Elastic Plastic Notes

LA-UR-03-0047 March 20, 2003
Approved for public release;

distribution is unlimited.
Title: Notes on Elastic-Plastic Flow
Author(s): Ralph Menikoff
Submitted to: T14 web site

URL: t14web.lanl.gov/Staff/rsm/preprints.html
#ElasticPlasticNotes
Los Alamos National Laboratory, an affirmative action/equal opportunity employer, is operated by the University of California for the U.S. Department of Energy
under contract W-7405-ENG-36. By acceptance of this article, the published recognizes that the U.S. Government retains a nonexclusive, royalty-free license
to publish or reproduce the published form of this contribution, or to allow others to do so, for U.S. Government purposes. Los Alamos National Laboratory
requests that the publisher identify this article as work performed under the auspices of the U.S. Department of Energy. Los Alamos National Laboratory
strongly supports academic freedom and a researchers right to publish; as an institution, however, the Laboratory does not endorse the viewpoint of a
publication or guarantee its technical correctness.
LA-UR-03-0047 March 20, 2003
Notes on Elastic-Plastic Flow
1 Kinematic Variables 1
1.1 Multiplicative Decomposition . . . . . . . . . . . . . . . . . . 2
1.2 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Transpose, Inverse and Transpose-Inverse . . . . . . . . . . . . 5
1.4 Strain Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.5 Volumetric and Deviatoric Components . . . . . . . . . . . . . 9
2 Hyper-Elasticity 11
2.1 Isotropic Material . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2 Illustrative Elastic Models . . . . . . . . . . . . . . . . . . . . 14
2.2.1 Linear Elasticity . . . . . . . . . . . . . . . . . . . . . 14
2.2.2 Non-linear Isotropic Model . . . . . . . . . . . . . . . . 15
2.3 Conservation Equations . . . . . . . . . . . . . . . . . . . . . 17
2.3.1 Eulerian Form . . . . . . . . . . . . . . . . . . . . . . . 17
2.3.2 Lagrangian Form . . . . . . . . . . . . . . . . . . . . . 18
2.4 Kinematic Equations . . . . . . . . . . . . . . . . . . . . . . . 20
2.5 Dissipation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.6 Wave Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.7 Shock Jump Conditions . . . . . . . . . . . . . . . . . . . . . 31
3 Plastic Strain Rate 33

3.1 Yield Condition . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.2 Associated Flow Rule . . . . . . . . . . . . . . . . . . . . . . . 37
3.2.1 Volume Preserving Plastic Flow . . . . . . . . . . . . . 40
4 Illustrative Constitutive Model 41

4.1 Specific Energy . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.2 Von Mises Yield Condition . . . . . . . . . . . . . . . . . . . . 43
4.3 Hohenemser-Prager rate . . . . . . . . . . . . . . . . . . . . . 45
4.4 Visco-Plastic Limit . . . . . . . . . . . . . . . . . . . . . . . . 46
5 Uniaxial Strain 51
5.1 Stress . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
5.2 Yield Condition . . . . . . . . . . . . . . . . . . . . . . . . . . 54
5.3 Plastic Strain Rate . . . . . . . . . . . . . . . . . . . . . . . . 54
Ralph Menikoff i comments to [email protected]

LA-UR-03-0047 March 20, 2003
5.4 One Dimensional Flow Equations . . . . . . . . . . . . . . . . 55
6 Wave Structure 58
6.1 Weak Purely Elastic Wave . . . . . . . . . . . . . . . . . . . . 59
6.2 Split Elastic-Plastic Wave . . . . . . . . . . . . . . . . . . . . 60
6.3 Strong Plastic Wave . . . . . . . . . . . . . . . . . . . . . . . 61
6.4 Visco-Elastic Wave . . . . . . . . . . . . . . . . . . . . . . . . 62
6.5 Partly and Fully Dispersed Waves . . . . . . . . . . . . . . . . 63
6.6 Entropy Anomalies . . . . . . . . . . . . . . . . . . . . . . . . 64
Appendix A: Useful Lemmas 68
Appendix B: Compliance and Stiffness Tensors 68

B.1 Isotropic Material . . . . . . . . . . . . . . . . . . . . . . . . . 72
B.2 Isotropic Sound . . . . . . . . . . . . . . . . . . . . . . . . . . 74
B.3 Isotropic Averages . . . . . . . . . . . . . . . . . . . . . . . . 74
Appendix C: Convexity of Specific Energy 76
Appendix D: Small Viscosity Limit 77
Appendix E: Lie Derivative 78
Acknowledgement 79
References 79
Index 83
Ralph Menikoff ii comments to [email protected]

LA-UR-03-0047 March 20, 2003
Notes on Elastic-Plastic Flow

Ralph Menikoff
A major complication of elastic-plastic flow, as compared to fluid flow,

is that it is formulated in terms of tensors rather than scalars. Here
the tensor notation is defined with attention to the spaces on which
the various tensors act and the metrics for the different frames. Ex-
amples are shown illustrating the small strain limit and the case of
uniaxial strain. The multiplicative decomposition formulation is used
to split the deformation gradient into elastic and plastic components.
Hyperbolic aspects of the elastic-plastic PDEs are emphasized.
1 Kinematic Variables
Elastic flow is formulated in terms of a time-dependent mapping from
the body reference frame to the spatial laboratory frame
(, t) : B S , ~ t) ,
xi = i (X,
where X~ B and ~x S. The particle velocity field in the reference frame

is a map from the body frame to the tangent space of the spatial frame
U ~ t) = T ~ S .
~ (X, (1.1)
(X,t)
t
Alternatively, the particle velocity can be expressed as a map from the spatial
frame to the tangent space of the spatial frame

~ 1 T~x S .
~u(~x, t) = U
It is important to note that the Lagrangian velocity U~ is a two-point rank-one

tensor field, i.e., the domain is B but the range is in T S.
Elastic-Plastic Notes 1 comments to [email protected]

1 Kinematic Variables Elastic-Plastic Flow
The deformation gradient is the derivative of the mapping, and is a

transformation from the tangent space of B to the tangent space of S
i
F = D , ~ t) =
Fi (X, : TX~ B T(
~ X,t)
~ S. (1.2)
X
The deformation gradient is a two-point rank-two tensor field. This is in
contrast to the rank-two stress and strain tensor fields defined later. For
ordinary tensor fields, the domain and range are the tangent or cotangent
spaces at the same point. Two-point tensors involve two different points, for
example F goes between X ~ X,
~ B and ~x = ( ~ t) S.
1.1 Multiplicative Decomposition
To motivate the decomposition of a flow into elastic and plastic compo-

nents, we consider expressing the mapping as the composition of maps,
= e p , from the reference frame to a plastic intermediate frame to the
spatial frame (see [15], [23, chpt. 9] and references therein)
p e
B ........................................................ PI ........................................................
Si
. ...
.......... ...
.... ...
...
...
... ..........
.
.

B ................................................................................................................................... Si
The chain rule for differentiation then leads to a multiplicative decompo-

sition for the deformation gradient, in terms of elastic and plastic deforma-
tions, Fe and Fp , respectively,
F = Fe Fp , Fi = (Fe )i A (Fp )A .
The deformation gradients Fe and Fp are local quantities that we consider

more fundamental than the mappings e and p .
Physically, plasticity is due to processes that occur on a microscopic scale,
such as the formation and movement of dislocations in a crystal. Disloca-
tions prevents the microscopic mappings from being continuous functions.
Consequently, the deformation gradients can be defined only as coarse-grain
averages of the local lattice distortions. Mathematically, as a result of the
averaging process the compatibility conditions needed to recover a mapping
1 Kinematic Variables 2 comments to [email protected]

Elastic-Plastic Flow 1.1 Multiplicative Decomposition
from the deformation gradient are not satisfied. Nevertheless, we assume the
multiplicative decomposition of the deformation gradients and consider the
mappings only as a convenient heuristic.
There is still a question with non-uniqueness of the decomposition; i.e.,
F = (Fe Q1 )(QFp ) for any invertible Q : T P T P. The non-uniqueness
is similar to the question of frame indifference discussed in a later section.
The difficulty is resolved in part by formulating plasticity in terms of total
strain and plastic strain rather than the deformation gradient alone. In
addition, the choice of the plastic strain rate is needed to define the plasticity
model.
1.2 Notation
The mathematical framework for elastic-plastic flow, or even elastic flow,

is differential geometry [11]. To indicate the space for the domain and range
of various transformations, superscripts (corresponding to contravariant
vectors in the tangent space T, called simply vectors) and subscripts (cor-
responding to covariant vectors in the cotangent space T , called simply
covectors) in lower case Greek letters are associated with B, upper case
roman letters with P and lower case roman letters with S. Vectors with upper
case letters are used to indicate points in the body frame (B or T B or T B)
while lower case letters indicate points in the spatial frame (S or T S or T S),
e.g., X~ B and ~x S.
To be more specific, we denote the basis elements of TX~ B by
~ =
E .
X
Then a vector in TX~ B may be written as
~ = U E
U ~ ,
where the summation over repeated indices, upper and lower, is assumed.
~ that are
Thus, it is the components of a vector relative to the basis E
denoted with superscripts.
Similarly, we denote the dual basis elements in T X~ B by
~ = dX ,
E
comments to [email protected] 3 1 Kinematic Variables

1.2 Notation Elastic-Plastic Flow
with dual pairing

~ , E
(E ~ ) = ,
where the Kronecker- is 1 if the indices are the same and 0 otherwise. Al-
~ E
ternatively, the duality can be expressed as E ~ is the identity operator
in TX~ B. An element of T X~ B may be written
V~ = V E
~ .
The components of a covector relative to the basis E ~ are denoted with

subscripts. The dual pairing of a vector in T X~ B with a vector in TX~ B can

be expressed in terms of their components as (V~ , U

~ ) = V U .
~ 7 X
Under a change of basis X ~ X),
~ 0 = ( ~ vectors and covectors transform
in a complementary manner,
U 7 (U 0 ) = (F) U

V 7 (V 0 ) = (FT ) V ,
where F = D,~ such that the dual pairing is invariant, i.e., V U = (V 0 ) (U 0 ) .

The purpose of subscripts and the superscripts is to provide a calculus for
keeping track of the way the representation of physical quantities transform
under change of basis. The number of upper and lower indices on the left
and right hand sides of a tensorial equation must match. This provides a
simple consistency check akin to dimensional analysis.
Inner products within TX~ B and T~x S are defined by metrics
~ , V~ i ~ = (GW
hW ~ , V~ ) ~ = G (X)W
~
V
X X
hw,
~ ~v i~x = (g w,
~ ~v )~x = gab (~x)wa v b ,
~ and g(~x) denote the metric maps between T and T in the

where G(X)
body frame and the space frame, respectively,
~ : T ~ B T ~ B
G(X) and ~ : T ~ B T ~ B ,
G1 (X)
X X X X
g(~x) : T~x S T ~x S and g 1 (~x) : T ~x S T~x S .
The metrics are independent of time but may vary from point to point in B or
S. The matrices G and gij are symmetric, consequently, hW~ , V~ i = hV~ , W
~i
and hw,
~ ~v i = h~v , wi.
~

Elastic-Plastic Flow 1.2 Notation
The following identities hold

G1 G = IT B = identity in T B, (G1 ) G =
G G1 = IT B = identity in T B, G (G1 ) =
g 1 g = IT S = identity in T S, (g 1 )ik gkj = ij
g g 1 = IT S = identity in T S, gik (g 1 )kj = i j
By convention, we denote (G1 ) by G and(g 1 )ij by g ij . Furthermore,

raising or lowering the indices of a variable indicates the implicit use of the
metric tensor to transform between T and T , e.g., ui v i = ui gij v j = ui g ij vj .
Finally, upper case bold letters denote tensors in B and lower case bold letters
denote tensors in S. A tilde over a letter is used to denote a tensor in the
plastic frame.
1.3 Transpose, Inverse and Transpose-Inverse
In addition to the derivative of the mapping, D, its transpose, inverse

and transpose-inverse are important for transforming tensors between the
domain and range of . In transforming tensors, it is helpful to keep track
of the domain and range of these additional operators. In terms of indices,
they can be written as follows:
i
F derivative, ~ t) =
Fi (X, X
:T ~
X B T ~x S
i
FT transpose, (FT ) (~x, t) = Fi 1 : T ~x S T X~ B
F1 inverse, (F1 ) i (~x, t) = (F1 ) i 1 : T ~x S T ~
X B

F T
transpose-inverse, (FT )i (X, ~ t) = (F1 ) i :T
~
X BT
xS
~
where ~x = (X,~ t) or X
~ = 1 (~x, t). Moreover, these operators satisfy the
following identities
F1 F = IT ~
X
B and F F1 = IT xS
~
FT FT = IT X~ B and FT FT = IT ~x S
Remark 1.1 In the language of differential geometry, the induced mapping

of tensors on B to tensors on S is referred to as the push-forward of the map

1.3 Transpose, Inverse and Transpose-Inverse Elastic-Plastic Flow
: B S and denoted . For a vector and covector

( U~ )(~x) = (FU~ ) 1 ,
( V~ )(~x) = (FT V~ ) 1 .
The induced mapping in the other direction, from tensors on S to tensors on B
is referred to as the pull-back and denoted . For a vector and covector
(~u)(X)~ = (F1~u) ,
~ = (FT ~v ) .
(~v )(X)
The inverse of the derivative map is simply the derivative of the inverse
map, F1 = D(1 ). The transpose can be defined in a co-coordinate free
manner by the relation
~ )~x = (FT ~v , W
(~v , FW ~ ) ~ = v Fi W ,
X i
where ~v T ~x S and W
~ T ~ B. Instead of the transpose, ref. [11] uses the
X

adjoint F , defined by the relation
~ , ~v i~x = hW
hFW ~ , F~v i ~ ,
X
and represented in coordinates by

(F ) a (~x, t) = (G 1 )(Fb 1 ) gba : T~x S TX~ B .
The transpose and adjoint operators are related by
FT g = G 1 F .
We note that the domain and range of F are the same as those for F1 .
1.4 Strain Tensors
In order to define a strain tensor, it is convenient to introduce several

additional tensors. The right and left Cauchy-Green tensors are
C = (FT g) F ~ t) = (gij ) Fi Fj
C (X, : TX~ B T X~ B
b = (FG1 ) 1 FT bij (~x, t) = (G Fi Fj ) 1 : T ~x S T~x S

Elastic-Plastic Flow 1.4 Strain Tensors
The Lagrangian strain tensor is defined by
E = 12 (C G) ~ t) = 1 (C G ) : T ~ B T ~ B .
E (X, (1.3)
2 X X
Transforming the strain tensor from the Lagrangian frame to the spatial
frame, E 7 FT EF1 , results in the Eulerian or Almansi strain tensor
e = 21 (g b1 ) eij (~x, t) = 12 [gij (b1 )ij ] : T~x S T ~x S . (1.4)
The inverse of the left Cauchy-Green tensor b1 is known as the Finger

tensor.
The tensors E and e represent the total strain. In analogy with Eq. (1.3),
the Lagrangian plastic strain tensor is defined by
Ep = 21 (Cp G) , ~ t) : T ~ B T ~ B ,
(Ep ) (X, (1.5)
X X
where heuristically
Cp = (Fp T G)
e F ,
p p
~ t) : T ~ B T ~ B
(Cp ) (X, X X
is the right Cauchy-Green plastic tensor, and G

e is the metric in the plastic
frame.
Similarly, in analogy with Eq. (1.4), the Eulerian elastic strain tensor
is defined by
ee = 21 (g be 1 ) , (ee )ij (~x, t) : T~x S T ~x S , (1.6)
where heuristically
1
be = (Fe G
e ) 1 F T ,
e e (be )ij (~x, t) : T ~x S T~x S
= FCp 1 FT
is the left Cauchy-Green elastic tensor.

The plastic strain tensor can be transformed to the spatial frame
ep = FT Ep F1 , ep (~x, t) : T~x S T ~x S (1.7)

= 21 (be 1 b1 )
= 12 (g b1 ) 12 (g be 1 )

1.4 Strain Tensors Elastic-Plastic Flow
Consequently, the total strain can be expressed as the super-position or

sum of elastic and plastic strains
e = ee + e p .
Similarly, the elastic strain tensor can be transformed to the body frame
Ee = FT ee F , Ee (X, ~ t) : T ~ B T ~ B (1.8)
X X
= 21 (C Cp )
= 12 (C G) 12 (Cp G)
Again the super-position of strains is satisfied
E = Ee + Ep .
Many theories of plasticity start out with the super-position assumption in-
dependent of the multiplicative decomposition. All theories are constructed
to have the same small strain limit. However, theories can differ greatly in
the large strain or non-linear regime.
Remarks:
1.2 One could also define elastic and plastic strains in the intermediate
plastic frame as
eee = 12 [(Fe T g) e Fe G]
e , ~ t) : T ~ P T ~ P
eee (X, X X
Ep = 2 [G (Fp G) p Fp ] , Ep (X, t) : TX~ P T X~ P

e 1 e T 1 1 e ~
Provided that the total strain is transformed to the plastic frame

e = F T EF 1
E p p
the super-position assumption Ee =e ee + E

e again holds. Super-position only
p
makes sense when the total, elastic and plastic strains are transformed to the
same frame.
1.3 The strain tensor defines a bilinear map from T B T B R by
~ 1, U
E(U ~ 2 ) = (EU
~ 1, U
~ 2) ,
where U ~ 2 T B. The bilinear form is symmetric E(U

~ 1, U ~ 1, U
~ 2 ) = E(U
~ 2, U
~ 1 ).
This is the coordinate free expression of the symmetry E = E .

