains
for i = j and the shear strains for i 6= j. That is, in the following symbolized
representation 2 3
⇤ 4 4
["] = 44 ⇤ 45
4 4 ⇤
the diagonal components (⇤) correspond to the normal strains, while the o↵-
diagonal components (4) correspond to shear strains.

In the geometrically linear theory, the trace of the strain tensor " measures the
volumetric strain
e := tr(") = "ii = "11 + "22 + "33 (III.4)
while the deviator of "
1
dev(") := " e1 (III.5)
3
represents the isochoric (volume-preserving) part of the deformation. The schematic
representations of these deformations are given in Figures III.2 and III.3, respec-
tively.
2 3
e2 e2
0 0
6 7
" = 40 05
0 0
e 6= 0
dev(") = 0
< 0 compression
e1 e1 > 0 expansion
expansion compression
volumetric deformation

Figure III.2.: Schematic demonstration of purely volumetric deformations where

dev " = 0

e2 2 3 e2 2 3
0 0 0 0
6 7 6 7
"=4 0 05 " = 40 05
0 0 0 0 0 0
e=0 e=0
e1 dev(") 6= 0 e1 dev(") 6= 0
simple shear in e1 -e2 pure shear in e1 -e2
volume-preserving deformations

Figure III.3.: Schematic demonstration of purely isochoric deformations where

tr " = 0

32
 2. Stress Tensor and Heat Flux

2. Stress Tensor and Heat Flux

Consider an arbitrary part P ⇢ B of the body B. The thermomechanical action
of the rest of the body B \ P in the vicinity is represented by the two fundamental
quantities of continuum thermomechanics: the stress traction vector t and the
(outward) heat flux h (scalar) as demonstrated in Figure III.4.
t
n
h
x
x

B
P⇢B P⇢B @P

Figure III.4.: Isolation of the part P cut out from the body B

These quantities are linear functions of the orientation of the cut (surface of P)
at x 2 @P, characterized by the unit surface normal n. These linear relationships
are expressed through the stress tensor and the heat flux vector q and are
defined by Cauchy’s theorem

t= n and h = q · n . (III.6)

In an infinitesimal volume element dV ⇢ B aligned with the coordinate axes, the

components of ( ij = ei · ej ) and q (qi = q · ei ) can be illustrated as in Figure
III.5 where 11 , 22 and 33 are normal stresses and 12 , 23 , 13 , 21 , 32 and 31
are shear stresses.
e3 e3

33 q3
32

31 23

13 22 q2
12 21 e2 e2
11 q1

e1 e1

Figure III.5.: Illustration of the components of the stress tensor and the heat
flux vector q

33
III. Thermodynamics of the 3-D Continuum

3. Balance Laws of Continuum Thermomechanics

For the cut out part P ⇢ B, depicted in Figure III.4, we define the following basic
physical quantities
Z Z
(i) Mass m := dm = ⇢ dV
P P
Z
(ii) Linear Momentum I := ⇢v dV
P
Z
(iii) Angular Momentum D 0 := x ⇥ ⇢v dV
P
Z
1
(iv) Kinetic Energy K := ⇢ |v|2 dV (III.7)
P 2
Z
(v) Internal Energy E := ⇢e dV
P
Z
(vi) Entropy H := ⇢⌘ dV
P
Z
(vi) Entropy Production := ⇢ dV
P

where the quantities e, ⌘, and are defined per unit mass. Note that the fields
⇢(x, t), u(x, t), v(x, t), e(x, t), ⌘(x, t), and (x, t) describe the density, displace-
ment, velocity, internal energy, entropy, and entropy production, respectively.

As mentioned in the preceding section, the isolated part P ⇢ B of the continuum

is loaded by the surface loads (the stress traction vector t and the heat flux h)
and the volume loads (the body force b (per unit mass) and the heat source r (per
unit mass)). The global quantities associated with these loads are
Z Z
(i) Mechanical Force F := ⇢b dV + t dA
P @P
Z Z
(ii) Mechanical Couple M 0 := x ⇥ ⇢b dV + x ⇥ t dA
P @P
Z Z
(iii) Mechanical Power P := ⇢b · v dV + t · v dA (III.8)
P @P
Z Z
(iv) Thermal Power Q := ⇢r dV h dA
P @P
Z Z
r h
(v) Entropy Power S := ⇢ dV dA
P ✓ @P ✓

The fundamental conservation laws of continuum mechanics set the relationships

34
 3. Balance Laws of Continuum Thermomechanics

between the physical fields (III.7) and the thermodynamic sources/loads (III.8).

dm
(i) Conservation of Mass (closed system) =0
dt
dI
(ii) Conservation of Linear Momentum =F
dt
dD 0
(iii) Conservation of Angular Momentum = M0 (III.9)
dt
d
(iv) Conservation of Energy (1st Law) (K + E) = P + Q
dt
dH
(v) Conservation of Entropy (2nd Law) = S 0
dt

Remark III.1 (Entropy Balance).

dH
=S+ (III.10)
dt
where H is the rate of entropy, S is the external entropy power, and is the rate
of internal entropy production.

These conservation equations are global integral statements written for the finite
volume P ⇢ B. The derivation of the local di↵erential forms can be carried out
through the following steps:
Step 1. Insert the Cauchy Theorem (III.6) into the surface integrals.

t= n and h = q · n .
R
Step 2. Transform the surface integrals @P . . . dA into volume integrals by using
the Gauss Integral Theorem
Z Z
@(•)
(•)ni dA = dV . (III.11)
@P P @xi

Step 3. Apply the localization theorem through

Z
lim (⇤) dV = 0 ! (⇤) = 0 . (III.12)
P! dV P

(i) Conservation of Mass

Z Z
d d
(m) = ⇢ dV = ⇢˙ dV = 0 , (III.13)
dt dt P P

which yields the local form of the conservation of mass

⇢˙ = 0 . (III.14)

35
III. Thermodynamics of the 3-D Continuum

(ii) Conservation of Linear Momentum

Z Z Z Z
d
⇢v dV = (⇢v ˙ + ⇢v̇) dV = ⇢b dV + t dA (III.15)
dt P P P @P

where the surface integral can be written as

Z Z Z
t dA = n dA = div dV . (III.16)
@P @P P

Insertion yields Z Z
˙⇢ + ⇢v̇) dV =
(⇢
⇢v (⇢b + div ) dV (III.17)
P P

where we substituted the conservation of mass (III.14). The localization of the

latter integral Z
lim (⇢v̇ div ⇢b) dV = 0 (III.18)
P! dV P
then yields the well-known local form of the conservation of linear momentum

⇢v̇ = ⇢a = div + ⇢b (III.19)

governing the local stress equilibrium at a material point P 2 B.

(iii) Conservation of Angular Momentum Without loss of generality, we

take the moment of linear momentum about the origin, thus the moment arm
vector is simply the position vector x.
Z Z Z
d d
D0 = x ⇥ ⇢v dV = x ⇥ ⇢b dV + x ⇥ t dA (III.20)
dt dt P P @P
| {z }
M0

where the surface integral can be expressed as

Z Z
x ⇥ t dA = ✏ijk xj tk dA
@P Z @P
(III.6)
= ✏ijk xj kl nl dA
@P
Z
(III.11) @✏ijk xj kl
= dV
P @xl
Z
@ kl
= (✏ijk jl kl + ✏ijk xj ) dV
P @xl
|{z}
(div )k
Z Z
= ✏ijk kj dV + x ⇥ div dV
P | {z } P
✏ijk T
jk
Z Z
T
= ✏: dV + x ⇥ div dV .
P P

36
 3. Balance Laws of Continuum Thermomechanics

We then have
Z Z
d
x ⇥ ⇢v dV = (v ⇥ ⇢v + x ⇥ ⇢v ˙ + x ⇥ ⇢a) dV
dt P
ZP Z Z (III.21)
T
= x ⇥ ⇢b dV + x ⇥ div dV + ✏| :{z } dV .
P P P
✏:

Incorporating the conservation of mass (III.14) and the equality v ⇥ ⇢v = 0 in the

latter, we obtain
Z Z
x ⇥ (⇢a div ⇢b) dV + ✏ : dV = 0 (III.22)
P P

The first integral vanishes identically due to the balance of linear momentum
(III.19) and leads us to the following identity through localization
Z
lim ✏: =0 !✏: =0 (III.23)
P! dV P

✏: = ✏ijk jk (III.24)
for i = 1 ✏123 ( 23 32 ) =0
for i = 2 ✏231 ( 31 13 ) = 0 (III.25)
for i = 3 ✏312 ( 12 21 ) = 0

Since ✏123 = ✏231 = ✏312 = 1, the equality (III.25) implies that

23 = 32 , 31 = 13 , 12 = 21 (III.26)

or more generally, the conservation of angular momentum requires that the stress
tensor is symmetric.
= T (III.27)

(iv) Conservation of Energy (1st Law of Thermodynamics)

d
[K + E] = P + Q (III.28)
dt

 Z
d 1 2
2
⇢ |v| dV + ⇢e dV
dt P
Z Z Z Z
= ⇢b · v dV + t · v dA + ⇢r dV h dA (III.29)
P @P P @P

37
III. Thermodynamics of the 3-D Continuum

The surface integrals transformed into volume integrals:

Z Z Z
(III.6)
v · t dA = v · ( n) dA = vi ij nj dA
@P P @P
Z Z
(III.11) @
= (vi ij ) dV = (vi ij ),j dV
P @xj P
Z Z (III.30)
@vi @ ij
= ij dV + vi dV
P @xj P @xj
Z Z
= rv : dV + v · div dV
P P
Z Z Z
(III.6)
h dA = q · n dA = div q dV (III.31)
@P @P P
T
Note that in the former, since = , we have

rv : = sym(rv) : (III.32)
1
and the definition of the strain tensor " := sym(ru) = 2
(ru + rT u) = implies
sym(rv) = ". ˙ Thus,
rv : = "˙ : = : "˙ . (III.33)
Taking the time derivative and substituting the above transformations, we obtain

Z Z
(III.14) Z Z
(III.14)
*
2
1
⇢˙ |v| dV + ⇢v · a dV + ⇢e✓
˙ dV + ⇢ė dV
2
PZ P Z P (III.34)
= (⇢b · v + : "˙ + v · div ) dV + (⇢r div q) dV
P P

Putting the terms together on one side, we get

Z
⇠⇠⇠: (III.19)
Z
⇠⇠ ⇠
v · (⇢a⇠ div
⇠ ⇢b) dV + (⇢ė : "˙ + div q) dV = 0 (III.35)
P ⇠⇠ P

where we have substituted the Balance of Linear Momentum (III.19) in the first
term. Finally, applying the localization theorem
Z
lim (⇢ė : "˙ ⇢r + div q) dV = 0 , (III.36)
P! dV P

we arrive at the local form of the energy balance, also known as the first law of
thermodynamics
⇢ė = : "˙ + ⇢r div q , (III.37)
which simply states that the rate of specific internal energy is equal to the sum of
the internal stress power (working) ( : ")˙ and the thermal power (⇢r div q).

38
 3. Balance Laws of Continuum Thermomechanics

(v) Conservation of Entropy (2nd Law of Thermodynamics)

Z
d
:= ⇢ dV = H S 0
P dt
Z Z Z (III.38)
d r h
= ⇢⌘ dV ⇢ dV + dA 0
dt P P ✓ @P ✓

Transforming the surface integral into the volume integral

Z Z Z
h (III.6) q·n qi n i
dA = dA = dA
@P ✓ @P ✓ @P ✓
Z
(III.11) @ ⇣ qi ⌘
= dV
P @xi ✓
Z Z (III.39)
1 @qi 1 @✓
= dV + qi dV
P ✓ @xi P ✓2 @xi
Z Z
1 1
= div q dV 2
q · r✓ dV ,
P ✓ P ✓

we arrive at
Z Z Z Z Z
r 1 1
⇢ dV = ⇢⌘˙ dV ⇢ dV + div q dV q · r✓ dV . (III.40)
P P P ✓ P ✓ P ✓2
The localization
Z
r 1 1
lim (⇢ ⇢⌘˙ + ⇢ div q + 2 q · r✓) dV = 0 (III.41)
P! dV P ✓ ✓ ✓
leads to
1 1
⇢ = ⇢⌘˙
(⇢r div q) q · r✓ 0. (III.42)
✓ ✓2
From the first law (III.37), we have ⇢r div q = ⇢ė : "˙

1 1
⇢ = ⇢⌘˙ (⇢ė ˙
: ") q · r✓ 0 (III.43)
✓ ✓2

Box III.1: Local Forms of Conservation Laws.

(i) Mass ⇢˙ = 0 (III.44)
(ii) Linear Momentum ⇢a = div + ⇢b (III.45)
T
(iii) Angular Momentum = (III.46)
(iv) Energy ⇢ė = : "˙ + ⇢r div q (III.47)
1 1
(v) Entropy ⇢ = ⇢⌘˙ (⇢ė ˙
: ") q · r✓ 0
✓ ✓2
(III.48)

39
III. Thermodynamics of the 3-D Continuum

4. Closure Problem
The conservation laws derived above (III.44-III.48) provide a number of conditions
and contain several unknown fields. These can be identified in the 3-D space as
follows.
Balance of # Eqn. Unknowns #
Mass 1 (III.44) Density ⇢ 1
Linear Momentum 3 (III.45) Displacement u 3
Angular Momentum 3 (III.46) Stress 9
Energy 1 (III.47) Temperature ✓ 1
Entropy 1 (III.48) Entropy ⌘ 1
P Heat flux q 3
9
Entropy production 1
P
19
Note that the mass-specific body force b(x, t) and the heat source r(x, t) are
assumed to be given. Comparing the total number of unknowns with that of
conditions, we note that 19 9 = 10 additional equations are needed to solve
the problem. These additional equations are called the constitutive (material)
equations that describe the material-specific stress response (6) , the entropy
⌘ (1) , and the heat conduction q (3) .
5. Material Equations in General Internal Variable Formulation
5.1 Concepts of Dissipation and Thermodynamic Consistency
The material (constitutive) equations are constructed so that they a priori satisfy
the entropy inequality (III.48) dictated by the second law of thermodynamics. In
order to recast this inequality into a rate-type energetic (power) expression, we
introduce the dissipation per unit mass D as

⇢D := ⇢✓ 0 with ✓ > 0 . (III.49)

Hence, weighting (III.48) with the absolute temperature ✓ > 0, we obtain

1
⇢D = ⇢✓⌘˙ (⇢ė ˙
: ") q · r✓ 0,
✓
1 (III.50)
= : "˙ ⇢ė + ⇢✓⌘˙ q · r✓ 0.
| {z } ✓
| {z }
local
conductive

The latter is referred to as the Clausius-Duhem Inequality (CDI) in the

literature on continuum thermodynamics. This inequality can be split into the
local dissipation Dloc and the dissipation due to conduction Dcon .

(CPI) ⇢Dloc := : "˙ ⇢ė + ⇢✓⌘˙ 0

1 (III.51)
(FI) ⇢Dcon := q · r✓ 0
✓

These are the so-called the Clausius-Planck Inequality (CPI) and the Fourier
Inequality (FI), respectively.

40
 5. Material Equations in General Internal Variable Formulation

Introducing the Helmholtz Free Energy through the Legendre transforma-

tion of the internal energy e as

:= e ✓⌘ , (III.52)

we arrive at the alternative representation of the Clausius-Planck Inequality

⇢Dloc = : "˙ ⇢ė + ⇢✓⌘˙ 0,

•

= : "˙ ⇢( + ✓⌘) + ⇢✓⌘˙ 0,

(III.53)
= : "˙ ˙
⇢ ˙ ⇢✓⌘ ⇢✓⌘˙ + ⇢✓⌘˙ 0,
⇢Dloc = : "˙ ⇢ ˙ ⇢⌘ ✓˙ 0.

The constitutive equations that fulfill the second law (III.50) are said to be ther-
modynamically consistent.

Remark III.2 (Decomposition of CDI into CPI and FI and the Fourier’s Law).
The split of CDI into CPI and FI, introduced in (III.51), results into a stronger
condition that dictates the fulfillment of the two inequalities Dloc 0 and Dcon 0,
separately. This is justified by the di↵erent and often decoupled natures of the local
and conductive dissipative mechanisms. While the local (intrinsic) dissipation Dloc
arises from frictional dissipative mechanisms taking place locally in a material
microstructure, the conductive dissipation Dcon results from di↵usion-type non-
local transport phenomena.
In thermomechanics, the heat conduction, an example for the latter, is described
by a material-specific constitutive equation for the heat flux vector q̂. Since the
existence of spatial temperature variation is a prerequisite for the development
of heat flow within the material, the heat flux is expected to be driven by the
temperature gradient r✓. The simplest relation that relates the heat flux vector q̂
to the the temperature gradient r✓ is the Fourier’s law

q̂ = kr✓ (III.54)

that simply states that the heat flux is proportional to the temperature gradient
through the second-order heat conduction tensor k. The minus sign accounts for
the opposite signs of the heat flux vector and the temperature gradient; that is,
heat flows from hotter parts towards colder parts. In fact, this is dictated by the
second law of thermodynamics through the Fourier’s inequality (III.51)2 ; that is,
inserting (III.54) into (III.51)2 , we obtain
1
⇢Dcon = q · r✓ 0,
✓
1
= ( kr✓) · r✓ 0, (III.55)
✓
1
= r✓ · kr✓ 0.
✓
Based on this result, we conclude that the Fourier’s law (III.54) is said to be
thermodynamically consistent for positive-definite conduction tensors k.

41
III. Thermodynamics of the 3-D Continuum

5.2 General Internal Variable Formulation

To characterize the behavior of an inelastic solid, in addition to the usual observ-
able and controllable external variables of state, i.e. the strain tensor " and the
absolute temperature ✓, we introduce additional variables of state. We denote the
generalized vector of internal variables by I 2 Rn that consists of n scalar field at
x 2 B at time t 2 R+ . Note that the internal variables of state may be identified
and measured but they are not coupled to any external force variable, which might
provide a means of control.

internal
A

"

Figure III.6.: Schematic illustration of internal variables I in the Maxwell model

The scalar, vectorial, or tensorial nature of an internal variable as well as its

physical character need to be specified. This identification is a challenging aspect
of phenomenological models. Examples of internal variables are

• Viscous strain tensor "v 2 R6

• Plastic strain tensor "p 2 R6
• Isotropic hardening variable ↵ 2 R
• Scalar damage variable d 2 R

The internal variables can be illustrated in a simple rheological model, the Maxwell
Model, where a dashpot is connected to a spring in series as showin in Figure III.6.
Apparently, the observable and controllable external variable " (strain) is inade-
quate for determining the thermomechanical state, i.e. the stress response, of the
model. The position of Point A (see Figure III.6), which may be observed but
cannot be controlled externally by prescribing either or ", has to be known.
The local thermomechanical state of the material is then given by

State(x, t) = {"(x, t), I(x, t), ✓(x, t)} (III.56)

at point x at time t 2 R+ . Therefore, the energetic state of a material is described

by the Helmholtz free energy function

= ˆ (", I, ✓) (III.57)

with : R6 ⇥ Rn ⇥ R ! R.
Furthermore, the evolution of internal variables is described by additional consti-
tutive functions
ˆ ",
İ = ⌃( ˙ ", I, ✓) (III.58)

42
 5. Material Equations in General Internal Variable Formulation

where ⌃ ˆ denotes the constitutive evolution equation depending upon the state
(III.56) and the rate ".˙ These functions are formulated in such a way that they a
priori fulfill the second law (III.50), in particular the Clausius-Planck Inequality
(III.53)4 .

5.3 Coleman’s Exploitation of the 2nd Law of Thermodynamics

Taking the time derivative of the free energy (III.57), we obtain
˙ = @" : "˙ + @I • İ + @✓ ✓˙ (III.59)

where @⇤ 4 is a short-hand notation alternative to @4 @⇤

. Moreover, the operator •
n
symbolizes an inner product in the general space R of internal variables. Insertion
of (III.59) into the Clausius-Planck Inequality (III.53) yields

⇢Dloc = [ ⇢@" ] : "˙ ⇢[⌘ + @✓ ]✓˙ ⇢@I • İ 0. (III.60)

According to Coleman’s reasoning, see Coleman & Gurtin (1967), the equation
˙ We then require
(III.60) should hold for all processes at arbitrary rates "˙ and ✓.
that the expressions in the brackets vanish identically. This results in the well-
established constitutive relations for the stress tensor and entropy

= ⇢@" ˆ (", I, ✓) and ⌘ = @✓ ˆ (", I, ✓) . (III.61)

The identities reduce the Clausius-Planck Inequality into the following form

⇢Dloc = ⇢@I ˆ • İ 0. (III.62)

Introducing the thermodynamic driving forces F 2 Rn conjugate to the internal

variables I 2 Rn as
F := ⇢@I ˆ (", I, ✓) , (III.63)
the local dissipation becomes

⇢Dloc = F • İ 0. (III.64)

This inequality serves as a constraint on the formulation of the evolution equations

(III.58) of the internal variables. They will be constrained such that CPI (III.64)
is a priori satisfied, i.e.
ˆ ",
⇢Dloc = F • ⌃( ˙ ", I, ✓) 0. (III.65)

Example III.1. (The Maxwell Model)

⇢ = ⇢ ˆ (", "v ) = 1
2
E(" "v ) 2 (III.66)

F= ⇢@"v ˆ = E(" "v )

(III.67)
= ⇢@" ˆ = E(" "v )
⇢Dloc = F "˙v 0 (III.68)

43
III. Thermodynamics of the 3-D Continuum

ˆ ", F
"˙v = ⌃( ˙ ", "v ) = (III.69)
⌘
for Newtonian fluids.
1
⇢Dloc = F 2 0 ! 8⌘ > 0 (III.70)
⌘

Remark III.3 (Thermodynamic Potentials of Thermoelasticity)). There are four

fundamental thermomechanical potentials frequently encountered in thermomechan-
ics literature. These are the internal energy e, the Helmholtz free energy , the
enthalpy h, and the Gibbs free energy g. The choice of an appropriate potential is
generally made according to circumstances of experiments or the properties of the
material at hand. In the context of continuum thermomechanics, these potentials
generally depend on a set of variables which is formed through a mutual combina-
tion between the sets { , "} and {⌘, ✓}, see Maugin (1999). As we have seen above,
in inelastic materials, these sets are further supplemented by additional internal
variables I, employed for the description of inelastic dissipative phenomena.
Looking at the time derivatives of the fields in the Clausius-Planck inequality,
one can readily conclude that the internal energy e can be considered as a ther-
modynamical potential depending primarily upon the strain " and the entropy
⌘, i.e. e = ê(", ⌘). We have then defined the Helmholtz free energy through
the partial Legendre transformation := e ✓⌘ implying the functional de-
pendency ˆ
= (", ✓). Similarly, we can also introduce the Gibbs free energy
g = ĝ( , ✓) := ⇢ 1 " and the enthalpy h = ĥ( , ⌘) := e ⇢ 1 ". Use of
the Gibbs free energy may be preferred to the others, for example, in modeling of
gaseous materials, on which some experiments are carried out under constant pres-
sure or temperature conditions. The Legendre transformations introduced above
can be summarized as follows

ê(", ⌘) = e = + ✓⌘ (Internal Energy)

ĥ( , ⌘) = h = e ⇢ 1 :" (Enthalpy) (III.71)
ĝ( , ✓) = g = ⇢ 1 :"=h ✓⌘ (Gibbs Free Energy)

Based on the definitions above, the Legendre transformations among the four ther-
modynamic potentials can be visually summarized through a commutative diagram
as in Figure III.7.
Using the constitutive relations obtained in (III.61) for the stress tensor and en-
tropy through the partial derivatives of the Helmholtz free energy with respect to
the strain tensor and temperature, respectively, we can readily obtain the following
results for the internal energy and the Gibbs free energy.

