Académique Documents
Professionnel Documents
Culture Documents
Biswajit Banerjee
Contents
1 Introduction 2
2 Conservation of Mass 3
3 Stress Tensors 4
4 Thermodynamic Potentials 4
4.1 Functional dependencies of ψ and e . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
7 Constitutive Relations 18
7.1 Stress power in current configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
7.2 Stress power in intermediate configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
7.3 Relation between Up , U , and Ue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
7.4 Rate equation for the total stretch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
7.5 Elastic constitutive relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
7.6 Plastic flow rules and evolution of internal variables . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1
7.6.1 Isotropic material with scalar internal variables . . . . . . . . . . . . . . . . . . . . . . . . 24
7.6.2 Isotropic material with some tensor internal variables . . . . . . . . . . . . . . . . . . . . . 26
7.7 Plastic Yield . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
7.7.1 Isotropic elastic material and scalar internal variables . . . . . . . . . . . . . . . . . . . . . 27
7.7.2 Isotropic elastic material and tensor internal variables . . . . . . . . . . . . . . . . . . . . . 27
7.8 Plastic flow potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
7.8.1 Isotropic material with scalar internal variables . . . . . . . . . . . . . . . . . . . . . . . . 28
8 Appendix 29
1 Introduction
The purpose of these notes is to provide some of the details of the formulation of thermoplasticity by Wright and
coworkers [SW01, Wri02]. We start with the assumption that the basics of thermoelasticity and the notation used
are known (the details can be found in my notes on thermoelasticity).
As the first point of departure from thermoelasticity, we consider the (now classical) multiplicative decomposition
of the deformation gradient into elastic and plastic parts:
F = Fe · Fp . (1)
Note that even though F represents a physical gradient, neither Fe or Fp has to be a gradient of a physical quantity
by itself. However, for our purposes, we require the additional restriction that
This decomposition is usually interpreted to mean that there is an unstressed intermediate configuration that can be
attained instantaneously by unloading elastically from the current configuration. Figure 1 shows a schematic of the
situation.
Clearly, the intermediate configuration can be identified only modulo a rigid rotation Q since
We will not concern ourselves with such issues here. Detailed discussions can be found in many places, for
example [SW01, NN04]. Instead, we will follow the lead of Wright [Wri02] and assume that the plastic
component of the deformation gradient is invariant under a rigid rotation Q. That is, if
then
Ferot = Q · Fe and Fprot = Fp . (5)
2
Current Configuration
Reference Configuration
F ρ
ρ
0
Fp ρ
I
Fe
Intermediate Configuration
2 Conservation of Mass
Experimental data strongly favors the assumption that plastic flow is volume preserving metals. However, in
macroscopic scale deformations, voids and other defects in the material may cause small volume changes. To keep
the analysis general we assume that plastic deformation is not necessarily volume preserving.
Thus there may be a change in volume in going from the initial to the intermediate configuration. Let the density in
the initial configuration be ρ0 , that in the intermediate configuration be ρI , and the density in the current
configuration be ρ.
Then the overall conservation of mass (from the initial to the current configuration) implies that
ρ det(F ) = ρ0 . (6)
Plugging in the decomposition for F gives
ρ det(Fe · Fp ) = ρ det(Fe ) det(Fp ) = ρ0 . (7)
If we define the density in the intermediate configuration via
ρ det(Fe ) =: ρI (8)
we get
ρI det(Fp ) = ρ0 . (9)
Equation (9) represents the conservation of mass from the initial to the intermediate configuration while equation
(8) represents the conservation of mass from the intermediate to the current configuration.
The quantity ρI is assumed to be an internal variable (or at least a function of other internal variables). In the
following we often replace ρI with q0 , where qi , i = 0 . . . n represents a set of internal variables.
3
3 Stress Tensors
Let σ be the Cauchy stress. Then the 2nd Piola-Kirchhoff stress is defined as
ρ0 −1
S= F · σ · F −T (10)
ρ
i.e., it is the pull-back of σ to the initial configuration. Now, let us plug in the decomposition of F into this formula
to get
ρ0 −1
S= F · Fe−1 · σ · Fe−T · Fp−T . (11)
ρ p
Define the 2nd P-K stress in the intermediate configuration as the pull-back of the Cauchy stress by the elastic part
of the deformation gradient, i.e.,
ρI −1
SI := F · σ · Fe−T . (12)
ρ e
Then we can write
ρ0 −1
S= F · SI · Fp−T . (13)
ρI p
This is equivalent to a pull-back by Fp of the 2nd P-K stress in the intermediate configuration to the initial
configuration. Note that if ρ0 = ρI , i.e., for isochoric plasticity, we can use the Kirchhoff stress
ρ0
τ := det(F ) σ = σ (14)
ρ
instead of the Cauchy stress without any effect on our analysis.