Elastic-Plastic Flow 1.5 Volumetric and Deviatoric Components
1.5 Volumetric and Deviatoric Components
The ratio J = VV0 of the specific volume in the spatial frame to the specific
volume in the reference frame is determined from the deformation gradient
by the relation

~ t) det F F = det G1 C
J2 (X,

1
= det [G + 2E]G

= det IT X~ B + 2EG1
For small strain, kEkG1 1, (see Remark 1.5 for definition of norm
and Remark 1.4 for definition of trace) we can write

J2 = 1 + 2 Tr E G1 + O 2 . (1.9)
Consequently, the change in the specific volume can be expressed as

V V0
= Tr E G1 + O 2 .
V0
We note that the relative change in specific volume or volumetric strain
is positive in expansion and negative in compression. In addition, similar
arguments can be used to yield
h i
J2 = det (IT X~ S 2eg 1 )1 ,
and for small strain to leading order (V V0 )/V0 = Tr (e g 1 ).

From the identity det(AB) = det(A) det(B) it follows that the specific
volume ratio can be expressed as
h i1/2
J = det(g )/ det(G) det(F) .
The determinant factors enter because (det G)1/2 d3 X is the volume element
in B, and (det g)1/2 d3 x is the volume element in S. When the metrics are
normalized, det(g) = det(G) = 1, the specific volume ratio can be expressed
simply as J = det(F). However, the metric still is needed in Eq. (1.9) to
relate J directly to the strain tensor.

1.5 Volumetric and Deviatoric Components Elastic-Plastic Flow
The strain tensor can be split into the sum of a volumetric component
and a deviatoric component, (see Remark 1.6 for definition of deviator)

E = 13 Tr E G1 G + devG (E) .
Transforming to the Eulerian strain, leads to a decomposition identical in

form
e = FT EF1 = 13 Tr (e b) b1 + devb1 (e)
when the transformation of the metric
b1 = FT GF1 : T S T S
is taken into account. Thus, b1 can be interpreted as the reference frame

metric transformed to the spatial frame.
Alternatively, the decomposition can be based on the metric g, i.e.,

e = 13 Tr e g 1 g + devg (e) .
Transforming the Eulerian strain to the Lagrangian frame gives the decom-
position
E = 13 Tr E C1 C + devC (E) .
Thus, C = FT gF can be interpreted as the spatial frame metric transformed

to the reference frame. Since the decomposition of a tensor is not unique, it
is important to specify the metric in the definition of the deviator.
Remarks:
1.4 The trace is defined by

Tr G1 E = Tr EG1 Eij Gji .
The contraction of upper and lower indices is invariant under any transfor-
mation. Hence the trace of an operator on T B or T B, such as the tensor
G1 E : T B T B or the tensor EG1 : T B T B, is invariant under a
change of basis.
1.5 The norm with respect to a metric is defined as

0 0
kEk2G1 G G E E0 0 .

Elastic-Plastic Flow 1.5 Volumetric and Deviatoric Components
The metric is needed in order for the norm to be independent of the choice
of basis. The norm also can be express in terms of the trace,

kEk2G1 = Tr E GT 1
EG 1
.
1.6 The deviator is defined by

devQ (E) E 13 Tr EQ1 Q .
Since physical space is three dimensional, Tr (IT B ) = Tr (IT B ) = 3. Conse-

quently, the deviator is a projection operator, i.e.,
devQ (devQ (E)) = devQ (E) .
Another important formula is for the transform of a deviator

FT devQ (E) F1 = devFT QF1 FT E F1 .
2 Hyper-Elasticity
A material model is called elastic if the stress is a function of the strain.
Some engineering models are defined in terms of the stress rate as a function
of stress, strain and strain rate. These models have the property that the
stress depends on the strain path. Material models for which the stress can
not be expressed in terms of only the current strain state are called hypo-
elastic. In order to have a thermodynamic description an energy function
is needed such that the stress is the derivative of the energy with respect to
the strain. Materials models with such an energy function are called hyper-
elastic.
Frame indifference requires that the energy function is invariant for all
isometries on T S, i.e., under the transformation F 7 Q F for all Q : T S
T S such that
QT g Q = g .
As a consequence the energy can depend on the material deformation only
through the strain tensor E and not on the deformation gradient by itself.
comments to [email protected] 11 2 Hyper-Elasticity

2 Hyper-Elasticity Elastic-Plastic Flow
A similar requirement, that the energy is invariant for all isometries on T P,

i.e., under the transformation Fp 7 Q Fp for all Q : T P T P such that
QT G e Q = G,e leads to the conclusion that the energy can depend on the
plastic deformation only through the plastic strain tensor Ep . Consequently,
for an elastic-plastic material we are led to assume that there is a function
for the specific energy depending on total strain, plastic strain and entropy
E = E(E, Ep , ) ,
where is the specific entropy. Furthermore, we assume the thermody-
namic relation
dE = V0 S : dE V0 Sp : dEp + T d , (2.1)
where T is the temperature, and S and Sp are the elastic and plastic stress
tensors in the Lagrangian frame. The : denotes contraction over a pair of
tensor indices, e.g., S : dE = Tr(ST dE) = S dE .
The second Piola-Kirchhoff stress tensor and the corresponding plas-
tic stress tensor are given by
E E
S = 0 , ~ t) =
S (X, 0 : T X~ B TX~ B
E E
E ~ t) = 0 E : T ~ B T ~ B
Sp = 0 , (Sp ) (X, X X
Ep (Ep )
The Eulerian or Cauchy stress tensor is given by
= J1 FSFT , ij (~x, t) = (J1 Fi Fj S ) 1 : T ~x S T~x S
or alternatively as [2]
E
= 2 .
g
Since the strain E is a symmetric matrix, it follows that both the second
Piola-Kirchhoff stress S and the Cauchy stress ij are symmetric matrices.
Remark 2.1 The symmetry of the stress tensor is required to ensure local
angular momentum conservation. This condition assumes that individual
atoms or molecules within a material have no angular momentum.
2 Hyper-Elasticity 12 comments to [email protected]

Elastic-Plastic Flow 2 Hyper-Elasticity
In analogy to the strain decomposition, the stress can be split into a

hydrostatic and deviatoric part
= P g 1 + devg1 () ,
where P = 13 Tr ( g) is the pressure. When strain is positive in expansion,

the stress is positive in tension, hence, the negative sign in the definition of
the pressure. In the Lagrangian frame, the pressure can be expressed as
P = 13 J1 Tr (S C). In addition,
J F1 devg1 () FT = devC1 (S) . (2.2)
This formula enables us to transform the stress deviator between the Eulerian
and Lagrangian frames.
2.1 Isotropic Material
A material is isotropic if its energy function is invariant for all isometries

on T B, i.e., under the transformation F 7 F Q for all Q : T B T B such
that
QT G Q = G .
The energy can then be expressed as function of the invariants of
G1 C : T B T B , (F F) = G C .
The invariants of a 3 3 matrix are the coefficients of its characteristic

polynomial
P () = 3 I1 2 + I2 I3
whose roots are the eigenvalues of the matrix. For G1 C, the invariants can
be expressed as
I1 = Tr(G1 C)
h i
1
I2 = 1
2
I12 Tr (G C) 2
I3 det(G1 C) = J2
Remark 2.2 Under a transformation the tensor G1 C 7 Q1 (G1 C)Q.

By Lemma A.5 the eigenvalues of the matrix G1 C and hence the invariants

2.1 Isotropic Material Elastic-Plastic Flow
of G1 C are unchanged by the transformation. From the definition of C it

follows that the eigenvalues of G1 C are real and positive. Consequently,
there are constraints on the invariants of G1 C.
Applying the chain rule
E E I1 E I2 E I3
= + +
C I1 C I2 C I3 C
and using Lemma A.1, the stress can be expressed as [23, p. 261, Eq. (7.1.108)]
! ! !
E E E E
1
VS
2 0
= + I1 G1 G1 C G1 + I3 C1
I1 I2 I2 I3
! ! !
1 E 1 E E E
2
V = I3 g + + I1 b bgb
I3 I1 I2 I2
2.2 Illustrative Elastic Models
Two example models for an isotropic elastic material are described. The
first model is known as linear elasticity. The second model is for a non-linear
isotropic material. For each model, the specific energy and the corresponding
stress tensors are presented.
2.2.1 Linear Elasticity
Linear elasticity represents the leading order expansion of the energy

function about the ambient state of an isotropic material,
h i2 h i
E = 12 V0 0 Tr G1 E + V0 G0 Tr (G1 E)2 ,
where 0 and G0 are the Lame coefficients. The second Piola-Kirchhoff

stress is then given by
E
S = 0
E

= K0 Tr G1 E G1 + 2 G0 devG1 G1 E G1 ,

Elastic-Plastic Flow 2.2 Illustrative Elastic Models
and the Cauchy stress is given by
= J1 F S FT
V0 V0
= K0 Tr (b e) b + 2 G0 devb (b e b) ,
V V
where K0 = 0 + 32 G0 , is the bulk modulus, and G0 is the shear modulus.

In the small strain limit b = g 1 + O kekg1 , and the Cauchy stress
reduces to

= K0 Tr g 1 e g 1 + 2 G0 devg1 g 1 e g 1 + O kek2g1 .
With the identity matrix for the spatial metric, gij = ij , this equation
corresponds to the standard form for the stress of a linearly elastic isotropic
material.
Remark 2.3 Linear elasticity is a reasonable description of many metals
at a stress low compared to the yield strength. It is frequently used in
engineering applications.
2.2.2 Non-linear Isotropic Model
As an illustrative example of a non-linear isotropic material, suppose the

energy has the form
1/3

E = 18 V0 K0 (I3 1)2 + 12 V0 G0 I1 I3 3 (2.3)
Then the stress is given by

S = 21 K0 I3 (I3 1) C1 + 2 G0 I3 1/3 devC1 G1 EC1
(2.4)
J = 21 K0 I3 (I3 1) g 1 + 2 G0 I3 1/3 devg1 beg 1
Despite its appearance, we note that the deviator term is a symmetric matrix,
due to the identity

1 1 1 1 1 1 1 1
G EC = 2
G (C G)C = 2
G C .

2.2 Illustrative Elastic Models Elastic-Plastic Flow
We also note that

G1 C1 = G1 (G + 2 E)1
= 2 G1 EG1 4 G1 EG1 EG1 +

which converges for G1 E < 12 . Thus, in this example the stress is really a
non-linear function of the strain.
In the limit of small strain, kEkG1 1,

I1 = 3 + 2 + O kEk2G1

I3 = 1 + 2 + O kEk2G1
C = G + O (kEkG1 )
b = g 1 + O (kEkG1 )

where = Tr G1 E . Moreover, the Cauchy stress can be expressed as
V V0 1

= K0 g + 2 G0 devg1 g 1 eg 1 + O kEk2G1 .
V0
This is the standard form for linear elasticity, in which the bulk and shear
moduli are taken to be constants. In general, for arbitrary strain, it is not
consistent with hyper-elasticity for the moduli to be constant [1, 24]. In
other words, linear elasticity is only valid for small strain. The illustrative
example is a simple model that is equivalent to linear elasticity for small
strain, yet consistent with hyper-elasticity for large strain.
Remarks:
2.4 Though isotropy is a convenient modeling assumption, in reality most
materials are not isotropic. Metals, for example, are polycrystalline and
crystals by their very nature are anisotropic. But at the cell size typically
used in an engineering simulation, metals behave as if they are isotropic
when their grains are randomly oriented. In other words, isotropy can be a
reasonable approximation on a coarse grain scale. However, there are circum-
stances when a material, even on a coarse scale, has a preferred direction.
This occurs, for example, due to the rolling of sheets or the drawing of wire.
In these cases the isotropic approximation breaks down. The non-uniform
distribution of grain orientations is referred to as texture in the materials
community.

Elastic-Plastic Flow 2.2 Illustrative Elastic Models
2.5 Even if a material is initially isotropic, the material does not remain
isotropic for non-hydrostatic deformations. The stress is still determined by
the isotropic specific energy for the reference state. However, the elastic
tensor for a particular strain state need not be isotropic. This is important
outside the domain of linear elasticity, i.e., large elastic strains, though plastic
deformation may mitigate the anisotropy of the deformed state by limiting
the non-hydrostatic component of the elastic strain tensor.
2.3 Conservation Equations
2.3.1 Eulerian Form
The conservation laws of mass, momentum and energy in the Eulerian

frame are expressed as

+ uj =0
t ;j

ui + ui uj ij =0 (2.5)
t ;j
h1 i
i
h i
2 ui u + E + 12 ui ui + E uj ui ij =0
t ;j
where = 1/V is the density, the subscript ;j denotes covariant differen-

tiation,
i
ui ;j = i
u + jk uk
xj

ui;j = ui ijk uk
xj
and !
a ab0 gcb0 gdb0 gdc
dc = 12 g d
+ c
b0 (2.6)
x x x
is the Christoffel symbol corresponding to the metric g. The covariant
derivative obeys the usual product rule for differentiation, e.g., (ui ij );j =
ui;j ij + ui ( ij ;j ). In addition, from Eq. (2.6) it follows that gik;j = 0, hence
gik (uk );j = ui;j .

2.3 Conservation Equations Elastic-Plastic Flow
The covariant derivative accounts for the curvature of the coordinate

system, i.e., if ~ei are basis vectors in T S then x
~ei k
j = ij ~
ek . This is illustrated
in the directional derivative w ~
~ and the divergence of the vector ~u = ui~ei ,
i

~
~ )(u
(w i
~ei ) = wj i
u + jk uk ~ei = wj ui ;j ~ei
x j
~ (ui~ei ) = i i k
u + ik u = ui ;i
xi
In addition, the covariant derivative of a tensor with contravariant indices is
ij j
ij ;k = k
+ kpi
pj + kp ip ,
x
and a tensor with covariant indices
p p
eij;k = eij ik epj jk eip .
xk
In particular, the divergence of the Cauchy stress is
ij j
ij ;j = j
i
+ kj kj + kj ik .
x
a a
The Christoffel symbol is symmetric in the lower indices, i.e., bc = cb .
The curvature tensor is defined in terms of the Christoffel symbol by
a a
db cb
ra bcd = a e
+ ce a e
db de cb .
xc xd
The non-commutivity of the covariant derivative,
ua ;bc ua ;cb = ra bcd ud ,
is characterized by the curvature tensor.
Remark 2.6 The Christoffel symbol is not a tensor, i.e., for the transformed
metric C = FT gF, the transformed Christoffel symbol 6= (F1 ) i jk
i
Fj Fk .
2.3.2 Lagrangian Form
The flow equations can be transformed from the Eulerian frame to the La-
grangian frame by applying the Piola identity. Define the Piola transform
of ~y T S 7 Y~ T B by
Y = J (F1 ) i y i . (2.7)

Elastic-Plastic Flow 2.3 Conservation Equations
Then the Piola identity is expressed as
Y ; = J y i ;i , (2.8)
where the covariant derivatives in T S and T B are based on Christoffel symbol

for the metrics g and G, respectively.
The conservation laws can be expressed in integral form. In fact, this is
how they are derived physically. For example, momentum conservation is
expressed as
Z I
d3 x ui + da nj ji = 0 , for any B,
t (,t) (,t)
~ t) | X
where (, t) = {~x = (X, ~ }. As a consequence of Piola identity
and Gauss law
I Z Z I
dA N Y~ = 3
d XY
; = 3
d xy i
;i = da n ~y .
() ()
This enables the surface terms in the integral form of the conservation laws
to be transformed readily between frames. Then re-expressing the conserva-
tion laws as differential equations leads to the Lagrangian form of the flow
equations
d

J J (F1 ) i U i =0
dt ;
d

0 U i Pi ; =0 (2.9)
dt
d 1

i i
0 U i U + E U i P =0
dt 2 ;

where d
=
= ~ is the convective derivative, and
+ ~u
dt t X t ~
~
x
E ~ t) = Fi S : T ~ B T ~ S
P = g 1 = FS , Pi (X, X X
F
is the first Piola-Kirchhoff stress tensor. In addition, mass conservation
is trivial, dtd 0 = 0, and has been replaced by volume conservation. Equa-
tion (2.9a) is derived by applying the Piola Identity to Lemma A.4. We
note that the first Piola-Kirchhoff stress is a two-point tensor and is not
represented by a symmetric matrix.