Internal Energy
e= + ✓⌘
˙ + ⇢✓⌘˙
⇢ė = ⇢ ˙ + ⇢✓⌘
(III.72)
= : "˙ ⇢⌘⇢ ⇢
˙⇢ + ⇢✓⌘˙
✓˙ + ⇢✓⌘ ⇢
⇢ė = : "˙ + ⇢✓⌘˙

44
 6. Initial Boundary-Value Problem (IBVP) of Thermoinelasticity

Internal ✓⌘ Helmholtz
Energy e = ê(", ⌘) = ˆ (", ✓) Free Energy

1 1
⇢ :" ⇢ :"

✓⌘ Gibbs
Enthalpy h = ĥ( , ⌘) g = ĝ( , ✓)
Free Energy

Figure III.7.: Commutative diagram describing the Legendre transformations

among the four fundamental thermomechanical potentials

Thus
ˆ = ⇢@" ê(", ⌘) and ✓ = @⌘ ê(", ⌘) (III.73)
Gibbs Free Energy
1
g= ⇢ :"
⇢ġ = ⇢ ˙ ˙ :" : "˙
(III.74)
:⇠
= ⇠⇠"˙ ⇢⌘ ✓˙ ˙ :" :⇠
⇠⇠"˙
⇢ġ = ": ˙ ⇢⌘ ✓˙

Thus
g = ĝ( , ✓), "= ⇢@ ĝ and ⌘= @✓ ĝ (III.75)
One can also derive similar constitutive relationships for the strain tensor and the
temperature through the appropriate partial derivatives of the enthalpy ĥ. (HE)

6. Initial Boundary-Value Problem (IBVP) of Thermoinelasticity

The mathematical description of an initial boundary-value problem (IBVP) of
thermoinelasticity consists of four sets of equations
1. Balance Equations 3. Boundary Conditions
2. Constitutive Equations 4. Initial Conditions
as oulined in Box III.1.
h
t
x2B x2B
@B✓
@Bt @Bh
@Bu
B B
Mechanical Thermal
Figure III.8.: Schematic representation of the boundary conditions of the mechan-
ical and thermal boundary-value problems

45
III. Thermodynamics of the 3-D Continuum

To formulate the boundary conditions, we decompose the boundary @B of the

solid body B ⇢ R3 as
• @Bu : the displacement field u is prescribed (Dirichlet/Essential BCs)
• @Bt : the stress traction t is prescribed (Neumann/Natural BCs)
• @B✓ : the temperature field ✓ is prescribed (Dirichlet/Essential BCs)
• @Bh : the heat flux h is prescribed (Neumann/Natural BCs)
as shown in Figure III.8. Note that the Dirichlet and Neumann boundaries of the
respective problems are complementary and disjoint:

@B = @Bu [ @Bt and @B = @B✓ [ @Bh ,

(III.76)
@Bu \ @Bt = ; and @B✓ \ @Bh = ; .

Box III.2: IBVP of Coupled Thermoinelasticity

1. Balance Equations

Linear Momentum ⇢a = div + ⇢b

Energy ⇢ė = : "˙ + ⇢r div q

2. Constitutive Equations

Stresses = ⇢@" ˆ (", I, ✓)

Entropy ⌘ = @✓ ˆ (", I, ✓)
Thermodynamic Force F = ⇢@I ˆ (", I, ✓)
Evolution Equations ˆ ",
İ = ⌃( ˙ ", I, ✓)
Heat Flux Vector ˙ ✓, r✓)
q = q̂(✓,
e.g. Fourier’s Law q = kr✓

3. Boundary Conditions

Displacement u = ū on @Bu
Traction t = t̄ = n on @Bt
Temperature ✓ = ✓¯ on @B✓
Heat Flux h = h̄ on @Bh

4. Initial Conditions

Displacement u(x, t0 ) = u0 (x) in B

Velocity v(x, t0 ) = v 0 (x) in B
Temperature ✓(x, t0 ) = ✓0 (x) in B

46
Chapter IV

Finite Element Discretization of Non-linear

Boundary-Value Problems

Probably, the most flexible and widely used spatial discretization technique for the
numerical solution of partial di↵erential equations (PDEs) is the Finite Element
Method (FEM). In FEM, we discretize the material body (spatial domain) B into
finite subdomains B e ⇢ B, the so-called finite elements as depicted in Figure
IV.1 and make an assumption about the variation of the state fields such as the
displacement field u(x, t) and the temperature field ✓(x, t) within the subdomain
B e through the shape functions N (x).

de1 de2 u(x)

x
Be ⇢ B B
Figure IV.1.: Discretization of a solid body B into finite elements B e for approxi-
mating the displacement field u(x, t) in the 1-D setting

Example IV.1. The basic FE approximation of the displacement field u(x, t) is

u(x, t) ⇡ N (x) de (t) in B e . (IV.1)

For 1-D problems with two-node elements, see Figure IV.2, we have

nnode
X=2 [N1 (x) N2 (x) ] " e #
| {z } d 1 (t)
u(x, t) ⇡ Ni (x) dei (t) = N (x) e
= N (x) de (t) (IV.2)
i=1
d2 (t)
| {z }
de

with linear shape functions

1
Ni (⇠) = 2
(1 + ⇠i ⇠) for i = 1, 2 (IV.3)

in terms of the natural coordinate ⇠ 2 [ 1, 1] as shown in Figure IV.3.

47
IV. Finite Element Discretization of Non-linear Boundary-Value Problems

u(x, t)

de2
de1

1 2

Figure IV.2.: FE approximation of the displacement field u(x, t) in 1-D

Be ⇢ B

Node 1 Node 2
⇠
1 0 +1
1
N1 = 2
(1 ⇠)
1

1
N2 = 2
(1 + ⇠)
h 1
t

Figure IV.3.: Schematic illustration of a two-node linear finite element along with
its shape functions Ni (⇠) formulated in terms of the natural coordi-
nate ⇠ 2 [ 1, 1]

In solid mechanics, generally the isoparametric map is used to express the

physical coordinates in terms of the natural coordinates; that is, the same shape
function as the ones used to interpolate the state fields are used,
nnode
X=2
x(⇠) ⇡ Ni (⇠) xei = N1 (⇠)xe1 + N2 (⇠)xe2
i=1
(IV.4)
= 1
2
(1 ⇠)xe1 + 1
2
(1 + ⇠)xe2 .

Since the map x(⇠) is bijective, it can be inverted to obtain

1
x (xe1 + xe2 )
⇠ = x 1 (x) = 2
(IV.5)
le /2
where le := xe2 xe1 denotes the length of the element. Therefore, the shape func-
tions can be expressed in terms of either ⇠ or x. Moreover, the integrals taken
over the element physical domain can be recast into the integrals over the parent
element Z Z +1
( ) dx = ( )J(⇠)d⇠ (IV.6)
Be 1
dx
where J(⇠) := d⇠ denotes the Jacobian, which is equal to J = le /2 for two-
node linear elements in 1-D. This transformation allows us to compute the volume

48
 1. Strong Form of a Nonlinear BVP (Mechanics)

integrals for the element vectors and matrices using numerical integration schemes
such as the Gauss quadrature in a modular fashion.

Based on this example, it is important to observe that the FE discretization intro-

duced in (IV.2) transforms a continuous field into a system of algebraic equations
in terms of the discrete nodal displacements (degrees of freedom) of a given FE
mesh and the associated shape functions.

In what follows, without loss of generality, we will focus on the procedure followed
to set up a FEM for a purely mechanical boundary-value problem (BVP). This
procedure involves the following steps:

(i) Description of the mechanical BVP in its strong form in terms of the
governing PDE and the associated boundary conditions.
(ii) Construction of the corresponding weak form through the Galerkin method.
(iii) Discretization of the weak form by a typical FE discretization.
(iv) Solution of the system of algebraic equations obtained by FEM through an
iterative technique.

1. Strong Form of a Nonlinear BVP (Mechanics)

The strong (di↵erential) form of the BVP of mechanics consists of the governing
di↵erential equation and boundary conditions. Restricting ourselves to the quasi-
static case, i.e. ⇢ü ⇡ 0, the strong form can be expressed as follows:

Find the displacement field u(x, t) such that the following equations hold.
8
>
> (a) Balance Equation div + ⇢b = 0 in B,
>
< (b) Material Equation = ˆ (") in B,
S) (IV.7)
>
> (c) Boundary Conditions u = ū on @Bu ,
>
:
n = t̄ on @Bt .

Therefore, the solution of the problem in (IV.7) determines the displacement field
u(x, t) of an elastic solid for a given external loading.

2. Weak Form of Nonlinear BVP

The weak form of the nonlinear elasticity problem (IV.7) can be constructed by
the Galerkin Method that involves the following steps:

1. Consider a field of test functions (variations) u(x, t) in B that satisfy the

homogeneous form of the essential BC’s; that is,

u(x, t) = u 2 H 1 | u = 0 on @Bu (IV.8)

where H 1 functions fulfill

Z
kr uk2 dV < 1 ”square integrable” . (IV.9)
B

49
IV. Finite Element Discretization of Non-linear Boundary-Value Problems

2. Multiply the governing di↵erential equation of the strong form (IV.7) by

u(x, t)
u · [div + ⇢b] = 0 in B . (IV.10)

3. Integrate the expression (IV.10) over the domain B

Z
u · (div + ⇢b) dV = 0 . (IV.11)
B

4. Integrate (IV.11) by parts and apply the Gauss Integral Theorem to the
term involving divergence; that is,
T
u · div = div( u) r( u) :
(IV.12)
[ ui ij,j = ( ui ij ),j ui,j ij ]

to obtain
Z Z
T
u·(div +⇢b) dV = [r( u) : u·⇢b div( u)] dV (IV.13)
B B

where
Z Z Z
T
div( u) dV = div( u ) dV = ( u ) · n dA . (IV.14)
B B @B

Thus, the integral becomes

Z Z
(r( u) : u · ⇢b) dV u · ( n) dA = 0 (IV.15)
B | {z }
" @B=@Bu [@Bt

where " = sym[r( u)]. Since u = 0 on @Bu and n = t̄ on @Bt , we arrive at

the following final integral expression for the weak form
Z Z Z
G(u, u) := ": dV u · ⇢b dV u · t̄ dA = 0 . (IV.16)
B B @Bt

This expression is also known as the Principle of Virtual Work (PVW) in elasticity.
According to this principle, the weak (integral) expression, obtained in (IV.16),
can also be expressed as

Virtual Work of Internal Forces ⌘ Virtual Work of External Forces

Gint = Gext

where Z
Gint (u, u) := ": dV ,
ZB Z (IV.17)
Gext (u, u) := u · ⇢b dV + u · t̄ dA .
B @Bt

Based on the Galerkin method (IV.16), the weak form of the BVP can be expressed

50
 2. Weak Form of Nonlinear BVP

as follows:
Find the displacement field u(x, t) such that for all functions u(x) with u = 0
on @Bu , the following equations hold
8
> (a) Galerkin-Type Weak Form
>
>
>
>
>
> G(u, u) = Gint (u, u) Gext (u, u) = 0
>
> Z Z Z
>
>
>
> = " : dV u · ⇢b dV u · t̄ dA = 0
>
>
>
> B B @Bt
<
W ) (b) Constitutive Equations (possibly inelastic)
>
>
>
> = ˆ (", I)
>
>
>
> ˆ I)
İ = ⌃(",
>
>
>
>
>
>
>
> (c) Essential Boundary Conditions
>
:
u = ū on @Bu
(IV.18)

Remark IV.1 (Variational Formulation of Elasticity). The total energy functional

of an elastic body can be expressed as

⇧(u) = ⇧int (u) + ⇧ext (u) (IV.19)

where Z
⇧int (u) := ⇢ ˆ (") dV
BZ Z (IV.20)
⇧ext (u) := u · ⇢b dV u · t̄ dA
B @B
The Principle of Minimum Potential Energy states that an elastic body subjected
to a given loading attains its equilibrium position by minimizing its total energy.
This principle can be mathematically expressed in terms of the variation of the total
energy functional through its Gateaux derivative in the direction of the variation
u of the displacement field; that is,

d
⇧ := [⇧(u + ✏ u)] = 0, (IV.21)
d✏ ✏=0

Z
d
⇧int = ⇢ ˆ (sym(ru) + ✏ sym(r u)) dV
d✏ B ✏=0
Z Z (IV.22)
= ˆ
⇢@" : " dV = : " dV
B B

with " = sym(ru), and

 Z Z
d
⇧ext = (u + ✏ u) · ⇢b dV (u + ✏ u) · t̄ dA
d✏ B @B e✏=0
Z Z (IV.23)
= u · ⇢b dV u · t̄ dA .
B @B

51
IV. Finite Element Discretization of Non-linear Boundary-Value Problems

Integration by parts

: ru = ij ui,j = ( ij ui ),j ij,j ui (IV.24)

results in
Z Z
⇧= u · (div + ⇢b) dV + u · ( n t̄) dA = 0 . (IV.25)
B | {z } @B | {z }
0 0

Hence, the Euler-Lagrange equations of the minimization problem considered above

are none other than the conservation of linear momentum and the Neumann
boundary conditions:

div + ⇢b = 0 in B and n = t̄ on @Bt . (IV.26)

3. Finite Element Discretization

Analogous to Example IV.1, we introduce the following approximation for the dis-
placement field u(x, t) within an element domain in the multi-dimensional setting

u(x, t) ⇡ N (x) de (t) in B e . (IV.27)

Computing the strain field from "(x, t) := sym(ru), we obtain

"(x, t) ⇡ B(x) de (t) in B e . (IV.28)

Similarly, the test function u(x) and its symmetric gradient " = sym(ru) are
expressed as
u ⇡ N (x) de and " = B de , (IV.29)
respectively. Also, observe that

u = N de = deT N T and " = B de = deT B T (IV.30)

where de is the vector of element nodal variations.

Remark IV.2 (Matrix Storage of Tensors: Voigt Notation). To save space in

memory, we store symmetric second-order tensors, e.g. the stress tensor and the
strain tensor ", in column vectors [ ], ["] 2 R6 . The key condition to be satisfied
is
!
: " = [ ]T ["] . (IV.31)
For the stresses, we have
2 3
11
2 3 6 7
6 22 7
11 12 13 6 7
=4 22 23
5 ! [ ]=6
6
33 7
7 (IV.32)
sym. 6 12 7
33 4 23
5
13

52
 3. Finite Element Discretization

and for the strains, we have

2 3
"11
2 3 6 "22 7
"11 "12 "13 6 7
6 "33 7
"=4 "22 "23 5 ! 6
["] = 6 7. (IV.33)
2 " 7
sym. "33 6 12 7
42 "23 5
2 "13

The factor “2” in the Voigt (Matrix) representation of shear strains is introduced
to account for symmetry, and to fulfill (IV.31) as shown below

:"= ij "ij = 11 "11 + 22 "22 + 33 "33

(IV.34)
+ 2 12 "12 + 2 13 "13 + 2 23 "23 .

In summary, a two-step procedure is followed for the transformation of the tensor

notation into the Voigt representation.
Step 1: Index Mapping
2 3
1 4 6 imap := [1, 2, 3, 1, 2, 1]T6⇥1
I = 4 4 2 55 , (IV.35)
6 5 3 jmap := [1, 2, 3, 2, 3, 3]T6⇥1

Step 2: Put the factor 2 in the matrix representation of shear strains to account
for (IV.31).
Tensor-to-Matrix

1 for I = 1 ,2 ,3 ,4 ,5 ,6
2 [ sigma ]( I ) = sigma ( imap ( I ) , jmap ( I ) )
3 end

Matrix-to-Tensor

1 for i = 1 ,2 ,3
2 for j = 1 ,2 ,3
3 sigma (i , j ) = [ sigma ]( I (i , j ) )
4 end
5 end

Likewise, symmetric fourth-order tensors possessing both major symmetry ijkl =

klij and minor symmetry ijkl = jikl = ijlk are stored as square matrices
[ ] 2 R6⇥6 , subject to the following energetic constraint
1 ! 1
= ": : " = ["]T1⇥6 [ ]6⇥6 ["]6⇥1 . (IV.36)
2 2
For the basic identity tensors,
!
" : : " = ["]T [ ]["] (IV.37)

53
IV. Finite Element Discretization of Non-linear Boundary-Value Problems

yields 2 3
1 0 0 0 0 0
60 1 0 0 0 0 7
6 7
60 0 1 0 0 0 7
[]=6
60
7. (IV.38)
6 0 0 1/2 0 0 77
40 0 0 0 1/2 0 5
0 0 0 0 0 1/2

Since [1] = [1 1 1 0 0 0]T , we have

2 3
1 1 1 0 0 0
61 1 1 0 0 07
6 7
61 1 1 0 0 07
[1 ⌦ 1] = 6
60
7. (IV.39)
6 0 0 0 0 077
40 0 0 0 0 05
0 0 0 0 0 0
1
The deviatoric projection tensor := 3
1 ⌦ 1, used to obtain dev(") = " : =
: ", can be expressed in the Voigt notation as
2 3
2/3 1/3 1/3 0 0 0
6 1/3 2/3 1/3 0 0 0 7
6 7
6 1/3 1/3 2/3 0 0 0 7
[ ]=6 6 0
7. (IV.40)
6 0 0 1/2 0 0 77
4 0 0 0 0 1/2 0 5
0 0 0 0 0 1/2

For isotropic elasticity, the moduli (elasticity tensor)

1
⇢ = ": :" ! = :" (IV.41)
2
where, as derived in (II.159),

= 1 ⌦ 1 + 2µ , (IV.42)

and in Voigt notation

2 3
+ 2µ 0 0 0
6 + 2µ 0 0 07
6 7
6 + 2µ 0 0 07
[ ]=6
6
7. (IV.43)
6 0 0 0 µ 0 077
4 0 0 0 0 µ 05
0 0 0 0 0 µ

Four-Node Quadrilateral Element for 2-D Continua (nen = 4, ndim = 2)

Here, nen denotes the number of element nodes, and ndim is the number of space
dimensions.
u(x, t) ⇡ N (x) de (t) (IV.44)

54
 3. Finite Element Discretization

d22
⌘
d32 d21
4 (-1,+1) 3 (+1,+1) 2
x̂(⇠, ⌘) 3 d31

⇠
d42
x2 4 d41
1 (-1,-1) 2 (+1,-1) d12
x1 d11
1
Figure IV.4.: Schematic representations of the isoparametric map x = x̂(⇠, ⌘) and
the 4-node quadrilateral element. The parent element depicted in
the natural coordinate system (left), a quadrilateral element in the
physical space (right)

2 3
d11
6 7
6 7
6 d12 7
6 7
6 7
6 d21 7
2 3 2 36 7
6 7
u1 (x, t) N1 0 N2 0 N3 0 N4 0 6 d22 7
u=4 5=4 56 7 (IV.45)
6 7
u2 (x, t) 0 N1 0 N2 0 N3 0 N4 6 d31 7
| {z }6
6
7
7
N (x)2⇥8 6 d32 7
6 7
6 7
6 d41 7
4 5
d42
| {z }
de8⇥1

The components of ["] are computed according to

✓ ◆
1 @ui @uj
"ij = sym(ru) = + . (IV.46)
2 @xj @xi

In particular, we have
@u1 @u2 @u1 @u2
"11 = , "22 = , and 2"12 = + . (IV.47)
@x1 @x2 @x2 @x1
P
n en P
n en
If u1 (x, t) = N I (x) dI1 and u2 (x, t) = N I (x) dI2 then
I=1 I=1

nen
X nen
X
@N I e @N I
"11 = dI1 = B(1, :) d , "22 = dI2 = B(2, :) de (IV.48)
@x1
I=1 | {z }
@x2
I=1 | {z }
I
N,1 I
N,2

55
IV. Finite Element Discretization of Non-linear Boundary-Value Problems

and nen
X
2"12 = (N,2I dI1 + N,1I dI2 ) = B(3, :) de . (IV.49)
I=1

as given in the following compact form

2 3
d11
6 7
6 7
6 d12 7
6 7
2 3 2 366 d21
7
7
" N,11
0 0 N,12 0 N,13 N,14 0 6 7
6 11 7 6 76 d22
7
6 7 6 76 7
["] = 6 "22 7 = 6 0 N,21 0 N,22 0 N,23 0 N,24 7 6 7 (IV.50)
4 5 4 566 d31
7
7
2 "12 N,21 N,11 N,22 N,12 N,23 N,13 N,24 N,1 6
4
6
7
7
| {z }66
d32 7
7
B(x)3⇥8 6 7
6 d41 7
4 5
d42
| {z }
de8⇥1

N1 N2
1 1

0 0
1 1
0.5 0.5
-1 -0.5 0 -1 -0.5 0
0 -0.5 ⌘ 0 -0.5 ⌘
0.5 1 -1 0.5 1 -1
⇠ ⇠

N3 N4
1 1

0 0
1 1
0.5 0.5
-1 -0.5 0 -1 -0.5 0
0 -0.5 ⌘ 0 -0.5 ⌘
0.5 1 -1 0.5 1 -1
⇠ ⇠

Figure IV.5.: Graphical illustration of the shape functions of the four-node quadri-
lateral element

Isoparametric Mapping Using the same shape functions used to interpolate

the displacement field u(x, t), shown in Figure IV.5,
1
N I (⇠, ⌘) = (1 + ⇠I ⇠)(1 + ⌘I ⌘) (IV.51)
4

56
 3. Finite Element Discretization

the physical coordinates can be interpolated through

nen
X
x(⇠, ⌘) = N I (⇠, ⌘)xI (IV.52)
I=1

as depicted in Figure IV.4.

2 3
x11
6 7
6 7
6 x12 7
6 7
6 7
6 x21 7
2 3 2 36 7
6 7
x N1 0 N2 0 N3 0 N4 0 6 x22 7
4 15 = 4 56 7 (IV.53)
6 7
x2 0 N1 0 N2 0 N3 0 N4 6 x31 7
| {z }6
6
7
7
N (⇠,⌘)2⇥8 6 x32 7
6 7
6 7
6 x41 7
4 5
x42
| {z }
xe8⇥1

Eight-Node Brick Element for 3-D Continua (nen = 8, ndim = 3) (HE)

1
N I (⇠, ⌘, ) = (1 + ⇠I ⇠)(1 + ⌘I ⌘)(1 + I ) (IV.54)
8

Incorporating the FE discretizations (IV.27)-(IV.29) in the Galerkin Functional

(IV.16), we obtain
nelem
G(u, u) = A
e=1
deT [f eint f eext ] = 0 (IV.55)

where Z
f eint := B T [ ˆ ] dV ,
Z Be Z (IV.56)
f eext := T
N ⇢b dV + T
N t̄ dA
Be @Bte

and we made use of u = N de = deT N T and [ "]T = [ de ]T [B]T .

nelem
The FE assembly procedure, denoted by Ae=1 , defines the associated global
vectors.
nelem nelem nelem
f int := A
e=1
f eint , f ext := A
e=1
f eext , and d := A
e=1
de . (IV.57)

Inserting the definitions (IV.57) into (IV.55), we obtain the global system of FE

57
IV. Finite Element Discretization of Non-linear Boundary-Value Problems

equations
G(u, u) = dT [f int (d) f ext ] = 0 . (IV.58)
For arbitrary nodal variations d of the nodal displacements, we end up with

f int (d) f ext = 0 . (IV.59)

For linear elasticity, we have

= : " or [ ] = [ ]["] (IV.60)

where is the constant elasticity tensor whose explicit form is given in (IV.42)
and (IV.43). Inserting this into (IV.56) along with ["] = Bde , we arrive at
Z
e e e e
f int := k d with k := B T [ ]B dV (IV.61)
Be

denoting the element sti↵ness matrix. Then the global system (IV.59) becomes

Kd = f ext (IV.62)
nelem nelem
where K := Ae=1 ke is the global sti↵ness matrix and d := Ae=1 de is the global
displacement vector. The solution of (IV.62) is obtained either by direct solvers
or more advanced iterative solvers for large systems such as conjugate gradient
and multi-grid solvers.