4 Thermodynamic Potentials
Following Wright [Wri02], we assume that the Gibbs free energy is regarded as fundamental (rather than the
Helmholtz free energy). Then the Gibbs free energy functional is a function of the 2nd P-K stress in the
intermediate configuration (SI ), the temperature (T ), and a set of internal variables (qj j = 0 . . . n). That is,
g = g(SI , T, qj ) (15)
The internal variables evolve only during the plastic part of the deformation. The intermediate configuration is
considered to be the reference configuration as far as any elastic deformations are concerned.
We assume that the elastic strain (Ee ) and the entropy (η) are given by (see [SW01] for a detailed justification of
this assumption and its connection to an instantaneous theromoelastic response)
∂g
E e = ρI
∂SI
(16)
∂g
η=
∂T
4
The elastic strain Ee is related to the elastic part of the deformation gradient by
1
Ee = (F T · Fe − 1 ) . (17)
2 e
Recall that for thermoelastic processes, the Helmholtz free energy (ψ) is defined as
ψ =e−T η (18)
where e is the internal energy, T is the temperature, and η is the entropy. Also, the Gibbs free energy is defined as
1 1
g = −e + T η + S : E = −ψ + S:E (19)
ρ0 ρ0
where ρ0 is the density in the initial configuration, S is the 2nd P-K stress and E is the Green strain.
For thermoplasticity, since all elastic processes are with respect to the intermediate configuration, we may write
1 1
g = −e + T η + SI : Ee = −ψ + SI : Ee . (20)
ρI ρI
Let us define
∂g ∂g
Q0 := −q0 + q0−2 SI : Ee and Qj := −q0 , j = 1...n . (30)
∂q0 ∂qj
Then we can write
n
X
dg = q0−1 Ee : dSI + η dT − q0−1 Q0 + q0−2 SI : Ee dq0 − q0−1
Qj dqj
j=1
n
X
de = q0−1 SI : dEe + T dη + q0−1 Q0 dq0 + q0−1 Qj dqj (31)
j=1
Xn
dψ = q0−1 SI : dEe − η dT + q0−1 Q0 dq0 + q0−1 Qj dqj .
j=1
6
We can also find other relations between these partial derivatives. For example, equating the partial derivative of
∂g ∂g
∂q0 with respect to SI with the partial derivative of ∂SI with respect to q0 , we have
∂2g
∂Ee −1 ∂Q0 ∂Ee
−q0−2 Ee + q0−1 = = − q0 −2 −2
+ q0 Ee + q0 SI : (34)
∂q0 ∂q0 ∂SI ∂SI ∂SI
or,
∂Ee −1 ∂Q0 ∂Ee
q0−1 = − q0 −2
+ q0 SI : (35)
∂q0 ∂SI ∂SI
or,
∂Ee ∂Q0 ∂Ee ∂Q0 ∂2g
=− − q0−1 SI : =− − SI : . (36)
∂q0 ∂SI ∂SI ∂SI ∂SI2
Similarly, for the other internal variables qj , j = 1 . . . n, we have
∂2g
∂η ∂Q0 ∂Ee
= = − q0−1 + q0−2 SI : (39)
∂q0 ∂q0 ∂T ∂T ∂T
or,
∂η ∂Q0 ∂Ee
q0 =− − q0−1 SI : . (40)
∂q0 ∂T ∂T
Similarly, derivatives with respect to the qj s lead to
∂η ∂2g ∂Qj
= = −q0−1 (41)
∂qj ∂qj ∂T ∂T
or,
∂η ∂Qj
q0 =− . (42)
∂qj ∂T
Several other relations can be derived based on mixed partial derivatives of g. For example,
∂2g
∂ ∂g ∂ ∂Ee
q0−1 Ee = q0−1
= = . (43)
∂T ∂SI ∂T ∂SI ∂T ∂T
Also,
∂2g
∂ ∂g ∂η
= = . (44)
∂T ∂SI ∂SI ∂T ∂SI
Therefore,
∂η ∂Ee
= q0−1 . (45)
∂SI ∂T
7
We can also show that (see Appendix)
1 ∂e(SI , T, q0 , qj )
∂η(SI , T, q0 , qj −1 ∂Ee
= − q0 SI :
∂T T ∂T ∂T
∂η(Ee , T, q0 , qj ) 1 ∂e(Ee , T, q0 , qj ) (46)
=
∂T T ∂T
∂η(Ee , T, q0 , qj ) −1 ∂SI
= q0 .