2.4 Kinematic Equations Elastic-Plastic Flow
2.4 Kinematic Equations
The conservations laws alone give a system of partial differential equations

(PDEs), either in Eulerian form Eq. (2.5) or in Lagrangian form Eq. (2.9),
that is not closed. In contrast to fluid flow, the stress depends on the strain
deviator in addition to the specific volume. Consequently, a kinematic equa-
tion for the evolution of the strain tensor is needed to complete the flow
equations.
Remark 2.7 Continuum mechanics can be formulated directly in terms of
the mapping by writing the momentum equation in the form of Newtons
second law (F = ma),
2 i i ~
0 = P D .
t2 X
This is well defined for smooth flows but not for discontinuous flows, because
the right hand side is a nonlinear function of X i . Discontinuities, known
as shock waves, commonly arise in the solution of nonlinear wave equations
even when the initial data is smooth. The strain tensor is introduced as an
independent variable in order to obtain a system of quasi-linear PDEs (i.e.,
first order and linear in all derivatives) and use the theory of hyperbolic
PDEs to regularize the continuum model.
In Lagrangian form, the evolution of the deformation gradient
d ~ , d i
F = DU F= Ui , (2.10)
dt dt X
is conveniently in conservation form. However, to recover the mapping

from the deformation gradient requires the compatibility conditions
i i 2

F =
F =
i . (2.11)
X X X X
From Eq. (2.10) it follows that if the compatibility conditions are satisfied
at some initial time then the compatibility conditions will be satisfied for
all time. Consequently, Eq. (2.10) is equivalent to 3 independent equations
and subsume the volume conservation equation (2.9a). The effective system
of PDEs has 7 independent equations. Hence, physically there are 7 wave

Elastic-Plastic Flow 2.4 Kinematic Equations
families. With Eq. (2.11) the full system has 6 additional physically irrelevant
waves. Later we show that the full system is indeed hyperbolic and that the
physical waves can be identified with 3 acoustic wave families (one quasi-
longitudinal and two quasi-transverse) in both the forward and backward
directions and a linearly degenerate mode for the advection of entropy.
Remark 2.8 When the continuum equations are discretized for numerical
simulations, the compatibility conditions may not be satisfied exactly. The
problem is analogous for incompressible fluid flow to enforcing the condition
that the divergence of the velocity vanishes. If the error in the compatibility
conditions increases with time of run then the simulations can be seriously
inaccurate or numerical instabilities may develop. Part of the problem is that
vorticity in the flow can lead to stirring on a fine scale and when resolution
of is lost, the discretization errors can become large.
The Eulerian form for the evolution of the deformation gradient is ob-
tained by applying the chain rule to Eq. (2.10)
d ~ u) f , d i
f = (~ f = ui ;j f j , (2.12)
dt dt
where f(~x, t) = F 1 : T~1 (~x,t) B T~x S. Plohr and Sharp [16] have shown
that this evolution equation may be recast in conservation form.
Lemma 2.1. The evolution of the deformation gradient can be expressed as
1 i

J f + J1 f i uj = J f u 1 j i
. (2.13)
t ;j ;j
Proof. From the mass conservation Eq. (2.5a) and Eq. (2.12), the left hand
side of Eq. (2.13) can be expressed as
1 i d i

J f + J1 f i uj = J1 f
t ;j dt
= J1 f j ui ;j
The right hand side of Eq. (2.13) can be expressed as

J1 f j ui = J1 f j ui ;j + J1 f j ui .
;j ;j

2.4 Kinematic Equations Elastic-Plastic Flow
Apply the Piola identity Eq. (2.8) to the second term on the right hand side,
we find
J f = J1 ; = 0 .
1 j
;j
Combining the above three equations yields Eq. (2.13).
Alternatively, the analog of Eq. (2.10) for the inverse map leads to the
evolution equation for f 1 [28]

(f 1 ) i + (f 1 ) j uj =0, (2.130 )
t ;i
which can be used in place of Eq. (2.13).

Remark 2.9 The time derivative in Eq. (2.13) does not enter in the combi-
nation t + ~u . Consequently, the Eulerian form of the evolution equation
for the deformation gradient is not Galilean invariant (~x 7 ~x ~u0 t and
~u 7 ~u + ~u0 ). However, solutions for initial data that satisfy the kinematic
compatibility condition Eq. (2.11) are Galilean invariant. This emphasises
the importance of the compatibility condition in the Eulerian formulation.
2.5 Dissipation
The external rate of working is given by V0 S : dE

dt
. Using Eq. (2.12) the
strain rate can be written as
dE
= FT dF , (2.14)
dt
where the rate of deformation tensor is

d= 1 ~ u) + [(g~
(g~ ~ u)]T , dij = 12 (ui;j + uj;i ) . (2.15)
2
Therefore, the external rate of working can be expressed as

!
dE
V0 Tr S = V0 Tr S FT d F
dt
= V :d (2.16)
dV
= P + V devg1 () : devg (d)
dt

Elastic-Plastic Flow 2.5 Dissipation
where in the last equation we substituted mass conservation equation ex-

pressed as
~ ~u = 1 d V .
V dt
The conservation form of the energy equation (2.5c) is expressed in terms

of the total energy, 12 |~u|2 + E. Substituting the mass and momentum equa-
tions yields an evolution equation for the specific internal energy
d
E = V :d . (2.17)
dt
Substituting the relation for the external work Eq. (2.16) and the rate
of change of internal energy Eq. (2.17) into the thermodynamic identity
Eq. (2.1) gives a constraint on the plastic strain rate
T dtd = V0 Sp : dtd Ep 0 . (2.18)
A constraint of this kind, obtained by requiring the local dissipation to be

positive, is known as the Clausius-Duhem inequality. Since we are ne-
glecting source terms, such as arise from heat conduction, in the absence of
plastic strain the flow is isentropic.
Even with plasticity, as discussed later, the plastic strain rate is a source
function. Hence, Eq. (2.18) is in characteristic form. Advection of entropy
corresponds to a degenerate wave family. This is analogous to the what
occurs for the Euler equations describing fluid flow.
Remarks:
2.10 It is instructive to transform the strain rate from the Lagrangian frame
to the Eulerian frame.
de dE ~ u)T e e (~
~ u)
= FT F1 (~
dt dt
~ u)T e e (~
= d (~ ~ u)
Consequently, the external rate of working is not equal to V : de

dt
.
2.11 The time derivative of the Eulerian strain can be expressed as a Lie
derivative, de
dt
= L~u (e). In effect, the time derivative is defined as the

2.5 Dissipation Elastic-Plastic Flow
transform of the derivative of the transformed quantity in the reference frame.

This is necessary for the theory to be frame invariant. The concept of a Lie
derivative is described in more detail in Appendix E.
2.6 Wave Analysis
The wave speeds for a hyperbolic system of PDEs are determined from
the eigenvalues of the flux matrix. Because of the complications that arise
from the compatibility relations, we shall use the Lagrangian equations and
then transform to the spatial frame.
A few preliminary definitions are needed. The fourth rank stiffness ten-
sor is defined by
S 2E
C = 0 .
E E E
It follows from the definition that the stiffness tensor has the following sym-
metry properties
0 0 0 0 0 0 0 0
C =C =C =C .
The stiffness tensor allows small changes in the stress to be related to small
changes in the strain
0 0
S = C : E , (S) = C (E)0 0 .
Hence,
d d
S = C: E
dt dt
= C : (FT dF) .
The transformed stiffness tensor

ipjq 0 0
c = J1 Fi 0 Fp Fj 0 Fq C ,
relates changes in the Cauchy stress to changes in the Eulerian strain

= c:e ,

Elastic-Plastic Flow 2.6 Wave Analysis
where
1 dS
= F FT
J dt
d ~ u) (~

~ u)T + Tr g 1 d ,
= (~
dt
dE
e = FT F1
dt
d ~ u)T e + e(~
~ u) .
= e + (~
dt
Consequently, the time derivative of the Cauchy stress can be expressed as
d
+ = c : d , (2.19)
dt
where

cijpq = cijpq + 1 ip jq
g + g +g + gip jq jp iq jp iq
ij g pq ,
2
and h i
= 1 ~ u) (~
(~ ~ u)T .
2
Remark 2.12 The form of the time derivative of the Cauchy stress given

by Eq. (2.19) is frame indifferent. The form of is known as Truesdells
objective derivative. There are other forms for objective time derivatives of
tensors, see for example [5] and [11, 1.6]. They are related to the choice
of frame in which the stiffness tensor is specified. The form of objective
derivative is important for numerical simulations based on hypo-elastic con-
stitutive models, especially when the shear component of the stiffness tensor
is assumed constant. A constant stiffness tensor is a useful approximation
for small strain but in general is not consistent with hyper-elasticity [1, 24].
The inverse of the stiffness tensor is called the compliance tensor and
allows small changes in the strain to be related to small changes in the stress
0 0
E = S : S , (E) = S0 0 (S) ,
where
0 0
C S0 0 = 1
2
+ .

2.6 Wave Analysis Elastic-Plastic Flow
The symmetry requirement, i.e., the symmetric identity operator on the right
hand side, complicates the computation of S from C. Details are given in
Appendix B.
An additional tensor, closely related to the stiffness tensor, is the acous-
tic tensor
P ij 0 0
A = g 1 , A = Fi 0 F j 0 C + g ij S
F
The acoustic tensor is a two-point rank-four tensor. However, in the spatial
frame, the acoustic tensor,
ipjq ij ipjq
a = J1 Fp Fq A =c + g ij pq ,
is a one-point rank-four tensor.
In the previous subsection we found that one characteristic is associated
with the advection of entropy. To find the other wave families, we take the
entropy to be constant and drop the energy equation. We analyze the system
of 12 PDEs consisting of the three momentum equation (2.9b) and the nine
kinematic equations (2.10)
d i i
U A j
0
Fj = 0
dt X
d i
F Ui =0
dt X
Here, the metric is assumed constant and the covariant derivative reduces to
the ordinary derivative in flat space.
We look for traveling wave solutions of the form W ~ = W~ () where
~ Ct, C is the wave speed, the direction of propagation N
= hN , Xi
is normalized to satisfy hN , N i = 1, and the vector of unknowns is
~ = (Fj =1 , Fj , Fj , U i )T ,
W with i, j = 1, 2, 3 .
=2 =3
Then the system of equations can be written as LW

~ = 0 with L in block
matrix form

CI 0 0 N1 I

0 CI 0 N2 I
d
L= ~ = ~0 .

0 0 CI N3 I
W
d

i =1 i =2 i =3
N A j N A j N A j 0 CI

The wave speeds are the solution to the equation obtained by setting the
determinant of L to zero. By adding multiples of the first three columns to
the fourth column the determinant equation reduces to
0 = det(L)

CI 0 0 0

0 CI 0 0

= det
0 0 CI 0

i =1 i =2 i =3
N A j N A j N A j 0 CI C 1 A(N )

= C 6 det A(N ) 0 C 2 I
i i
where A(N ) j = N A j N . Thus, there are six trivial wave speeds, C = 0,
and three pairs of non-trivial waves, C, determined by the equation

i 2 i
det A(N ) j 0 C j =0. (2.20)
We show that the trivial waves are ruled out by the compatibility condi-
tion Eq. (2.11), and that each pair of non-trivial waves corresponds to an
acoustic wave family.
First we examine the non-trivial waves. Let U ~ be an eigenvalue of A(N )
with eigenvalue 0 C 2 . Then a solution to the linearized equations is given by
~ =U
U ~ W ()
~0 + U
Fi = (F0 )i N (Ui /C)W ()
for any smooth function W (). We note that the deformation gradient
caused by the wave Fi = N (Ui /C)W () satisfies the compatibility
condition Eq. (2.11) and is physically admissible.
~ = 0 and
The eigenvector corresponding to the eigenvalue C = 0 have U
the change in the deformation gradient satisfying
i
j
N A j F =0, for i = 1, 2, 3. (2.21)
Since there are three linear equations for nine unknowns, Fj , there are
six families of solutions corresponding to the six fold degeneracy of the zero

eigenvalue. Moreover, if Fj satisfies the compatibility condition Eq. (2.11)

then it has the form Uj N and Eq. (2.21) implies that Uj is an eigenvector
of A(N ) with 0 eigenvalue. Conversely, if zero is not an eigenvalue of A(N )
then the solutions to Eq. (2.21) do not satisfy the compatibility condition
and are physically inadmissible.
The results of the wave analysis can be summarized as follows. Provided
that A(N ) is a positive matrix for all N , the Lagrangian flow equations (2.9)
and (2.10) are a hyperbolic system of PDEs. The system has one degen-
erate wave family corresponding to the advection of entropy and three pair
of non-degenerate acoustic waves. The remaining six degenerate waves are
physically irrelevant for solutions to initial value problems in which the initial
data satisfies the compatibility condition Eq. (2.11).
Physical experiments measure waves speeds in the spatial frame. There-
fore, it is important to transform the wave speeds we derived in the body
frame into the spatial frame. The form of a traveling wave variable in the
spatial frame is
= hn, ~xi ct = nk (F0 )k X ct
where the propagation direction n is normalized to satisfy hn, ni = 1. Re-
placing 7 , the equation for the sound speed Eq. (2.20) becomes

det a(n)ij gjk c2 i k = 0 , (2.22)
ipjq ipjq
where a(n)ij = np a nq = c(n)ij + nn g ij , c(n)ij = np c nq and nn =
n n. Or in terms of the stiffness tensor

det c(n)ij gjk (c2 nn ) i k = 0 .
Hence the eigenvalues of c(n) must be greater than nn . We note that in

compression nn is positive.
Remark 2.13 The direction of propagations in the body frame and the spa-
~ . However, the spatial normalization is
tial frame are related by ~n = FT N
equivalent to N C N = 1. The normalization of the propagation direction
affects the magnitude of the eigenvalues.
Let ~u be an eigenvector of a(n)g. The deformation gradient of the wave
has the form
(F 1 )i () = (F0 )i + ui nk (F0 )k w() ,

where F0 is the deformation gradient unperturbed material. It follows that

the increment of the strain tensor due to the wave is

eij = 1
2
ui (b1 k k 1
0 )jk n + n (b0 )ki uj w() + O w
2
.
About the ambient state, b1

0 = g and the increment of strain due to the
wave is simply e u n + n u , where u = gu and n = gn.

The physical interpretation of the acoustic wave families is based on the

eigenvectors. Typically, the eigenvector corresponding to the largest eigen-
value is approximately parallel to the propagation direction, i.e., u n 1.
This represents a longitudinal wave and corresponds to the hydrodynamics
wave associated with changes in specific volume. Moreover, the two slower
waves are associated with shear strength and their eigenvectors are approxi-
mately perpendicular to the direction of propagation, i.e., u n 0. These
represent transverse waves. They require shear stress and hence do not occur
for ordinary fluid flow. In terms of the strain tensor, for a longitudinal wave
e n n, and for a transverse wave n e n 0. It is important to note
that in contrast to fluid flow, even the longitudinal sound speed may depend
on the direction of propagation.
Remark 2.14 For an isotropic material, the sound speeds are independent
of the direction of wave propagation. Moreover, the two transverse sound
speed have the same magnitude. It is shown in subsection B.2 that the
sound speeds are given by
c2long = K + 43 G
c2tran = G
Since the moduli are positive, the longitudinal sound speed is greater than
the transverse sound speed.
For the elastic flow equations to be hyperbolic, the sound speeds must be
real. Hence, for any direction n, the eigenvalues of a(n)g must be positive
definite. This is equivalent to what is known as strong ellipticity,
~ ) k~n k2 km
~ , a :~n m
(~n m ~ k2
~ T S and some > 0. In fact, is just the minimum of c2 over
for any ~n , m
all directions. Strong ellipticity can be expressed as a restricted convexity
condition (rank-1 convexity) on the specific energy E, see [11, p. 19]. Details
are given in Appendix C.

Remarks:
2.15 We note that for zero stress, the leading order expansion of the specific
energy is
E = 21 E : C : E + O kEk3G1 .
In order to be an energy minimum equilibrium state, C must be positive def-

inite, and hence A would be positive definite. In general, this is a stronger
condition than is needed for the flow equations to be hyperbolic, i.e., hy-
perbolicity requires only strong ellipticity which is a weaker condition than
positivity of the acoustic tensor.
2.16 Uniaxial flow reduces to a one-dimensional problem. For some mate-

rials, a state in tension is reached at which a continued increase in strain re-
sults in a decrease in stress, i.e., a reversal of the sign of d(stress)/d(strain).
For an elastic material, a stress-strain relation of this type implies a loss of
hyperbolicity. A non-monotonic variation of stress with strain can result
from either thermal softening (decrease in the yield strength with temper-
ature) or a phenomenon known as dilatancy. Physically, dilatancy is due
to the nucleation and growth of cracks. The phenomenon of crack growth
is an unstable process. To describe unstable phenomena, requires degrees of
freedom for the state of the material in addition to the total strain tensor
(such as plastic strain or volume fraction of cracks) plus rate equations for the
evolution of the additional internal degrees of freedom. Consequently, loss of
hyperbolicity of the elastic flow equations is associated with a breakdown of
the physical assumptions underlying the continuum model.
2.17 Plasticity models have been proposed to describe dilatancy, by ap-

propriately shaping the yield surface in conjunction with a rate independent
associated flow rule. These models have had problems with instabilities. The
choice of rate independent plasticity is probably the dominant cause of the
difficulties. Very likely a rate dependent plastic strain rate would allevi-
ate the ill-posedness associated with the instabilities and lead to plasticity
models that could, at least qualitatively, describe dilatancy.
2.18 Plastic flow can be thought of as a non-equilibrium process that relaxes

the stress deviator toward its equilibrium value. As is typical with relaxation
processes, a non-zero plastic relaxation time can be expected to lower the
equilibrium sound speeds. For example, the longitudinal sound speed of an

isotropic material drops from [(K+ 43 G)/]1/2 in the elastic regime to (K/)1/2
in the plastic regime. In addition, since plastic flow is dissipative, traveling
waves in the plastic regime can be expected to decay, i.e., rather than an
elastic wave w = w(n ~x ct), a traveling plastic wave would have the form
w = w(n ~x ct) exp(t/ ).
2.7 Shock Jump Conditions
For a shock propagating in the direction N , it follows from the Lagrangian

flow equations (2.9) and (2.10) that the shock jump conditions are given by
M (U i ) = (Pi ) N

M 1
U Ui
2 i
+ E = (Ui Pi ) N (2.23)
M (Fi ) = 0 (U i ) N
where M = 0 Us is the mass flux through the shock front, and Us is the
shock velocity in the body frame. The jump of a variable U across the shock
front is denoted (U ) = U1 U0 . In addition, the propagation direction is
normalized to satisfy N N = 1.
Using the relation
(AB) = A(B) + (A)B ,
where a bar over a variable denotes the average, i.e., A = 12 (A0 + A1 ), the
jump in the strain is given by
j i
h i
(E ) = 1
2
(Fi ) gij F + F gij (Fj )
i i

M (E ) = 21 0 (Ui ) N F + F N (2.24)
Together with a constitutive relation, P = F S(E, E), the jump conditions

determine a one parameter curve for the shock locus of each wave family.
Although the constitutive equation is a function of strain, we note that the
deformation gradient is needed to determine the shock locus.
In the previous subsection, we have shown that there are three pair of
acoustic wave families corresponding to a quasi-longitudinal wave and two
quasi-transverse waves. The quasi-longitudinal wave satisfies (U~ ) N

2.7 Shock Jump Conditions Elastic-Plastic Flow

~ ) and is the analog of a gas dynamic shock wave. The quasi-transverse
(U

waves satisfy (U ~ ) N (U
~ ) and are associated with shear stress.
Physically, the admissible shock waves are restricted to be entropy in-

creasing. Algebraic manipulations to eliminate the velocity leads to the ana-
log of the Hugoniot equation for gas dynamic shocks
(E) = V0 Pi gij (Fj ) . (2.25)
The Hugoniot equation can be used to show that for weak shocks the entropy
change is third order. However, in contrast to the scalar gas dynamic case,
the Hugoniot equation is not sufficient by itself to determine the shock locus.
Remark 2.19 Using the relation
P = FS + 14 (F)(S)
and the symmetry of S, the Hugoniot equation can be expressed as
(E) = V0 S (E ) + 41 V0 (Fi ) (S ) gij (Fj ) .
Consequently, the natural generalization of the gasdynamic Hugoniot equa-

tion is not true, i.e., (E) 6= V0 S : (E).
The shock jump conditions can also be expressed in the spatial frame.
From equations (2.5) and (2.13) the jump conditions can be written as
M (ui ) = ( ij ) nj

M 1
u ui
2 i
+ E = (ui ij ) nj (2.26)
M (f i ) = ( f j ui ) nj
where M = (us n ~u) is the mass flux through the shock front, us is the
shock velocity in the spatial frame and the propagation direction is normal-
ized to satisfy ni ni = 1. It can be shown using the Piola identity that the
Lagrangian and Eulerian form of the jump conditions are equivalent [16].