4. Iterative Solution Techniques for Nonlinear Problems

Recall the discrete equilibrium equation (IV.59)

f int (d) = f ext (IV.63)

where f int is a nonlinear function of d

nelem Z
f int (d) := A
e=1
f eint , f eint :=
Be
B T [ ˆ (")] dV . (IV.64)

As opposed to (IV.60), the stress tensor = ˆ (") now a nonlinear funciton of the
strain tensor "; that is, the internal force vector f int is a nonlinear function of the
nodal displacement vector d
f int 6= Kd . (IV.65)
Therefore, the solution of the nonlinear system (IV.63) must be obtained through
an iterative technique. The most widely used method, Newton’s method, is based
on the linearization of (IV.63).

To this end, let {di }i=1,2,3,... be a sequence of iterates of the nodal displacement
vector at time t and define the residual vector r(d)

r(d) := f int (d) f ext . (IV.66)

58
 4. Iterative Solution Techniques for Nonlinear Problems

The linearization of the residual vector (IV.66) about d = di yields

@r
Lin r := r(di ) + · (d di ) . (IV.67)
d=di @d di
| {z }
di

Setting the linearized residual vector r to zero and solving this equality for the
new iterate di+1 , we get the following typical Newton-type update equation

di+1 = di K i 1ri (IV.68)

where
r i := r(di ) = f int (di ) f ext ,
@r @f int (d) (IV.69)
K i := = .
@d d=di @d d=di

Here, r i is the residual vector and K i is the tangent sti↵ness matrix.

The Newton Algorithm works as follows:

1. Set iteration index i = 1 and initialize the nodal displacement vector di =0.
2. Compute the global residual r i and the global tangent K i at d = di .
3. Update the nodal displacement vector by

di+1 = di K i 1ri (IV.70)

4. If kr i k > TOL, set i = i + 1 and GOTO 2

EXIT otherwise.
The Newton algorithm converges quadratically in the neighborhood of the solution
where we observe the following pattern of the Euclidean norm of the residual krk:
kr i+0 k = ... ⇥10+2
kr i+1 k = ... ⇥10+1
kr i+2 k = ... ⇥10 2
kr i+3 k = ... ⇥10 4
kr i+4 k = ... ⇥10 8
kr i+5 k = ... ⇥10 16
Starting from Iteration i + 2, we start to observe quadratic convergence.

Observe that the key finite element matrices r i and K i are assembled from their
element counterparts; that is,
nelem nelem
ri = A
e=1
r ei and K i = Ak
e=1
e
i (IV.71)

where Z
r ei := f eint f eext = B T [ ˆ i ] dV f eext ,
Be
Z (IV.72)
@f eint
kei := = B [ ˆi ]B dV
T
@de Be

59
IV. Finite Element Discretization of Non-linear Boundary-Value Problems

are formulated in terms of the nonlinear stresses and the tangent moduli
@ˆ
= ⇢@" ˆ and 2
:= ⇢@"" = (IV.73)
@"
evaluated for the strain interpolated in the element domain

"i = Bdei in B e ⇢ B (IV.74)

and Z
@
kei = @de f eint = B T [ ˆ i ] dV
@de Be
Z Z (IV.75)
T @ ˆ @"
= B e dV = B T [ ˆ i ]B dV
B e @" @d B e

where we made use of the result @de " = B that we know from (IV.74).

Based on the derivations given above, it is important to note that a typical implicit
FE code must be provided with the nonlinear stress tensor ˆ (") and the tangent
moduli := @@"ˆ that directly enter the lines of the FE code where the element
residual vector and the tangent sti↵ness matrix are computed and assembled as
shown in (IV.72) and (IV.71), respectively.

Remark IV.3 (Newton’s Method: Error Analysis). Suppose that we look for a
root of the nonlinear equation f (x) = 0 whose Taylor series expansion about x=xi
is
f 00 (⇠)
f (xi+1 ) = f (xi ) + f 0 (xi )(xi+1 xi ) + (xi+1 xi )2 (IV.76)
| 2! {z }
R1

where xi < ⇠ < xi+1 . If xr = xi+1 (true root), we have

0 f 00 (⇠)
0 = f (xi ) + f (xi )(xr xi ) + (xr xi )2
2
0 ⇡ f (xi ) + f 0 (xi )(xi+1 xi ) Newton’s Approx. 1st -Order
(IV.77)
f 00 (⇠)
0⇡0 + f 0 (x) (xr xi+1 ) + (xr xi )2
| {z } 2 | {z }
✏AT,i+1 ✏2AT,i

Introducing the absolute true errors ✏AT,i+1 := |xr xi+1 | and ✏AT,i := |xr xi |,
we end up with
f 00 (⇠) 2
0 = f 0 (xi )✏AT,i+1 + ✏ (IV.78)
2 AT,i
which yields
f 00 (⇠) 2
✏AT,i+1 = ✏ . (IV.79)
2f 0 (xi ) AT,i
The latter implies that if the iterations have resulted in an approximate solution
xi , close enough to the true root, ✏AT,i+1 is proportional to ✏2AT,i thereby indicating
the quadratic convergence.

60
Chapter V

Linear and Nonlinear Isotropic Elasticity

Elasticity, by definition, describes non-dissipative (⇢Dloc = 0) material response

that does not require introduction of any internal variables. Therefore, the compu-
tation of elastic stress response is mere a function evaluation; that is, we compute
the stress tensor and the tangent moduli for a given strain tensor without caring
much how that strain state has been arrived at.

1. Constitutive Formulation of Elasticity

Focusing on entirely isothermal problems, i.e. under isothermal conditions where
✓ is constant, a constitutive model of elasticity is completely described by the free
energy function
⇢ = ⇢ ˆ (") . (V.1)
From the Coleman’s exploitation method (III.60, III.61), we have

= ⇢@" ˆ , (V.2)

which is none other than a function evaluation of the derivative ⇢@" ˆ for a given
strain tensor ". The rate of the stress tensor is related to that of the strain tensor

˙ = : "˙ (V.3)

by the fourth-order elasticity tensor (moduli)

2 ˆ
:= ⇢@"" (") . (V.4)

Recall that these two tensors enter the algorithmmic structure of a typical implicit
FE code to calculate the element internal force vector f eint and the element tangent
sti↵ness matrix ke (IV.72). The latter are then assembled to construct the global
residual vector and the tangent sti↵ness matrix (IV.71).

Remark V.1. The elasticity tensor (tangent moduli) possesses both minor sym-
metry and major symmetry. While the former results from the symmetry of the
strain tensor " = "T , the latter is due to the second order derivative of the free
energy with respect to the same tensor ". These symmetry types can be expressed
in indicial notation as follows

ijkl = jikl = ijlk (Minor sym.) and ijkl = klij (Major sym.) . (V.5)

61
V. Linear and Nonlinear Isotropic Elasticity

2. Representation of Isotropic Free Energy Functions

An elastic material is called isotropic if the free energy function ˆ satisfies the
following condition

⇢ ˆ (Q"QT ) = ⇢ ˆ (") 8Q 2 SO(3) (V.6)

where Q is a proper orthogonal (rotation) tensor belonging to the Lie group SO(3)

SO(3) := Q | QT Q = 1 and det Q = +1 . (V.7)

Note that SO(3) stands for the Special Orthogonal group that contains all possible
rotations in 3-D.

Remark V.2 (Physical Interpretation of Isotropy). The condition of isotropy,

mathematically formulated in (V.6), can be understood as an invariance condition
for the free energy with respect to the orientation of the material in 3-D. That
is, if we had a sheet of material from which we could punch out specimens of
random orientation, when the material is isotropic, the energy required to deform
a specimen would be insensitive with respect to the orientation of that specimen as
depicted in Figure V.1.

For all ↵
↵

⇢ (") "
x2
x1

Figure V.1.: Physical interpretation of isotropic elasticity through the invariance

of deformation energy with respect to the orientation ↵ of specimens

In order to understand the consequences of the restriction (V.6), consider the

spectral representation of "
3
X
"= ✏i n i ⌦ n i . (V.8)
i=1

This representation implies none other than a diagonalization of the strain tensor
" with respect to the principal bases ni ⌦ ni , as shown in Figure V.2. Insertion
of (V.8) into (V.6) yields
3
! ! 3
!
X X
⇢ˆ Q ✏i n i ⌦ n i Q T = ⇢ ˆ ✏i n i ⌦ n i 8Q 2 SO(3) (V.9)
i=1 i=1

62
 2. Representation of Isotropic Free Energy Functions

which is equivalent to
3
! 3
!
X X
⇢ˆ ✏i n̄i ⌦ n̄i = ⇢ˆ ✏i n i ⌦ n i 8Q 2 SO(3) (V.10)
i=1 i=1

where n̄i := Qni for i = 1, 2, 3. For arbitrary rotations Q, (V.10) can only be
satisfied if the free energy ˆ is independent of eigenvectors ni . Therefore, the free
energy of an isotropic material can depend only on the eigenvalues of "; that is,

⇢ = ⇢ ¯ (✏1 , ✏2 , ✏3 ) . (V.11)

@B n2
n2
n1 n1
"(x, t)
x2B ✏3
n3 ✏2
n3 ✏1

B
Figure V.2.: Schematic representation of the principal strains and the associated
principal directions

This is the so-called Principal Strain Representation of isotropic elasticity.

Owing to the characteristic equation (II.100), the principal strains depend on the
principal invariants. Hence, the free energy (V.11) can be alternatively represented
by
⇢ = ⇢ ˜ (I1 , I2 , I3 ) (V.12)
where recall that
1
I1 := tr("), I2 := [I12 tr("2 )] and I3 := det(") . (V.13)
2
The latter is referred to as the Invariant Representation of isotropic elasticity.
The expressions (V.11) and (V.12) serve as the two alternative formulations of
isotropic elasticity.
2.1 Representation of Stresses
The derivative of (V.6) with respect to the strain tensor " yields the isotropic
invariance conditions for the stress tensor through the chain rule

(") = ⇢@" ˆ = ⇢@(Q"QT ) ˆ : @" (Q"QT )

= (Q"QT ) : @" (Q"QT ) (V.14)
= QT (Q"QT )Q
or,
Q (")QT = (Q"QT ) (V.15)

63
V. Linear and Nonlinear Isotropic Elasticity

Remark V.3. Note that for linear elasticity where the stress tensor is linearly
dependent on the strain tensor through a constant elasticity tensor = : ", the
constraint of isotropy on the stress tensor, obtained in (V.15), can be expressed as

Q (")QT = Q( : ")QT = : (Q"QT )

(V.16)
Qai ( ijkl "kl )Qbj = abkl (Qki "ij Qlj )

for all Q 2 SO(3). Writing the latter condition for several selected values of
Q 2 SO(3), one can reduce the number of independent components of the elasti-
ciy tensor from twenty one to two; that is, while a general anisotropic material
model of linear elasticity without any material symmetries has twenty one inde-
pendent elastic constants theoretically, this number reduces to two for an isotropic
linear elastic model with infinitely many material symmetries. These two material
constants can be any binary combination of the following material parameters: the
Lamé constant , the shear modulus µ, the Young’s modulus E, the Poisson’s ratio
⌫, the bulk modulus  where the last four constants have clear physical interpreta-
tions.

For the representation in terms of the principal strains (V.11), we have

X3
@ ¯ @✏i
= ⇢@" ¯ (✏1 , ✏2 , ✏3 ) = ⇢ . (V.17)
i=1
@✏i @"

Using the well-known result @" ✏i = ni ⌦ ni (HE), the latter can be expressed as

3
X
= i ni ⌦ ni with i := ⇢@✏i ¯ . (V.18)
i=1

Analogous to (V.8), i=1,2,3 are called the principal stresses and ni=1,2,3 are the
principal stress directions identical to that of the strain tensor. This is a well-
known consequence of isotropic elasticity where and " are co-axial; thus they
commute, i.e. " = " (HE).
For the alternative formulation (V.12) in terms of principal invariants, we have

X3
@ ˜ @Ii
= ⇢@" ˜ (I1 , I2 , I3 ) = ⇢ . (V.19)
i=1
@I i @"

Making use of the well-known results, addressed also in Assignment 1,

1
@" I1 = 1, @" I2 = I1 1 ", @" I3 = I3 " (V.20)

in the stress expression, we get

@˜ @˜ @˜
=⇢ 1+⇢ (I1 1 ") + ⇢ (I3 " 1 ) . (V.21)
@I1 @I2 @I3
Collecting the common scalar multipliers of each tensor in a single term, we arrive

64
 2. Representation of Isotropic Free Energy Functions

at
= ˆ1 1 + ˆ2 " + ˆ3 " 1
(V.22)
where
˜ ˜ @˜ ˜
ˆ1 := ⇢ @ + I1 ⇢ @ , ˆ2 := ⇢ , and ˆ3 := I3 ⇢ @ . (V.23)
@I1 @I2 @I2 @I3
This is a typical expression for an isotropic second-order tensor function.

2.2 Representation of Elasticity Moduli

The second derivative of the isotropy constraint (V.10) with respect to the strain
tensor " leads us to the isotropy restriction on the tangent moduli
T
ijkl (Q"Q ) = Qia Qjb Qkc Qld abcd (") . 8Q 2 SO(3) (V.24)

Recalling the definitions (V.3, V.4)

2
= ⇢@"" (") = @" ˆ (") (V.25)

and using the representation of the free energy in (V.11) and the associated stress
function (V.17), we obtain
" 3 #
X
= @" = @" i ni ⌦ ni
i=1
3 X
3 3
(V.26)
X X
= Cij ni ⌦ ni ⌦ nj ⌦ nj + i @" (ni ⌦ ni )
i=1 j=1 i=1

where the second term on the right-hand side can be expressed as

3
X 3 X
X 3
1 i j
i @" (ni ⌦ ni ) = ni ⌦ nj ⌦ (ni ⌦ nj + nj ⌦ ni ) . (V.27)
i=1 i=1 j6=i
2 ✏i ✏j

Insertion of the latter into (V.26) yields

3 X
X 3
= Cij ni ⌦ ni ⌦ nj ⌦ nj
i=1 j=1
3 X
3
(V.28)
X 1 i j
+ ni ⌦ nj ⌦ (ni ⌦ nj + nj ⌦ ni )
i=1 j6=i
2 ✏i ✏j

where
i := ⇢@✏i ¯ and Cij := @✏j i . (V.29)
Observe that (V.28) is valid only for distinct values. To circumvent the equal
eigenvalues, numerical perturbation techniques are devised.
In elasticity theory, the spectral representation is extremely convenient, and hence
popular, because

65
V. Linear and Nonlinear Isotropic Elasticity

(i) constitutive functions ¯ , i, Cij are formulated in R3 where the physical

interpretation is easier,
(ii) in the spectral space, these expressions are often quite descriptive,
(iii) the extension of the objects and to R6 can be coded once for all times
and would hold for all isotropic models.

In the case of the invariant representation (V.12), (V.22), one order higher deriva-
tive of (V.22) is required; that is,
" 3 #
X @ ˜ @Ii
:= @" = @" ⇢ . (V.30)
i=1
@Ii @"

Expanding the latter through the chain rule, we arrive at

3 X
X 3 3
X
2
:= C̃ij (@" Ii ⌦ @" Ij ) + ˜i @"" Ii (V.31)
i=1 j=1 i=1

where
@ ˜ (I1 , I2 , I3 )
˜i := ⇢@Ii ˜ (I1 , I2 , I3 ) and C̃ij := ⇢ . (V.32)
@Ii @Ij
The second-order derivatives of the invariants are the following fourth-order ten-
sors
2 2 2 1 1
@"" I1 = 0 , @"" I2 = 1 ⌦ 1 , and @"" I3 = I3 " ⌦" I3 " 1 (V.33)

with
1 1
ijkl = ( ik jl + il jk ) and ( " 1 )ijkl = ("ik1 "jl1 + "il 1 "jk1 ) . (V.34)
2 2
The constitutive modeling in terms of the invariants then focuses on the represen-
tations ˜ , ˜i , C̃ij . Apparently, that is not as descriptive as the principal strain
representation.

3. Linear Isotropic Elasticity

In linear isotropic elasticity, the free energy function is a quadratic function of "
1
⇢ ˆ (") = (tr ")2 + µ " : " . (V.35)
2
that involves just two constant material parameters and µ, which are called the
Lamé constants. The first and second derivatives of the free energy yields the
stress tensor and the elastic moduli

= ⇢@" ˆ = tr(") 1 + 2µ " ,

(V.36)
= @" = 1 ⌦ 1 + 2µ .

Note that the strain-dependent terms in (V.35) can be expressed as tr(") = I1

and " : " = tr("2 ) = I12 2I2 . The free energy and the stress tensor can then be

66
 4. Decoupled Volumetric-Isochoric Elasticity

equivalently expressed in terms of the principal invariants

1 2
⇢ ˜ (I1 , I2 ) = I + µ (I12 2I2 ) ,
2 1 (V.37)
= ˜1 1 + ˜2 (I1 1 ") ,

where
˜1 := ⇢@I1 ˜ = I1 + 2µ I1 and ˜2 := ⇢@I2 ˜ = 2µ . (V.38)
Moreover, from (V.28), we get

+ 2µ 0
C̃ij = . (V.39)
0 0

Expressing the invariants in terms of the principal strains, we obtain

I1 = tr(") = ✏1 + ✏2 + ✏3 , I12 2I2 = ✏21 + ✏22 + ✏23 . (V.40)

Inserting the results (V.40) into (V.35), we obtain the free energy function in
terms of the principal strains
1
⇢ ¯ (✏1 , ✏2 , ✏3 ) = (✏1 + ✏2 + ✏3 )2 + µ (✏21 + ✏22 + ✏23 ) . (V.41)
2
The stresses and moduli in the principal strain space then become
2 3
(✏1 + ✏2 + ✏3 ) + 2µ ✏1
i :=
4 (✏1 + ✏2 + ✏3 ) + 2µ ✏2 5 (V.42)
(✏1 + ✏2 + ✏3 ) + 2µ ✏3

and 2 3
+ 2µ
Cij := 4 + 2µ 5. (V.43)
+ 2µ

4. Decoupled Volumetric-Isochoric Elasticity

Many materials exhibit completely di↵erent volumetric and isochoric (volume pre-
serving) response. This motivates us to a priori decompose the free energy into
its volumetric and isochoric parts. The kinematic departure point is the split of
the strain tensor " into volumetric and deviatoric parts.
1
" = "vol + dev(") = e 1 + "0 (V.44)
3
where
1
e := tr(") and "0 := dev(") := " e1. (V.45)
3

67
V. Linear and Nonlinear Isotropic Elasticity

Insertion of the decomposition (V.44) into (V.35) yields

✓ ◆ ✓ ◆
1 1 1
⇢ ˆ (") = 2
e +µ 0
e1 + " : e1 + " 0
2 3 3
✓ ◆
1 1 (V.46)
= + µ e2 + µ "0 : "0
2 3
1
=  e2 + µ " 0 : " 0
2
where  := + 2µ/3 denotes the bulk modulus and µ stands for the shear modulus.

Generalizing the form (V.46) of linear elasticity toward non-linear elasticity, we

have
⇢ ˆ (") = ⇢ ˆ vol (e) + ⇢ ˆ iso ("0 ) . (V.47)
The stresses are then represented by

= ⇢@" ˆ = ⇢@e vol @" e + ⇢@"0 iso : @" "0 . (V.48)

Using the results

1
@" e = 1 and @" "0 = = 1 ⌦ 1, (V.49)
3
we end up with
= ⇢@e vol 1 + ⇢@"0 iso : . (V.50)
| {z } | {z }
spherical (hydrostatic) stress deviatoric stress

A further di↵erentiation yields the decoupled representation of the moduli.

2
:= @" = ⇢@ee vol 1⌦1+ : ⇢@"20 "0 iso : . (V.51)

In linear elasticity (V.46), we have

= e1 + 2µ"0 : = e1 + 2µ"0 (V.52)

where we have used the property

1 1 0
"0 : = "0 : ( 1 ⌦ 1) = "0 : (" : 1) 1 = "0
3 3
and
=  1 ⌦ 1 + 2µ . (V.53)

68
Chapter VI

Viscoelasticity: Formulation and

Numerical Implementation

Viscoelasticity refers to the time- and rate-dependent response of materials that

may result either from the internal non-equilibrium mechanisms of deformation
that continue to evolve despite of the ceased macroscopic external loading as ob-
served, for instance, in elastomers, or from the existence of a fluid phase, can be
liquid or gas, in the material such as concrete, biological tissue, sponge, and soil to
mention a few. The time-dependent response of viscoelastic materials can be ac-
counted for through micromechanical models where the internal constituents and
the mechanisms of their non-equilibrium deformation are incorporated explicitly.
Alternatively, the phenomenological models of viscoelasticity can be developed at
macro scale to take into account the time-dependent behavior through a set of
internal variables such as the viscous strain and the non-equilibrium stress that
describe the overall non-equilibrium response of the above-mentioned materials
collectively in a homogenized manner. In this chapter, we will restrict ourselves to
a great extent to the latter approach to the constitutive modeling of viscoelastic
material response and elaborate on the theoretical, numerical, and algorithmic
aspects of the phenomenological viscoelasticity.

1. Motivation: One-Dimensional Representations of Linear Viscoelas-

ticity
In this section, we introduce the di↵erent representations of linear viscoelasticity in
the one-dimensional setting to motivate the theoretical and algorithmic treatment
of viscoelasticity in the three-dimensional space in the upcoming sections.

1.1 Representation A: Generalized Internal Variable Formulation

Consider the one-dimensional rheological generalized Maxwell model depicted
in Figure VI.1. The local state of the material is described by the total strain "
and the internal variables ↵i=1,...,n that correspond to the viscous strains in the
Maxwell branches
State(x, t) = {", ↵1 , . . . , ↵n } . (VI.1)
The free energy expressing the stored energy in the Hookean elements (springs)
n
X1
1
⇢ ˆ (", ↵1 , . . . , ↵n ) = E0 "2 + Ei (" ↵i )2 (VI.2)
2 i=1
2

69
VI. Viscoelasticity: Formulation and Numerical Implementation

External
Internal E0

⌘1 E1

↵1
⌘n En

↵n

"
Figure VI.1.: The generalized Maxwell model of linear viscoelasticity

yields
n
X
= ⇢@" ˆ =E0 " + Ei (" ↵i ) ,
i=1 (VI.3)
i = ⇢@↵i ˆ = Ei (" ↵i ) .
The following equations then describe Representation A
n
X
(A1) = E0 " + Ei (" ↵i ) , (VI.4)
i=1
(A2) i = Ei (" ↵i ) , (VI.5)
(A3) ↵˙ i = i /⌘i with ↵i (t = 0) = 0 . (VI.6)

1.2 Representation B: In Terms of Internal Forces

Inserting (VI.6) into (VI.5), we obtain Representation B
n
X
(B1) = E0 " + i , (VI.7)
i=1

(B2) ˙ i + 1 Ei i = Ei "˙ with i (t = 0) = 0 (VI.8)

⌘i
where the viscous strains ↵i are suppressed and the thermodynamic forces i
conjugate to the viscous strains ↵i appear as the internal variables.

1.3 Representation C: Convolution Integral Representation

The exact integration of the linear evolution (VI.6)
1 1
↵˙ i + ↵i = " , ↵i (t = 0) = 0 , ⌧i := ⌘i /Ei (VI.9)
⌧i ⌧i
results in Z t ✓ ◆
1 t s
↵i (t) = exp "(s)ds (HE) . (VI.10)
⌧i 0 ⌧i

70
 2. Three-Dimensional Representation of Linear Viscoelasticity

Integrating the latter expression in (VI.10) by parts, we then arrive at

 ✓ ◆ s=t Z t ✓ ◆
t s t s
↵i (t) = exp "(s) exp "(s)ds
˙ . (VI.11)
⌧i s=0 0 ⌧i

For "(t = 0) = 0, we obtain

Z t ✓ ◆
t s
↵i (t) = "(t) exp "(s)ds
˙ (VI.12)
0 ⌧i

whose incorporation in the stress expression (VI.4) yields

Z t
(C1) (t) = Ê(t s)"(s)ds
˙ (VI.13)
0
✓ ◆
Pn t
where Ê(t) := E0 + i=1 Ei exp denotes the relaxation function.
⌧i

2. Three-Dimensional Representation of Linear Viscoelasticity

In this section, we extend the representations of linear viscoelasticity outlined in
the one-dimensional setting in the preceding section towards the formulation in
the three-dimensional framework.