∂Ee ∂T
8
Substituting (55) into (49) and reverting to ρI ≡ q0 leads to
n
ρ X q · ∇T
Qj q̇j + SI : Ėe − σ : ∇v ≤ . (56)
ρI T
j=0
Since
1
σ : ∇v = σ : [∇v + (∇v)T ] = σ : d (57)
2
we can write the entropy inequality as
n
ρ ρ X q · ∇T
σ :d− SI : Ėe − Qj q̇j − ≥0. (58)
ρI ρI T
j=0
ρ
σ:d= SI : Ėe (59)
ρI
since these terms represent the stress power. Also, if the internal variables do not evolve during such deformations,
we have
n
ρ X
Qj q̇j = 0 (60)
ρI
j=0
q · ∇T
≤0. (61)
T
Also, in the event that the temperature gradient vanishes, we must have
n
ρ ρ X
σ :d− SI : Ėe − Qj q̇j ≥ 0 . (62)
ρI ρI
j=0
Equations (61) and (62) represent a split of the entropy inequality into purely mechanical and purely thermal parts.
Recall that
1
d= [∇v + (∇v)T ] and ∇v = Ḟ · F −1 . (63)
2
Therefore,
1
d= Ḟ · F −1 + F −T · Ḟ T . (64)
2
Also,
F = Fe · Fp . (65)
Hence,
Ḟ = Ḟe · Fp + Fe · Ḟp ; Ḟ T = FpT · ḞeT + ḞpT · FeT ; F −1 = Fp−1 · Fe−1 ; F −T = Fe−T · Fp−T . (66)
9
Plugging these into the expression for d, we have
1h i
d= (Ḟe · Fp + Fe · Ḟp ) · (Fp−1 · Fe−1 ) + (Fe−T · Fp−T ) · (FpT · ḞeT + ḞpT · FeT )
2 (67)
1h i
= Ḟe · Fe−1 + Fe · Ḟp · Fp−1 · Fe−1 + Fe−T · ḞeT + Fe−T · Fp−T · ḞpT · FeT
2
or,
1h i 1h i
d= Ḟe · Fe−1 + (Ḟe · Fe−1 )T + Fe · Ḟp · Fp−1 · Fe−1 + (Fe · Ḟp · Fp−1 · Fe−1 )T . (68)
2 2
Define
1h i
de := Ḟe · Fe−1 + (Ḟe · Fe−1 )T
2 (69)
1h i
dp := Fe · Ḟp · Fp−1 · Fe−1 + (Fe · Ḟp · Fp−1 · Fe−1 )T .
2
Then we can write
d = de + dp . (70)
Now, form equation (12), we have
ρI −1 ρ
SI = F · σ · Fe−T =⇒ σ= Fe · SI · FeT . (71)
ρ e ρI
We can show that (see Appendix:item 5 for details)
ρ
1 1
σ:d= SI : FeT · Ḟe + ḞeT · Fe + FeT · Fe · Ḟp · Fp−1 + Fp−T · ḞpT · FeT · Fe . (72)
ρI 2 2
Recall that
1
Ee = (F T · Fe − 1 ) . (73)
2 e
Hence,
1
De := Ėe = (Ḟ T · Fe + FeT · Ḟe ) . (74)
2 e
Also, define
1 T
Dp := Fe · Fe · Ḟp · Fp−1 + Fp−T · ḞpT · FeT · Fe . (75)
2
Then we can write equation (72) as
ρ ρ
σ:d= SI : Ėe + SI : Dp . (76)
ρI ρI
We can also show that
ρ 1
SI : Dp = tr (Fe−1 · σ · Fe−T ) · (FeT · Fe · Ḟp · Fp−1 + Fp−T · ḞpT · FeT · Fe )
ρI 2
1 −1 1
= tr Fe · σ · Fe · Ḟp · Fp−1 + tr Fp−T · ḞpT · FeT · σ · Fe−T
2 2 (77)
1 1
= tr Fe · Ḟp · Fp · Fe · σ + tr σ · Fe−T · Fp−T · ḞpT · FeT
−1 −1
2 2
1
= σ : (Fe · Ḟp · Fp−1 · Fe−1 + Fe−T · Fp−T · ḞpT · FeT ) .
2
10
From the definition (69) we then have
ρ
SI : Dp = σ : dp . (78)
ρI
The entropy inequality in equation (62) may now be expressed as
n
ρ X
σ : dp − Qj q̇j ≥ 0 . (79)
ρI
j=0
where
1
Fe · Lp · Fe−1 + (Fe · Lp · Fe−1 )T ; Lp := Ḟp · Fp−1 .
dp = (80)
2
The quantity dp can be interpreted as a rate of plastic deformation.
Note that
1
tr Fe · Lp · Fe−1 + tr Fe−T · LTp · FeT
tr (dp ) =
2
1
tr Fe−1 · Fe · Lp + tr FeT · Fe−T · LTp
= (81)
2
1
tr (Lp ) + tr LTp = tr (Lp ) .