Elastic-Plastic Flow 3 Plastic Strain Rate
3 Plastic Strain Rate

Experiments indicate that for small strain the response of many solids
is reversible. This is referred to as the elastic regime. In contrast, for large
shear strain, irreversible processes take place which limit the shear stress.
The material response that limits the shear stress is referred to as the plas-
tic regime. The plastic strain tensor is the minimal number of degrees of
freedom that needs to be added in order to extend elastic flow to the plas-
tic regime in a physically consistent manner. Models with the addition of
only a plastic strain tensor can describe the dominant effects of plasticity.
But other internal degrees of freedom are need to account for other observed
plastic flow phenomenon, such as work hardening.
In this section a form for the plastic strain rate is presented under the
assumption that the specific energy of a material can be split into elastic and
plastic components
E = Ee (E
b , ) + E (E ) ,
e p p (3.1)
where the choice of elastic strain in terms of the total strain and plastic
strain E
b (E, E ) is discussed below. We note that the Piola-Kirchhoff stress,
e p
S = 0 E
E
depends only on the elastic energy while the plastic stress can be
written as the sum of two terms Sp = S0p + S00p , where
Ee
S0p = 0 ,
Ep
Ep
S00p = 0 ,
Ep
associated with the elastic and plastic components of the energy.
For simplicity, the plastic specific energy often is neglected. This is a rea-
sonable approximation for the mechanical properties of stiff material, those
for which the component of stress from compression dominates the thermal
component of stress. Experiments on metals [26], however, show that the
plastic energy is about 10 percent of the external work in the plastic flow
regime. Even for stiff materials, when thermal properties are important or
for problems in which the material undergoes a large number of cycles, the
plastic component of the energy can not be ignored.
Remark 3.1 The specific energy resulting from plastic work maybe due
comments to [email protected] 33 3 Plastic Strain Rate

3 Plastic Strain Rate Elastic-Plastic Flow
to additional internal degree of freedom, such as those associated with work

hardening, and only indirectly with the plastic strain.
b = 1 (C C ). Hence,
Many plasticity models use as the elastic strain Ee 2 p
based on super-position of strains the elastic energy is frequently taken to
have the functional form
Ee = Ee (E Ep , ) . (3.2)
As a consequence of this modeling assumption, the elastic component of

the plastic stress tensor and the Piola-Kirchhoff stress tensor are the same
S0p = S.
Instead, we use Cp 1 C as a measure for elastic strain. This is motivated
by analogy with the energy for an isotropic material. With the composition
of maps, = e p , the natural extension for an elastic-plastic isotropic
material is to take the elastic energy to be a function of the invariants of
1 1
G e =G
e C
e
e F T gF
e e

= Fp Cp 1 C Fp 1
e 1 C
By Lemma A.5 the invariants of Cp 1 C are the same as those of G e .
e
Consequently, we take
Ee = Ee (Cp 1 C, ) . (3.3)
Since Cp 1 C can be expressed in terms of E and Ep , this form satisfies the

requirement of frame indifference. Moreover, though S and S0p are related,
they are not equal.
3.1 Yield Condition
We assume that the flow is elastic, dtd Cp = 0, when Y(C,

b Cp ) < Y , where
Y is the yield strength and Y(C, Cp ) is a yield function. Both the yield
b
strength and the yield function have dimensions of stress. The time derivative
of the yield function has two contributions
d b b
Y dC Y dCp
b
dt
Y = C : dt + Cp
: dt .
3 Plastic Strain Rate 34 comments to [email protected]

Elastic-Plastic Flow 3.1 Yield Condition
Plastic flow occurs when the contribution from the total strain rate, dtd C,
raises the value of the yield function to the yield strength. For many ap-
plications rate independent plasticity models are used. In these models,
the plastic strain rate dtd Cp is chosen such that the yield function is never
exceeded, i.e.,
d b
dt
Y 0 , if Yb = Y .
This determines the magnitude of Cp , though the direction must be specified
by a model.
Remark 3.2 The term rate independent refers to the lack of an explicit
time constant. For simple problems, such as uniaxial strain in which the
total strain increases monotonically, the plastic dissipation depends only on
the strain and is independent of strain rate.
As an extension of elastic flow, rate independent plasticity has two prob-
lems. From Eqs. (2.14) and (2.15) the total strain rate is dtd C = 2 FT dF.
Hence the plastic strain rate is no longer equal to a source term, but depends
on dij = 12 (ui;j + uj;i ). Consequently, just because the elastic flow equations
are hyperbolic, it does not follow that the extended system of PDEs remains
hyperbolic. In fact for some choices of the direction of the plastic strain rate
and yield function, the rate independent equation have elliptic regions. The
initial value problem for these models can be regularized by adding shear
layers when the solution runs into an elliptic region [20]. Another problem
arises even for rate independent plasticity models in which the extended sys-
tem of PDEs remains hyperbolic. The equation for plastic strain is not in
conservation form. Hence there are not sufficient jump conditions to de-
fine strong shock waves. However, fully dispersed weak shocks (for which
the plastic dissipation provides the required shock entropy) are well defined.
Consequently, a rate independent plasticity model only provides solutions for
the limited class of initial value problems which do not lead to strong shocks.
Here, we consider rate dependent plasticity. In these models the value
of the yield function is allowed to exceed the yield strength. The plastic
strain rate serves to relax the yield function back to the yield surface. A rate
dependent plasticity model is determined by specifying a plastic strain rate
such that three conditions are satisfied:
i. Plastic flow only occurs outside the yield surface, i.e.,

d
C = 0 if Y(C,
dt p
b Cp ) < Y .

3.1 Yield Condition Elastic-Plastic Flow
ii. Plastic flow is dissipative, i.e.,

Sp : dtd Cp 0.
iii. Plastic flow reduces the yield function, i.e.,

b
Y
: d C 0.
Cp dt p
The system of PDEs for rate dependent plasticity is hyperbolic whenever the
elastic system of PDEs is hyperbolic.
This is because the plastic rate equation is in characteristic form. Since
source terms do not affect the flux matrix, the wave speeds of the augmented
elastic-plastic system are the same as the original elastic system with added
degeneracy for the convective wave speed (which contains the entropy wave
family). Hence, the augmented system has real wavespeeds. To remain hy-
perbolic, a complete set of eigenvectors of the flux matrix are required. De-
spite the degeneracy in the eigenvalues, the eigenvectors do remain complete.
To see this we note that the characteristic form of the PDEs is determined
by the left eigenvalues of the flux matrix. Adding rate equations do not
affect the characteristic form of the original system and hence the eigenvec-
tors of the acoustic wave families. Furthermore, the eigenvectors of the rate
equations depends on the directions associated with the new variables and
are linearly independent of the eigenvectors associated with the wave fami-
lies of the original system. Consequently, the elastic-plastic flow equations
are hyperbolic. The same argument applies if additional internal degrees of
freedom, along with rate equations for their evolution, are added in order to
describe the response of a material more realistically. Isotropic or kinematic
work hardening parameters are examples of other internal degrees of freedom
that are commonly added.
As a consequence of the plastic strain rate equation, the physically inter-
esting plastic waves are partly dispersed, i.e., a composite wave consisting
of a discontinuous shock followed by a narrow layer in which the yield func-
tion relaxes to the yield surface. Unlike fluid dynamics, the end state of a
plastic wave depends on the source term for the plastic strain rate and is
not fully determined by the Hugoniot jump conditions. For special cases,
such as uniaxial strain, the jump conditions plus the yield condition suffice
to determine the end state. But in general the jump in the plastic strain
across a plastic wave can only be determined by integrating the ODEs for
the profile of a traveling wave. This is similar to what occurs for the system

Elastic-Plastic Flow 3.1 Yield Condition
of PDEs describing reactive flow with more than one species and multiple
reactions. The change in equilibrium composition for flow through a nozzle,
such as a jet engine, depends on the reaction rates relative to the flow rate.
Remark 3.3 Though strong shock waves are well define for rate dependent
plasticity, the relaxation zone may be very narrow. Problems can occur in
numerical simulations when the cell size is insufficient to resolve the wave
profile.
3.2 Associated Flow Rule
An associated flow rule relates the plastic strain rate to the yield func-
tion. Though the plastic strain can be based on non-associative flow rules
(see for example [21]), here we consider only the flow rule obtained from
what is referred to as the maximum dissipation postulate. In the rate in-
dependent case, C is varied over the yield surface, Y = Y(C,
b Cp ), to find an
extremum in the plastic dissipation, Eq. (2.18), for fixed Cp and dtd Cp . The
extremum occurs when (see [22] and references therein)
Sp d Yb
: Cp = , (3.4)
C dt C
where is a Lagrange multiplier which is determined from the constraint
that C lies on the yield surface. For the moment, let us assume that S C
p
is everywhere invertible on the space of symmetric rank-two tensors. Then

the extremum is unique. Furthermore, by the implicit value theorem we can
express the yield function as Yb = Y(Sp (C, Cp ), Cp ), and invert Eq. (3.4) to
obtain the plastic stress rate
d Y
Cp = . (3.5)
dt Sp
This equation has the interpretation that the plastic strain rate is normal to
the yield surface when the yield function is expressed in terms of the plastic
stress.
In addition, we assume that for fixed Cp the interior of the yield surface,
{Sp | Y(Sp , Cp ) Y }, is convex and contains the point Sp = 0. From
the geometry of the yield surface, it can be shown that the extremum of

3.2 Associated Flow Rule Elastic-Plastic Flow
plastic dissipation is a maximum and that the maximum is positive. Since

the dissipation can be expressed as
d Y
Sp : Cp = Sp : ,
dt Sp
Y
and Sp
is an outward normal, it follows that > 0. In the rate independent
d
case, setting dt
Y = 0 determines the Lagrange multiplier to be
Y Sp d
S p
: C : dt C
= Y Sp Y Y Y
. (3.6)
Sp
: C :
p Sp
+ Cp
: S p
From the existence of an extremum and > 0, it follows that the denomi-
nator is non-zero and negative.
Remarks:
3.4 The associated flow rule applies when the yield surface is smooth. For
some simple models, such as the maximum shear stress criterion of Tresca,
the yield surface has kinks or points at which the normal is discontinuous.
Models for crystal plasticity assume the plastic strain is due to slip along
preferred crystal planes. This also results in a yield surface with corners. At
corners the gradient of the yield surface is ill defined. The plastic strain rate
can be defined as a convex linear combination of the normals to the adjacent
facets.
3.5 For computational efficiency, algorithms based upon rate independent

plasticity frequently substitute the plastic strain rate Eqs. (3.5) and (3.6)
directly into an equation for the stress rate with a constant shear modulus
!
d d d
dev () = 2 G dev E Ep .
dt dt dt
This results in a hypo-elastic model. Though useful for some engineering
applications, such a model is neither thermodynamically consistent nor frame
indifferent. Simulation results for algorithm of this type should be interpreted
with care.
3.6 The compatibility condition for the existence of a mapping can be ex-
pressed in terms of the strain tensor rather than the deformation gradient

Elastic-Plastic Flow 3.2 Associated Flow Rule
Eq. (2.11). In general, Ep does not satisfy the compatibility condition. Con-
sequently, p and hence Fp can not exist. For crystal plasticity, at the molec-
ular scale, one expects p to be discontinuous due to dislocations and grain
boundaries. On the macroscopic scale the discontinuities are smeared out us-
ing some coarse grain averaging procedure to give a continuous plastic strain
tensor. The effect of the coarse graining can be thought of as converting
the discontinuities to source terms that prevent the compatibility relations
from being satisfied. Consequently, Ep should be thought of as an internal
degree of freedom that characterizes some aspects of a material not included
in elastic models. The plastic mapping may be used as a heuristic to moti-
vate the structure of the model equations but should not be taken literally
and proofs for properties of the model must not rely on the existence of p
or even Fp .
To generalize to the rate dependent case, we assume that any level set of
the yield function satisfies the geometric conditions of a yield surface (i.e.,
the interior of Y = Y1 for any Y1 > Y is convex and contains the point
Sp = 0), and that the scale factor is a state function, = (C, Cp , ) > 0,
to be specified by the plasticity model. We note that has dimensions of
inverse time, and in order that there is no plastic flow in the elastic regime
we require = 0 for Y(Sp , Cp ) Y . The construction of a rate dependent
model based on the maximum dissipation postulate guarantees that plasticity
is dissipative and that plastic flow relaxes the value of the yield function
toward the yield surface.
Remarks:
3.7 The condition that every level set of the yield function satisfy the geo-
metric conditions for a yield surface is a weak assumption. The same condi-
tion is needed to account for work hardening.
3.8 Typically, the scale factor is a monotonically increasing function of

the distance to the yield surface, i.e., Y Y . Larger values of correspond
to a smaller time constant for the relaxation to the yield surface. Rate
independent plasticity can be viewed as the limit in which the time constant
goes to zero.