2.1 Volumetric-Isochoric Split

In most viscoelastic materials such as polymers, the viscous e↵ect is restricted to
the isochoric part of the deformation, while the volumetric response remails elastic.
The kinematic basis for a decoupled model is a decomposition of the strain tensor
into the volumetric and isochoric parts
1
" = e 1 + "0 , (VI.14)
3
which are expressed in terms of the dilatation e and the deviatoric strain tensor
"0 . The latter are defined as

e := tr(") and "0 = dev(") . (VI.15)

2.2 Representation A: Internal Variable Formulation

Motivating from the rheological element in the one-dimensional setting, shown in
Figure VI.1, the local state of a viscoelastic material of the generalized Maxwell
model is governed by the strain tensor " and n second-order symmetric tensors of
strain-like internal variables ↵1 , . . . , ↵n , i.e.

State(x, t) = {", ↵1 , . . . , ↵n } . (VI.16)

The free energy is then assumed to have the following decoupled form

⇢ ˆ (", ↵1 , . . . , ↵n ) = ⇢ ˆ vol (e) + ⇢ ˆ iso ("0 , ↵1 , . . . , ↵n ) . (VI.17)

71
VI. Viscoelasticity: Formulation and Numerical Implementation

For materials with quadratic energy storage functions, being analogous to (VI.2),
we have the following expression
X n
1
⇢ ˆ = e2 + µ0 "0 : "0 + µi ("0 ↵i ) : ("0 ↵i ) (VI.18)
2 i=1

in terms of the bulk modulus  = E/[3(1 2⌫)], the equilibrium shear modulus
µ0 , and the viscous shear moduli µi=1,...,n associated with n Maxwell branches.
The stresses and the internal forces that are defined by

= ⇢@" ˆ and i := ⇢@↵i ˆ (VI.19)

have the specific form

" n
#
X
(A1) = e1 + 2µ0 "0 + 2µi ("0 ↵i ) : (VI.20)
i=1

1
where := 3
1 ⌦ 1 is the fourth-order deviatoric projection tensor and the
internal forces

(A2) i = 2µi ("0 ↵i ) (VI.21)

governing the linear evolution equation

1
(A3) ↵˙ i = i with ↵i (t = 0) = 0 . (VI.22)
⌘i
The latter is based on the linear stress-strain rate relationship of Newtonian fluids
through the coefficients of viscosity ⌘i=1,...,n of the Maxwell branches, symbolically
shown in Figure VI.1. Therefore, the governing equations of Representation A in
the three-dimensional setting are given in (VI.20)-(VI.22).

Remark VI.1. We observe that the trace of (VI.22) has the form
1 (VI.21) 1
tr(↵˙ i ) = tr( i ) = tr(↵i ) (VI.23)
⌘i ⌧i

with ⌧i := ⌘i /2µi denoting the relaxation time of the ith Maxwell branch. This
means
d 1
[tr(↵i )] + tr(↵i ) = 0 with tr(↵i (t = 0)) = 0 (VI.24)
dt ⌧i
Therefore, the evolution equation (VI.22) with the initial condition ↵i (t = 0) = 0
ensures that the internal variable tensor ↵i is traceless (deviatoric). For this
reason, the deviatoric projection in (VI.20) is actually not needed.

2.3 Representation B: In Terms of Internal Forces

Insertion of (VI.21) into (VI.20) yields
n
X
0
(B1) = e1 + 2µ0 " + i (VI.25)
i=1

72
 2. Three-Dimensional Representation of Linear Viscoelasticity

and the incorporation of (VI.21) in (VI.22) for expressing the rates ↵˙ i in terms of
the rates ˙ i leads us to the evolution equation for the internal stresses

(B2) ˙i + 1 i = 2µi "˙ 0 with i (t = 0) = 0 (VI.26)

⌧i
where ⌧i := ⌘i /2µi . Hence, the governing equations of Representation B in the
three-dimensional setting are given in (VI.25)-(VI.26).

2.4 Representation C: Convolution Integral Representation

Insertion of (VI.21) into (VI.22) yields the evolution equation
1 1
↵˙ i + ↵i = " 0 with ↵i (t = 0) = 0 . (VI.27)
⌧i ⌧i
We first observe that
 ✓ ◆ ✓ ◆
d t 1 t
exp ↵i (t) = exp "0 (t) (VI.28)
dt ⌧i ⌧i ⌧i

can be conceived as an alternative form of (VI.27) with a prefactor exp(t/⌧i ) and

⌧i := ⌘i /2µi . Integration of (VI.28) over s 2 [0, t]
Z t  ✓ ◆ Z t ✓ ◆
d s 1 s
exp ↵i (s) ds = exp "0 (s)ds (VI.29)
0 ds ⌧ i 0 ⌧ i ⌧ i

then yields
✓ ◆ ✓ ◆ Z t ✓ ◆
t 0 *0 1 s
exp ↵i (t) exp ↵i (0) = exp "0 (s)ds (VI.30)
⌧i ⌧i 0 ⌧ i ⌧ i

for ↵i (0) = 0. Solving the last expression for ↵i (t), we get

✓ ◆Z t ✓ ◆
t 1 s
↵i (t) = exp exp "0 (s)ds
⌧i 0 ⌧i ⌧i
Z t ✓ ◆ (VI.31)
1 t s
= exp "0 (s)ds .
⌧i 0 ⌧i

Integrating (VI.31) by parts, we then obtain

Z t ✓ ◆
t s
↵i (t) = "0 (t) exp "˙ 0 (s)ds . (VI.32)
0 ⌧i

Insertion of (VI.32) into (VI.20) finally leads us to the convolution representation

of the stress tensor
Z t
(C1) = e1 + 2µ̂(t s)"˙ 0 (s)ds (VI.33)
0

73
VI. Viscoelasticity: Formulation and Numerical Implementation

Pn ⇣ ⌘
t
where µ̂(t) := µ0 + i=1 µi exp ⌧i
stands for the relaxation function for the
shear modulus and has the property

lim µ̂(t) = µ0 (VI.34)

t!1

i.e. µ0 is the shear modulus of our model in its equilibrium state.

3. Stress Updates and Consistent Tangent Moduli in Linear

Viscoelasticity
We consider a typical time step [tn , tn+1 ] where all variables at time tn are known
and construct algorithms for the update of internal variables and the computa-
tion of the stresses n+1 and the consistent moduli n+1 at time tn+1 for the
deformation-driven setting.

3.1 Algorithmic Representation of Evolution Di↵erential Equations

Let us recall Representations A and B of linear viscoelasticity
8 n
9
>
> X >
>
>
> = e1 + 2µ0 "0 + 2µi ("0 ↵i )>
>
>
> >
>
< i=1 =
(A) i = 2µi ("
0
↵i ) (VI.35)
>
> >
>
>
> >
>
>
> i >
>
: ↵˙ i = with ↵(t = 0) = 0 ;
⌘i

8 n 9
> X >
>
> = e1 + 2µ0 " +0 >
>
< i =
(B) i=1 (VI.36)
> ✓ ◆ >
>
> 1 >
>
: ˙ i = 2µi "˙ 0 i with i (t = 0) = 0;
⌘i

where e := tr(") and "0 := dev(").

xGauss
Storage of n ⇥ 6 variables
at each Gauss point of FE mesh

Figure VI.2.: Schematic representation of Gauss quadrature (integration) points

on a FE mesh

Within the framework of an implicit FEM, these representations govern the lo-
cal material response at an integration point of a typical finite element mesh as
schematically depicted in Figure VI.2.

74
 3. Stress Updates and Consistent Tangent Moduli in Linear Viscoelasticity

For Representation B, the set of internal variables at each Gauss quadrature (in-
tegration) point of the mesh is

I(x, t) = { 1, 2, . . . , n} (VI.37)

that represents the local database for the inelastic response at an integration
point. For a deformation-driven stress update algorithm within a time step t :=
tn+1 tn , we have
• all the variables at time tn are known,
• the total strain tensor "n+1 given (" = B de ), and
• the algorithm computing the internal state variables (↵i ’s in Representation
A, i ’s in Representation B at time tn+1 as illustrated in Figure VI.3.

Given are "n+1 : Prescribed "n+1 : Input

{"n , I n } I n+1 : Updated by algorithm
t In I n+1 : Output
ALGO
a6
input
tn tn+1
Figure VI.3.: Schematic description of the algorithmic update of internal variables

The simplest unconditionally stable single-step update, the Backward Euler Scheme
when applied to (VI.36)2 gives
" #
1 1 1
i
( n+1 i
n ) = 2µi ("0n+1 "0n ) i
n+1 (VI.38)
| t {z } | t {z } ⌘ i
⇡ ˙ n+1
i ⇡"˙0n+1

for i = 1, . . . , n. Multiplying (VI.38) with t = tn+1 tn , we get

2µ t
i
n+1
i
n = 2µi ("0n+1 "0n ) i
n+1 , (VI.39)
⌘i
which yields the sought linear update equation
✓ ◆
2µ t 0
1+ i i
n+1 = n + 2µi ("n+1 "0n ) . (VI.40)
⌘i

The latter can be recast into a more compact form

i
n+1 = ci1 i
n + ci2 ("0n+1 "0n ) (VI.41)

where ci1 := (1 + 2µi t/⌘i ) 1 and ci2 := 2µi ci1 . This is a first-order accurate single-
step numerical integration algorithm for linear viscoelasticity. The total stress
tensor is then computed by function evaluation at time tn+1
n
X
n+1 = en+1 1 + 2µ0 "0n+1 + i
n+1 . (VI.42)
| {z } i=1
equilibrium part | {z }
viscous overstress part

75
VI. Viscoelasticity: Formulation and Numerical Implementation

Insertion of (VI.41) into (VI.42) yields

n
X
n+1 = en+1 1 + 2µ0 "0n+1 + ci1 i
n + ci2 ("0n+1 "0n ) . (VI.43)
i=1

The consistent tangent moduli are then obtained by

d n+1
n+1 := . (VI.44)
d"n+1
Making use of now the well-known results
1
@" e = 1 , @" "0 = = 1 ⌦ 1, (VI.45)
3
we obtain n
X
n+1 = 1 ⌦ 1 + 2µ0 + ci2 . (VI.46)
| {z }
i=1
elastic | {z }
viscous

Box VI.1: Stress Update Algorithm for Linear Viscoelasticity

1 2 n
1. Given are the database {", , ,..., } at time tn and the current
total strain "n+1
2. Decompose the strains

en+1 = tr("n+1 ) and "0n+1 = dev("n+1 ) (VI.47)

3. Setup the algorithmic constants

1
ci1 = and ci2 = 2µi ci1 (VI.48)
1 + 2µi t/⌘i

and update the internal variables

i
n+1 = ci1 i
n + ci2 ("0n+1 "0n ) (VI.49)

4. Compute the total stress tensor

n
X
n+1 = en+1 1 + 2µ0 "0n+1 + i
n+1 (VI.50)
i=1

5. Compute the tangent moduli

" n
#
X
n+1 = 1 ⌦ 1 + 2µ0 + ci2 (VI.51)
i=1

76
 3. Stress Updates and Consistent Tangent Moduli in Linear Viscoelasticity

3.2 Algorithm for Representation C in Terms of Convolution Integral

Equation
Recall the convolution integral representation (Representation C) in (VI.33)
8 Z t 9
>
> = e1 + 0
2µ̂(t s)"˙ (s)ds >
>
>
< >
=
0
(C) Xn (VI.52)
>
> >
>
>
: with µ̂(t) := µ0 + µi exp( t/⌧i )>
;
i=1

where ⌧i := ⌘i /2µi . This expression may be recast into the form

n
X Z t ✓ ◆
0 t s
= e1 + 2µ0 " + i with i := 2µi exp "˙ 0 (s)ds . (VI.53)
i=1 0 ⌧i

This is none other than an integral representation of the internal forces in contrast
to Representation B. The key point of the algorithmic setting is the approximation
of the integral in (VI.53) within the interval [tn , tn+1 ].
Incorporating the property exp(a + b) = exp(a) exp(b), we obtain
tn+1Z
=tn + t ✓ ◆
tn + t s
i
n+1 = 2µi exp "˙ 0 (s)ds
⌧i
0
Z tn ✓ ◆ ✓ ◆
t tn s
= 2µi exp exp "˙ 0 (s)ds
0 ⌧i ⌧i
Z tn+1 ✓ ◆
tn+1 s
+ 2µi exp "˙ 0 (s)ds
tn ⌧i
✓ ◆ Z tn ✓ ◆
t tn s
= exp 2µi exp "˙ 0 (s)ds
⌧i ⌧i
|0 {z }
i
n
Z tn+1 ✓ ◆
tn+1 s
+ 2µi exp "˙ 0 (s)ds
tn ⌧i

and thus
✓ ◆ Z tn+1 ✓ ◆
t tn+1 s
i
n+1 = exp i
n + 2µi exp "˙ 0 (s)ds . (VI.54)
⌧i tn ⌧i

Observe that so far no approximation has been introduced. Therefore, this ex-
pression is still exact. The integral over the time step [tn , tn+1 ] in (VI.54) may now
be approximated by an algorithm. For this purpose, we consider the midpoint

77
VI. Viscoelasticity: Formulation and Numerical Implementation

rule; that is,

Z tn+1 ✓ ◆
tn+1 s
2µi exp "˙ 0 (s)ds
tn ⌧i
" !✓ ◆#
tn +tn+1
tn+1 2 "0n+1 "0n
⇡ t 2µi exp (VI.55)
⌧i t
✓ ◆
t
= 2µi exp "0n+1 "0n
2⌧i

Inserting this result into the update equation (VI.54), we get

✓ ◆ ✓ ◆
t t
i
n+1 = exp
i
n + 2µi exp "0n+1 "0n . (VI.56)
⌧i 2⌧i

This algorithm can be recast into the identical form as the algorithm outlined for
Representation B in (VI.41)
i
n+1 + ci2 ("0n+1 "0n )
= ci1 i
n (VI.57)
⇣ ⌘ ⇣ ⌘
t t
where ci1 = exp ⌧i
and c i
2 = 2µ i exp 2⌧i
. This is a second-order accurate
one-step algorithm for linear viscoelasticity. Observe that the structure of this
algorithm retains the preceding algorithm summarized in Box VI.1. The only
change is in the parameters ci1 and ci2 and it results in a more accurate algorithm
of second-order.

Remark VI.2. The algorithmic formulation of Representation A follows directly

from that of Representation B. To this end, the implicit update of the viscous strain
can be obtained as
↵in+1 = ↵in + t↵˙ in+1
2µi ("0n+1 ↵in+1 )
= ↵in + t
⌘i (VI.58)
 1
t2µi 2µi 0
↵in+1 = 1+ ↵in + t "
⌘i ⌘i n+1

Substituting this result into the algorithmic expression of the stress tensor and the
latter into that of the consistent tangent moduli, we arrive at
n
X
n+1 = en+1 1 + 2µ0 "0n+1 + 2µi ("0n+1 ↵in+1 ) ,
i=1

n+1 = @"n+1 n+1 (VI.59)

n
X 
t2µi /⌘i
= 1 ⌦ 1 + 2µ0 + 2µi 1 .
i=1
1 + t2µi /⌘i

78
 4. Representation of Nonlinear Viscoelasticity

4. Representation of Nonlinear Viscoelasticity

Two sources of physically based nonlinearities may be conceived in viscoelastic
solids.

(i) The equilibrium elastic response is non-linear; i.e. the free energy is not a
quadratic function of the strains anymore: Non-Hookean Response.
(ii) The evolution equations of internal variables are nonlinear: Non-Newtonian
Response.

viscous overstress (non-equilibrium)

response

non-linear
equilibrium
response

"
Figure VI.4.: Schematic illustration of nonlinear viscoelastic response where the
nonlinearity arises from the nonlinear equilibrium response

4.1 Viscoelasticity with Nonlinear Equilibrium Elastic Response

A material with nonlinear equilibrium response and linear viscous response whose
schematic stress-strain response is depicted in Figure VI.4 can be represented by
8 X n 9
>
> >
>
>
> = ⇢@e ˆ vol (e)1 + ⇢@"0 ˆ iso (" ) : +
0
i >
>
>
< | {z } i=1
>
=
(B)nonlinear
nonlinear elastic | {z } (VI.60)
>
>  linear overstress >
>
>
> 1 i >
>
>
: with ˙ i = 2µi "˙ 0 for i = 1, . . . , n > ;
⌘i

where ˆ vol (e) and ˆ iso ("0 ) are nonlinear energy storage functions. Note that the
evolution equation of the viscous non-equilibrium stress tensor i is still linear
and thus it can be recast into a convolution representation.
8 n 9
> X >
>
> = ⇢@e ˆ vol (e)1 + ⇢@"0 ˆ iso (" ) : +
0 >
>
< i =
(C)nonlinear i=1
> Z t ✓ ◆ >
>
> t s 0 >
>
: with i = 2µ i exp "˙ (s)ds for i = 1, . . . , n ;
0 ⌧ i
(VI.61)

79
VI. Viscoelasticity: Formulation and Numerical Implementation

4.2 Viscoelasticity with Nonlinear Evolution Equations

Incorporation of a spectrum of relaxation times through several linear Maxwell
branches requiring i = 1, . . . , n equations (VI.60) of Representation B, can be
alternatively achieved by introducing nonlinear evolution equations
˙ = f̂ ("˙ 0 , ) . (VI.62)
| {z }
nonlinear function

For example, we may substitute for the linear evolution equations (VI.22) of Rep-
resentation A for the internal variables ↵i ’s a nonlinear evolution equation
n
X
= ⇢@" ˆ = e1 + 2µ0 "0 + 2µi ("0 ↵i ) ,
i=1

i = 2µi ("0 ↵i ) , (VI.63)

1 i
↵˙ i = f ( i ) = k i k i .
⌘i k ik

Observe that this model is thermodynamically consistent for ⌘i > 0. From the
local forms of the Clausius-Planck Inequality, we have
n
X
⇢Dloc = i : ↵˙ i
i=1
n
(VI.63)3X 1 i
= i : k ik i (VI.64)
i=1
⌘i k ik
n
X 1 i +1
= k ik .
i=1
⌘i

Recall that k i k2 = i : i . Using the Backward-Euler Scheme for the update of

the internal variables within a typical time step [tn , tn+1 ], we obtain

↵in+1 = ↵in + t↵˙ in+1 . (VI.65)

Inserting the evolution equation (VI.63)3 , we end up with the following nonlinear
update equation
i
1 n+1
↵in+1 = ↵in + t i
n+1
i
i
. (VI.66)
⌘i n+1

This nonlinear equation necessitates a local iterative solution scheme to compute

↵in+1 . These aspects will be elaborated on in Exercise 3.

80
Chapter VII

Rate-Independent and Rate-Dependent

Elastoplasticity: Formulation and
Numerical Implementation

In contrast to viscoelasticity, covered in the preceding chapter, the rate-independent

elastoplasticity corresponds to inelastic material behavior where the dissipative
response arises from rate-independent frictional internal mechanisms such as the
plastic slip phenomenon due to dislocation motion in crystalline materials and the
frictional relative slip between the particles constituting granular materials. More-
over, elastoplasticity is typically characterized by a stress threshold, the so-called
the yield stress y0 as shown in Figure VII.1. The elastoplastic material response
with the rate-dependent yield stress and the post-yield behavior is referred to as
rate-dependent elastoplasticity, also known as elastoviscoplasticity.

y0 Fracture

"
"p
F = A0

Figure VII.1.: Elastoplastic uniaxial stress-strain response of a metallic bar in ten-

sion. The elastic response is bounded by the yield stress y0

1. Ideal Elastoplasticity
In classical rate-independent plasticity, stresses are bounded by an admissible
stress space, the so-called the elastic domain . The simplest one-dimensional
representation may be given by means of the model of ideal elastoplasticity as
depicted in Figure VII.2. Here, the elastic energy storage is governed by the Hooke
element (spring), while the rate-independent frictional dissipation is described by
the so-called St.Venant element, a friction element with the stress threshold y0 .
The St.Venant element remains stationary as long as the magnitude of the stress is

81
VII. Rate-Independent and Rate-Dependent Elastoplasticity

lower than the stress threshold y0 , whereas the purely dissipative frictional sliding
gets initiated as the stress equals y0 , see Figure VII.2.
y0 E

"p "e
"
Figure VII.2.: Schematic representation of ideal elastoplasticity

A typical uniaxial stress-strain diagram of an ideal elastoplastic material, i.e. with-

out any hardening, is shown in Figure VII.3.

y0

"
"p "e
"

y0

Figure VII.3.: The stress-strain response of an ideal elastoplastic material

The state of an ideal elastoplastic material is then described by the two field
variables
State(x, t) = {", "p } (VII.1)
where the total strain tensor " is the external variable and the plastic strain
tensor "p 2 R6 is the symmetric internal variable, which remains constant upon
unloading.

1.1 Elastic Stress Response

We assume a typical dependence of the free energy on the state variables

= ˜ (", "p ) = ˆ (" "p ) = ˆ ("e ) (VII.2)

where the elastic strain tensor is defined as

"e := " "p (VII.3)

in the small strain setting. Starting with the Clausius-Planck Inequality (III.53)
under isothermal conditions

⇢Dloc = : "˙ ⇢˙ 0, (VII.4)

82
 1. Ideal Elastoplasticity

we obtain
⇢Dloc = [ ⇢@"e ˆ ] : "˙ + ⇢@"e ˆ : "˙ p 0. (VII.5)
Through the Coleman’s exploitation method, we get

= ⇢@"e ˆ ("e ) (VII.6)

the local dissipation then reduces to

⇢Dloc = : "˙ p 0 (VII.7)

Remark VII.1. If = ˜ (", "p ) can be recast into the functional form ˆ (" "p ) =
ˆ ("e ), the thermodynamic force conjugate to the plastic strain tensor (internal
variable) and the stress tensor come out identical; that is,

= ⇢@" ˜ = ⇢@"e ˆ : @" "e = ⇢@"e ˆ ,

|{z}
(VII.8)
= ⇢@"p ˜ = ⇢@"e ˆ : @"p "e = ⇢@"e ˆ ,
| {z }

thus, = .

1.2 Plastic Flow Response

We need to specify an additional material equation for the evolution of the plastic
strain tensor "˙ p , also known as the plastic flow rule. A canonical formalism can
be achieved by exploiting the Principle of Maximum Dissipation (PMD)
subject to the constraint bounding the current stress state by an elastic domain

:= { 2 R6 | ˆ  0} (VII.9)

where ˆ( ) is the yield criterion function. Note that the yield criterion ˆ( ) =
0 defines a hypersurface in the stress space, which is assumed to be convex for
the appropriate use of the Lagrange Method.
The flow rule can be obtained by an extremum principle that maximizes the
dissipation for all admissible ⇤ 2 and the current stresses 2 @ ,
⇤ ⇤
: "˙}p
| {z
p
| {z: "˙} 8 2 ,
⇤
⇢Dloc ⇢Dloc (VII.10)
⇤
⇢Dloc ⇢Dloc .

The principle of maximum dissipation states that the projection of any admissible
stress tensor ⇤ 2 onto the flow direction "˙ p is always less than that of the cur-
rent stresses tensor on the current direction of plastic flow. The PMD (VII.10),
when formulated mathematically as a constrained optimization problem through
the Lagrange multiplier method, induces the normality structure for "˙ p where
the direction of the plastic flow turns out to be the normal direction @ ˆ( ) of
the convex yield surface as schematically illustrated in Figure VII.4.

83
VII. Rate-Independent and Rate-Dependent Elastoplasticity

"˙p = @ ˆ( )

⇤
O
⇢Dloc = : "˙ p
⇤ ⇤
⇢Dloc = : "˙ p
@ := { 2 R6 | ˆ = 0}
Figure VII.4.: Graphical illustration of the principle of maximum dissipation

We like to maximize the dissipation ⇢Dloc = : "˙ p , or equivalently minimize

⇢Dloc , under the constraint that lies in the elastic domain 2 . The La-
grange function then reads

L( , ) = : "˙ p + ˆ( ) ! stationary. (VII.11)

The necessary conditions for the solution of the constrained optimization problem
(VII.10) are the Karush-Kuhn-Tucker (KKT) optimality conditions.