=
2
Therefore,
tr (dp ) = tr Ḟp · Fp−1 (82)
ρ0
J˙ = J tr (d) ; J := det(F ) = . (84)
ρ
Similarly, let us define
Jp (Fp ) := det(Fp ) . (85)
If the tensor Fp is invertible, then the directional derivative of Jp is given by (see for instance [Gur81] p. 23)
11
Also, from (9), we have
ρ0
Jp = det(Fp ) = . (89)
ρI
Hence,
ρ0
J˙p = tr (dp ) . (90)
ρI
Now, !
dJp d ρ0 ρ0
J˙p = = =− ρ̇I . (91)
dt dt ρI ρ2I
Comparing equations (90) and (91), we get
ρ̇I
tr (dp ) = − (92)
ρI
which has a form similar to
ρ̇
tr (d) = − . (93)
ρ
These relations indicate that the quantity dp may be considered to be a plastic rate of deformation tensor just as d is
considered to be the total rate of deformation tensor.
The Cauchy stress can be decomposed into volumetric and distortional components as
1
σ= tr (σ) 1 + s = −p 1 + s . (94)
3
It may not be obvious that we can do the same for the plastic rate of deformation dp . The question arises that if we
decompose dp into volumetric and deviatoric parts as
1
dp = tr (dp ) 1 + ηp (95)
3
then is ηp truly a distortional term or does it contain a volumetric component too? Indeed, it can be shown that the
deviatoric part of dp does not contain any volumetric terms (see Appendix: item 6) .
Then we can write
1
σ : dp = (−p 1 + s) : tr (dp ) 1 + ηp
3
*0 1
*0 (96)
= −p tr (dp ) − p
tr (η
p) + tr (dp )
tr (s)
+ s : ηp
3
ρ̇I q̇0
= −p tr (dp ) + s : ηp = p + s : ηp = p + s : ηp .
ρI q0
Recall the entropy inequality in equation (79)
n
ρ X
σ : dp − Qj q̇j ≥ 0 (97)
ρI
j=0
12
which can be written (with ρI ≡ q0 ) as
n
q̇0 ρ X
σ : dp − ρ Q0 − Qj q̇j ≥ 0 (98)
q0 q0
j=1
or,
n
ρ̇I ρ X
s : ηp + (p − ρ Q0 ) − Qj q̇j ≥ 0 . (100)
ρI ρI
j=1
This is another form of the Clausius-Duhem inequality. From equation (100) we can see that for us to have a well
posed problem it is necessary that rate equations for ηp , ρ̇I , and q̇j must be provided. Clearly these constitutive
relations must be prescribed in such a way that the inequality in equation (100) is never violated.
or,
n n
p X ∂ρI ρ X
s : ηp + q̇j − Qj q̇j ≥ 0 (105)
ρI ∂qj ρI
j=1 j=1
or,
n
" #
ρX p ∂ρI
s : ηp − Qj − q̇j ≥ 0 . (106)
ρI ρ ∂qj
j=1
13
6 Energy Equation for Thermoplasticity
Recall that the balance of energy for thermoelastic processes is given by
ρ ė = σ : ∇v + ρ s − ∇ · q . (107)
ρ
SI : Ėe = σ : d − σ : dp . (111)
ρI
Substituting (111) into (110) gives
n
ρ X
ρ T η̇ + Qj q̇j + σ : d − σ : dp − σ : ∇v − ρ s + ∇ · q = 0 . (112)
ρI
j=0
Recall that
σ : ∇v = σ : d . (113)
Hence,
n
ρ X
ρ T η̇ + Qj q̇j − σ : dp − ρ s + ∇ · q = 0 . (114)
ρI
j=0
Recall from equations (33) that the entropy η can be derived from a thermodynamic potential, i.e.,
∂g ∂ψ
η= = (115)
∂T ∂T
where, from (32),
g = g(SI , T, q0 , qj ) ; ψ = ψ(Ee , T, q0 , qj ) , j = 1 . . . n . (116)
There can be two forms of the energy equation based on which variables we consider to be independent.
14
1. If we consider the Helmholtz free energy to be fundamental, then we have
η = η(Ee , T, q0 , qj ) , j = 1 . . . n . (117)
∂ψ ∂ψ ∂ψ ∂ψ
= q0−1 SI ; = −η ; = q0−1 Q0 ; = q0−1 Qj . (119)
∂Ee ∂T ∂q0 ∂qj
Therefore,
∂η ∂ ∂ψ ∂SI
=− = −q0−1
∂Ee ∂T ∂Ee ∂T
(120)
∂η ∂ ∂ψ ∂Qj
=− = −q0−1 .