3.2 Associated Flow Rule Elastic-Plastic Flow
3.2.1 Volume Preserving Plastic Flow
Based on the underlying dislocation mechanism for crystal plasticity,

many plasticity models take the plastic strain to be volume preserving, i.e.,
det(G1 Cp ) = 1. By Lemma A.3, the condition dtd det(G1 Cp ) = 0 places the
constraint on the strain rate that

Tr Cp 1 dtd Cp = 0 . (3.7)
Hence, if the plastic strain is volume preserving the plastic strain rate is the
deviator with respect to Cp of some tensor. To enforce the condition that
plastic strain is volume preserving, it is natural to take the specific energy
to be a function of Jp2/3 Cp , i.e.,
1 2/3
E = Ee (J2/3
p Cp C) + Ep (Jp Cp ) ,
where J2p = det(G1 Cp ). With this functional form the energy is independent
of the plastic volume and it is straight forward to show that the plastic stress
is deviatoric, i.e., devCp 1 (Sp ) = 0.
The plastic strain rate of Eq. (3.5) is volume preserving when the yield
function has the form
Y = Y(devCp 1 (Sp ) , Cp ) . (3.8)
However, to verify the plastic strain rate satisfies the required properties
of a plasticity model, we need to reexamine the steps in the derivation of
Eq. (3.5). In particular, the assumption that SC
p
is invertible is not true
when Sp is deviatoric.
With the volume preserving form of the yield function, Eq. (3.8), the
extremum condition for the dissipation, Eq. (3.4), can be written as
Sp d Y

: Cp =0.
C dt Sp Cp
We note that the term in parenthesis is deviatoric. Provided that the null
space of S
C
p
is limited to the deviatoric component, we can conclude that
Eq. (3.5) remains valid for the volume preserving plasticity.
Remark 3.9 For porous materials or materials with cracks, the volume
preserving assumption of the plastic deformation is not even approximately

Elastic-Plastic Flow 3.2 Associated Flow Rule
valid. Thus, the volume preserving constraint is not a physical law, but
merely an approximation that leads to a useful description of many materials,
in particular, metals.
4 Illustrative Constitutive Model

A constitutive material model, needed to complete the elastic-plastic flow
equations, is composed of three items:
i. A stress-strain relation, which for thermodynamic consistency, is de-

fined by a specific energy function.
ii. A yield function Y(Sp , Cp ), and a rule for the direction of the plastic
strain rate, such as the associated flow rule, Eq. (3.5).
iii. A scalar function (C, Cp , ) which specifies the rate at which the yield
function relaxes toward the yield surface.
This section provides a simple illustrative example of all three components

of a constitutive model for an isotropic material.
4.1 Specific Energy
We assume that Ep = 0 and use a variant of Eq. (2.3) for Ee ,

1/3
Ee = Ehydro (V, ) + 12 V0 G0 (I1 I3 3) , (4.1)
where G0 is the shear modulus at the initial state,

and the invariants are
1 1
I1 = Jp Tr Cp C and I3 = det Jp Cp C = J2 . When the elastic
2/3 2/3
1
energy depends on the invariants of J2/3 p Cp C, the plastic stress is related
to the elastic stress deviator as follows.
Lemma 4.1. The component of the plastic stress from the elastic energy
and the Piola-Kirchhoff stress are related by

devCp 1 S0p Cp = devC1 (S) C . (4.2)
comments to [email protected] 41 4 Illustrative Constitutive Model

4.1 Specific Energy Elastic-Plastic Flow
1
Proof. Let I1 ,I2 , I3 be the invariants of J2/3
p Cp C Applying the chain rule,
the stress tensors are given by
Ee I1 Ee I2 Ee I3
20 S0p =
I1 Cp I2 Cp I3 Cp
Ee I1 Ee I2 Ee I3
20 S = + +
I1 C I2 C I3 C
From the definition of the invariants, it is straight forward to show that the
invariants satisfies
! !
Ij Ij
devCp 1 Cp = devC1 C.
Cp C
Consequently, the stress tensors satisfy Eq. (4.2).
The stress stress corresponding for the model, Eq. (4.1), is given by

S = P (V, ) J C1 + G0 (Jp /J)2/3 devC1 Cp 1 ,
G0 (4.3)
= P (V, ) g 1 + (Jp /J)2/3 devg1 (be ) ,
J
where P = V
Ehydro is the hydrostatic pressure, and be = FCp 1 FT . In
addition, the plastic stress is given by

Sp = G0 (Jp /J)2/3 devCp 1 Cp 1 CCp 1
(4.4)
= G0 (Jp /J)2/3 devI Cp 1 C Cp 1 .
Remarks:
4.1 It is interesting to note that Eq. (4.3) for the stress can be expressed as

S = P (V, ) J C1 + 2 G0 (Jp /J)2/3 devC1 Cp 1 Ee C1 .
This is equivalent to Eq. (2.4) with Ee substituted for the strain E and J2/3
p Cp
substituted for the metric G.
4.2 If the parameter G0 is a constant then the mean pressure corresponds

to the hydrostatic pressure and can be expressed as P = Phydro (V, e eshear )
4 Illustrative Constitutive Model 42 comments to [email protected]

Elastic-Plastic Flow 4.1 Specific Energy
where the shear energy eshear is a function of only the strain. This is a
simple, thermodynamically consistent and computationally efficient way to
add strength to a hydrostatic equation of state. Moreover, the material
temperature is determined from the hydrostatic equation of state in the usual
manner. In practice there is no data with which to define the temperature
any better. On the other hand, the shear modulus is expected to decrease
as T % Tmelt .
4.3 A more general functional form for the stress can be obtained in terms
of the Helmhotz free energy, e (Cp 1 C, T ) = Ee (Cp 1 C, ) T . Letting the
shear modulus be a function of T and V ,
1/3
e = hydro (V, T ) + 12 V0 G(V, T ) (I1 I3 3) ,
and there are contributions to the entropy from both the shear strain and
the specific volume. In this case the Cauchy stress is given by

h i
1
= P (V, T ) + 1
2
2/3
3 (Jp /J) Tr(Cp C) V0 G(V, T ) g 1
V
G(V, T )
+ (Jp /J)2/3 devg1 (be ) .
J
We note that a consequence of the dependence of G on V is an additional
hcontribution to the pressure.
i The additional pressure term is proportional to
3 (Jp /J)2/3 Tr(Cp 1 C) , i.e., the norm of the deviator of the elastic strain.
As is shown in a later section, the strain deviator is limited by the yield
condition, and the additional pressure term is O ([Y /G]2 ). Typically, Y is
1 percent of G, and the additional pressure term is neglected.
4.2 Von Mises Yield Condition
The simplest yield criterion is known as the von Mises yield condition.
It defined by the yield function
q
3
Y(Sp , Cp ) = dev 1 (Sp ) . (4.5)

2
Cp Cp
Substituting Eqs. (4.2) and (2.2), the yield function in the spatial frame can
be written as q
Y = 32 J kdevg1 ()kg . (4.5)

4.2 Von Mises Yield Condition Elastic-Plastic Flow
q
Remark 4.4 The factor of 32 in the yield function is standard and is
chosen to simplify formula involving uniaxial strain. For uniaxial strain with
the identity spatial metric, the deviator of the strain tensor has the form
dev(e) = 3el diag[ 2, 1, 1 ], and its norm is kdev(e)k2 = 23 2el . For
a simple isotropic material dev() = 2 G0 dev(e), and the yield condition
reduces to el 2 YG0 .
Plastic flow limits the yield function to be Y . Y . It follows from the

from Eq. (4.5) that the stress deviator is limited by the yield strength. Next
we show that plastic flow also limits the deviator of the elastic strain. Sub-
stituting Eq. (4.4) for the plastic stress, the yield condition function becomes
2/3 ( 2 1/2
)
Jp
q
3 1
Y = 2
G0 Tr devI Cp C . (4.6)
J
Then the yield condition, Y Y , gives the inequality
2
J
4/3
1
Tr devI Cp C 2
3
(Y /G0 )2 .
Jp
Hence
2/3
J

Cp 1 C = IT B + O (Y /G0 ) . (4.7)
Jp
Since we are using Cp 1 C as a measure of the elastic strain, the deviator of
the elastic strain is limited by the yield condition to O (Y /G0 ). Typically,
Y is about one percent of G0 . Hence, plastic flow limits the elastic strain
deviator to be small.
The gradient of the yield surface is given by
3
Y 2
Cp devCp 1 (Sp ) Cp
= . (4.8)
Sp Y
Then from Eq. (3.5), the plastic strain rate has the form
3
d 2
Cp devCp 1 (Sp ) Cp
Cp = . (4.9)
dt Y
Using Eq. (4.4) for the plastic stress, the plastic strain rate reduces to
2/3
d G0 Jp

Cp = 32 Cp devI Cp 1 C . (4.9)
dt Y J

Elastic-Plastic Flow 4.2 Von Mises Yield Condition
Remark 4.5 The plastic strain rate equation (4.9) can be expressed as
d
Cp FT g devg1 () g)F (Cp 1 C)1
dt
Aside from the factor Cp 1 C, which to O (Y /G0 ) is proportional to the iden-
tity matrix, this is the Prandle-Reuss form for the plastic strain rate.
4.3 Hohenemser-Prager rate
It is natural to take the proportionality factor as a function of the distance

to the yield surface. The simplest possibility is the Hohenemser-Prager
rate [7] (see also, [10, p. 105]) that defines the proportionality factor as

1 (Y Y ) , for Y > Y

=

0, otherwise
where is a parameter with dimensions of dynamic viscosity (pressuretime,

commonly measured in cgs unit of Poise = micro-bars) that sets the scale
for the relaxation rate, and Y is the yield strength.
Combined with the von Mises yield, above the yield surface (Y > Y ), the
plastic strain rate is given by
d Y Y

3
Cp = 2
Cp devCp 1 (Sp ) Cp . (4.10)
dt Y
Substituting Eq. (4.4), we obtain
d Y Y

Cp = 3
2
G0 (Jp /J)2/3 Cp devI Cp 1 C . (4.11)
dt Y

It can easily be verified that Tr Cp 1 dtd Cp = 0, and hence the plastic strain
rate is volume preserving. Moreover, from Eqs. (4.11) and (4.4) the plastic
dissipation,
d Y Y
2
Sp : Cp = 3
G20 (Jp /J)4/3 devI Cp 1 C

2
dt Y I

4.3 Hohenemser-Prager rate Elastic-Plastic Flow
is explicitly positive.
For small , if the yield function were substantially to exceed the yield
surface then the plastic strain rate would be very large and would rapidly
increase Cp in order to decrease the stress and drive the yield function back to
the yield surface. However, as the stress approaches the yield surface 0,
i.e., the time constant for relaxation increases and in fact becomes infinite
on the yield surface.
4.4 Visco-Plastic Limit
We next show an important consequence of the behavior of the time

constant is that in the limit of small and slow total strain rate, the von
Mises yield strength with the Hohenemser-Prager plastic strain rate reduces
to a visco-plastic model [23], i.e., time independent elastic-plasticity plus
a viscous component to the shear stress. To obtain this result we need to
analyze the behavior of the proportionality constant for the plastic strain
rate.
We begin by computing the time derivative of the proportionality con-
stant.
d 1 d Y d 2
= Y = ln Y
dt dt 2 dt
Y d
2 d

= ln Tr devI Cp 1 C ln(J)4/3
2 dt dt

1 d 1
Tr dev C C C C

Y
I p dt p
= 2

1

Tr devI Cp C
(4.12)

1
Tr devI Cp C Cp 1 dtd C

d

2
+ 2 3
ln(J)
dt

devI Cp 1 C

Tr


Elastic-Plastic Flow 4.4 Visco-Plastic Limit
By Eq. (4.9)
d 1 d

Cp C = Cp 1 Cp Cp 1 C
dt dt
G0
= 32 (Jp /J)2/3 devI Cp 1 C Cp 1 C
Y
Then from Eq. (4.7), to leading order in Y /G0 , the first term in curly brackets
on the right hand side of Eq. (4.12) is

1
Tr devI Cp C d
C 1
dt p
C G0

2 = 32 1 + O (Y /G0 )
Tr

devI Cp 1 C Y
From Eqs. (2.14), (4.3) and (4.6) the second term in curly brackets on the
right hand side of Eq. (4.12) is

Tr devI Cp 1 C Cp 1 d
dt
C G0

1 T
2 = 3 (Jp /J)2/3 Tr FCp C devC1 (S) F d
Tr devI Cp C

1 Y 2
G0

= 3J devg1 () : d 1 + O (Y /G0 )
Y 2
From Lemma A.4 the third term in curly brackets on the right hand side of
Eq. (4.12) is
d ~ ~u .
23 ln J = 23
dt
Combining the three terms, Eq. (4.12) becomes
" #
d G0 G0 2 J
+ 32 = 32 devg (d) : devg1 () + O (Y /G0 ) .
dt Y
This form of equation is analyzed in Appendix D. In the limit 0,
Y = Y + O () and to leading order in Y /G0
2J
= devg (d) : devg1 () + O () . (4.13)
Y
Y 2 Y 2
Since 2Y
= , from Eq. (4.13) and (4.5) we obtain

3
2
J2 kdevg1 ()k2g = Y 2 + 4 J devg (d) : devg1 () + O 2

4.4 Visco-Plastic Limit Elastic-Plastic Flow
Consequently, in the limit of small , the stress can be split into an elastic
component and a viscous component
el + vis .
The elastic component el is given by Eq. (4.3) with the elastic strain such
that the elastic stress is restricted to the yield surface, i.e., rate independent
plasticity. The viscous component of the stress is given by
vis = 43 J1 g 1 devg (d) g 1 .
The viscous stress accounts for the leading order effect of plastic flow, namely,
the relaxation of the stress deviator to the yield surface.
The portion of the stress-strain work, Eq. (2.17), from the stress exceeding
the yield surface, i.e., dtd E d : el = d : vis , is dissipative. The viscous
dissipation corresponds to the plastic work from the stress exceeding the
yield surface. This is in additon to the plastic dissipation corresponding to
rate independent plasticity for which the stress is on the yield surface.
Remark 4.6 From the above derivation, the viscous approximation for
plastic flow is only valid when the viscous stress is small, i.e., k vis kg1 Y .
The visco-plastic constitutive model can be represented by a mechanical
system as shown in figure 1. The mechanical system is a composite made
up of three types of elements: (i) Spring with a linear stress-strain relation
= K, or more generally, dtd = K dtd . (ii) Dashpot with stress-strain
relation = dtd . (iii) Slider with stress-strain relation dtd = 0 if < Y .
More elaborate models allow for a history dependent relaxation to the yield
surface. The generalized models are represented by a sequence of springs
(Gi ) and dashpots (i ) elements in parallel. In effect, each spring represents
an additional internal degree of freedom. The sequence of springs and dash-
pots are an approximation to the Laplace transform of the history response
function. They allow empirical models to account for the observed frequency
dependent viscous response of materials such as polymers. The additional
internal degrees of freedom represent continuum variables needed to account
for aspects of the the short wavelength structure that are lost in coarse grain
or homogenized models of heterogeneous materials.

Elastic-Plastic Flow 4.4 Visco-Plastic Limit
Figure 1: Mechanical representation of visco-plastic constitutive model. The

springs K and G represent the response of the volumetric and deviatoric compo-
nents of the stress, respectively, to the strain. The slider (Y ) allows plastic flow
above the yield surface, and the dashpot () accounts for the relaxation of the
stress to the yield surface.
Remarks:
4.7 Plasticity is the underlying mechanism for viscosity in a solid. This
is very different than the viscous mechanisms in liquids and gases. Conse-
quently, the coefficient of viscosity for a solid can be very much larger than
for a liquid or gas; for metals viscous coefficients are typically in the range
of 100 to 1000 Poise, where as the viscous coefficient for liquids and gases is
typically measured in centi-poise.
4.8 Typically, the yield strength decreases with temperature. This is re-
ferred to as thermal softening, and can result in an instability leading to the
formation of shear layers. Morover, the limit Y 0 corresponds to visco-
elastic flow and can be used as a model for liquids. In this case, the material
does support shear strain, but only as a transient response. On a time scale
set by the ratio of the shear viscosity to the shear modulus, the shear strain
relaxes towards zero.
4.9 Viscosity and rate dependent plasticity lead to different shock wave
structures. With rate dependent plasticity, split waves occur consisting of an
elastic precursor followed by fully dispersed plastic wave. With viscosity the
shock can be partly dispersed but would not split.
4.10 For numerical simulations, rate dependent plasticity circumvents the
problems with shock waves due to the equations for rate independent plas-

4.4 Visco-Plastic Limit Elastic-Plastic Flow
ticity not being fully in conservation form. However, for a strong shock the
wave profile can be very narrow and cause problems with resolution. In ad-
dition, it should be noted that the small elastic strain approximation is not
valid in the profile of a strong shock.
4.11 The rate for crystal plasticity is determined by the motion of disloca-
tions, see for example [8], [9] and references therein. At low deformation rates
stress (d small) the plastic rate is dominated by thermal activation for the
movement of a pinned dislocation. This gives rise to the phenomena creep.
At high stress rates, in contrast, the plastic strain rate is dominated by the
drag on a dislocation as it moves through the crystal lattice. However, there
remains questions about the rate at which dislocations are nucleated and
grow. In addition, the initial number of dislocations is sensitive to impurities
and sample preparation.
4.12 The measured rise time for shock profiles in many metals can be fit
with a plastic strain rate in which (kdevg1 () Y kg )2 , see [25]. In
this case, the form in which rate independent approximation accounts for
the effect of plasticity would be different than for the Hohenemser-Prager
rate model.
4.13 The observed phenomenolgy of plastic flow is quite rich. The basic
difficulty with modeling the phenomena stems from the fact that materials
are not really homogeneous. As a result, internal variables (in addition to
the plastic strain tensor) are needed to characterize the state of the material,
along with corresponding rate equation for their dynamics. Many choices are
possible for internal variables. Common internal variables range from a sim-
ple isotropic work hardening parameter, to a back-stress tensor for kinematic
work hardening, to elaborate calibrations such as the mechanical threshold
stress (MTS) model of Follansbee & Kocks [3]. The difficulty in developing
plasticity models lies in the fact that the distribution of inhomogeneities (dis-
locations, cracks, grain boundaries), unlike the Maxwell distribution in gas
dynamics, is not necessarily highly peaked. Consequently, simple mean field
theory for coarse grained average quantities is not sufficient. As yet, there
is single no theory which can explain all the phenomena, let alone is agreed
upon by a majority of researchers.

Elastic-Plastic Flow 5 Uniaxial Strain
5 Uniaxial Strain
We chose coordinate systems with identity metrics for the body, plastic
and spatial frames; i.e., G, = , , G
e
AB = AB , and gij = ij . Uniaxial
strain reduces the flow to a one-dimensional problem by restricting the total
deformation gradient to have the form
F = diag[ exp(), 1, 1 ] ,
where = ln(V /V0 ) is a measure of the total strain, and diag[ d1 , d2 , d3 ]

denotes a diagonal matrix with diagonal elements d1 , d2 and d3 . Moreover,
~ = (u(x, t), 0, 0) are assumed to be functions of
= (x, t) and the velocity U
only one spatial variable. Consequently, the compatibility conditions for the
deformation gradient are trivially satisfied.
We further assume that the plastic deformation gradient has the form
Fp = diag[ exp( 23 p ), exp( 13 p ), exp( 13 p ) ] ,
where p is a measure of the plastic strain. The elastic deformation gradient

is then given by

1
Fe = exp 3
diag[ exp( 23 e ), exp( 13 e ), exp( 13 e ) ]
where e = p is a measure of the elastic strain. We note that J = exp()

and Jp = 1. This choice reduces the plastic strain tensor to have one inde-
pendent scalar degrees of freedom, p , compared to the six associated with a
symmetric rank-two tensor.
Remark 5.1 In general, if the material properties are anisotropic then the
elastic and plastic deformation gradients need not remain diagonal. Anisotropic
uniaxial flow remains a one-dimensional problem, i.e., functional dependence
on only one spatial variable. However, additional degrees of freedom are
needed to determine the components of the elastic and plastic strain tensors.
It follows from the form of the deformation gradients that
C = diag[ exp(2 ), 1, 1 ] ,

Cp = exp 32 p diag[ exp(2p ), 1, 1 ] .
comments to [email protected] 51 5 Uniaxial Strain

5 Uniaxial Strain Elastic-Plastic Flow
Moreover, the elastic strain is characterized by
Cp 1 C = J2/3 diag[ exp( 43 e ), exp( 23 e ), exp( 23 e ) ] .
Since diagonal matrices commute, in this case the elastic left Cauchy-Green
tensor be has the same matrix elements as Cp 1 C (though they act on dif-
ferent spaces). To leading order, the elastic strain can be expressed as

Cp 1 C = J2/3 IT B + 32 e J2/3 diag[ 2, 1, 1 ] + O 2e .
Later we will have use of the trace and the deviator of the elastic strain.
These are given by

Tr Cp 1 C = exp( 23 e ) exp(2e ) + 2 J2/3

= 3 J2/3 + O 2e ,
exp(2e ) 1
devIT B (Cp 1 C) = exp( 23 e ) J2/3 diag[ 2, 1, 1 ]
3
= 23 e J2/3 diag[ 2, 1, 1 ] + O 2e .
We note that the deviator is proportional to e .