@ L=0 "˙ p = @ ˆ ,
(VII.12)
0 , ˆ( )  0 , ˆ( ) = 0 .

The flow rule of ideal elastoplasticity "˙ p = @ ˆ is normal to the yield surface in
the stress space along with the KKT optimality conditions. The thermodynamical
consistency of the model can be shown by inserting (VII.12) into (VII.7)

⇢Dloc = : "˙ p = : @ ˆ 0. (VII.13)

For the rate-independent theory, the yield criterion function ˆ( ) can be expressed
as
ˆ( ) = ˆ( ) C (VII.14)
where C is a constant related to the initial yield stress. The function ˆ( ) is called
the level set function, which is homogeneous of degree one; that is,

ˆ(↵ ) = ↵ ˆ( ) 8↵ 2 R+ (VII.15)

resulting in
: @ ˆ = ˆ( ) . (VII.16)
Use of (VII.16) in (VII.13) yields

⇢Dloc = :@ ˆ= : @ ˆ = ˆ( ) (VII.17)
| {z }
ˆ( )

and when ˆ( ) = 0, i.e. plastic flow occurs, we have ˆ( ) = C, thus

⇢Dloc = C 0. (VII.18)

84
 1. Ideal Elastoplasticity

This condition is always fulfilled since 0 in (VII.12) and C 2 R+ . As expected,

the normality rule based on the PMD is thermodynamically consistent.

Box VII.1: Model of Ideal Elastoplasticity

1. Kinematics "e := " "p

2. Free Energy = ˆ ("e )
3. Stresses = ⇢@"e ˆ
4. Yield Criterion = ˆ( )  0
5. Flow Rule "˙ p = @ ˆ( )
6. Loading/Unloading (KKT) 0, ˆ( )  0, ˆ( ) = 0

1.3 Tangent Moduli (Prandtl-Reuss Tensors)

We seek for the relationship between ˙ and ". ˙ Starting from the stress expression,
we obtain
˙ = ⇢@"2e "e ˆ : "˙ e
(VII.3) 2
= ⇢@"e "e ˆ : ("˙ "˙ p ) (VII.19)
(VII.12) 2
= ⇢@"e "e ˆ : ("˙ @ ˆ) .
For an elastic step, the plastic parameter follows from KKT conditions that

= 0 for an elastic step (VII.20)

yielding
˙ = ⇢@"2e "e ˆ : "˙ (VII.21)
identical to elasticity. For a plastic step ( > 0), however, we can compute the
plastic parameter in terms of "˙ from the so-called consistency condition that
requires the vanishing rate of the yield criterion function for a persistently yielding
elastoplastic material; that is,
ˆ = 0, ˆ=0 ! ˙ = 0. (VII.22)

This condition can be exploited as follows

(VII.19)
˙ =@ ˆ: ˙ = @ ˆ : ⇢@"2e "e ˆ : ("˙ @ ˆ) = 0 . (VII.23)

Solving the latter equality for , we obtain

@ ˆ : ⇢@"2e "e ˆ : "˙

= . (VII.24)
@ ˆ : ⇢@"2e "e ˆ : @ ˆ

Insertion of (VII.24) into (VII.19) yields the rate form

ep
˙ = : "˙ (VII.25)

85
VII. Rate-Independent and Rate-Dependent Elastoplasticity

where
1
ep
:= e
( e
: @ ˆ) ⌦ ( e
: @ ˆ) (VII.26)
D
and
e
:= ⇢@"2e "e ˆ and D := @ ˆ : e
: @ ˆ. (VII.27)
ep
The Prandtl-Reuss Tensor governs the stress rate response in plasticity.

Remark VII.2. Observe that the underlined term in (VII.26) is active only in
case of a plastic loading, i.e. as ˆ( ) = 0 and > 0.

1.4 Specification of a Model Problem: von Mises Plasticity

Consider one of the most important models of plasticity, namely the von Mises-
type J2 -theory that specifies ˆ( ) for the following isotropic case.
Assuming the yield criterion function ˆ is an isotropic function of , we have

= ˆ(I , II , III ) (VII.28)

with
1
I := tr( ) , II := (I 2 tr( 2
)) and III := det( ) (VII.29)
2
denoting the principal invariants of the stress tensor. Decomposing the stress
tensor additively into volumetric and deviatoric parts
0 0
= p1 + or := p1 , (VII.30)

we can introduce the invariants p, J2 , and J3 .

1 1 02 1 03
p := tr( ) , J2 := tr( ) and J3 := tr( ) (VII.31)
3 2 3
where p is the negative pressure and 0 := dev . The yield criterion function
can then be alternatively formulated in terms of these newly introduced invariants

= ˜(p, J2 , J3 ) . (VII.32)

The von Mises theory of plasticity assumes a simplified representation of (VII.32)

only in terms of the second basic invariant J2 of the deviatoric stress tensor 0
r
p 2
( ) = ˆ(J2 ) = 2J2 y0  0 (VII.33)
3
1 0 0
Inserting J2 := 2
: into (VII.33), we obtain
r
ˆ( ) = k k 0 2
y0  0 . (VII.34)
3

86
 1. Ideal Elastoplasticity
q
2
Remark VII.3 (On the factor 3
in the yield criterion function). Note that
the material parameter, the stress threshold y0 is none other than the yield stress
measured in a uniaxial test. Therefore, the stress tensor and its deviatoric part
corresponding to the uniaxial stress state at the onset of plastic flow are given by
2 3 2 3
y0 2y0 /3
=4 0 5; 0
=4 y0 /3 5. (VII.35)
0 y0 /3
0
The norm of the deviatoric stress tensor can then be calculated as
p r
y 0 6 2
k 0 k = [22 + ( 1)2 + ( 1)2 ]1/2 = y0 = y0 . (VII.36)
3 3 3
q
This result explains the factor 23 in the von Mises yield criterion function. There-
fore, if one wants to use the yield stress measured through another experiment than
the uniaxial one, this factor should be adjusted accordingly. (HE) Derive this factor
for the yield stress y0eqb obtained from the equibiaxial tension test.
Observe that the von Mises (J2 ) theory neglects the dependence on pressure. The
geometric visualization of the von Mises surface ˆ = 0 in the principal stress space
3
X
= ↵ n↵ ⌦ n↵ (VII.37)
↵=1

takes a cylindrical shape with the axis associated with the hydrostatic stress state
as shown in Figure VII.5. As depicted, the von Mises criterion bounds 0 but

3
A 2

p1
Hydrostatic Axis A
1

Figure VII.5.: Graphical illustration of the von Mises yield surface

not the pressure. The cross-section of the elastic domain gives the so-called ⇡-
plane deviatoric projection where the von Mises criterion appears as a circle in
the deviatoric space as illustrated in Figure VII.6. The evolution law then gets
the form 0
p
"˙ = @ = ˆ (VII.38)
k 0k
where signifies the amount of the plastic flow and 0 / k 0 k represents the di-
rection of the flow. The von Mises plasticity can be combined with an isotropic

87
VII. Rate-Independent and Rate-Dependent Elastoplasticity

3
Hydrostatic axis
"˙ p = @ ( )
q
2
y
3 0
0

1 2
q
2
= || 0 || y
3 0
=0

Figure VII.6.: Graphical illustration of the deviatoric plane (⇡-plane) of the von
Mises yield surface

elasticity law
1
⇢ ˆ ("e ) =  tr2 ("e ) + µ "e0 : "e0 (VII.39)
2
yielding the the pressure and the stress deviator
0
p =  tr("e ) and = 2µ"e0 . (VII.40)

Box VII.2: von Mises Plasticity

1. Kinematics "e := " "p

2. Free Energy ⇢ ˆ ("e ) = 12  tr2 ("e ) + µ"e0 : "e0
3. Stresses =  tr("e )1 + 2µ"e0
q
4. Yield Criterion ˆ = k 0k 2
y 0
3 0

5. Flow Rule "˙ p = @ ˆ( )

6. Loading/Unloading (KKT) 0, ˆ( )  0, ˆ( ) = 0

We note that the model is characterized by the three material parameters, namely
the bulk Modulus , the shear Modulus µ, and the yield stress y0 . Moreover, it is
important to observe that the von Mises theory describes the isochoric flow only
✓ 0
◆
p p
ė := tr("˙ ) = tr =0 ep = 0 . (VII.41)
k 0k

Therefore, the trace of the elastic strain tensor is equal to that of the total strain
tensor; that is,
*0
tr("e ) = tr(") tr("p ) = tr(") , (VII.42)
which implies that the volumetric response of the von Mises material remains
elastic.

88
 1. Ideal Elastoplasticity

1.5 Stress Update Algorithm for von Mises Plasticity: The Radial
Return Algorithm
As in the algorithmic treatment of viscoelasticity, we consider a typical time inter-
val [tn , tn+1 ] and integrate the evolution equations within this interval as schemat-
ically depicted in Figure VI.3 for a set of generic internal variables I. The in-
troduction of an elastic trial state yields the canonical radial return scheme, the
well-known algorithm of computational plasticity. The integration of the flow rule
using the backward (implicit) Euler scheme yields
0
"pn+1 = "pn + n+1
n+1
0
(VII.43)
n+1

with the incremental KKT conditions

n+1 0, ˆn+1  0 , n+1 n+1 =0 (VII.44)

where n+1 := n+1 [tn+1 tn ]. Inserting (VII.44) into the stress expression
0
n+1 = 2µ"e0n+1 = 2µ("0n+1 "p n+1 ) (VII.45)

we obtain the following representation

0
0 n+1
n+1 = 2µ("0n+1 "p n ) 2µ n+1 0
(VII.46)
n+1

where the first part defines the trial stress tensor

0 trial
n+1 = 2µ("0n+1 "p n ) , (VII.47)

which is obtained by advancing the total strain "n+1 at frozen plastic strain "pn .
0 tr(ial)
Note that n+1 is known in a deformation driven algorithm at time tn+1 . Thus
(VII.46) becomes
0
0 0 tr n+1
n+1 = n+1 2µ n+1 0
. (VII.48)
| {z } n+1
trial stress | {z }
softening stress

In order to find out whether the incremental step is accompanied by plastic flow
or not, we need to check the current yield criterion
r
ˆn+1 = 0 2
n+1 y0  0 . (VII.49)
3
0
To compute n+1 , we recast (VII.48) into the form
0 0 tr 0
0 n+1 0 tr n+1 n+1
n+1 0
= n+1 0 tr
2µ n+1 0
(VII.50)
n+1 n+1 n+1

or
0 0 tr
n+1 nn+1 = n+1 ntr
n+1 2µ n+1 nn+1 (VII.51)

89
VII. Rate-Independent and Rate-Dependent Elastoplasticity

with 0 0 tr
n+1 n+1
nn+1 := 0
and ntr
n+1 := 0 tr
. (VII.52)
n+1 n+1

From this, we may deduce the following results

0 0 tr
n+1 = n+1 2µ n+1 and nn+1 = ntr
n+1 (VII.53)

for the norm of the stress tensor and the normal flow direction. Thus, the current
yield criterion function may be written in the form
r
0 tr 2
n+1 = n+1 y0 2µ n+1  0 , (VII.54)
3
which can be recast into the expression
tr
n+1 = n+1 2µ n+1 0 (VII.55)

in terms of the trial value of the yield criterion function

r
tr 0 tr 2
n+1 := n+1 y0 . (VII.56)
3
Owing to the fact that n+1 0 and µ > 0, we have
tr
n+1 n+1 (VII.57)

Hence, we decide the type of the incremental loading based on the trial value
(
tr  0 elastic step
n+1 (VII.58)
> 0 plastic step
tr
where n+1 acts as an indicator function. For an elastic step with n+1 = 0, we
have
0 0 tr
n+1 = n+1 ; (VII.59)
that is, the trial step is the final stress state. No correction is needed.
For a plastic step, however, we have to compute n+1 from an algorithmic consis-
tency condition
tr
n+1 = n+1 2µ n+1 = 0 (VII.60)
yielding the result for the incremental plastic flow n+1

tr
n+1
n+1 = . (VII.61)
2µ

Insertion of this result into the stress expression (VII.46) along with (VII.48)2

90
 1. Ideal Elastoplasticity

leads us to 0 tr
0 0 tr n+1
n+1 = n+1 2µ n+1 0 tr
n+1
" # (VII.62)
tr
n+1 0 tr
= 1 0 tr n+1 ,
n+1

which finally yields

q
2
3
y0
0 0 tr
n+1 = 0 tr n+1 . (VII.63)
n+1

Therefore, the final stresses are obtained by scaling the trial stresses. This step is
called the radial return mapping as depicted in Figure VII.7.

Plastic Corrector: Return Mapping

3 0 trial
n+1

0
n+1
q
2
y
3 0 0 Elastic Predictor
n

1 2

q
0 2
= || || y
3 0
=0

Figure VII.7.: Graphical illustration of the radial return mapping algorithm

The final stress tensor n+1 is located on the yield surface and have the value
r r
0 2 2
n+1 = y0 nn+1 = y0 ntr
n+1 (VII.64)
3 3
and
0
n+1 = en+1 1 + n+1 . (VII.65)

The sensitivity of the algorithmic stresses (VII.65) with respect to the current
total strain "n+1
ep
n+1 = n+1 : "n+1 (VII.66)
is governed by the fourth-order elasto-plastic tangent moduli tensor

ep d n+1
n+1 := , (VII.67)
d"n+1

which is consistent with the employed update scheme. Based on the radial return
mapping algorithm, introduced above, we have
0 tr tr
n+1 = en+1 1 + n+1 2µ n+1 nn+1 . (VII.68)

91
VII. Rate-Independent and Rate-Dependent Elastoplasticity

ep
For an elastic step ( n+1 = 0), coincides with the elasticity tensor
ep e
n+1 ⌘ n+1 =  1 ⌦ 1 + 2µ (VII.69)
1
with = 3
1 ⌦ 1.
For a plastic step ( n+1 > 0), on the other hand, we have

ep e tr
n+1 = n+1 @"n+1 [2µ n+1 nn+1 ]

e tr tr
(VII.70)
= n+1 nn+1 ⌦ @"n+1 (2µ n+1 ) 2µ n+1 @"n+1 nn+1 .
| {z } | {z }
T1 T2

For term T1 , we make use of the persistency condition along with the implicit
function theorem
n+1 = 0 ) d"n+1 n+1 = 0. (VII.71)
Recalling the algorithmic expression for the yield criterion function
r
0 tr 2
n+1 = k n+1 k y0 2µ n+1 ,
3
| {z }
tr
n+1

its total derivative with respect to the total strain tensor d"n+1 n+1 yields
tr
nn+1 : 2µ @"n+1 (2µ n+1 ) =0

leading us to the result

tr
@"n+1 (2µ n+1 ) = 2µnn+1 . (VII.72)
The second term T2 can be derived straightforwardly
" #
0 tr
tr n+1
@"n+1 nn+1 = @"n+1 0 tr k
k n+1
" #
0 tr tr
n+1 @ 0 n+1
= @ 0 tr
0 tr k
:
n+1 k n+1 @"n+1 (VII.73)
1 tr tr
= 0 tr k
[ nn+1 ⌦ nn+1 ] : 2µ
k n+1
2µ tr tr
= 0 tr k
[ nn+1 ⌦ nn+1 ].
k n+1

Inserting the results obtained in (VII.72) for T1 and (VII.73) for T2 into (VII.70),
we obtain
ep e tr tr 2µ tr tr
n+1 = n+1 2µ nn+1 ⌦ nn+1 2µ n+1 0 tr k
[ nn+1 ⌦ nn+1 ],
k n+1

92
 1. Ideal Elastoplasticity

which can be further recast into the following more compact form
ep tr tr
n+1 = 1 ⌦ 1 + 2µ 1 2µ 2 nn+1 ⌦ nn+1 (VII.74)

where the coefficients 1 and 2 are defined as

tr
2µ n+1 n+1
1 = 2 =1 tr =1 0 tr k
k 0 n+1 k k n+1

Box VII.3: Update Algorithm for Ideal von Mises Elastoplasticity

1. Given is the database of history variables {"pn } at time tn and the advance
total strain "n+1 .
2. Elastic Predictor (Trial) Step: Compute predictor values
r
0 tr p 0 tr 2
n+1 = 2µ("0n+1 "0 n ) and tr
n+1 =k n+1 k y0
3
3. Algorithmic check for plastic loading: The step is elastic for
tr
IF ( n+1  0) THEN
"pn+1 = "pn
0 0 tr
n+1 = n+1
0
n+1 = 2µ
GOTO 5
ENDIF
4. Plastic Corrector (Radial Return Mapping): Compute the incremental
plastic parameter
tr
n+1 = n+1 /2µ

and perform radial return mapping

0 tr
0 0 tr n+1
n+1 = n+1 2µ n+1 0 tr k
k n+1
| {z }
tr
nn+1

and get the deviatoric part of consistent tangent moduli

0 tr tr
n+1 = 2µ 1 2µ 2 nn+1 ⌦ nn+1
tr 0 tr
with 1 = 2 =1 n+1 /k n+1 k

5. Add volumetric (spherical) contributions to stresses and moduli

0
n+1 = en+1 1 + n+1

0
n+1 = 1 ⌦ 1 + n+1

93
VII. Rate-Independent and Rate-Dependent Elastoplasticity

2. Rate-Independent Plasticity with Hardening

In this section, we extend the von Mises-type J2 theory of ideal elastoplasticity,
covered in the preceding section, towards the von Mises plasticity incorporating
isotropic and kinematic hardening.

2.1 General Considerations

The hardening phenomena in plasticity describe the evolution of the elastic do-
main (admissible stress space) of the yield stress during the course of plastic
deformation. The physical reasons for hardening in crystalline materials are the
creation and accumulation of line defects (dislocations) and micro-stress fields dur-
ing plastic flow. The accumulated dislocations act as obstacles on the glide planes
against the motion of dislocations. An increasing number of forest dislocations
represents obstacles against the motion of a dislocation in a considered slip plane,
see Figure VII.8.

⌧ Forest
dislocations

Glide distortion

Glide plane

⌧
Figure VII.8.: Graphical illustration of the plastic slip and accumulation of dislo-
cations as obstacles against plastic flow

From the phenomenological point of view, we incorporate the e↵ects of harden-

ing at the macro-level by introducing additional internal variables. The classical
approaches in metal plasticity describe the following e↵ects: ‘

1. Isotropic Hardening: The affine isotropic expansion of the elastic domain

is described by a scalar stress-like variable as depicted in Figure VII.9

q y(↵)
q 2
2
y 3 y1
3 0

y0
y0
1 2 ↵
=0
Figure VII.9.: Graphical representation of isotropic hardening

94
 2. Rate-Independent Plasticity with Hardening

and mathematically described by

r
2
= ˆ( 0 , ) = || 0 || (y0 + )  0 . (VII.75)
3

2. Kinematic Hardening: Rigid movement of the initial elastic domain, de-

scribed by a second-order stress-like tensorial variable , often referred to
as the back stress as illustrated in Figure VII.10

3
0
0 ⇠ :=

q
2
y
3 0
q
2
y
3 0

1 2

Figure VII.10.: Graphical representation of kinematic hardening

and mathematically expressed as

r
2
= ˆ( , ) = ||
0 0
|| y0  0 . (VII.76)
3

3. Combined Isotropic-Kinematic Hardening: The combination of (1) isotopic

and (2) kinematic hardening is often suitable for the description of cyclic
processes in metal plasticity, see Figure VII.11.
q
2
3
3

q
2
y
3 0

1 2

Figure VII.11.: Graphical representation of combined isotropic-kinematic harden-

ing

The combined isotropic-kinematic hardening can be incorporated in the von

95
VII. Rate-Independent and Rate-Dependent Elastoplasticity

Mises yield criterion function as follows

r
2
= ˆ( 0 , , ) = || 0
|| (y0 + )  0 . (VII.77)
3

4. Distortion of yield surface: Owing to the texture developments in metals,

the shape of yield surface may transform into completely anisotropic form as
schematically illustrated in Figure VII.12 in comparison with the combined
isotropic-kinematic hardening.
q
2
3
3

q
2
y
3 0

1 2

Figure VII.12.: Graphical representation of distortional hardening

A possible way to model this phenomenon can be through the introduction

of a fourth-order tensor in the computation of the norm, i.e.
r
p 2
= ˆ( , , , ) = ( 0
0
): :( 0 ) (y0 + ) . (VII.78)
3
The fourth-order tensor evolves as plastic deformation accumulates start-
ing from identity, i.e. (t = 0) = .

2.2 Combined Linear Isotropic-Kinematic Hardening in von Mises

Plasticity
We consider the combination of linear isotropic and kinematic hardening. The
state of the hardening material can be described by the state variables

State(x, t) = {", "p , ↵, ↵} (VII.79)

that denote

"(x, t) : the total strain tensor (external) ,

"p (x, t) : the plastic strain tensor (internal) ,
↵(x, t) : the scalar strain-like variable for isotropic hardening (internal) ,
↵(x, t) : the second-order strain-like tensor for kinematic hardening (internal) .

96
 2. Rate-Independent Plasticity with Hardening

The free energy is assumed to have a decoupled form comprised of the stress-
producing elastic part and the micro-mechanical part; that is,

= ˆ e (" "p ) + ˆ p (↵, ↵) . (VII.80)

| {z } | {z }
macro micro
stresses stresses
The micro stress response is decoupled from the macro stress response. This is a
typical assumption in metal plasticity. Introducing the elastic strain tensor

"e := " "p , (VII.81)

we have the following quadratic expression for the free energy

1 1 1
⇢ = ⇢ ˆ ("e , ↵, ↵) = ee 2 + µ "e0 : "e0 + H↵ : ↵ + h↵2 (VII.82)
|2 {z } |2 {z 2 }
⇢ˆe ⇢ˆp

where ee := tr("e ) and "e0 := dev("e ). Observe that the free energy function of
elastoplasticity with combined linear isotropic-kinematic hardening is character-
ized by the bulk modulus , the shear modulus µ of the macro-response and the
kinematic hardening modulus H and the isotropic hardening modulus h of the
micro-response.
The usual exploitation of the Clausius-Planck Inequality yields the stress tensor

= ⇢@"e ˆ = ee 1 + 2µ"e0 (VII.83)

and the stress-like internal variable tensors conjugate to ↵ and ↵

:= ⇢@↵ ˆ = H↵ and := ⇢@↵ ˆ = h↵ , (VII.84)

respectively. The local dissipation can then be simplified to

2 3 2 3
p
6 7 6 "˙ 7
6 7 6 7
⇢D := : "˙ p : ↵˙ ↵˙ = F • İ = 6 7 • 6 ↵˙ 7 0. (VII.85)
4 5 4 5
↵˙
| {z } | {z }
F İ

The evolution equations for the internal variables {"p , ↵, ↵} are obtained through
a generalization of the principle of maximum dissipation. Considering the elastic
domain
:= {( , , ) 2 R6+6+1 | ˆ( , , )  0} (VII.86)
based on the yield criterion function constructed in the extended stress space
r
2
= ˆ( , , ) = k 0
k (y0 + ) . (VII.87)
3

97
VII. Rate-Independent and Rate-Dependent Elastoplasticity
q
2
0
3
3

⇠
q
2
y
3 0

1 2

Figure VII.13.: Graphical representation of combined isotropic-kinematic harden-

ing

Defining the so-called relative stress tensor

0
⇠ := , (VII.88)

shown also in Figure VII.13, we have

r
2
= ˜(⇠, ) = k⇠k (y0 + ) .
3
Formulating the principle of maximum dissipation
⇤ ⇤ ⇤ ⇤ ⇤ ⇤
: "˙ p : ↵˙ ↵˙ : "˙ p : ↵˙ ↵˙ 8{ , , }2 (VII.89)
| {z } | {z }
⇢D ⇢D ⇤

through the associated Lagrange functional as a constraint optimization problem

L( , , , ) = : "˙ p + : ↵˙ + ↵˙ + ˆ( , , ) ! stationary (VII.90)

leads us to the evolution equations for the internal variables

"˙ p = @ = ⇠/k⇠k

↵˙ = @ = ⇠/k⇠k (VII.91)
p
↵˙ = @ = 2/3

along with the loading-unloading conditions

0,  0, = 0.