∂qj ∂T ∂qj ∂T
Also, the specific heat at constant strain is defined as
∂e(Ee , T, q0 , qj )
Cv := (121)
∂T
which implies that
∂ ∂ψ ∂η ∂η
Cv = (ψ + T η) = +η+T =T . (122)
∂T ∂T ∂T ∂T
Hence,
∂η Cv
= . (123)
∂T T
Plugging (120) and (123) into (118) gives
n
∂SI Cv X ∂Qj
η̇ = −q0−1 : Ėe + Ṫ − q0−1 q̇j . (124)
∂T T ∂T
j=0
or,
n
∂SI X ∂Qj
ρ Cv Ṫ = ρ T q0−1 : Ėe + ρ q0−1 T − Qj q̇j + σ : dp + ρ s − ∇ · q (126)
∂T ∂T
j=0
Using ρI ≡ q0 and rearranging gives us the new form of the energy equation:
n
ρ ∂SI ρ X
∂Qj
ρ Cv Ṫ = −∇ · q + ρ s + T : Ėe + T − Qj q̇j + σ : dp . (127)
ρI ∂T ρI ∂T
j=0
15
For a thermoelastic process, the last two terms above evaluate to zero and we are left with the energy
equation for thermoelasticity.
If the heat flux q can be derived from a temperature potential T as
q = −κ · ∇T (128)
where κ is the thermal conductivity tensor, and if the coefficient of thermal stress is defined as
∂SI
βS := (129)
∂T
then we can express the energy equation as
n
ρ ρ X
∂Qj
ρ Cv Ṫ = ∇ · (κ · ∇T ) + ρ s + T βS : Ėe + T − Qj q̇j + σ : dp . (130)
ρI ρI ∂T
j=0
The coefficient βS can be difficult to determine for the general anisotropic case.
Recall from equation (96) that
q̇0 ρ̇I
σ : dp = p + s : ηp = p + s : ηp (131)
q0 ρI
Using this relation, we can provide another form of the energy equation:
n
ρ ρ X ρ̇I
∂Qj
ρ Cv Ṫ = ∇ · (κ · ∇T ) + ρ s + T βS : Ėe + T − Qj q̇j + p + s : ηp . (132)
ρI ρI ∂T ρI
j=0
η = η(SI , T, q0 , qj ) , j = 1 . . . n . (133)
∂η ∂Ee
= q0−1
∂SI ∂T
∂η ∂Q0 ∂Ee
= −q0−1 − q0−2 SI : (135)
∂q0 ∂T ∂T
∂η ∂Qj
= −q0−1 .
∂qj ∂T
16
Plugging the relations in equations (135) and (48) into (134), we get
n
1
−1 ∂Ee −1 ∂Ee −1 ∂Q0 −1 ∂Ee −1
X ∂Qj
η̇ = q0 : ṠI + Cp − q0 SI : Ṫ − q0 + q0 SI : q̇0 − q0 q̇j
∂T T ∂T ∂T ∂T ∂T
j=1
(137)
or,
n
1
∂Ee ∂Ee ∂Ee X ∂Qj
η̇ = Cp − q0−1 SI : Ṫ + q0−1 : ṠI − q0−1 SI : q̇0 − q̇j
T ∂T ∂T ∂T ∂T
j=0
(138)
n
1
∂Ee ∂Ee X ∂Qj
= Cp − q0−1 SI : Ṫ + q0−1 : ṠI − q0−1 SI q̇0 − q0−1 q̇j .
T ∂T ∂T ∂T
j=0
Plugging the expression in equation (139) into the energy equation (114) gives
n
! !
1 ∂Ee ρ ∂Ee ρ̇I ρX ∂Qj
ρ Cp − SI : Ṫ + T : ṠI − SI − T q̇j
ρI ∂T ρI ∂T ρI ρI ∂T
j=0
n
(140)
ρX
+ Qj q̇j − σ : dp − ρ s + ∇ · q = 0
ρI
j=0
or,
! !
1 ∂Ee ρ ∂Ee ρ̇I
ρ Cp − SI : Ṫ = −∇ · q + ρ s + T : ṠI − SI
ρI ∂T ρI ∂T ρI
n (141)
ρX
∂Qj
+ Qj − T q̇j − σ : dp .
ρI ∂T
j=0
This is a slightly more complicated version of the energy equation for thermoplasticity and the form in
equation (127) is preferable since strain rates are more accessible than stress rates.
If the heat flux q is related to the temperature gradient by
q = −κ · ∇T (142)
and if the coefficient of thermal expansion is defined as
∂Ee
αE := (143)
∂T
then the energy equation can be written as
! !
1 ρ ρ̇I
ρ Cp − SI : αE Ṫ = −∇ · q + ρ s + T αE : ṠI − SI
ρI ρI ρI
n (144)
ρX
∂Qj
+ Qj − T q̇j − σ : dp .