The great simplification of uniaxial strain is that diagonal matrices com-
mute, and that the inverse of a diagonal matrix is obtained merely by invert-
ing each diagonal element. We next specialize the illustrative constitutive
model of the previous section to the case of uniaxial strain. The three parts
of the model, (i) isotropic material stress, (ii) von Mises yield condition, and
(iii) Hohenemser-Prager rate for plastic strain, are evaluated in turn.
5.1 Stress
We begin by evaluating the deviatoric term needed for the Piola-Kirchhoff

stress

devC1 Cp 1 = Cp 1 1
3
Tr Cp 1 C C1

= devIT B Cp 1 C C1

= 23 e J2/3 diag[ 2, 1, 1 ] C1 + O 2e .
5 Uniaxial Strain 52 comments to [email protected]

Elastic-Plastic Flow 5.1 Stress
Therefore the Piola-Kirchhoff stress, Eq. (4.3), is given by

h i
S = P (V, ) J IT B + G0 J2/3 devIT B Cp 1 C C1
h i
= P (V, ) J diag[ 1, 1, 1 ] + 23 e G0 diag[ 2, 1, 1 ] C1 + O 2e .
and the Cauchy stress is given by

= P (V, ) IT S + G0 J5/3 devIT S FCp 1 FT
h i
= P (V, ) IT S + G0 J5/3 FCp 1 FT 1
3
Tr(Cp 1 C) IT S
G0
1
= P (V, ) IT S + 3
[exp(2e ) 1] exp( 23 e ) diag[ 2, 1, 1 ]
J
G0
= P (V, ) diag[ 1, 1, 1 ] + 23 e diag[ 2, 1, 1 ] + O 2e .
J
The plastic stress, Eq. (4.4), is then

Sp = G0 J2/3 devIT B Cp 1 C Cp 1

= 23 e G0 diag[ 2, 1, 1 ] Cp 1 + O 2e .
The non-hydrostatic shear energy is given by
Eshear = Ee Ehydro
1/3

= 12 V0 G0 I1 I3 3
h i
1
= V0 G0
2
exp( 23 e ) exp(2e ) + 2 3 .
To leading order in e , the shear energy is given by
Eshear = 32 V0 G0 2e .
The sameh expression is obtained when the shear energy for linear elasticity,
i2
V0 G0 Tr devI (Ee ) , is specialize to the case of uniaxial strain.
2
Remark 5.2 For uniaxial strain, the strain tensor, E = diag[ J 21 , 0, 0 ],
has only one non-zero diagonal element. The stress tensor is also diagonal
but all three diagonal elements are non-zero. This is in contrast to uniaxial
stress in which only one diagonal component of the stress is non-zero, but
the strain tensor would have all non-zero diagonal elements.

5.2 Yield Condition Elastic-Plastic Flow
5.2 Yield Condition
Next we evaluate the von Mises yield function, Eq. (4.5).
Y 2 = 32 J2 kdevI ()k2I
2
= 32 G20 J4/3 FCp 1 FT 13 Tr Cp 1 C I

I
= G20 [exp(2e ) 1] 2
exp( 43 e )

= 4 2e G20 + O 3e .
Hence,
Y = 2 |e | G0 + O 2e ,
Y
and on the yield surface |e | = 2 G0
.
5.3 Plastic Strain Rate
The plastic strain rate direction for the flow rule associated with the von
Mises yield stress, Eq. (4.8), is
Y G0 2/3
= 3
2
J Cp devI Cp 1 C
Sp Y
1 G0 h i
= 2
exp(2e ) 1 exp( 23 e ) Cp diag[ 2, 1, 1 ] .
Y
This is compatible with the time derivative for the assumed form of Cp
d
Cp = 23 dtd p Cp diag[ 2, 1, 1 ] .
dt
Therefore, the plastic strain rate reduces to a scalar equation. In the plastic
regime, Y > Y ,
d Y Y G0 h
3
i

dt p
= 4
exp(2e ) 1 exp( 23 e )
Y
3 G0 sgn(e )
h i h i2
Y Y Y
=2 |e | 2 G0 exp( 3 G0 ) + O |e | 2 G0
,

where sgn(e ) = 1 or 1 when e > 0 or e < 0, respectively.

Elastic-Plastic Flow 5.4 One Dimensional Flow Equations
5.4 One Dimensional Flow Equations
The fully non-linear elastic-plastic flow equations for uniaxial strain with
isotropic constitutive relations are summarized below

+ u =0
t x

2 xx
u + u =0
t x

(5.1)
xx
E + E u u =0
t x

p + u p = Rp
t x
where E = 21 u2 + e is the total specific energy and e is the specific internal

energy. This system of equations corresponds to the one-dimensional Euler
equations for fluid flow in which the pressure is replaced with the longitudinal
component of the stress tensor xx plus the addition of a rate equation for
the plastic strain variable p .
The constitutive relations are given by
xx = P + 0

P = Phydro V, e Eshear
(5.2)
0 = 2 G0
3J
[exp(2e ) 1] exp( 32 e )
n h i o
Eshear = 12 V0 G0 exp( 23 e ) exp(2e ) + 2 3
where J = V /V0 and e = ln(V /V0 ) p . This material model corresponds to

adding a simple shear strength to a hydrostatic pressure in a thermodynam-
ically consistent manner. To leading order, the shear stress and shear energy
are given by
4 e G0
0 = 3
+ O 2e ,
J

Eshear = 23 V0 G0 2e + O 3e .

5.4 One Dimensional Flow Equations Elastic-Plastic Flow
The yield function and rate for the plastic strain variable are given by
Y = G0 |exp(2e ) 1| exp( 23 e )
(Y Y )+ G0 h i (5.3)
Rp = 3
4
exp(2e ) 1 exp( 23 e )
Y
where the subscript + denotes the positive part, i.e., f+ = f for f > 0 and
0 otherwise. To leading order the yield function and plastic strain rates are
given by

Y = 2 |e | G0 + O 2e ,
G0 sgn(e ) h i h i2
Rp = 3
2
|e | Y
2 G0 +
exp( 3 YG0 ) +O |e | Y
2 G0
.

Remarks:
5.3 For uniaxial elastic flow, the Hugoniot jump conditions, Eq. (2.26),
reduce to the standard shock equations for a fluid with P replaced by xx .
Moreover, the differential thermodynamic relation, Eq. (2.1), reduces to
de = Sxx dExx + T d = xx dV + T d .
Again this is the same as for a fluid with P replaced by xx . Consequently,

e
xx = V and the analogy between uniaxial elastic flow and one-dimensional
fluid flow is complete.
5.4 The system of equations for uniaxial elastic-plastic flow has the same
form as those for reactive flow. Plasticity corresponds to an endothermic re-
action. Hence, the phenomena of a detonation wave does not occur. Instead,
the onset of plastic flow has an effect on shock waves similar to a phase tran-
sition; i.e., split wave occur consisting of an elastic precursor followed by a
plastic wave. However, unlike a phase transition the plastic flow is dissipative
and hence irreversible. This shows up in experimental wave profiles in which
the split wave is followed by a release wave resulting from a decrease in the
driving pressure.
5.5 In the limit of small e , shear energy affects the pressure by an amount
2
P = Eshear = 23 G0 ,
V
0 e

Elastic-Plastic Flow 5.4 One Dimensional Flow Equations
where is the Gruneisen coefficient. In the elastic regime, Phydro K0 e .

G0
Typically, 1 and G0 . K0 . Hence, P/P K 0
e = O (Y /K0 ) and
consequently the shear energy has a small effect. Frequently, the shear energy
is neglected in numerical simulations.
The rate depend plasticity equations are in conservation form. The steady
waves of interest are partly dispersed shock waves. They are composed of
a discontinuity in which the stress exceeds the yield surface, followed by a
relaxation zone in which the stress returns to the yield surface. The dis-
continuity satisfies the usual Hugoniot jump conditions. Since the plastic
strain obeys a rate equation, the jump in the plastic strain variable, p , is
zero. When the state ahead of the wave is isotropic, the plastic strain at the
end state of the dispersed wave is determine by the yield condition. In this
special case, one could use rate independent plasticity if the structure of the
shock profile is not of interest.
More generally, if the material is precompressed in one direction (say x)
and subsequently shocked in a different direct (say y) then the plastic variable
behind the wave is determined by the dynamics and not solely by the yield
surface. In this case rate independent plasticity is ill-posed. Precompression
results in an anisotropic state. For anisotropic constitutive properties, either
the stress-strain relation or the yield surface, the rate depend plasticity for-
mulation is needed to resolve the ambiguity in the plastic strain tensor for
the end states of wave locus.
For very strong waves, the yield function can exceed the yield strength
by a large amount within the very narrow wave profiles. in this regime,
the small elastic strain approximation breaks down. the constitutive equa-
tions to leading order in e , while a valid approximation for small e , are
not thermodynamically consistent for large e . this inconsistency may cause
instabilities in the wave profile for strong waves. Though the fully non-linear
constitutive relations (5.2) and the plastic strain rates (5.3) appear compli-
cated, they are easily evaluated numerically, and the undesirable side-effects
of thermodynamic inconsistencies can readily be avoided.

6 Wave Structure Elastic-Plastic Flow
Figure 2: Elastic-plastic Hugoniot locus. Black line corresponds to weak elastic

shocks. Green dotted line corresponds to unstable shock waves. Solid circle marks
the Hugoniot elastic limit (HEL). Green dashed line is the segment of the Rayleigh
line through the HEL to the start of the stable strong shock waves, which is marked
by an open circle. Blue line corresponds to the second (plastic) shock of the split
waves that replace the unstable shocks. Red line corresponds to strong plastic
waves.
6 Wave Structure
The elastic-plastic flow equations have a more complex wave structure

than the usual discontinuous shocks that occur in the solution to the Euler
equations with a convex equation of state. The transition from elastic to
plastic flow has a similar effect as a non-convex equation of state due to a
phase transition. In both cases there is a discontinuity in the sound speed
at the phase boundary that gives rise to split waves. Furthermore, the
source terms in the equation for the plastic strain have a similar effect as
occurs when the Euler equations are extended to acount for a relaxation
phenomena. As a result shock waves in the plastic regime are either fully- or
partly-dispersed and not discontinuous.
The uniaxial strain model of the previous section is used to illustrate
the wave structures that can occur in the solution to the elastic-plastic flow
equations. With parameters in the constitutive relations corresponding to
HMX (the explosive cyclo-tetramethelene-tetranitromene but treated here
as a non-reactive solid) the principal Hugoniot locus is shown in figure 2
6 Wave Structure 58 comments to [email protected]

Elastic-Plastic Flow 6 Wave Structure
projected in the (V, xx )-plane and in the (up , us )-plane.

Next, the profiles for each type of wave, driven by a piston in a numerical
simulation, are illustrated. For the calulations the AMRITA environment
developed by James Quirk [17, 18] is used in conjunction with an EOS plu-
gin [13] for the constitutive properties. A second order (both space and
time) Godunov algorithm in Lagrangian coordinates is used. Moreover, an
adaptive mesh refinement solver allows the profiles to be resolved efficiently.
6.1 Weak Purely Elastic Wave
Figure 3: Wave profile for weak elastic wave.
Weak shock waves are in the purely elastic regime, i.e., the plastic strain
rate is zero. The shock profile is discontinuous and corresponds to an ordinary
fluid dynamic shock wave.
comments to [email protected] 59 6 Wave Structure

6.2 Split Elastic-Plastic Wave Elastic-Plastic Flow
6.2 Split Elastic-Plastic Wave
Figure 4: Wave profile for split elastic-plastic wave.
As the wave strength increases above the Hugoniot elastic limit shown
in figure 2, the shock speed decreases and the shock wave is unstable. It
breaks up into a split wave. The lead wave is purely elastic and takes the
material upto the yield surface. Plastic flow occurs in the following wave.
The limiting of the shear stress above the yield surface results in a discon-
tinuous decrease in the longitudinal sound speed. Consequently, the plastic
wave speed is less than the speed of the lead wave, which is at the Hugo-
niot elastic limit. We show in a subsequent subsection that the plastic wave
is fully dispersed. That is to say, the plastic flow provides all of the dissi-
pation required by the Hugoniot jump conditions and the plastic profile is
continuous.

Elastic-Plastic Flow 6.3 Strong Plastic Wave
6.3 Strong Plastic Wave
Figure 5: Wave profile for strong plastic wave.
As the wave strength is further increased, the speed of the plastic wave
increases until it overtakes the elastic wave. The split waves are then re-
placed by strong plastic waves. The profile of a strong plastic wave is partly
dispersed and consists of a lead shock followed by a continuous relaxation
region to the state on the Hugoniot locus. As seen in figure 5, across the
lead shock the plastic strain is continuous but the elastic strain is discontin-
uous. The plastic flow provides only part of the dissipation required by the
Hugoniot jump conditions.

6.4 Visco-Elastic Wave Elastic-Plastic Flow
6.4 Visco-Elastic Wave
Figure 6: Wave profile for viscous wave, Y = 0.
An intesting special case occurs when the yield strength is zero. This
corresponds to a visco-elastic material. In this case there are no split waves.
Weak waves are fully dispersed as in the plastic wave of a split wave, and
strong waves are partly dispersed as in strong plastic waves. The fully dis-
persed waves have a profile simpilar to what occurs in fluid dynamics when
a viscous stress, Q = x u, is added to the pressure. However, there is an
important difference for the numerics. The visco-elastic model is hyperbolic
and the stable time step is the minimum of the CFLcondition and the time
x
constant of the plastic strain rate, i.e., t < min c
, . In contrast a
viscous shock profile arises from a parabolic model and the stable time step
2
for an explicit algorithm is t < (x)

. As the resolution is increase the vis-
cous time step is proportional to (x)2 , whereas the visco-elastic time step
is proportional to x. Consequently, it is computational less expensive to
resolve shock profiles with a visco-elastic model.

Elastic-Plastic Flow 6.5 Partly and Fully Dispersed Waves
Figure 7: Hugoniot loci in (V, P )-plane. Blue curve is frozen locus, red curve is
equilibrium locus. Dotted and dashed black lines are the Rayleigh line for strong
plastic wave and second shock of split wave, respectively.
6.5 Partly and Fully Dispersed Waves
The two types of plastic wave profiles, fully- and partly-dispersed, can be
understood by examining the frozen and equilibrium Hugoniot loci shown in
figure 7. For the frozen locus the plastic strain is held fixed, while for the
equilibrium locus the plastic strain is determined by the condition that the
stress lies on the yield surface. Also shown in the figure are the Rayleigh
lines for weak and strong plastic waves. The Rayleigh line is the cord in
the (V, xx )-plane from the initial to final shock states. It follows from the
jump conditions that its slope is proportional to the square of the wave speed.
Moreover, for a steady traveling wave, the projection of the wave profile in
the (V, xx )-plane lies on the Rayleigh line.
For a weak plastic wave, the Rayleigh line (black dotted line in figure 7)
lies between the equilibrium and frozen Hugoniot loci. Consequently, a shock
is not possible and the wave profile must be continuous. In contrast, for a
strong plastic wave, the Rayleigh line (black dashed line in figure 7) intersects
the frozen Hugoniot locus. An analysis of the ODEs for the steady wave
profiles shows that the plastic flow doesnt provide enough dissipation at the

6.5 Partly and Fully Dispersed Waves Elastic-Plastic Flow
initial state for the profile to follow the Rayleigh line. Instead, the profile
consists of a shock to the point at which the Rayleigh line intersects the
frozen Hugoniot locus followed by a continuous variation along the segment
of the Rayleigh line to the final state on the equilibrium Hugoniot locus.
For the visco-elastic model, the elastic segment on which the frozen and
equilibrium loci overlap shrinks to zero. A fully dispersed wave profile occurs
when the wave speed lies between the equilibrium and frozen sound speeds.
A partly dispersed profile occurs for stronger waves with a wave speed greater
than the frozen sound speed.
6.6 Entropy Anomalies
An important consequence of the length scale associated with a dispersed

wave profile is that an entropy anomaly results from any transient in which
a wave profile is formed or changes [12]. The piston driven waves in the
examples of the previous subsections serve to illustrate the anomaly when a
wave profile is formed. Moreover, the reflection of the wave off the wall at
the end of the domain serves to illustrate the anomaly when the wave profile
changes.
For the visco-elastic case of the previous subsection the wave profiles of
the stress, particle velocity and entropy are shown in figure 8 at a time well
after a steady-state shock profile has been reached. Behind the wave front we
see that the stress and particle velocity are constant. The entropy anomaly
is illustrated by the large increase in entropy near the piston. We note that
the extent of the anomalous entropy region is on the same order as the width
of the wave profile.
Lagrangian time histories of the stress for a point near the piston and a
point overtaken by the steady-state wave are shown in figure 9. Since the
piston is impulsively started, the stress rises much more rapidly, near the
piston than in the steady-state wave. The rapid rise in stress generates more
entropy, and explains the sign of the entropy anomaly shown in the profile
of figure 8.
For the Euler equations, this entropy anomaly in numerical solutions
would be a startup error associated with the artificial shock width of a
capturing algorithm. Here the wave profile is determined by the material

Elastic-Plastic Flow 6.6 Entropy Anomalies
model and is fully resolve. The anomaly is in fact part of the solution to the
PDEs. It would not shrink in spatial extent with a finer grid.
Figure 8: Wave profiles of stress, velocity and entropy for impulsively started
piston driven visco-elastic wave.
A) x(0) = 0.1 mm B) x(0) = 1.0 mm

Figure 9: Time history of stress for Lagrangian points showing incident shock for
visco-elastic material. A) Transient startup for second cell from piston. B) Steady
state shock profile.