In this associated framework of the von Mises Plasticity, we have

↵˙ = "˙ p =) ↵ = "p for ↵(t = 0) = "p (t = 0) = 0 .

98
 2. Rate-Independent Plasticity with Hardening

This is the so called Prager-type hardening that defines the back-stress to be

proportional to the current plastic strain, i.e.

= H↵ = H"p . (VII.92)

In addition, we have
k"˙ p k = k@⇠ k .
By setting k@⇠ k = 1 according to the von Mises criterion, we observe

k"˙ p k = ;

that is, the Lagrange multiplier describes the amount of the plastic flow. We
interpret the evolution of the scalar variable for isotropic hardening as follows
r r r
2 2 p 2p p p
↵˙ = = k"˙ k = "˙ : "˙ .
3 3 3
This is the classical definition of the equivalent plastic strain. Therefore,
Z tr
2 p
↵(t) = "˙ (s) : "˙ p (s) ds, (VII.93)
0 3
yields the so-called accumulated plastic strain, which is also known as the plastic
arc length.

Box VII.4: Combined Linear Isotropic-Kinematic Hardening Model

of von Mises Plasticity

1. Kinematics "e := " "p

1 1 1
2. Free Energy ⇢ = e2 + µ"e0 : "e0 + H↵ : ↵ + h↵2
2 2 2
e0
3. Stresses = e1 + 2µ"
4. Back-stresses = H↵
5. Isotropic Hardening stress = h↵ r
2 0
6. Yield Criterion Function = k⇠k (y0 + )  0 , ⇠ :=
3
7. Flow Rule "˙ p = ⇠/k⇠k
8. Evolution kin. hardening ↵˙ = r ⇠/k⇠k
2
9. Evolution iso. hardening ↵˙ =
3
6. Loading/Unloading (KKT) 0, ˆ( )  0, ˆ( ) = 0

99
VII. Rate-Independent and Rate-Dependent Elastoplasticity

2.3 Stress Update Algorithm for Hardening Plasticity

We extend the radial return mapping scheme of ideal elastoplasticity to incorporate
hardening e↵ects. For this purpose, we start by integrating the evolution equations
using the implicit (backward) Euler scheme
9
>
"n+1 = "pn + n+1 nn+1 >
p >
>
>
>
=
↵n+1 = ↵n + n+1 nn+1 > (VII.94)
>
>
q >
>
↵n+1 = ↵n + n+1 3 > 2 ;

along with the incremental loading/unloading conditions

n+1 0, n+1  0, n+1 n+1 =0 (VII.95)

for the incremental plastic parameter n+1 := n+1 (tn+1 tn ). Moreover, carefully
observe that we have introduced
⇠n+1 0
nn+1 = with ⇠n+1 := n+1 n+1 (VII.96)
k⇠n+1 k

in terms of the relative stress ⇠. The database of history variables, given by

{"pn , ↵n , ↵n }, is assumed to be known. In a deformation-driven process, which is
the case in an implicit FEM, we advance the total strain "n+1 and compute the
stresses
0
0
n+1 = 2µ"en+1 = 2µ ("0 n+1 "pn+1 ) ,
n+1 = H↵n+1 , (VII.97)
n+1 = h↵n+1 .

We introduce the following trial state by assuming the step to be elastic, i.e.
assuming n+1 = 0
0 tr
n+1 = 2µ("0n+1 "pn ) ,
tr
n+1 = H↵n , (VII.98)
tr
n+1 = h↵n .

Insertion of (VII.94) and (VII.98) into (VII.97) leads us to

0 0 tr
n+1 = n+1 2µ n+1 nn+1 ,
tr
n+1 = n+1 +H n+1 nn+1 , (VII.99)
q
tr 2
n+1 = n+1 +h n+1 3
.

Having the definitions (VII.99) at hand, we can now express the current value of

100
 2. Rate-Independent Plasticity with Hardening

the relative stress tensor ⇠ as

0 0 tr tr
⇠n+1 = ( n+1 n+1 ) =( n+1 n+1 ) (2µ + H) n+1 nn+1 .
| {z }
tr
⇠n+1

Analogous to the case of ideal elastoplasticity, we conclude that

tr tr tr
nn+1 = nn+1 = ⇠n+1 /k⇠n+1 k,
(VII.100)
tr
k⇠n+1 k = k⇠n+1 k (2µ + H) n+1 .

We may then express the current yield criterion function in the following algorith-
mic form
r
0 2
n+1 = k n+1 n+1 k (y0 + n+1 )
3 r r !
tr 2 tr 2
= k⇠n+1 k (2µ + H) n+1 y0 + n+1 + h n+1
3 3 (VII.101)
r ✓ ◆
tr 2 tr 2
= k⇠n+1 k y0 + n+1 2µ + H + h n+1
3 3
| {z }
tr
= n+1

in terms of the trial value of the yield criterion function

r
tr tr 2 tr
n+1 := k⇠n+1 k (y0 + n+1 ). (VII.102)
3
tr
We note that n+1 serves as an indicator for the incremental loading in the sense
8
<
tr  0 : elastic step ,
n+1 = (VII.103)
: > 0 : elastic plastic step .

For an elastic step, we have n+1 = 0 and the final state at tn+1 is identical with
the trial state; that is,
0 0 tr tr tr
n+1 = n+1 , n+1 = n+1 , n+1 = n+1 ,...

For an elastic-plastic step, on the other hand, we have to compute the plastic
incremental parameter n+1 from the consistency condition n+1 = 0,

tr 2
n+1 = n+1 (2µ + H + h) n+1 = 0. (VII.104)
3
For the case of linear hardening the consistency equation is linear in n+1 . Hence,
we obtain the following closed-form expression for the incremental plastic flow

101
VII. Rate-Independent and Rate-Dependent Elastoplasticity

parameter
tr
n+1
n+1 = . (VII.105)
2
2µ + H + h
3
Observe that in the case of non-linear hardening, the expression (VII.105) must
be replaced by an iterative Newton scheme for n+1 . The insertion of (VII.105)
into (VII.99) gives the update equations for the stresses
tr
0 tr n+1
0
n+1 = n+1 2µ n tr ,
2 n+1
2µ + H + h
tr
3
tr n+1
n+1 = n+1 + H n tr ,
2 n+1 (VII.106)
2µ + H + h
3r
tr
tr n+1 2
n+1 = n+1 + h ,
2 3
2µ + H + h
3
which can be considered as a generalized radial return mapping for the combined
linear isotropic-kinematic hardening. Based on the known stresses (VII.106), we
update the strain-like internal variables based on the inverse representation of the
stress functions
"pn+1 = "pn + ( 0 tr
n+1
0
n+1 )/2µ ,

↵n+1 = (VII.107)
n+1 /H ,
↵n+1 = n+1 /h .
p
Insertion
p of (VII.106) into (VII.107) indicates that ↵n+1 = "n+1 and ↵n+1 =
↵n + 2/3 n+1 . In an efficient coding, we would not store ↵n+1 separately. A
geometrical interpretation of the generalized radial return mapping algorithm is
depicted in Figure VII.14 where note that the increment of the backstresses occurs
in the trial normal. Hence, the center of the von Mises circle moves incrementally
tr
in the direction of nn+1 .
To complete the algorithmic formulation, we need to derive the the consistent
tangent moduli associated with the deviatoric part of the stresses, i.e.
0
ep0 d n+1
n+1 = . (VII.108)
d"n+1

Starting from the algorithmic stress update equation (VII.106)1 , we obtain

tr0
ep0 d n+1 tr tr
n+1 = nn+1 ⌦ @"n+1 (2µ n+1 ) 2µ n+1 @"n+1 (nn+1 ).
d"n+1 | {z } | {z } (VII.109)
| {z } T1 T2
e0 =2µ

For the term T2 , we have

✓ tr
◆
tr ⇠n+1 2µ tr tr
@"n+1 nn+1 = @"n+1 tr
= tr
[ nn+1 ⌦ nn+1 ] (VII.110)
k⇠n+1 k k⇠n+1 k

102
 2. Rate-Independent Plasticity with Hardening

3
tr0
n+1
0
n+1
q
2
3
( n+1 n) = h 23 n+1
kinematic
movement of
the yield surface 0
n
n+1
tr
n+1 n =H n+1 nn+1

n
n+1

1 2

Figure VII.14.: Graphical representation of combined isotropic-kinematic harden-

ing

in analogy to ideal elastoplasticity. The term T1 is obtained through the consis-

tency condition; that is, taking the total derivative of
r
0 tr tr 2 tr 2
n+1 = k n+1 n+1 k (y0 + n+1 ) (2µ + H + h) n+1 =0
3 3
| {z }
tr
n+1

with respect to the current strain tensor "n+1 , we arrive at

tr 2
2µnn+1 (2µ + H + h)@"n+1 n+1 = 0.
3
Solving the latter for T1 , we obtain
2µ
@"n+1 (2µ n+1 ) = n tr . (VII.111)
(1 + H/2µ + h/3µ) n+1

Insertion of the expressions found for T1 and T2 into (VII.109) gives

ep0 2µ tr tr 2µ tr tr
n+1 = 2µ nn+1 ⌦ nn+1 2µ n+1 tr
[ nn+1 ⌦ nn+1 ],
(1 + H/2µ + h/3µ) k⇠n+1 k

which can be recast into the form

ep0 tr tr
n+1 = 2µ 1 2µ 2 nn+1 ⌦ nn+1 (VII.112)

where

103
VII. Rate-Independent and Rate-Dependent Elastoplasticity

tr
1 n+1
1 := 1 tr k
,
1 + H/2µ + h/3µ k⇠n+1
" # (VII.113)
tr
1 n+1
2 := 1 tr k
.
1 + H/2µ + h/3µ k⇠n+1

3. Rate-Dependent Elastoplasticity (Viscoplasticity)

3.1 Perzyna-type Viscoplasticity for the von Mises Model
We consider a class of viscoplastic models, which take into account an experimen-
tally observed overstress phenomenon which depends on the strain rate ". ˙ Note
that the viscous overstress increases for an increasing rate of strain; i.e. the faster
the specimen is deformed, the sti↵er its stress response becomes, see Figure VII.15.

"˙

"

Figure VII.15.: Graphical representation of rate-dependent elastoplasticity (vis-

coplasticity). The arrow denotes the increasing loading rate

The class of Perzyna-type viscoplasticity can be motivated by an approximate so-

lution of the principle of maximum dissipation. This results in only a minor change
of the structure of the flow rule. Recall the principle of maximum dissipation for
ideal plasticity (VII.10)
⇤ ⇤
: "˙}p
| {z
p
| {z: "˙} 8 2 .
⇢Dloc ⇤
⇢Dloc

implying (i) the convexity of and (ii) the normality rule based on an elastic
domain
={ 2 R6 | ˆ( )  0}
with the von Mises yield criterion function
r
2
= ˆ( ) = k 0 k y0  0 .
3

104
 3. Rate-Dependent Elastoplasticity (Viscoplasticity)

Recall that the exact solution of the minimum principle based on the Lagrangian

L( , ) = : "˙ p + ˆ( ) ! stationary

yields the classical flow rule and the loading-unloading conditions

"˙ p = @ ˆ( ) with 0,  0, = 0. (VII.114)

The Perzyna-type model of viscoplasticity can be motivated by an approximate

solution of the constraint optimization problem of PMD by a penalty method.
Note that the penalty method does not introduce an additional unknown variable,
like , as opposed to the Lagrange method but involves a penalty parameter that
controls the rate-dependency by penalizing the constraint condition. To this end,
we define the penalty function
1 ˆ+
P( ) = : "˙ p + [ ( )]2 ! stationary (VII.115)
2⌘
in terms of the overstress function
8
< ˆ
ˆ+ ( ) = ( ) if > 0,
(VII.116)
: 0 otherwise.

We then obtain the evolution equation

1 ˆ+
"˙ p = ( )@ ˆ( ) . (VII.117)
⌘

Interpreting the term ˆ+ ( )/⌘ as the amount of viscoplastic flow, we get

1 ˆ+
"˙ p = @ ˆ( ) with := ( ). (VII.118)
⌘

Comparison of (VII.114) and (VII.118) indicates that

(i) In rate-independent plasticity, the plastic parameter is governed by the

Karush-Kuhn-Tucker-equations and the stresses are strictly bounded by the
elastic domain, see Figure VII.16 (left).
(ii) In viscoplasticity, however, the plastic parameter is determined by the
additional constitutive equation := ˆ+ ( )/⌘ and stresses are unbounded,
see Figure VII.16 (right).

The penalty parameter ⌘ can be interpreted as an additional material constant

governing the viscosity of the material against the plastic flow.

105
VII. Rate-Independent and Rate-Dependent Elastoplasticity

3 3

"˙ p = @
with := ˆ+ /⌘
"˙ p = @ 0
ˆ=0
0
ˆ=0
ˆ+

1 2 1 2

Ideal Plasticity Ideal Viscoplasticity

Figure VII.16.: Graphical comparison of rate-independent ideal elastoplasticity

(left) and ideal viscoplasticity in the ⇡-plane

For ideal plasticity, we summarize the rate-independent and rate-dependent elasto-

plastic response in Box VII.5.

Box VII.5: Rate-Independent and Rate-Dependent Plasticity

1. Kinematics "e := " "p

2. Free Energy = ˆ ("e )
3. Stresses = ⇢@"e ˆ ("e )
4. Yield Criterion Function = ˆ( )
5. Flow Rule "˙ p = @ ˆ( )
6 Loading/Unloading
6.1. Rate-Indepedendent 0, ˆ( )  0, ˆ( ) = 0
1
6.2. Rate-Dependent := ˆ+ ( )
⌘

The viscoplastic model considered so far constitutes a linear relationship between

the overstress function + and the amount of viscoplastic flow
1 + +
= ( ) () =⌘ , (VII.119)
⌘
which corresponds to the generalization of Newton’s viscosity law = ⌘ "˙ p . In
order to fit real data, we usually consider a Norton-type power law (similar to
creep)
1
= ( + )m (VII.120)
⌘
where m is an additional material parameter. In this non-linear context, due to the
power term, we must formulate the overstress function + in a non-dimensional

106
 3. Rate-Dependent Elastoplasticity (Viscoplasticity)

format. That is, for example, for the von Mises plasticity we recast
r
ˆ( ) = k 0 k 2
y0
3
into r
˜( ) = 3 k 0k
1  0.
2 y0

3.2 Numerical Implementation of Perzyna-Type Viscoplasticity

In the numerical and algorithmic treatment of the Perzyna-type viscoplasticity,
we have only a minor modification to the radial return mapping algorithm of the
rate-independent plasticity. This modification concerns the computation of the
incremental plastic parameter

n+1 := t n+1

and its derivative with respect to the total strain tensor contributing to the elasto-
plastic tangent moduli. The equality due to the consistency condition n+1 =
tr
n+1 2µ n+1 = 0 does not hold in the rate-dependent theory. Here, the in-
cremental plastic parameter n+1 must be identified by integrating its specific
evolution equation
n+1 1 +
n+1 = = ,
t ⌘ n+1
which can be recast into
+ ⌘
n+1 n+1 =0
t
and thus we have
tr ⌘
n+1 2µ n+1 n+1 =0 (VII.121)
t
leading us to
tr
n+1
n+1 = ⌘ . (VII.122)
2µ +
t
The radial return mapping algorithm for viscoplasticity then becomes
0 tr
0 0 tr n+1
n+1 = n+1 2µ n+1 0 tr k
,
k n+1

which yields
2 3
tr
0 6 1 n+1 7 0 tr
n+1 = 41 ⌘ k 0 tr k 5 n+1 . (VII.123)
1+ n+1
2µ t

Compare this result carefully with the stress return mapping (correction) formula
(VII.62) and note that it provides a radial return to an overstress of the yield
criterion which may be conceived as a time-dependent (fictive) isotropic hardening
e↵ect as visually demonstrated in Figure VII.17.

107
VII. Rate-Independent and Rate-Dependent Elastoplasticity

3 3

⌘ 0 tr
n+1 t n+1
=0 n+1
0 tr
n+1 0
n+1
0 0 0
n n+1 n

1 2 1 2

Ideal Plasticity Ideal Viscoplasticity

Figure VII.17.: Graphical comparison of the radial return mapping algorithm for
rate-independent ideal elastoplasticity (left) and ideal viscoplas-
ticity

We also need a slight modification in the consistent tangent moduli. This is

concerned with the computation of the sensitivity of the plastic parameter n+1
with respect to the total strain tensor "n+1 . For this purpose, let us introduce the
residual function
tr ⌘
rn+1 := n+1 2µ n+1 n+1 = 0 .
t
For the vanishing total derivative of the latter with respect to the total strain
tensor "n+1 , we arrive at
✓ ◆
⌘ tr
1+ @ (2µ n+1 ) = 2µ nn+1 . (VII.124)
2µ t "n+1

Inserting this result into (VII.70) yields the solution of the consistent moduli
ep tr tr
= 1 ⌦ 1 + 2µ 1 2µ 2 nn+1 ⌦ nn+1 (VII.125)

with the modified coefficients

2µ n+1
1 := 1 tr ,
k 0 n+1 k
(VII.126)
1 2µ n+1
2 := ⌘ tr .
1+ k 0 n+1 k
2µ t

Comparing the expressions obtained for the Perzyna-type viscoplasticity with that
of the rate-independent ideal elastoplasticity, we observe:
• The modifications needed to obtain the Perzyna-type model of viscoplasticity
are minor.
• We recover the rate-independent algorithmic structure simply by setting
⌘ = 0.

108
 4. Associative and Non-Associative Plasticity

Remark VII.4. Perzyna-type viscoplasticity with mixed linear isotrop-

ic/kinematic hardening (HE) The specific form of the free energy, the yield
criterion function, and the overstress function are given by the following expres-
sions

= ⇢ ˆ e ("e ) + ⇢ ˆ p (↵, ↵) = 1 0 0 1 1
⇢ 2
e2 + µ"e : "e + 2
H↵ : ↵ + 2
h↵2 ,
q
= ˆ( , , ) = k 0
k 2
(y
3 0
+ ),

+ 1 +
=: ⌘ () := ,
⌘
respectively. The free energy is a function of the elastic strains "e and the strain-
like internal variables for isotropic hardening ↵ and kinematic hardening ↵. The
stress-like fields and are the corresponding internal forces, which are defined
by := ⇢@↵ ˆ and := ⇢@↵ ˆ .
The elastic material parameters for this model are the bulk modulus  and the
shear modulus µ. The three plastic material parameters are the isotropic hardening
modulus h, the kinematic hardening modulus H, and the initial yield stress y0 . The
viscosity of the plastic flow is governed by the penalty parameter ⌘.

a) Solve the principle of maximum plastic dissipation by a penalty approach in

order to derive the flow rules for the internal variables.
b) Derive the stress update algorithm using the implicit (backward) Euler inte-
gration scheme.
c) Give a graphical interpretation of one step of the return algorithm in the
deviator (⇡) plane.
d) Derive the algorithmic tangent moduli in the form

n+1 = 1 ⌦ 1 + 2µ 1 2µ 2 nn+1 ⌦ nn+1

and identify the constants 1 and 2.

e) Implement the material model into FEAPpv.

4. Associative and Non-Associative Plasticity

So far, we have considered only the models of associative plasticity, which are
based on the principle of maximum dissipation yielding the flow rules normal to
the yield surface ˆ = 0. In order to describe the experimentally observed elasto-
plastic response of frictional granular materials, one needs to extend this setting
toward the so-called non-associative framework by introducing a third constitutive
function ˆ( ), namely the plastic potential, which governs the flow direction in-
dependent of the yield criterion function. Therefore, non-associative flow response

109
VII. Rate-Independent and Rate-Dependent Elastoplasticity

is described by two functions

"˙ p = @ ˆ( )
(VII.127)
ˆ( )  0, ˆ( ) = 0 or 1
0, := [ ˆ+ ( )]
⌘
with

ˆ( ) : the yield criterion function governing the loading condition ,

ˆ( ) : the plastic potential governing the flow direction .

A classical model of non-associative plasticity is the Drucker-Prager Model in soil

mechanics where the yield threshold depends on the pressure;
r
ˆ( ) = 3 k 0 k Hs  0 (VII.128)
2
where
0 1
:= dev( ) and s := tr( ) .
3
That is, the cylindrical yield surface of the von Mises model becomes a conical
yield surface in the Drucker-Prager model with a growing diameter (hardening)
in the increasing direction of pressure thereby accounting for the pressure-induced
confinement e↵ect on the motion of particles forming a frictional granular material.
However, the flow direction is governed by the von Mises-type cylinder
r
3 0
ˆ( ) = k k c with c = constant . (VII.129)
2
While the flow-direction is independent of the pressure s, the loading/unloading
depends critically on the pressure.
With the plastic potential ˆ at hand, the dissipation reads

⇢Dloc = : "˙ P = : @ ˆ( ) 0, (VII.130)

which provides a constraint for the formulation of ˆ. The basic structure of ideal
non-associative plasticity is summarized in Box VII.6.
The key consequence of the non-associated theory of plasticity is the non-symmetry
of the elastoplastic tangent moduli in both continuous and algorithmic setting. In
the former, we have

˙ = ep
: "˙ = ⇢@"2e "e ˆ : ["˙ @ ˆ( )] . (VII.131)

From the consistency condition

˙ = @ ˆ : ˙ = @ ˆ : ⇢@ 2e e ˆ : "˙ @ ˆ : ⇢@"2e "e ˆ : @ ˆ = 0 , (VII.132)
" "

110
 4. Associative and Non-Associative Plasticity

Box VII.6: Basic Structure of Ideal Non-Associative Plasticity

1. Kinematics "e := " "p

2. Free Energy = ˆ ("e )
3. Stresses = ⇢@"e ˆ ("e )
4. Yield Criterion Function = ˆ( )
5. Plastic Potential Function = ˆ( )
6. Flow Rule "˙ p = @ ˆ( )
7 Loading/Unloading
7.1. Rate-Indepedendent 0, ˆ( )  0, ˆ( ) = 0

1 ˆ+
7.2. Rate-Dependent := ( )
⌘

we obtain the plastic parameter

1 ˆ
= @ : ⇢@"2e "e ˆ : "˙ (VII.133)
D
with
D := @ ˆ : ⇢@"2e "e ˆ : @ ˆ .
Inserting (VII.133) into (VII.131), we arrive at

ep
˙ = : "˙ ,
1 (VII.134)
ep
= ⇢@"2e "e ˆ (⇢@ 2e e ˆ : @ ˆ) ⌦ (@ ˆ : ⇢@"2e "e ˆ )
D ""

where the unsymmetry in the tangent is apparent. The models of non-associative

plasticity thus result in unsymmetric sti↵ness matrices within the framework of
implicit finite element method. In numerics, it is well-known that the solution of
algebraic equation systems involving unsymmetric coefficient matrices are compu-
tationally more expensive; i.e. they simply need a longer computing time.

111
 Middle East Technical University
Department of Civil Engineering
Division of Construction Materials

Assoc.Prof.Dr.-Ing. S. Göktepe
1956 [email protected]

CE 7018 · Computational Inelasticity

–Exercise 1–
December 1, 2020

Linear and Non-linear Elasticity

Q1. Stress Analysis of a Screw-Wrench. In the file, called Iwrench, a complete bending problem
of a screw-wrench is given along with its discretization, loading, boundary conditions and the
material specification (Linear elasticity with the material parameters κ = 175 000, µ = 80 000)
in compatible units. For this linear elastic material calculate the solution for the applied loading
and produce the following plots.

a) Mesh, boundary conditions and applied load in one plot.

b) Contour plots of the components σ11 , σ2 , and σ12 of the stress tensor σ.
c) Deformed and undeformed configurations in one plot.