ρI ∂T
j=0
17
7 Constitutive Relations
To close the system of equations in thermoplasticity, we have to provide constitutive relations that can be used to
determine the plastic part of the deformation. The usual situation is that the total deformation gradient is known
and the decomposition of the deformation gradient into elastic and plastic parts have to be computed.
For elastically anisotropic materials, the orientation of the material in the intermediate configuration needs to be
known before an the correct elastic stress-strain relation can be used. We will avoid the related complications for
now and deal only with the elastically isotropic materials. However, it is important to know what the fuss is all
about.
Consider an elastic material. The deformation gradient F can be decomposed into a pure stretch and a rotation in
two ways, i.e.,
Fe = R · Ue = Ve · R where R · RT = RT · R = 1 . (145)
A schematic of the polar decomposition theorem (and what the decomposition means in terms of Lagrangian triads)
is shown in Figure 2. Now, suppose that we rotate the reference configuration by a rotation Q so that the
N
2
n
2
R n
3
N N
3 1
n
1
V U
F
λ2 N2
λ2 n2
λ3 n
λ3 N 3
3 λ 1N
1 λ1 n
1
R
In this situation, the actual amount of stretch remains unchanged. However, the rotation is felt by the eigenvector
triads of the stretch tensor and also be the original rotation tensor. Such a rotation of the reference configuration
18
should not affect the Cauchy stress whether the material is isotropic or not. Let us now check how the 2nd P-K
stress transforms under a rotation of the reference configuration. Recall that
ρ0 −1
S= F · σ · Fe−T . (147)
ρ e
ρ0 ρ0 −T
S rot = (Fe · QT )−1 · σ · (Fe · QT )−T = Q · Fe−1 · σ · Fe−T · Q−1 = Q · S · QT . (148)
ρ ρ
Therefore, the 2nd P-K stress S also transforms in the same way as the elastic stretch tensor Ue .
Now, suppose that the constitutive model of the material relates the 2nd P-K stress to the elastic stretch via
S = C(Ue ) . (149)
or,
Q · S · QT = Crot (Q · Ue · Q) =⇒ S = QT · Crot (Q · Ue · Q) · Q . (151)
Therefore the two constitutive models are related by
or
Crot (Uerot ) = Q · Crot (QT · Uerot · Q) · QT . (153)
In general, Crot 6= C except for special values of the rotation Q. However, if we know the constitutive relation in
one configuration, we can evaluate the response in a rotated configuration by
P λ2 n2
1
N N
3 P 1
3
λ 1n
1
Up
Ve
p
λ2 P
2
p
λ2 p
2
N
2
n
Q n 2
3
N p
3 λ 1 P1
p
λ3 p3
p N p
λ 3 P3 1 n λ1 p1
1
Figure 3: A decomposition of the deformation gradient into a plastic stretch Up , and rigid rotation Q, and an elastic
stretch Ve [NN04], p. 251).
In the figure, Ni are the principal directions of U and ni are the principal directions of V where
F = R · U = V · R. The triad Pi represents the principal directions of the plastic stretch Up .
There are clearly two intermediate configurations in this case. The first intermediate configuration is achieved by
purely plastic loading from the unloaded configuration. The second intermediate configuration is achieved by a
pure rotation from the first (no elastic or plastic deformations). The final configuration is achieved by a purely
elastic deformation from the second configuration. The internal variables do not evolve in this phase. For further
discussions in the context of anisotropic elastic behavior see [NN04].
Also note that from (155) we have
Ve−1 · F = Q · Up . (156)
Since the right hand side consists of a rotation and a pure stretch, we can interpret the product to be the result of a
polar decomposition of the left hand side. This implies that if we know Ve and F then we can find Up using a polar
decomposition. An additional implication is that det(Ve−1 ) = 1/ det(Ve ) > 0, i.e., det(Ve ) > 0. Another
implication is that Q is fully determined if Ve and F are known.
20
For elastically isotropic materials, the Cauchy stress (σ) is an isotropic function of the left stretch (Ve ). Then the
elastic constitutive equation may be written as
σ = C(Ve ) where C(Q · Ve · QT ) = Q · C(Ve ) · QT (157)
for any rigid motions Q.
Also, σ and Ve are coaxial in the sense that their eigenvectors point in the same directions (they have the same
principal directions).
Let us now check how the rotation Q in the decomposition (155) and its rate Q̇ affect the stress power in this
situation. The total stress power is given by
1
P=σ:d= σ : [∇v + (∇v)T ] . (158)
2
Since σ and Ve are coaxial, we can show that stress power can be expressed as (see Appendix, item 8)
1h
σ:d= σ : (V̇e · Ve−1 + Ve−1 · V̇e )+
2 i (159)
σ : (Ve · Q · U̇p · Up−1 · QT · Ve−1 + Ve−1 · Q · Up−1 · U̇p · QT · Ve ) .