6.6 Entropy Anomalies Elastic-Plastic Flow
The reflection of a wave off a wall is equivalent to the symmetric collision

of two waves of the opposite family. Thus, the reflection of a shock serves
to illustrate the affect when waves interact. The wave profiles for the visco-
elastic case are shown in figure 10 at a time well after the lead shock has
reflected off the wall at the end of the computational domain. Behind the
reflected wave front, again, the stress and particle velocity are constant.
The startup entropy anomaly persists since in that region entropy is merely
advected. Now, however, there is an additional entropy anomaly at the wall.
Lagrangian time histories of the stress for a point near the wall and a
point overtaken by the steady-state wave are shown in figure 11. Away from
the wall, the time history shows a double rise, corresponding to the passage of
the incident shock and then the reflected shock. Near the wall, however, the
incident and reflected wave profiles overlap. Consequently, the two shocks
are not distinguishable and there is no plateau in the stress history. Instead
the stress history has an appearance similar to that of a single shock to the
final pressure.
The fact that the anomaly is an entropy increase is compatible with a
general property of shock waves, namely, the triple shock entropy theorem
[4]. The theorem states that the entropy behind a sequence of two shocks
is less than the entropy behind a single shock to the same final pressure.
The profile near the wall is expected to be more spread out than a single
steady shock. Consequently, the entropy behind a single shock should be an
upperbound on the entropy anomaly.
A similar entropy anomaly can occur in the numerical solution of the
Euler equations. It is exemplified by the Noh problem [14], and often refered
to as excess wall heating. We note that at fixed pressure, the energy
increases with entropy. Thus, the excess heating is cause by the entropy
anomaly. Excess heating results in a higher temperature. For reactive flows,
the excess heating, if sufficiently large, could initiate a detonation wave and
thus complete change the solution to the problem.
With a dispersed wave, the entropy anomaly is part of the solution to
the PDEs and would not change once the wave profiles are resolved. The
anomaly is a numerical error when the width of the wave is artificial, such as
occurs in a shock capturing algorithm. The numerical error can be thought
of as due to the artificial length scale, and mimics a physical effect when the
width of the wave is determined by the material model.

Elastic-Plastic Flow 6.6 Entropy Anomalies
Figure 10: Wave profiles of stress, velocity and entropy for visco-elastic wave after
it reflects from wall at x = 5.
A) x(0) = 4.0 mm B) x(0) = 4.9 mm

Figure 11: Time history of stress for Lagrangian points showing incident and
reflected shock for visco-elastic material. A) Steady state profiles for both incident
and reflected shocks. B) Transient next to wall in region where incident and
reflected shock profiles overlap.

Appendix A Elastic-Plastic Flow
Appendix A: Useful Lemmas

Lemma A.1.
C
(det C) = (det C)C1

Lemma A.2. (Q : C )C1 = C1 Q C1

Lemma A.3. d
dt
det C = 0 iff Tr C1 dtd C = 0 iff d
dt
C = devC ( dtd C).

~ ~u .
Lemma A.4. J1 dtd J = Tr g 1 d =
Lemma A.5. The eigenvalues of Q C Q1 are the same as those of C for

any invertible matrix Q.
Lemma A.6. If G is a positive symmetric matrix then it can be written in
the form G = Q Q where Q is positive and symmetric.
Appendix B: Compliance and Stiffness Tensors

The stress S and strain E are of symmetric rank-two tensors. They are
linearly related by the rank 4 stiffness and compliance tensors, C and S,
respectively,
S = C:E and E = S:S .
,
i.e., S = C E where , , , = 1, 2, 3, and similarly E = S, S .
By introducing Voigt notation the stress and strain tensors can be treated
as vectors, and the stiffness and compliance tensors as matrices. Since the
stress and strain tensors can be represented by a symmetric matrix they have
six independent entries that can be labeled as follows:

1 6 5
Voigt indices = 6 2 4 ,

5 4 3
i.e., the mapping from a Voigt index to a tensor index is given by V (1) =
1, 1; V (2) = 2, 2; V (3) = 3, 3; V (4) = 2, 3; V (5) = 3, 1; V (6) = 1, 2. The Voigt
vector corresponding to the strain tensor is defined as

~E = (E11 , E22 , E33 , 2 E23 , 2 E31 , 2 E12 )T ,
Appendix B 68 comments to [email protected]

Elastic-Plastic Flow Appendix B
and the vector corresponding to the stress tensor is

~S = (S11 , S22 , S33 , 2 S23 , 2 S31 , 2 S12 )T .

The 2 is introduced in order that S : E = ~S ~E.
Remark B.1 The standard convention is
~Sstd = (S11 , S22 , S33 , S23 , S31 , S12 )T ,

~Estd = (E11 , E22 , E33 , 2 E23 , 2 E31 , 2 E12 )T .
This non-symmetric treatment of stress and strain has side affects on the
compliance matrix. It is not suited to the spectral decomposition of the
elastic tensor. Also, it affects the coefficients in the formulae for the Reuss
average shear modulus.
V (a),V (b)
The stiffness tensor can be expressed as a 6 6 matrix C b
a,b = C
where a, b = 1, 2, 3, 4, 5, 6. Similarly, for the compliance tensor Sa,b = SV (a),V (b) .
b
Both C
b and S b are symmetric matrices. From the symmetry, we obtain
! !
~S = I3 0 b I3 0
C ~E ,
0 2 I3 0 2 I3
where I3 is the 3 3 identity matrix. Similarly,

! !
~E = I3 0 b I3 0
S ~S .
0 2 I3 0 2 I3
It follows that ! !
b = I3
S
0 b 1 I3
C
0
,
1 1
0 I
2 3
0 I
2 3
whereas the original 4th rank tensors satisfy the relation

,
S, C = 1
2
+ .
For linear-elasticity, C is constant and the energy is given by
E = 12 E : C : E = 12 Tr(SE) .
comments to [email protected] 69 Appendix B

Appendix B Elastic-Plastic Flow
Alternatively, we could have started by identifying a rank-two tensor S

with a 9 dimensional vector S

S1 S6 S5
S
S9 S2 S4 .

S8 S7 S3
Then symmetric rank-two tensors are mapped into Voigt vectors by the 6 9
transformation matrix
!
I3 0 0
U= 1 I3 1 I3
.
0 2 2
The transformation has the following properties UUT = I6 and UT U =

sym(I9 ). Consequently, the map is an isometry from symmetric rank-two
tensors to vectors of rank 6, while the antisymmetric rank-two tensor are in
the kernel (null space) of U.
We note that ~S = US. Furthermore,
S = UT CU
b E ,
b
where the effective 6 6 stiffness matrix is given by

! !
I3 0 b I3 0
C
b = C .
b
0 2 I3 0 2 I3
Let the 6 6 Hermitian matrix C

b have the spectral decomposition
b
6
X
C
b = i~ei ~ei .
b
i=1
Then the spectral decomposition of the rank 4 stiffness tensor, see for example
[27], is given by
6
X
C= i ei ei ,
i=1
where ei = UT ~ei is the symmetric rank-two tensor corresponding to the Voigt

vector ~ei .

Elastic-Plastic Flow Appendix B
About the ambient equilibrium state, E = 0 and S = 0, the spectral

decomposition of the stiffness tensor gives the leading order term in the
expansion of the stress-strain relation
6
X
S= i (E : ei )ei ,
i=1
and the specific energy

6
1 X
E = E0 + V0 i (E : ei )2 .
2 i=1
At a stable equilibrium point, the energy must be a minimum. Hence, the

i are all positive. From linear algebra, a necessary and sufficient coefficient
for a Hermitian matrix, C, b to be positive is that the determinants of all the
b
principle minors are positive. This places constraints on the coefficients of

the stiffness tensor at a stable equilibrium state.
Remark B.2 A stiffness tensor with low symmetry, for example correspond-
ing to a monoclinic crystal, does not necessarily have I3 as an eigenfunction
of C. In this case, the standard decomposition of the specific energy as the
sum of volumetric and deviatoric components, E(E) = E1 (V ) + E2 (dev E), is
not possible.
The spectral decomposition also provides bounds on the acoustic wave
speeds. For propagation in the direction n, the wave speed c for the ambient
equilibrium state satisfies the condition that c2 is an eigenvalue of the 3 3
Hermitian matrix n C n. (In general, the matrix in Eq. (C.1) needs to
be considered.) By the Rayleigh-Ritz principle the maximum sound speed is
given by
h i
c2max = max m n : C : m n
m

h i sym(m n) : C : sym(m n)
max || sym(m n)||2 max
m m || sym(m n)||2

(~e, C~
b e)
b
max
~e ||~e||2
max ,

Appendix B Elastic-Plastic Flow

1
where sym(m n) = 2
m n + n m and we have used the fact that
h i
|| sym(m n)||2 = 1
2
1 + (m n)2 1 .
Similarly, the minimum sound speed is given by

h i
c2min = min m n : C : m n
m

h i sym(m n) : C : sym(m n)
min || sym(m n)||2 min
m m || sym(m n)||2

(~e, C~
b e)b
12 min
~e ||~e||2
12 min ,
where we have used the fact that

h i
|| sym(m n)||2 = 1
2
1 + (m n)2 1
2
.
Therefore, independent of the direction of propagation, the sound speeds are

bounded by
1

2 min
c2 max .
For an isotropic material, the lower bound is sharp for 3K > 2G (Poisson
ratio > 0) and for 3K = 2G ( = 0) both bounds are sharp.
B.1 Isotropic Material
The specific energy of a linearly elastic isotropic material is given by
1
0 E = (Tr E)2 + G||E||2
2
1
= K(Tr E)2 + G|| dev E||2 ,
2
where = K 32 G is the Lame coefficient, K is the bulk modulus and G is

the shear modulus.

Elastic-Plastic Flow B.1 Isotropic Material
Then the stress-strain relations have the simple form (which is a special
case of the spectral decomposition with eigenvalue 3 K and fivefold degener-
ate eigenvalue 2 G)
S = K(Tr E)I3 + 2G dev E

1 1
E= (Tr S)I3 + dev S .
9K 2G
The stiffness and compliance tensors are given by
,
C = + G( + ) ,

1 1

1 (B.1)
S, = + ( + ) .
9K 6G 4G
Here for simplicity we have assumed that the metric is the identity matrix,
i.e., G = . The corresponding Voigt matrices have the form
! !
b = Cu 0
C and b = Su 0
S ,
0 C` 0 S`
where the u block represents the indices 1, 2, 3 and the ` block represents
the indices 4, 5, 6. The components of the stiffness matrix are given by
K + 43 G K 32 G K 32 G

2 4 2

K 3 G K + 3 G K 3 G
Cu = and C` = G I3 ,
K 23 G K 23 G K + 34 G
and the components of the compliance matrix are given by
1 1 1 1 1 1

9K
+ 3G 9K 6G 9K 6G
1 1 1 1 1 1
1
Su =
9K 6G 9K
+ 3G 9K
6G
and S` = I3 .
1 1 1 1 1 1
4G
9K
6G 9K
6G 9K
+ 3G

B.2 Isotropic Sound Elastic-Plastic Flow
B.2 Isotropic Sound
Let N be the direction of propagation for an acoustic wave. From the

isotropic stiffness tensor Eq. (B.1), we obtain

C = N C N = G + ( + G)N N .
In the ambient equilibrium state the acoustic wave speeds, c2 , and the
direction of the displacement vectors are the eigenvalues and eigenfunctions
of C. It is easy to verify that N is an eigenvector with eigenvalue + 2 G =
K + 34 G. This corresponds to a longitudinal sound wave. Furthermore, any
T orthogonal to N is an eigenvector with eigenvalue G. These correspond to
transverse shear waves.
As expected for an isotropic material, the wave speeds are independent
of the direction of propagation. Moreover, in the plane orthogonal to the
propagation direction, there is no preferred direction for the transverse wave.
Consequently, the eigenvalue for a shear wave is doubly degenerate. We also
note that the longitudinal sound speed is greater than the transverse shear
wave speed.
B.3 Isotropic Averages
A coarse grain average of a polycrystalline material results in an isotropic

response when the crystals are randomly oriented. Two approximations are
instructive. Assuming a uniform strain leads to the Voigt average [29] and
represents an upper bound [6], while assuming uniform stress leads to the
Reuss average [19] and represents a lower bound [6].
The Voigt average corresponds to averaging the stiffness tensor over all
orientations
,
Z
d 0 0 ,0 0
hC iV = R 0 R 0 C (RT )0 (RT )0 ,
4
where R is the rotation matrix corresponding to the solid angle . Since the
measure is invariant under the rotation group (Haar measure on a compact

Elastic-Plastic Flow B.3 Isotropic Averages
group), the average projects out the isotropic representation:

K G
C : C K C : C G
hCi = 2 C + 2 C ,

K G
C C

where
K
(C ), =
G 2
(C ), = + + .
3
It follows that the Voigt average of the bulk and shear moduli are given by

1
KV = 9
C
b +C
11
b +C
22
b +2 C
33
b +C
12
b +C
23
b
31

1 b C
GV = 15
C
b +C
11
b +C
22 33
b +C
12
b +C
23
b
31 + 3 C44 + C55 + C66
b b b .
The Voigt average bulk modulus KV also corresponds to the hydrostatic

modulus, 91 Tr(C : I3 ), i.e., E I3 .
Similarly the Reuss average corresponds to averaging the compliance ten-
sor over all orientations. The average compressibilities are given by
1

= S b +S
11
b +S
22
b +2 S
33
b +S
12
b +S
23
b
31
KR
1

4 b
= 15 S11 + S22 + S33 S12 + S23 + S31 + 3 S44 + S55 + S66 .
b b b b b b b b
GR
Remarks:
1
B.3 Typically, the Reuss average is written in terms of C
b rather than S.
b
With this notation the last 3 terms in the formula for GR are multiplied by
a factor of 43 instead of 3.
B.4 The sound speeds computed from the Voigt and Reuss averages, c2long =
K + 43 G and c2trans = G, are upper and lower bounds for average material
but not necessarily on the sound speed for a given propagation direction.
The spectral decomposition provides bounds on the latter.