Q2. Beam-Column Connection.The following beam-column connection is to be analyzed for

three different discretizations of 5, 20, and 125 four-node plane-strain elements. Geometry and
boundary conditions of the current problem and 20-element discretization are depicted below.
The whole rightmost edge is subjected to vertical displacement of u = 0.1. Create three input
files for these three different meshes and use the linear elastic material model already employed
in Question 1 to
3

A 11 12 17 18 19 20
1
9 10 13 14 15 16

7 8
3
5 6
u
3 4

1 2

a) Generate the input files for all discretizations.

b) Plot the vertical displacement of Point A over the total number of active degrees of
freedom (number of equations) of the whole structure.
c) Plot the contours of σ11 and σ22 on the deformed structure for the finest discretization.
d) Plot the field of nodal displacement vectors on the undeformed mesh for the finest 125-
element discretization.
Q3. Non-linear Hyperleasticity.The free energy function of a non-linear elasticity material model
is given by
1
ρΨ(ε) = κ tr2 [ε] + µ̂(ε)|| dev[ε]||2
2
where the non-linear shear modulus µ̂(ε) is defined as
! " #$
|| dev[ε]||
µ̂(ε) := µ0 + (µ∞ − µ0 ) 1 − exp − .
η

Moreover, κ denotes the bulk modulus and µ0 , µ∞ , and η are the material parameters describing
the strain-dependent shear modulus.

a) Derive the stress tensor σ and the elasticity moduli C by taking the appropriate derivatives
of the free energy function.
b) Implement the model into FEAPpv by coding the model into the file “nlinelas.f”.

HE. Linear and Non-linear Bending of a Cantilever Beam.The cantilever beam shown below is
subjected to a point load applied at its lower right corner. The beam is to be discretized into
totally 160 elements through 20 divisions in the x-direction and 8 divisions in the y-direction.
y

100

30
x

a) The structure is to be analyzed up to Pmax = 500 in 100 loading steps with the nonlinear
elastic model described in Q3 and the linear elastic material model. Note that for different
material models you need to modify the makefile. For the non-linear analysis use the
material parameters: κ = 175000, µ0 = 10000, µ∞ = 190000, η = 0.001.
For the linear elastic analysis two limiting cases of shear modulus value are to be con-
sidered; that is, µ = µ0 and µ = µ0 + µ∞ and the bulk modulus is kept unaltered
κ = 175000.
Compare the load vs. displacement plots of the lower right corner obtained from the three
analyses by plotting the three different load-deflection curves in one diagram. Note that
the FEAPpv command for generating the time vs. displacement data of a specified node
is discussed on Page 7 of the quick FEAPpv manual.
b) For the nonlinear analysis, plot the contours of σ11 and σ22 on the deformed beam.

2
Results
Q1. Stress Analysis of a Screw-Wrench.

Time = 1.00E+00
STRESS 1

-7.83E+03
-5.71E+03
-3.58E+03
-1.46E+03
6.65E+02
2.79E+03
4.91E+03
7.04E+03
Current View
Min = -7.83E+03
X = 7.50E-01
Y =-8.00E+00

Max = 7.04E+03
X = 7.50E-01
Y = 8.00E+00

Time = 1.00E+00
STRESS 2

-1.13E+03
-7.88E+02
-4.43E+02
-9.81E+01
2.47E+02
5.91E+02
9.36E+02
1.28E+03
Current View
Min = -1.13E+03
X = 3.00E+00
Y =-8.00E+00

Max = 1.28E+03
X = 3.00E+00
Y = 8.00E+00

Time = 1.00E+00
STRESS 4

-1.48E+03
-9.88E+02
-4.92E+02
4.50E+00
5.01E+02
9.97E+02
1.49E+03
1.99E+03
Current View
Min = -1.48E+03
X = 4.75E+00
Y = 6.00E+00

Max = 1.99E+03
X =-1.50E+00
Y =-8.00E+00

Time = 1.00E+00

Time = 1.00E+00

3
Q2. Beam-Column Connection
FEAP - Beam-Column Connection batch
1,1,1,2,2,4 dt,,1.d-2
prop,,1
PARAmeter ! end
s = 10 ! 1, 2, 5, 10
r = 20 ! 2, 4, 10, 20 batch
n2 = (s+1)*(r+1)+1 opti
e2 = s*r+1 tol,,1.d-22
n3 = (r+1)*(s+1)+(s+1)*(s+1)+1
e3 = s*r + s*s +1 nopr
na = (r+1)*(s+1)+(s+1)*(s+1)-s loop,,100
time
BLOCk1 loop,,50
CART, s, r, 1, 1 ,1, 0 tang,,1
1 0.0 0.0 next
2 1.0 0.0 plot,wipe
3 1.0 2.0 plot,defo
4 0.0 2.0 plot,stre,1
next
BLOCk2
CART s, s, n2, e2, 1, 0 plot,post
1 0.0 2.0 plot,wipe
2 1.0 2.0 plot,defo
3 1.0 3.0 plot,stre,1
4 0.0 3.0 plot,post

BLOCk3 plot,post
CART r, s, n3, e3, 1, 0 plot,wipe
1 1.0 2.0 plot,defo
2 3.0 2.0 plot,stre,2
3 3.0 3.0 plot,post
4 1.0 3.0
plot,post
EBOUndary plot,wipe
1 3.0 1 1 plot,defo
2 0.0 1 1 plot,mesh
plot,disp,all
EDISplacement plot,post
1 3.0 0.0 -0.1
disp,,na
MATErial,1 end
SOLId ! inte
UCON ELAS stop
175000.0 80000.0 q
n
END
TIE

INTEractive
STOP

4
 Time = 1.00E+00

STRESS 1 STRESS 2

-6.46E+03 -4.16E+03
-4.77E+03 -3.31E+03
-3.07E+03 -2.46E+03
-1.38E+03 -1.60E+03
3.16E+02 -7.52E+02
2.01E+03 1.01E+02
3.71E+03 9.54E+02
5.40E+03 1.81E+03
Current View Current View
Min = -6.46E+03 Min = -4.16E+03
X = 3.00E+00 X = 9.89E-01
Y = 2.90E+00 Y = 1.79E+00

Max = 5.40E+03 Max = 1.81E+03

X = 3.00E+00 X =-8.46E-03
Y = 1.90E+00 Y = 2.00E+00

Time = 1.00E+00 Time = 1.00E+00

Variation of uy of Point A with Mesh Refinement

0.0065

0.006
Vertical Disp. of Point A

0.0055

0.005

0.0045

0.004

0.0035
0 200 400 600 800 1000 1200
Number of DOF

5
Q3. Non-linear Hyperleasticity.

c$Id: umati1.f,v 1.1 2004/01/11 19:11:33 rlt Exp $

subroutine umati1(type,vv, d, ud, n1,n3)

c * * F E A P * * A Finite Element Analysis Program

c.... Copyright (c) 1984-2004: Regents of the University of California

c All rights reserved
c-----[--.----+----.----+----.-----------------------------------------]

c Inputs:
c type - Name of material model
c vv(5) - Command line real data
c d(*) - Program material parameter data

c Outputs:
c ud(*) - Material parameter data for model
c n1 - Number of history items/point (time dependent)
c n3 - Number of history items/point (time independent)
c-----[--.----+----.----+----.-----------------------------------------]
implicit none

include ’iofile.h’
logical pcomp, pinput, errck
character type*15
integer n1,n3, ii
real*8 vv(5),d(*),ud(*)

c Set command name

if(pcomp(type,’mat1’,4)) then ! Default form: DO NOT CHANGE
type = ’NLEL’ ! Specify new ’name’
c Input user data and save in ud(*) array
else ! Perform input for user data
n1 = 0
errck = pinput(ud(1),4)
write( *,2000) n1
write(iow,2000) n1
write( *,2001) (ud(ii), ii=1,4)
write(iow,2001) (ud(ii), ii=1,4)
endif
2000 format(
* 10x,’MATERIAL MODEL FOR FEAP ----- (c) S. Goktepe, METU ’/
* 10x,’[ ] [ ] Non-linear Elasticity for CE7018 Ex#1 ’/
* 10x,’[ ] [ ] UMAT1............... Version 04/09/2013’/
* 10x,’[ ] [n1 ] material history variables .........’,i11)
2001 format(
* 10x,’[ 1] [kappa] e: bulk modulus kappa = ’, e12.5/
* 10x,’[ 2] [ mu0] e: shear modulus mu_0 = ’, e12.5/
* 10x,’[ 3] [mu_00] e: shear modulus mu_00 = ’, e12.5/
* 10x,’[ 4] [ eta] e: saturation rate eta = ’, e12.5)
c
return
end
c
c

6
 subroutine umatl1(eps,theta,td,d,ud,hn,h1,nh,ii,istrt, sig,dd,isw)

c-----[--.----+----.----+----.-----------------------------------------]
c Input:
c eps(*) - Current strains at point (small deformation)
c - Deformation gradient at point (finite deformation)
c theta - Trace of strain at point
c - Determinant of deforamtion gradient
c td - Temperature change
c d(*) - Program material parameters (ndd)
c ud(*) - User material parameters (nud)
c hn(nh) - History terms at point: t_n
c h1(nh) - History terms at point: t_n+1
c nh - Number of history terms
c ii - Current point number
c istrt - Start state: 0 = elastic; 1 = last solution
c isw - Solution option from element

c Output:
c sig(*) - Stresses at point.
c N.B. 1-d models use only sig(1)
c dd(6,*) - Current material tangent moduli
c N.B. 1-d models use only dd(1,1) and dd(2,1)
c-----[--.----+----.----+----.-----------------------------------------]
implicit none
include ’tdata.h’
include ’pmod2d.h’
c
integer nh,istrt,isw, ii
real*8 td
real*8 eps(*),theta(*),d(*),ud(*),hn(nh),h1(nh), sig(*),dd(6,*)
c
integer i,j
real*8 xi(6), xii(6,6), xpp(6,6), third, twothird
real*8 xkappa, xmu_0, xmu_00, eta
real*8 e, dev(6), ndev, xmu, xmup, xmupp

c Compute and output stress (sig) and (moduli)

c... define identity tensors

parameter ( third =-0.3333333333333333d0,

& twothird = 0.6666666666666667d0)
data xi/1.d0, 1.d0, 1.d0, 0.d0, 0.d0 ,0.d0/
data xii/1.d0, 0.d0, 0.d0, 0.d0, 0.d0, 0.d0,
& 0.d0, 1.d0, 0.d0, 0.d0, 0.d0, 0.d0,
& 0.d0, 0.d0, 1.d0, 0.d0, 0.d0, 0.d0,
& 0.d0, 0.d0, 0.d0, 5.d-1,0.d0, 0.d0,
& 0.d0, 0.d0, 0.d0, 0.d0, 5.d-1,0.d0,
& 0.d0, 0.d0, 0.d0, 0.d0, 0.d0, 5.d-1/

data xpp/twothird, third, third, 0.d0, 0.d0, 0.d0,

& third, twothird, third, 0.d0, 0.d0, 0.d0,
& third, third, twothird, 0.d0, 0.d0, 0.d0,
& 0.d0, 0.d0, 0.d0, 5.d-1, 0.d0, 0.d0,
& 0.d0, 0.d0, 0.d0, 0.d0, 5.d-1, 0.d0,
& 0.d0, 0.d0, 0.d0, 0.d0, 0.d0, 5.d-1/
C
C... Get the material parameters from ud-field
xkappa = ud(1) ! bulk modulus kappa
xmu_0 = ud(2) ! ground shear modulus mu_0
xmu_00 = ud(3) ! steady-state shear modulus mu_00
eta = ud(4) ! saturation rate eta
c

7
c.. trace of strains
e = eps(1) + eps(2) + eps(3)

c.. Account for the Voigt storage

eps(4) = eps(4)/2.d0
eps(5) = eps(5)/2.d0
eps(6) = eps(6)/2.d0

c.. deviator of strains and its norm

do i = 1,6
dev(i) = eps(i) - e*xi(i)/3.d0
enddo
ndev = dev(1)*dev(1) + dev(2)*dev(2) + dev(3)*dev(3)
& + 2.d0*(dev(4)*dev(4) + dev(5)*dev(5) + dev(6)*dev(6))
if (ndev.le.1.d-13) ndev = 1.d-12
ndev = dsqrt(ndev)

c.. compute the non-linear shear modulus mu(eps) and its derivatives
xmu = xmu_0 + (xmu_00-xmu_0)*( 1.d0 - dexp(-ndev/eta) )
xmup = (xmu_00-xmu_0)*dexp(-ndev/eta) / eta
xmupp = - xmup / eta

c.. compute stresses

do i = 1,6
sig(i) = xkappa*e*xi(i) + (2.d0*xmu + xmup*ndev)*dev(i)
enddo
c
c.. compute moduli
do i = 1,6
do j = 1,6
dd(i,j) = xkappa*xi(i)*xi(j) + (2.d0*xmu + xmup*ndev)*xpp(i,j)
& + (3.d0*xmup/ndev + xmupp)*dev(i)*dev(j)
enddo
enddo
end

8
 Middle East Technical University
Department of Civil Engineering
Division of Construction Materials

Assoc.Prof.Dr.-Ing. S. Göktepe
1956 [email protected]

CE 7018 · Computational Inelasticity

–Exercise 2–
December 15, 2020

Linear Viscoelasticity in Terms of Internal Variables

Consider the generalized Maxwell of linear viscoelasticity that is represented in terms of strain-like
internal variables (Representation A). The free energy function is given by

!
n
ρΨ = ρΨ̂(ε, α1 , . . . , αn ) = 1
2 κe2 + µ0 ε! : ε! + µi (ε! − αi ) : (ε! − αi )
i=1

where e := tr(ε), ε! := dev(ε) and {αi }i=1,...,n denote the strain-like tensorial internal variables.
The material parameters used are the bulk modulus κ, the elastic shear modulus µ0 , the shear moduli
{µi }i=1,...,n , and the viscosities {ηi }i=1,...,n of the Maxwell branches.

a) Derive the continuous formulation of the stress tensor σ and the internal forces β i . Make use
of the volumetric-isochoric decoupling.

b) Write down the linear evolution equation for the internal variables αi .

c) Show the thermodynamic consistency of this model for material parameters ηi ≥ 0 with
i = 1, . . . , n.

d) Derive the algorithmic representation of the current stress tensor σ n+1 and the consistent
tangent moduli Cn+1 by using the backward (implicit) Euler integration scheme.

e) Implement the material routine (linve_A.f) for the case of n = 1.

f) Analyze the given structure to perform a qualitative creep- and relaxation tests. Plot strain
and stress distributions at representative instants.

Relaxation Test
Creep Test
p(t)

h u(t)
h

l l
Discretization and material parameters

FEAP creep test FEAP relaxation test

1,1,1,2,2,4 1,1,1,2,2,4

PARAmeter PARAmeter
r = 20 r = 20
s = 12 s = 12
xl = 50. xl = 50.
xh = 30. xh = 30.
ri = r+1 ri = r+1
si = s+1 si = s+1
no = ri*s/2 no = ri*s/2

BLOCk BLOCk
4,r,s,1,1,1,0 4,r,s,1,1,1,0
1 0.0, 0.0 1 0.0, 0.0
2 xl, 0.0 2 xl, 0.0
3 xl, xh 3 xl, xh
4 0.0, xh 4 0.0, xh

EBOUn EBOUn
1 0.0 1 1 1 0.0 1 1
1 xl 1 0

FORCE DISP
ri, ri,100.0,0.0 ri, ri,xl/100.,0.0
ri*si,0,100.0,0.0 ri*si,0,xl/100.0,0.0

MATErial,1 MATErial,1
SOLId SOLId
PLANE STRAIN PLANE STRAIN
UCON LVEA UCON LVEA
175000., 40000.0, 40000., 1000. 175000., 40000.0, 40000., 1000.

END END

2
Results of the Creep Test: ε11 distributions at t = 0.01, 0.11, 0.12, 0.22 s.
STRESS 1

1.59E-04
2.00E-05
1.16E-04
2.12E-04
3.08E-04
4.04E-04
5.00E-04
3.51E-04
Current View
Min = 1.59E-04
X = 0.00E+00
Y = 1.50E+01

Max = 3.51E-04
X = 5.00E+01
Y = 2.31E-03

Time = 1.00E-02
STRESS 1

1.92E-04
2.00E-05
1.16E-04
2.12E-04
3.08E-04
4.04E-04
5.00E-04
5.19E-04
Current View
Min = 1.92E-04
X = 0.00E+00
Y = 1.50E+01

Max = 5.19E-04
X = 5.00E+01
Y = 4.19E-03

Time = 1.10E-01
STRESS 1

3.37E-05
2.00E-05
1.16E-04
2.12E-04
3.08E-04
4.04E-04
5.00E-04
1.70E-04
Current View
Min = 3.37E-05
X = 0.00E+00
Y = 1.50E+01

Max = 1.70E-04
X = 5.00E+01
Y = 1.90E-03

Time = 1.20E-01
STRESS 1

3.51E-07
2.00E-05
1.16E-04
2.12E-04
3.08E-04
4.04E-04
5.00E-04
2.95E-06
Current View
Min = 3.51E-07
X = 0.00E+00
Y = 1.50E+01

Max = 2.95E-06
X = 5.00E+01
Y = 3.00E+01

Time = 2.20E-01

3
Results of the Relaxation Test: σ11 distributions at t = 0.01, 0.11, 0.12, 0.22 s.
STRESS 1

1.84E+03
2.00E+03
2.10E+03
2.20E+03
2.30E+03
2.40E+03
2.50E+03
2.88E+03
Current View
Min = 1.84E+03
X = 1.01E+01
Y = 6.76E-02

Max = 2.88E+03
X = 0.00E+00
Y = 3.00E+01

Time = 1.00E-02
STRESS 1

1.18E+03
2.00E+03
2.10E+03
2.20E+03
2.30E+03
2.40E+03
2.50E+03
2.17E+03
Current View
Min = 1.18E+03
X = 1.01E+01
Y = 9.07E-02

Max = 2.17E+03
X = 0.00E+00
Y = 0.00E+00

Time = 1.10E-01
STRESS 1

-7.62E+02
-7.00E+02
-6.60E+02
-6.20E+02
-5.80E+02
-5.40E+02
-5.00E+02
-6.21E+02
Current View
Min = -7.62E+02
X = 2.50E+00
Y = 1.29E-02

Max = -6.21E+02
X = 0.00E+00
Y = 1.50E+01

Time = 1.20E-01
STRESS 1

-1.83E+00
-7.00E+02
-6.60E+02
-6.20E+02
-5.80E+02
-5.40E+02
-5.00E+02
-6.22E-01
Current View
Min = -1.83E+00
X = 0.00E+00
Y = 3.00E+01

Max = -6.22E-01
X = 7.50E+00
Y = 4.41E-05

Time = 2.20E-01

4
 Middle East Technical University
Department of Civil Engineering
Division of Construction Materials

Assoc.Prof.Dr.-Ing. S. Göktepe
1956 [email protected]

CE 7018 · Computational Inelasticity

–Exercise 3–
December 15, 2020

Non–linear Viscoelasticity in Terms of Internal Variables

Consider a non-linear viscoelastic material model of the generalized Maxwell type, which is represented
in terms of internal variables (Representation A). The free energy function is given by

!
n
ρΨ = ρΨ̂(ε, α1 , . . . , αn ) = 1
2 κe2 + µ0 ε! : ε! + µi (ε! − αi ) : (ε! − αi )
i=1

with the internal forces and the non-linear viscous evolution equations
1 β
β i = −ρ∂αi Ψ̂ = 2µi [ε! − αi ] , α̇i = ||β i ||δi i
ηi ||β i ||

where e := tr(ε), ε! := dev(ε) and {αi }i=1,...,n are the tensorial strain-like internal variables. The
material parameters used are the bulk modulus κ, the elastic shear modulus µ0 , the shear moduli
{µi }i=1,...,n , the viscosities {ηi }i=1,...,n , and the power terms {δi }i=1,...,n .

a) Derive the continuous formulation of the stresses σ and the internal forces β i .

b) Express the restrictions on the material parameters in order to have a thermodynamically

consistent model.

c) Construct the algorithmic setting for the update of the internal variables αi and their iterative
computation by the local Newton method for the fully implicit backward Euler integration
scheme.

d) Derive the algorithmic representation of the stresses σ n+1 and the consistent tangent moduli
Cn+1 for the chosen integration scheme.

(HE) Construct the algorithmic formulation of the model using the explicit (forward) Euler method
and implement the model into FEAPpv for the case of n = 1.

(HE) Analyze the boundarqqy value problem, already considered in Exercise#2, for the numerical
computations. Perform displacement controlled cyclic tests for different values of the material
parameters δ and η in order to investigate the sensitivity of stress-strain response of the model
to those parameters.
 Middle East Technical University
Department of Civil Engineering
Division of Construction Materials

Assoc.Prof.Dr.-Ing. S. Göktepe
1956 [email protected]

CE 7018 · Computational Inelasticity

–Exercise 4–
January 5, 2021

Von Mises plasticity with non-linear isotropic hardening

Consider an extension of the ideal von Mises plasticity toward a von Mises model with non-linear
isotropic hardening. The two constitutive equations that describe this material behavior are the free
energy function Ψ and the yield criterion function φ:

ρΨ = ρΨ̄(ε, εp , α) = ρΨ̂(εe , α) = ρΨ̂e (εe ) + ρΨ̂p (α) ,

φ = φ̂(σ, β) .