The first term in this expression can be interpreted as the elastic stress power while the second term is the plastic
stress power. Also observe that there are no terms containing the rate of rotation Q̇ in the expression.
This makes life easier for us because we can now assign the entire rotation Q to the elastic part of the deformation
gradient, i.e., we can write
Fe = Ve · Q and Fp = Up . (160)
Now, a polar decomposition of Fe also gives us
Fe = Q · Ue . (161)
Hence, we can express the decomposition (155) alternatively as
F = Q · Ue · Up and Ve = Q · Ue · QT . (162)
If we know Ue and Q, we can compute the Cauchy stress for isotropic elastic materials using the constitutive
relation. The decomposition in (162) thus treats the plastic part of the deformation as a pure stretch and simplifies
things considerably.
If we substitute Ve = Q · Ue · QT in equation (159) we get (see Appendix, item 9)
1 h n o
σ:d= σ : Q · (U̇e · Ue−1 + Ue−1 · U̇e ) · QT
2 n oi (163)
+ σ : Q · (Ue · U̇p · Up−1 · Ue−1 + Ue−1 · Up−1 · U̇p · Ue ) · QT .
1h i
de = Q · (U̇e · Ue−1 + Ue−1 · U̇e ) · QT
2 (165)
1h i
dp = Q · (Ue · U̇p · Up−1 · Ue−1 + Ue−1 · Up−1 · U̇p · Ue ) · QT .
2
Hence we can write the stress power in equation (163) as
P = σ : d = σ : (de + dp ) . (166)
We can show that the elastic and plastic rates of deformation transform in the same way as the Cauchy stress under
rigid body motions and hence are objective rates.
Alternatively, we can show that the stress power in the intermediate configuration can also be decomposed into a
product of the tensor SI and rates of deformation in the intermediate configuration.
We can show that the stress power can be written as (see Appendix, item 11 fro details).
where
1 h i
De = U̇e · Ue + Ue · U̇e
2 (168)
1 h i
Dp = Ue2 · U̇p · Up−1 + Up−1 · U̇p · Ue2 .
2
If we push forward the rates of deformation De and Dp by Fe , we get
1 h i
Fe−T · De · Fe−1 = Q · Ue−1 · U̇e · QT + Q · U̇e · Ue−1 · QT = de
2 (169)
1h i
Fe−T · Dp · Fe−1 = Q · Ue · U̇p · Up−1 · Ue−1 · QT + Q · Ue−1 · Up−1 · U̇p · Ue · QT = dp .
2
Given any deformation gradient F , we can eliminate the rotation by using a deformation measure of the form
C = F T · F = UT · U = U2 (170)
where U is a pure stretch. Substituting the decomposition
F = Q · Ue · Up (171)
into (170) gives
U 2 = (Up · Ue · QT ) · (Q · Ue · Up ) = Up · Ue · Ue · Up . (172)
Hence,
Ue · U 2 · Ue = (Ue · Up · Ue ) · (Ue · Up · Ue ) = (Ue · Up · Ue )2 . (173)
This leads to
Up = Ue−1 · (Ue · U 2 · Ue )1/2 · Ue−1 . (174)
22
7.4 Rate equation for the total stretch
Recall that
U2 = F T · F . (175)
Differentiation with respect to time gives us
d
U 2 = Ḟ T · F + F T · Ḟ .
(176)
dt
A bit of algebra shows that (see Appendix, item 12)
d
(U 2 ) = Up · (De + Dp ) · Up . (177)
dt
If the history of SI , T , q0 , and qj are known then we can find De as shown in the following section. We can then
solve for Ue using the definition of De . Since Up can be expressed as a function of U and Ue , we can use (177) to
solve for U . Note that this approach is different from that used in most numerical codes where the stretch U is
assumed to be known and it is the stresses that are computed.
To determine the elastic response of the material we need constitutive relations. In most computations, we usually
know the total deformation gradient. Suppose, for the time being, that we also know Ve or Ue . Then our goal is to
compute the Cauchy stress σ (or the 2nd P-K stress SI ).
The elastic part of the rate of deformation in the intermediate configuration De is given by (see Appendix, item 13
for proof)
n
!
ρ̇I X ∂Qj
De = Ėe = ρI S : ṠI − SI + αe Ṫ − q̇j (178)
ρI ∂SI
j=0
where the fourth-order elastic stiffness tensor and the second-order thermal expansion tensor are defined as
∂2g ∂Ee
S := ; αe := . (179)
∂SI2 ∂T
We also need a relation (flow rule) to determine Dp . The usual assumption is that Dp also depends on the same
variables as the Gibbs function, i.e.,
Dp = D
b p (SI ; T, q0 , qj ) j = 1...n . (180)
The evolution equations for the internal variables are also assumed to have the same dependencies, i.e.,
23
7.6.1 Isotropic material with scalar internal variables
If the material is isotropic and all the internal variables are scalars, then the functions D
b p and qbi are isotropic
functions of the stress SI .