Appendix C Elastic-Plastic Flow
Appendix C: Convexity of Specific Energy

The PDEs for elastic flow are hyperbolic if and only if the acoustic wave
speeds are real. From Eq. (2.20) this is equivalent to the matrix
i 0 0 j0
A(N ) j = Fi 0 N C N (FT ) 0 gj 0 j + (N S N ) i j
having positive eigenvalues for all N . Here we formulate the hyperbolicity

condition in terms of a requirement on the specific energy E(E, ).
By Lemma A.5 the condition for real sound speeds is equivalent to the
matrix
00 i 00 0 00
(F1 ) i A(N ) j Fj 00 = N C N C 0 00 + (N S N ) 00 (C.1)
having positive eigenvalues. From the definitions of the stress tensor S and
the stiffness tensor C, the hyperbolicity condition can be expressed as the
condition on E that the matrix
2E E

1
N N C + 2
N N ITX~ B (C.2)
CC C
has positive eigenvalues for all N . Alternatively, elastic flow is hyperbolic if
and only if
2E E ~ ~ 2

~ ~
(N M ) : ~ ~ 1 ~
: (N M ) > 2 N N M (C.3)
CC C C
~ T ~ B where C is the metric used for inner products. It is

~ ,N
for all M X
natural to use C as the metric in the body frame since it is the sound speeds
in the spatial frame that are physical, see Remark 2.13.
E
When the stress S = 20 C vanishes, this is equivalent to rank-1 con-
~
vexity. Namely, E E + t [N M ~ +M ~ N ~ ] is a convex function of t.
It is the analog of the hyperbolicity condition for fluid dynamics, that is,
2E
(c)2 = V 2 > 0 if and only if E(V, ) is a convex function of V . We note
that stress is negative in compression and causes the hyperbolicity condition

to be stronger than rank-1 convexity.
A slightly stronger condition than rank-1 convexity is
2
(N ~ ) : E : (N
~ M ~ M
~ ) kN k kM k , (C.4)
C C
CC
Appendix C 76 comments to [email protected]

Elastic-Plastic Flow Appendix C
E 2
where > 0 and again C is used for the metric. When CC is replaced by
the acoustic tensor A this is the condition of strong ellipticity. Then can
be identified with the minimum of c2 over all directions and polarizations.
Consequently, for Eq. (C.4), is the minimum over all directions N of 41 (c2
V0 Sn ) where Sn = N S N is the normal stress. Strong ellipticity is a
condition needed for existence of solutions in elasto-statics, for example, see
[11]. It is also sufficient for hyperbolicity.
Appendix D: Small Viscosity Limit

The rate equation has the form
d A B(t)
+ = .
dt
This can be written as
d B(t)

exp At

= exp At

.
dt
The solution is given by
Z t dt0 0
(t) = B(t0 ) exp A(tt) .
0
In the limit of small , the exponential is highly peaked. If B(t) varies slowly
on the time scale /A, i.e.,
d
BB ,
A dt
then to leading B can be taken out of the integral, yielding
Z
dt0 t 0
(t) B(t) exp A(tt)
0
Z 0
dt0 0 B(t)
B(t) exp At =
A
Consequently, in the limit of small viscosity
B(t)
(t) = + O () .
A
comments to [email protected] 77 Appendix D

Appendix E Elastic-Plastic Flow
Appendix E: Lie Derivative

To illustrate the concept of a Lie derivative we begin by considering the
particle acceleration, which formally is defined by
d ~

~a = ~u = U 1 .
dt t
The difficulty is that the standard definition of the derivative,
~ ~ (X,
U ~ t + t) U
~ (X,
~ t)
U = lim ,
t t0 t
(X,t+t) S and U (X, t)

is not well defined because U~ (X,
~ t + t) T ~ ~ ~
T(X,t)
~ S. Though the tangent spaces TX~ S are finite dimensional, and hence
isomorphic, the isomorphism is not unique.
In this case, we can use the pull-back and push-forward maps of the flow
to provide the needed isomorphism between the tangent spaces.
TX~ B
....... .....
..... .....
.
.....
.
..
.
.....
.....
.....
.....
.....
.....
.... .....
..... .....
..... ......
..
...... ........
..
T(X,t+t)
~ S ............................................................. T(X,t)
~ S
Then the time derivative, defined as,
~ ~ t) (X,
(X, ~ t + t)U
~ (X,
~ t + t) U
~ (X,
~ t)
U = lim ,
t t0 t
is well defined. On a flat space, metric gij = ij , this gives the standard
definition for the convective derivative, dtd = t

+ ~u .
This idea can be generalized to be independent of the reference frame B.
First, for a vector field ~u(~x) T~x S we define a flow, ~u , by the differential
equation

(~x, t) = ~u ~u (~x, t)
t ~u
with the initial condition ~u (~x, 0) = ~x. Then the push-forward maps
the tangent space at ~x onto the tangent space at (~x, t), and the pull-back
Appendix E 78 comments to [email protected]

Elastic-Plastic Flow Appendix E
maps the tangent space at (~x, t) onto the tangent space at ~x. The Lie
derivative of a tensor field, see e.g., [11, 1.6], such as the Eulerian strain,
is given by
d

L~u e = ( ~u ) (~x, t) ( ~u ) e(~x, t) .
dt
In effect, one transforms into a common vector space in order to take the
derivative and then transforms back to the desired space.
The Lie derivative allows evolution equations to be defined in a mannar
such that frame indifference is satisfied, see [11, Box 6.1, p. 99]. In particular,
we have defined the plastic strain rate in the Lagrangian frame. It can then
be transformed into the Eulerian frame. The theory is physically the same
when viewed in any inertial frame.
Acknowledgement
The author wishes to thank his colleauge Bradley Plohr. Discussion with
him on elastic-plastic flow have greatly contributed to the authors under-
standing and is the basis for these notes.
References
[1] J. K. Dienes, On the analysis of rotation and stress rate in deforming
bodies, Acta Mechanica 32 (1979), 217232. 16, 25
[2] T. C. Doyle and J. L. Ericksen, Nonlinear elasticity, Advances in Me-
chanics IV, Academic Press, 1956. 12
[3] P. S. Follansbee and U. F. Kocks, A constitutive description of the de-
formation of copper based on the use of the mechanical threshold stress
as an internal state variable, Acta Metall. 36 (1988), 8193. 50
[4] L. F. Henderson and R. Menikoff, Triple shock entropy theorem and its
consequences, J Fluid Mech 366 (1998), 179210. 66
[5] W. Herrmann, Elastic-plastic constitutive relastions at large strain,
Shock Compression of Condensed Matter 1991 (S. C. Schmidt, R. D.
comments to [email protected] 79 References

References Elastic-Plastic Flow
Dick, J. W. Forbes, and D. G. Tasker, eds.), Elsevier Science Publishers,

1992, pp. 379382. 25
[6] R. Hill, The elastic behaviour of a crystalline aggregate, Proc. Roy. Soc.
London A65 (1952), 349354. 74
[7] K. Hohenemser and W. Prager, Beitragzur mechanik desbildsamen ver-

haltensvonflubstahl (Contribution to the mechanics of the malleable be-
havior of cast iron), Z. Angew. Math. Mech. 12 (1932), 114. 45
[8] J. N. Johnson, Micromechanical considerations in shock compression

of solids, High-Pressure Shock Compression of Solids (J. R. Asay and
M. Shahinpoor, eds.), Springer-Verlag, 1993. 50
[9] J. N. Johnson and D. L. Tonks, Dynamic plasticity in transition from

thermal activation to viscous drag, Shock Compression of Condensed
Matter 1991 (S. C. Schmidt, R. D. Dick, J. W. Forbes, and D. G.
Tasker, eds.), Elsevier Science Publishers, 1992, pp. 371378. 50
[10] J. Lubliner, Plasticity theory, Macmillan Publishing, 1990. 45
[11] J. E. Marsden and T. J. R. Hughes, Mathematical foundations of elas-

ticity, Printice-Hall, 1983. 3, 6, 25, 29, 77, 79
[12] R. Menikoff, Errors when shock waves interact due to numerical shock
width, SIAM J Sci Comp 15 (1994), 12271242. 64
[13] R. Menikoff and J. J. Quirk, Versatile user-friendly EOS pack-

age, Tech. Report LA-UR-01-6222, Los Alamos National Labo-
ratory, 2001, http://t14web.lanl.gov/Staff/rsm/preprints.html
#EOSpackage. 59
[14] W. F. Noh, Errors for calculations of strong shocks using artificial vis-
cosity and an artificial heat flux, J Comp Phys 72 (1987), 78120. 66
[15] B. J. Plohr and D. H. Sharp, A conservative Eulerian formulation of the

equations for elastic flow, Advances in Applied Mathematics 9 (1988),
481499. 2
[16] , A conservative formulation for plasticity, Advances in Applied

Mathematics 13 (1992), 462493. 21, 32
References 80 comments to [email protected]

Elastic-Plastic Flow References
[17] J. J. Quirk, Amrita - a computational facility for cfd modelling, p. ?,

von Karmen Institute, 1998, http://www.galcit.caltech.edu/jjq/
vki/vki:cfd29::jjq l1.ps.gz. 59
[18] , AMR sol: Design principles and practice, p. ?, von Kar-
men Institute, 1998, http://www.galcit.caltech.edu/jjq/vki/
vki:cfd29::jjq l2.ps.gz. 59
[19] A. Reuss, Berechnung der fliebgrenze von mischkristallen auf grund
der plastizitatsbedingung fur einkristalle (Calculation of the plastic flow
boundary of mixed crystals based on the plasticity condition for single
crystals), Z. Angew. Math. Mech. 9 (1929), 4958. 74
[20] D. G. Schaeffer, Amathematical model for localization in granular flow,
Proc. R. Soc. London (1992), 217250. 35
[21] M. Scheidler and T. W. Wright, A continuum framework for finite vis-
coplasticity, International Journal of Plasticity 17 (2001), 10331085.
37
[22] J. C. Simo, A framework for finitstrain elastoplasticity based on max-
imum plastic dissipation and the multiplicative decomposition: Part I
continuum formulation, Computer Methods in Applied Mechanics and
Engineering 66 (1988), 199219. 37
[23] J. C. Simo and T. J. R. Hughes, Computational inelasticity, Springer-
Verlag, 1998. 2, 14, 46
[24] J. C. Simo and K. S. Pister, Remark on rate constitutive equations
for finite deformation problems: Computational implications, Computer
Methods in Applied Mechanics and Engineering 46 (1984), 201215. 16,
25
[25] J. W. Swegle and D. E. Grady, Shock viscosity and the prediction of
shock wave rise times, J. Appl. Phys. 58 (1985), 692701. 50
[26] G. I. Taylor and H. Quinney, The latent energy remaining in a metal
after cold working, Proc. Roy. Soc. London A143 (1933), 307326. 33
[27] P. S. Theocaris and D. P. Sokolis, Spectral decomposition of the linear
elastic tensor for monoclinic symmetry, Acta Cryst. A55 (1909), 635
647. 70
comments to [email protected] 81 References

References Elastic-Plastic Flow
[28] J. Trangenstein and P. Colella, A higher-order Godunov method for mod-

eling finite deformation in elastic-plastic solids, Comm. Pure Appl. Math
44 (1991), 41100. 22
[29] W. Voigt, Lehrbuch der krystall physik (Textbook on crystal physics),

Teubner, Leipzig, 1928. 74
References 82 comments to [email protected]

Elastic-Plastic Flow Index
Index
:, contraction, 12 J, ratio of specific volumes VV0 , 9
A, acoustic tensor, 26 K, bulk modulus, 15
b, left Cauchy-Green tensor, 6 , Lame coefficient, 14
be , elastic tensor, 7 , coef. of dynamic viscosity, 45
B, body frame, 2 kk, norm, 11
T B, tangent space, 2 P , pressure, 13
T B, cotangent space, 2 P, first Piola-Kirchhoff stress, 19
C, right Cauchy-Green tensor, 6 P, plastic frame, 2
Cp , plastic tensor, 7 T P, tangent space, 2
T P, cotangent space, 2
C , stiffness tensor, body frame, 24 , mapping, B S, 1
c, stiffness tensor, spatial frame, 24 , mass density, 17
d, rate of deformation, 22
S, compliance tensor, 25
, Kronecker symbol, 4
S, spatial frame, 2
dev, deviator, 11
T S, tangent space, 2
E, specific energy, 12, 33
T S, cotangent space, 2
Ee , elastic energy, 33
S, second Piola-Kirchhoff stress, 12
Ep , plastic energy, 33
Sp , plastic stress, 12
e, Eulerian strain tensor, 7
, Cauchy stress tensor, 12
ee , elastic tensor, 7
T , temperature, 12
ep , plastic tensor, 8
Tr (), trace, 10
E, Lagrangian strain tensor, 7
~u, particle velocity, spatial frame,
Ep , plastic tensor, 7
1
, small strain parameter, |V VV 0|
,9 ~
0 U , particle velocity, body frame, 1
, entropy, 12
V , specific volume, 9
F, deformation gradient , 2 V0 , value in body frame, 9
Fe , elastic deformation, 2
Fp , plastic deformation, 2
acoustic tensor, A , 26
f = F 1 , 21
adjoint, 6
G, shear modulus, 15
Almansi strain tensor, e, 7
g, spatial frame metric, 4
associated flow rule, 37
G, body frame metric, 4
G,
e plastic frame metric, 7
bulk modulus, K, 15
a
dc , Christoffel symbol, 17
I, identity operator, 5 Cauchy-Green tensor
comments to [email protected] 83 Index

Index Elastic-Plastic Flow
left, b, 6 invariants
right, C, 6 tensor, 13
characteristic polynomial, 13 inverse, 5
Christoffel symbol, 17 isotropic material, 13
Clausius-Duhem inequality, 23
compatibility condition, 20 Lame coefficient, 14
Lie derivative, 23, 79
compliance tensor, S , 25 linear elasticity, 14, 16
contravariant index, 3 loss of hyperbolicity, 30
convective derivative, d/dt, 19
cotangent space, 3 mapping, , 1
covariant differentiation, 17 metric, 4
covariant index, 3 body frame, G, 4
crystal plasticity, 3840, 50 plastic frame, G,
e 7
curvature tensor, 18 spatial frame, g, 4
multiplicative decomposition, 2
deformation gradient, F, 2
deviator, 11 norm, kk, 11
dilatancy, 30
particle velocity
Elasticity, 11 body frame, U ~, 1
hyper, 11 spatial frame, ~u, 1
hypo, 11 Piola identity, 19
linear, 14 Piola transform, 19
entropy inequality, 23 Piola-Kirchhoff tensor
first, P, 19
Finger tensor, b1 , 7 second, S, 12
flow equations Prandle-Reuss plastic strain rate,
Eulerian, 17 45
Lagrangian, 19 pressure, P , 13
frame indifference, 11
rank-1 convexity, 76
Hohenemser-Prager rate, 45 rate independent plasticity, 35, 38
Hugoniot elastic limit, 58, 60 rate of deformation tensor d, 22
Hugoniot equation, 32 Rayleigh line, 58, 63
hyper-elastic, 11 Reuss average, 74
hyperbolicity condition, 29
hypo-elastic, 11, 38 shear modulus, G, 15
small strain, 16
inner product, 4 sound speed
Index 84 comments to [email protected]

Elastic-Plastic Flow Index
isotropic, 29
specific energy, E, 12
spectral decomposition, 70
stiffness tensor, C , 24
strain tensor
Eulerian, e, 7
Lagrangian, E, 7
stress tensor
Cauchy, , 12
linear elastic, 16
Piola-Kirchhoff, S, 12
viscous, vis , 48
strong ellipticity, 29, 77
subscript, covariant index, 3
summation convention, 3
super-position of strains, 8, 34
superscript, contravariant index, 3
tangent space, 3
texture, 16
thermodynamic identity, 12
trace, 10
transpose, 5
Tresca yield condition, 38
uniaxial strain, 51
uniaxial stress, 53
viscous stress, 48
Voigt average, 74
Voigt notation, 68
von Mises yield condition, 43
yield condition
von Mises, 43
yield function, 34
yield strength, 34
comments to [email protected] 85 Index

Elastic Plastic Notes

Uploaded by

Copyright:

Available Formats

Elastic Plastic Notes

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Elastic Plastic Notes

Uploaded by

Copyright:

Available Formats

LA-UR-03-0047 March 20, 2003

Approved for public release;

Title: Notes on Elastic-Plastic Flow

Author(s): Ralph Menikoff

Submitted to: T14 web site

Notes on Elastic-Plastic Flow

3 Plastic Strain Rate 33

4 Illustrative Constitutive Model 41

Ralph Menikoff i comments to [email protected]

5.4 One Dimensional Flow Equations . . . . . . . . . . . . . . . . 55

Appendix A: Useful Lemmas 68

Appendix B: Compliance and Stiffness Tensors 68

Appendix C: Convexity of Specific Energy 76

Appendix D: Small Viscosity Limit 77

Appendix E: Lie Derivative 78

Ralph Menikoff ii comments to [email protected]

Notes on Elastic-Plastic Flow

A major complication of elastic-plastic flow, as compared to fluid flow,

where X~ B and ~x S. The particle velocity field in the reference frame

It is important to note that the Lagrangian velocity U~ is a two-point rank-one

Elastic-Plastic Notes 1 comments to [email protected]

The deformation gradient is the derivative of the mapping, and is a

1.1 Multiplicative Decomposition

To motivate the decomposition of a flow into elastic and plastic compo-

The chain rule for differentiation then leads to a multiplicative decompo-

The deformation gradients Fe and Fp are local quantities that we consider

1 Kinematic Variables 2 comments to [email protected]

The mathematical framework for elastic-plastic flow, or even elastic flow,

To be more specific, we denote the basis elements of TX~ B by

comments to [email protected] 3 1 Kinematic Variables

with dual pairing

The components of a covector relative to the basis E ~ are denoted with

be expressed in terms of their components as (V~ , U

where F = D,~ such that the dual pairing is invariant, i.e., V U = (V 0 ) (U 0 ) .

~ and g(~x) denote the metric maps between T and T in the

1 Kinematic Variables 4 comments to [email protected]

The following identities hold

By convention, we denote (G1 ) by G and(g 1 )ij by g ij . Furthermore,

1.3 Transpose, Inverse and Transpose-Inverse

In addition to the derivative of the mapping, D, its transpose, inverse

Remark 1.1 In the language of differential geometry, the induced mapping

comments to [email protected] 5 1 Kinematic Variables

: B S and denoted . For a vector and covector

and represented in coordinates by

1.4 Strain Tensors

In order to define a strain tensor, it is convenient to introduce several

b = (FG1 ) 1 FT bij (~x, t) = (G Fi Fj ) 1 : T ~x S T~x S

1 Kinematic Variables 6 comments to [email protected]

The Lagrangian strain tensor is defined by

e = 21 (g b1 ) eij (~x, t) = 12 [gij (b1 )ij ] : T~x S T ~x S . (1.4)

The inverse of the left Cauchy-Green tensor b1 is known as the Finger

is the right Cauchy-Green plastic tensor, and G

ee = 21 (g be 1 ) , (ee )ij (~x, t) : T~x S T ~x S , (1.6)

is the left Cauchy-Green elastic tensor.

ep = FT Ep F1 , ep (~x, t) : T~x S T ~x S (1.7)

comments to [email protected] 7 1 Kinematic Variables

Consequently, the total strain can be expressed as the super-position or

Again the super-position of strains is satisfied

Ep = 2 [G (Fp G) p Fp ] , Ep (X, t) : TX~ P T X~ P