For the model problem at hand, these constitutive functions take the following specific forms
1
ρΨ̂e (εe ) = 2 κ [tr(εe )]2 + µ dev(εe ) : dev(εe )
ρΨ̂p (α) = (y∞ − y0 )(− ω1 + α + ω1 exp(−ωα))
!
φ̂(σ, β) = "σ " " − 23 (y0 + β) with β := ρ∂α Ψ̂p (α)

where (·)" := dev(·) denotes the deviatoric part of a tensor (·). The material behavior is governed by
the five material parameters κ, µ, y0 , y∞ , and ω.

a) Plot the actual yield stress y0 + β over the plastic arclength α. Derive the expressions for the
stresses σ and the isotropic hardening contribution β.

b) Use the principle of maximum plastic dissipation to obtain the evolution equations for the
internal variables εp and α.

c) Summarize the material model in a constitutive box.

d) Derive the algorithmic stresses through the radial return stress update algorithm.

e) Derive the algorithmic (consistent) tangent moduli and implement the model into FEAPpv as
the material routine nlinep.f.

f) Analyze a perforated plate whose right edge is subjected to a horizontal deformation upto
umax = 0.5. Generate the contour plots for σ11 and the plastic arc length α. Use the following
parameters in the numerical simulation:

κ = 175 · 103 , µ = 80 769, y0 = 200, y∞ = 400, and w = 10 .

 subroutine umatl1(eps,theta,td,d,ud,hn,h1,nh,ii,istrt, sig,dd,isw)

c-----[--.----+----.----+----.-----------------------------------------]
c Purpose: User Constitutive Model

c Input:
c eps(*) - Current strains at point (small deformation)
c - Deformation gradient at point (finite deformation)
c theta - Trace of strain at point
c - Determinant of deforamtion gradient
c td - Temperature change
c d(*) - Program material parameters (ndd)
c ud(*) - User material parameters (nud)
c hn(nh) - History terms at point: t_n
c h1(nh) - History terms at point: t_n+1
c nh - Number of history terms
c ii - Current point number
c istrt - Start state: 0 = elastic; 1 = last solution
c isw - Solution option from element

c Output:
c sig(*) - Stresses at point.
c N.B. 1-d models use only sig(1)
c dd(6,*) - Current material tangent moduli
c N.B. 1-d models use only dd(1,1) and dd(2,1)
c-----[--.----+----.----+----.-----------------------------------------]
implicit none
include ’tdata.h’
include ’pmod2d.h’
c
integer nh,istrt,isw, ii
real*8 td
real*8 eps(*),theta(*),d(*),ud(*),hn(nh),h1(nh), sig(*),dd(6,*)
c
integer i,j, nitl
real*8 xi(6), xii(6,6), xpp(6,6), third, twothird
real*8 xkappa, xmu, y0, y00, w
real*8 e, dev(6), phitr
real*8 sigtr(6), betatr, xnorm_sg, xnorm, gamma, xn(6)
real*8 beta, betap, res, a, fac1, fac2, tol, alpha

c Compute and output stress (sig) and (moduli)

c... define identity tensors

parameter ( third =-0.3333333333333333d0,

& twothird = 0.6666666666666667d0)
data xi/1.d0, 1.d0, 1.d0, 0.d0, 0.d0 ,0.d0/
data xii/1.d0, 0.d0, 0.d0, 0.d0, 0.d0, 0.d0,
& 0.d0, 1.d0, 0.d0, 0.d0, 0.d0, 0.d0,
& 0.d0, 0.d0, 1.d0, 0.d0, 0.d0, 0.d0,
& 0.d0, 0.d0, 0.d0, 5.d-1,0.d0, 0.d0,
& 0.d0, 0.d0, 0.d0, 0.d0, 5.d-1,0.d0,
& 0.d0, 0.d0, 0.d0, 0.d0, 0.d0, 5.d-1/

data xpp/twothird, third, third, 0.d0, 0.d0, 0.d0,

& third, twothird, third, 0.d0, 0.d0, 0.d0,
& third, third, twothird, 0.d0, 0.d0, 0.d0,
& 0.d0, 0.d0, 0.d0, 5.d-1, 0.d0, 0.d0,
& 0.d0, 0.d0, 0.d0, 0.d0, 5.d-1, 0.d0,
& 0.d0, 0.d0, 0.d0, 0.d0, 0.d0, 5.d-1/
C
C... Get the material parameters from ud-field
xkappa = ud(1) ! bulk modulus kappa

2
 xmu = ud(2) ! ground shear modulus mu
y0 = ud(3) ! initial yield stress y0
y00 = ud(4) ! saturated yield stress y00
w = ud(5) ! saturation parameter w

c
c.. trace of strains
e = eps(1) + eps(2) + eps(3)
c.. account for Voigt storage
eps(4) = eps(4)/2.d0
eps(5) = eps(5)/2.d0
eps(6) = eps(6)/2.d0

c.. deviator of strains

do i = 1,6
dev(i) = eps(i) - e*xi(i)/3.d0
enddo

c.. compute the trial state

do i = 1,6
sigtr(i) = 2.d0*xmu*(dev(i) - hn(i))
enddo
betatr = (y00-y0)*(1.d0 - dexp(-w*hn(7)))
xnorm = xnorm_sg(6,sigtr,sigtr)
phitr = xnorm - dsqrt(2.d0/3.d0)*(y0+betatr)

c... CHECK plastic flow

if (phitr.le.0.d0) then ! ELASTIC STEP

do i = 1,6
h1(i) = hn(i)
sig(i) = xkappa*e*xi(i) + sigtr(i)
do j = 1,6
dd(i,j) = xkappa*xi(i)*xi(j) + 2.d0*xmu*xpp(i,j)
enddo
enddo
h1(7) = hn(7)

else ! PLASTIC STEP

c
tol = 1.d-12
gamma = 0.1d0
nitl = 0
do i = 1,6
xn(i) = sigtr(i) / xnorm
enddo

101 continue
nitl = nitl + 1
alpha = hn(7) + dsqrt(2.d0/3.d0)*gamma
beta = (y00-y0)*(1.d0 - dexp(-w*alpha))
betap = w *(y00-y0)*dexp(-w*alpha)
res = xnorm - 2.0*xmu*gamma - sqrt(2.d0/3.d0)*(y0+beta)
a = -2.d0*xmu - 2.d0*betap/3.d0
gamma = gamma - res / a
c... check the convergence
c write(*,’(a10,e12.5,a10,i5,a10,e12.5)’)
c & "time=", ttim," iter=",nitl, " res=", res
if ((nitl.lt.50).and.(dabs(res).gt.tol)) go to 101
if (nitl.eq.50) write(*,*)’>No Convergence! |res|=’, dabs(res)
c
fac1 = 1.d0 / a

3
 fac2 = gamma / xnorm

do i = 1,6
h1(i) = hn(i) + gamma*xn(i)
sig(i) = xkappa*e*xi(i) + 2.d0*xmu*(dev(i)-h1(i))
do j = 1,6
dd(i,j) = xkappa*xi(i)*xi(j)
& + 2.d0*xmu*(1.d0 - 2.d0*xmu*fac2)*xpp(i,j)
& + 4.d0*xmu*xmu*(fac1+fac2)*xn(i)*xn(j)
enddo
enddo
h1(7) = hn(7) + dsqrt(2.d0/3.d0) *gamma
endif

c... store the plastic arc length as sig(4) for post-processing

if (isw.eq.8) sig(4) = h1(7)
end

σ11 distribution at u = 0.5.

STRESS 1

-2.46E+02
-5.81E+01
1.30E+02
3.18E+02
5.06E+02
6.94E+02
8.82E+02
1.07E+03
Current View
Min = -2.46E+02
X = 3.26E+00
Y = 1.13E+00

Max = 1.07E+03
X = 0.00E+00
Y = 2.93E+00

Time = 1.00E+00

α distribution at u = 0.5.

STRESS 4

0.00E+00
4.11E-02
8.21E-02
1.23E-01
1.64E-01
2.05E-01
2.46E-01
2.87E-01
Current View
Min = 0.00E+00
X = 4.43E+00
Y = 3.91E-01

Max = 2.87E-01
X = 0.00E+00
Y = 2.93E+00

Time = 1.00E+00

4
 Middle East Technical University
Department of Civil Engineering
Division of Construction Materials

Assoc.Prof.Dr.-Ing. S. Göktepe
1956 [email protected]

CE 7018 · Computational Inelasticity

–Homework 1–
Due: November 3, 2020

Tensor Analysis

1. Write out the following expressions in indicial notation.

a) C = AT B
b) A = (a ⊗ b)(c ⊗ d)
c) α = A : BC
d) β = (w ⊗ z) : (a ⊗ b ⊗ c ⊗ d) : (u ⊗ v)

2. Show that the following equalities hold.1

1
a) tr(dev(A)) = 0 with dev(A) := A − 3 tr(A) 1
b) A : (BC) = (B T A) : C = (AC T ) : B
c) u × (v × w) = (u · w) v − (u · v) w
d) (u × v) × w = (u · w) v − (v · w) u

3. Let A be given as
A = α(1 − e1 ⊗ e1 ) + β(e1 ⊗ e2 + e2 ⊗ e1 )
where α, β are scalars and ei=1,2,3 denote the Cartesian basis vectors.

a) Express the components of A with respect to the Cartesian bases in matrix notation.
b) Determine the principal invariants I1 , I2 , and I3 of the tensor A. 2
c) Obtain the eigenvalues λi=1,2,3 of A through the characteristic equation.
d) Determine the eigenvectors ni=1,2,3 of A corresponding to its principal values λi=1,2,3 .

4. For the symmetric second-order tensor A considered in Question 3:

a) Obtain the rotation tensor Q such that nk = Qek holds and show that QQT = 1.
b) Show that the components of A with respect to its eigenvector bases can be obtained
through QT AQ.
c) Prove that the the principal invariants I1 , I2 , and I3 are insensitive to the chosen basis
system, with respect to which the tensor is represented; that is, Ii (QT AQ) = Ii (A) hold
for i = 1, 2, 3.

1
In Parts c and d, you may want to use the (!−δ) identity: !ijk !pqk = δip δjq − δiq δjp .
2
Recall that the principal invariants of a tensor A are defined as I1 := tr A, I2 := 12 (I12 −tr(A2 )), and I3 := det(A).
5. The spectral representation of a symmetric second-order tensor A

!
3
A= λα N α ⊗ N α
α=1

is given in terms of its eigenvalues λα=1,2,3 and eigenvectors N α=1,2,3 . Show that the inverse
A−1 of the tensor A can be represented in the spectral form as

!
3
A−1 = λ−1
α N α ⊗ N α.
α=1

Hint: You may use the Cayley-Hamilton Theorem to express A−1 in terms of A and its
principal invariants I1 , I2 , and I3 as A−1 = I3−1 [A2 − I1 A + I2 1] and simplify the right-hand
side of this equality by substituting the spectral representation of A. Recall that the principal
invariants can also be expressed as I1 (A) = λ1 + λ2 + λ3 , I2 (A) = λ1 λ2 + λ2 λ3 + λ1 λ3 , and
I3 (A) = λ1 λ2 λ3 .

6. Derive the following derivatives of the principal invariants of a symmetric tensor A with
respect to the tensor A itself.

∂I1 ∂I2 ∂I3

a) ∂A I1 = =1 b) ∂A I2 = = I 1 1 − AT c) ∂A I3 = = I3 A−T
∂A ∂A ∂A

7. A scalar-valued isotropic tensor function Ψ(A) of a symmetric tensor A can be expressed in

terms of its principal invariants I1 , I2 , and I3

Ψ = Ψ(A) = Ψ̂(I1 , I2 , I3 ).

a) Obtain the tensor B := ∂A Ψ̂(I1 , I2 , I3 ) through the chain rule

!
3
B= Ψ̂α ∂A Iα = β1 1 + β2 A + β3 A−1
α=1

where Ψ̂α := ∂Iα Ψ for α = 1, 2, 3. Express the coefficients βα=1,2,3 in terms of the
principal invariants I1 , I2 , and I3 of A and the derivatives Ψ̂α=1,2,3 .
b) Show that the tensors A and B commute; that is, AB = BA.

8. Take the following derivatives

a) P = ∂A dev(A) b) J = ∂A A−1

with dev(A) := A − 13 tr(A) 1. Note that the latter derivative can be derived by taking the
identity AA−1 = 1 as a departure point.

2
 Middle East Technical University
Department of Civil Engineering
Division of Construction Materials

Assoc.Prof.Dr.-Ing. S. Göktepe
1956 [email protected]

CE 7018 · Computational Inelasticity

–Homework 2–
Due: December 16, 2020

Conservation Laws and Hyperelasticity

1. The stress field σ(x, t) at a material point x ∈ B and time t ∈ R+ is specified as

 
γ 0 1
σ(x, t) =  0 −βx2 + x3 g(x2 , x3 ) 
1 g(x2 , x3 ) x2 − αx3

in terms of the scalar constants α, β, γ and the function g(x2 , x3 ) where x1 , x2 , and x3 denote
the spatial coordinates with respect to the Cartesian coordinate system.

a) For the quasi-static case (ρü ≈ 0) and in the absence of external forces (ρb = 0),
determine the form of the function g(x2 , x3 ) so that the balance of linear momentum is
fulfilled identically.
b) With the function g(x2 , x3 ) obtained in Part a, determine the traction vector t on a
surface with normal n = √13 (e1 + e2 + e3 ).

2. Recall that the local form of the balance of energy is given by

ρė = σ : ε̇ − div(q) + ρr .

a) Taking the local form of the first law of thermodynamics as a departure point and using
the Legendre transformation e := Ψ + θη, derive the following transient heat conduction
equation of coupled thermoinelasticity

ρc θ̇ = (F − θ∂θ F ) · İ + θ∂θ σ : ε̇ − div(q) + ρ r

where c := −θ∂θθ2 Ψ denotes the mass-specific heat capacity. Recall that in thermoinelas-

ticity the Helmholtz free energy depends upon the strain tensor ε, the absolute temper-
ature θ, and the set of internal variables I, i.e. Ψ = Ψ̂(ε, θ, I). Accordingly, the ther-
modynamic internal forces F , energy-conjugate to the internal variables I, are defined
through F := −ρ∂I Ψ̂(ε, θ, I) so that the local dissipation (Clausius-Planck Inequality)
boils down to ρDloc = F · İ ≥ 0.
b) For rigid heat conduction problems, where ε = 0 and I = 0, the transient heat conduction
equation, derived in Part a, reduces to ρc θ̇ = −div(q)+ρr and the Helmholtz free energy
function depends only upon the absolute temperature θ, i.e. Ψ = Ψ̂(θ). Should the mass-
specific heat capacity c := −θ∂θθ2 Ψ be constant, show that the corresponding Helmholtz

free energy can be expressed as

% & '(
θ
Ψ = Ψ̂(θ) = c (θ − θ0 ) − θ ln .
θ0
3. Consider a free energy function of non-linear isotropic elasticity

ρΨ̂(J1 , J2 , J3 ) = a0 + a1 J1 + a2 J12 + a3 J13 + a4 J1 J2 + a5 J2 + a6 J3 ,

formulated in terms of the basic invariants of the strain tensor ε

J1 := tr(ε), J2 := 1
2 tr(ε2 ), J3 := 1
3 tr(ε3 ) .

The constants ai=0,1,2,...,7 are the seven material parameters.

a) Show that the following identities between the principal invariants Ii=1,2,3 and the basic
invariants Ji=1,2,3 of the strain tensor ε hold

1
I2 = 1
2 J12 − J2 and I3 = J13 − J1 J2 + J3 .
6

b) Derive the expressions for the stress tensor σ and the tangent moduli C for the given free
energy Ψ = Ψ̃(ε) = Ψ̂(J1 , J2 , J3 ).
c) Determine the material constants ai=0,1,2,...,7 so that the isotropic linear elasticity model
is reproduced by the current model. Recall the expressions of the stresses and the moduli
for the linear elastic response are

σ = λ tr(ε)1 + 2µ ε , C = λ 1 ⊗ 1 + 2µ I .

d) Implement this non-linear elasticity model into FEAPpv as a user-defined material code
nlinelas2.f.
e) Analyze the beam-column connection problem (Q2) discussed in Exercise 1 and plot the
σ11 and σ22 contours on the deformed shape for the 125-element discretization. In the
FE analysis use the material constants

a0 = 0, a1 = 0, a2 = 60833.33, a3 = 5000, a4 = 5000, a5 = 160000, a6 = 5000

given in appropriate units.

2
d) ​The implemented non-linear elasticity model ‘​nlinelas2.f​’ is provided in the assignments
section.

e) ​The result of the analysis of beam-column connection problem is represented by plotting

σ​11​ and σ​22

Figure 1: ​Plot of σ​11​ contour of the deformed shape of beam-column connection where σ​11
represented as Stress 1
Figure 2: ​Plot of σ​22​ contour of the deformed shape of beam-column connection where σ​22
represented as Stress 2
 Middle East Technical University
Department of Civil Engineering
Division of Construction Materials

Assoc.Prof.Dr.-Ing. S. Göktepe
1956 [email protected]

CE 7018 · Computational Inelasticity

–Homework 3–
Due: December 29, 2020

Linear Viscoelasticity: Convolution Integral Representation.

Consider the generalized Maxwell of linear viscoelasticity that is represented in terms of strain-like
internal variables. The free energy function is given by

!
n
ρΨ = ρΨ̂(ε, α1 , .., αn ) = 1
2 κe2 + µ0 ε! : ε! + µi (ε! − αi ) : (ε! − αi )
i=1

with the internal forces and the linear viscous evolution equations
1
β i = −ρ∂αi Ψ̂ = 2µi [ε! − αi ] , α̇i = β
ηi i

where e := tr(ε), ε! := dev(ε) and {αi }i=1..n denote the tensors of internal variables. The material
parameters used are the bulk modulus κ, the elastic shear modulus µ0 , the shear moduli {µi }i=1..n
and the viscosities {ηi }i=1..n .

a) Derive the convolution integral representation of the internal variables αi starting from the
given evolution equation.

b) Give an expression for the stresses in terms of the internal forces β i . Integrate the expression
for the internal forces β i in the time interval [0, tn + ∆t].

c) Perform a numerical integration of the integral expression by applying the midpoint rule.

d) Derive the algorithmic representation of the stresses σ n+1 and the consistent tangent moduli
Cn+1 .

e) Summarize the material model in an algorithmic box.

f) Implement the material routine for the case of n = 1.

g) Carry out the creep and relaxation analyses conducted for the BVP in Exercise#2. Plot the
strain and stress contours at the time instants chosen in Exercise#2.
f) ​The implemented material routine is provided as ‘linve_B.f’ in the assignments
section.

g)​ Results of the Creep Test: ε​11​ distributions at t = 0.01, 0.11, 0.12, 0.22s.
Results of the Relaxation Test: σ​11​ ​distributions at t = 0.01, 0.11, 0.12, 0.22s.
 Middle East Technical University
Department of Civil Engineering
Division of Construction Materials

Assoc.Prof.Dr.-Ing. S. Göktepe
1956 [email protected]

CE 7018 · Computational Inelasticity

–Homework 4–
Due: January 19, 2021

von Mises Plasticity with Linear Isotropic and Kinematic Hardening

Consider an extension of the ideal von Mises plasticity to the von Mises plasticity with combined
linear isotropic and linear kinematic hardening. The two constitutive equations that describe this
material model are the free energy function Ψ and the yield criterion function φ:

ρΨ = ρΨ̄(ε, εp , α, α) = ρΨ̃(εe , α, α) = ρΨ̂e (εe ) + ρΨ̂p (α, α) ,

φ = φ̂(σ, β) .

The specific forms of these functions are given as

1
ρΨ̂e (εe ) = 2 κ [tr(εe )]2 + µ dev(εe ) : dev(εe ) with εe := ε − εp
1
ρΨ̂p (α, α) = Hα : α + 12 hα2
2
!
φ̂(σ, β, β) = "σ ! − β" − 23 (y0 + β) with β := ∂α Ψ̂p (α, α) and β := ∂α Ψ̂p (α, α)

where (·)! denotes the deviator of a quantity. The material behavior is governed by the material
parameters κ, µ, y0 , H, and h.

a) Derive the expressions for the stresses σ and the isotropic and kinematic hardening stresses β
and β, respectively.

b) Through the principle of maximum plastic dissipation, obtain the evolution equations for the
internal variables εp , α, and α.

c) Summarize the material model in a constitutive box.

d) Use the radial return stress update algorithm to compute the algorithmic stresses.

e) Give a graphical interpretation of the radial return mapping algorithm.

f) Derive the consistent tangent moduli and implement the model into FEAPpv as the material
routine linep.f.

g) Analyze the perforated plate problem of Exercise 4 whose right edge is subjected to a horizontal
deformation up to umax = 0.5. Generate the contour plots for σ11 and the plastic arc length
α. Use the following parameters in the numerical simulation: κ = 175 · 103 , µ = 80 769, y0 =
200, H = 1500, and h = 1000 .

h) Assume that the perforated plate considered in Part g is made up of a thermoplastic material
and the deformation is conducted at a very high rate, i.e. under adiabatic conditions, (div(q) ≈
0), and in the absence of an external heat source (ρ r = 0). Furthermore, it is reasonable to
assume that the thermoleastic heating is negligible compared to the elastoplastic heating; that
is, θ∂θ σ : ε̇$(F −θ∂θ F )·İ . Moreover, for moderate temperature changes, the thermodynamic
forces can be considered to be insensitive to thermal changes ∂θ F ≈ 0. Hence, the transient
heat conduction equation derived in Problem 2a of Homework 2 boils down to the following
ordinary differential equation
ρc θ̇ = ρDloc = F · İ .
Store the local relative absolute temperature ϑ := θ−θ0 as a history variable, update its current
value ϑn+1 by integrating the ordinary differential equation above through the backward Euler
scheme, and plot the temperature distribution for the boundary-value problem described in
Part g. Comment on the results obtained. Take ρc = 1 in your numerical analysis.

2
The implemented material routine is provided in assignments section as linep.f.

The contour plots for sigma_11 and plastic arc length α are represented below as Stress 1 and
Stress 4 respectively.
h) ​The temperature distribution of the boundary value problem is represented as Stress 4 in the
following figure.

Since the plastic arc length i.e. Stress 4 in the second figure demonstrates the plastic
deformation amount of the plate, the red coloured area can be considered as the stress
concentration points since the plastic deformation at its highest on this area. Moreover, as
stated in part b, local dissipation is directly related to the plastic deformation and hardening
amounts; therefore, temperature change begins where the plastic deformation starts which can
be seen in the above figure.
Note that the reason most of the plates temperature did not change is that there is not
enough time for conduction to the whole plate. If enough time is provided the temperature
increase will be seen on the whole plate. However, the highest temperature will always be the
stress concentration point as long as deformation continues.
 Middle East Technical University
Department of Civil Engineering
Division of Construction Materials

Assoc.Prof.Dr.-Ing. S. Göktepe
1956 [email protected]

CE 7018 · Computational Inelasticity

–Homework 4–
Due: January 26, 2021

Superimposed Linear Viscoplasticity

Alternative to the approach of Perzyna-type viscoplasticity, linear elastoviscoplasticity can be modeled
by superimposing linear viscoelasticity and ideal elastoplasticity; that is, ρΨ̂ = ρΨ̂ve + ρΨ̂ep . The
viscoelastic part of the free energy and the linear evolution equation for the viscous strain tensor are
given by
1
ρΨ̂ve (ε, εv ) = 1
2 κ1 (tr(ε))2 + µ0 ε! : ε! + µ1 (ε! − εv ) : (ε! − εv ), ε̇v = 2µ1 (ε! − εv ) .
η
For the elastoplastic part, the free energy and the yield criterion functions are defined as
!
2
ρΨ̂ep (ε, εp ) = 12 κ2 (tr(ε))2 + µ2 (ε! − εp ) : (ε! − εp ), φ(σ) = "σ ! " − y0
3
where (·)! := dev(·) denotes the deviator of a quantity.

a) Superimpose both parts of the free energy function and derive the continuous form of the
total stresses σ. Hint: Treat both branches separately and independently, and sum up the
viscoelastic and elastoplastic stress tensors to obtain the total stress response.

b) Compute the algorithmic stresses σ n+1 and the consistent tangent moduli Cn+1 using a fully
implicit integration scheme. Hint: Again treat both branches separately, perform the fully
implicit integration for the internal variables as usual.

c) Summarize the material model in a box and implement the material model into FEAPpv as a
material routine linvp.f. You may take the material routine of Exercise #2 as a departure
point and add the elastoplastic part into it.

d) Analyze the perforated plate problem of Exercise #4 at the three different rates of horizontal
deformation u̇ = 0.5, 0.05, and 0.005 s−1 applied to its right edge upto umax = 0.5. Generate
the contour plots for σ11 for each rate of deformation. Furthermore, for Node 1 of the mesh
provided, draw the horizontal reaction vs. displacement curves on the same diagram for the
three different loading rates. Comment on the results obtained.

Hint: You may want to use the input file Ipplate, provided along with this assignment. The
latter generates the two output files Ppplatea.rea and Ppplatea.dis where the horizontal
reaction at Node 1 and its displacement are reported, respectively, along with the corresponding
time values in the first column.
In the FE analyses, use the following values of the material parameters: κ1 = 75 000, µ0 = 1000,
µ1 = 20 000, η = 200 000, κ2 = 100 000, µ2 = 20 000, and y0 = 200 in compatible units.
 The material model box provided above is implemented into FEAPpv as a material routine linvp.f
and provided in Assignment 5 section. The contour plots for σ​11 ​at each deformation rate are represented
from Figure 1 to Figure 3.

d) To make an accurate comparison between the three different rates of horizontal deformation 0.5, 0.05,
and 0.005 s​-1 are applied to its right edge up to u​max​= 0.5 with the same number of loops. Therefore, the
total deformation time of these deformation rates are determined as 1.0, 10.0, 100.0 s respectively with
corresponding time steps.

Figure 1: ​σ​11 contour

​ plot of the material model at 0.5 deformation rate
Figure 2: ​σ​11 contour
​ plot of the material model at 0.05 deformation rate

Figure 3: ​σ​11 contour

​ plot of the material model at 0.005 deformation rate
 As represented in the third and fifth section of the material model box the viscoelastic stress is
directly related to viscoelastic strain rate which is consistent with Newton’s viscosity law. Moreover, the
plastic strain rate is also directly related to plastic deformation amount. Therefore, it is only natural to
observe lower stress results with decreasing deformation rates as represented in Figure 1 to Figure 3 as
with decreasing maximum stress values.

Figure 4: ​Horizontal reaction vs Displacement curves of Node 1 at different deformation rates

This phenomenon is also represented in Figure 4 where the horizontal reaction at Node 1 increases with
increasing deformation rate.