Recall that
ρI
SI = Je Fe−1 · σ · Fe−T and Fe = Ve · Q ; Je = = det(Fe ) = det(Ve ) . (182)
ρ
Therefore we can write
SI = Je QT · Ve−1 · σ · Ve−1 · Q . (183)
Also, recall that
dp = Fe−T · Dp · Fe−1 = Ve−1 · Q · Dp · QT · Ve−1 . (184)
Hence, from (180) and relations (184) and (183), we have
dp = Ve−1 · Q · D
b p Je QT · V −1 · σ · V −1 · Q; T, q0 , qj · QT · V −1 .
e e e (185)
b p · QT = D
Q·D bp . (186)
Therefore,
dp = Ve−1 · D
b p Je QT · V −1 · σ · V −1 · Q; T, q0 , qj · V −1 .
e e e (187)
The isotropy of the material also implies that
The representation theorem for isotropic tensor valued functions of second order tensors implies that, since dbp is an
isotropic function of σ, we can write (see Appendix, item 15 or [Gur81], p. 233 for a proof)
24
where ϕ1 , ϕ2 , ϕ3 are scalar valued functions of the invariants of σ, the temperature, and the internal variables.
The flow rule is commonly expressed as a function of the pressure and the invariants of the deviatoric stress.
Recall that the Cauchy stress can be decomposed as
1
σ= tr (σ) + s = −p 1 + s . (194)
3
Hence,
σ 2 = σ · σ = p2 1 − 2 p s + s · s . (195)
Therefore, the flow rule (193) can be written as
1 qb0 1 qb0
1 1
s − tr s2 1
2 2 2
dp = − 1 + (ϕ1 − 2 p ϕ2 ) s + ϕ2 =− 1 + ξ1 s + ξ2 s − tr s 1
3 ρI 3 3 ρI 3
(204)
or,
1
s − tr s2 1
2
ηp = ξ1 s + ξ2 . (205)
3
25
Entropy inequality for isotropic materials with scalar internal variables. Recall from (79) that the entropy
inequality can be written as
n
ρ X
σ : dp − Qj q̇j ≥ 0 (206)
ρI
j=0
or alternatively from (100) as
n
ρ̇I ρ X
s : ηp + (p − ρ Q0 ) − Qj q̇j ≥ 0 . (207)
ρI ρI
j=1
If we plug the expression for ηp from (205) into (207) we get (using the substitutions ρ̇I = qb0 , q̇j = qbj )
n
qb0 ρ X
2 1 2
s : ξ1 s + ξ2 s − tr s 1 + (p − ρ Q0 ) − Qj qbj ≥ 0 (208)
3 ρI ρ I
j=1
or,
n
0 qb0 ρ X
1 2
ξ1 s : s + ξ2 (s · s) : s − tr s + (p − ρ Q0 )
tr (s)
*
− Qj qbj ≥ 0 (209)
3 ρI ρ I
j=1
or,
n
2
3
qb0 ρ X
ξ1 tr s + ξ2 tr s + (p − ρ Q0 ) − Qj qbj ≥ 0 . (210)
ρ I ρI
j=1
If the internal energy and the Gibbs free energy do not depend upon the plastic volume change we get, using (106),
n
" #
ρ X p ∂ρ I
ξ1 tr s2 + ξ2 tr s3 −
Qj − qbj ≥ 0 . (211)
ρI ρ ∂qj
j=1
Equations (210) and (211) can be used to restrict the possible forms of ξ1 and ξ2 . If the latter, the first two terms
represent the rate of deviatoric plastic work. The terms containing p represents the rate of volumetric plastic work.
The terms containing Qj represent the rate of storage of cold work.
Recall that the 2nd P-K stress SI refers to the intermediate configuration and is unaffected by rigid body rotations.
If we define the yield function as
fI (SI ; T, qj ) ≤ 0 (212)
then the yield surface is also unaffected by rigid body rotations.
Recall that
ρI −1 ρI T
SI = Fe · σ · Fe−T = Q · Ve−1 · σ · Ve−1 · Q . (213)
ρ ρ
Hence the yield surface may alternatively be expressed as
!
ρI T
fI Q · Ve−1 · σ · Ve−1 · Q; T, qj ≤ 0 . (214)
ρ
26
7.7.1 Isotropic elastic material and scalar internal variables
If the material is isotropic and all the internal variables are scalars, then fI is an isotropic function of SI . Also, fI
is unaffected by rigid body rotations. Hence we can write the yield function as
!
ρI
fI · Ve−1 · σ · Ve−1 ; T, qj ≤ 0 (215)
ρ
or
ρI
fI Je Ve−1 · σ · Ve−1 ; T, qj ≤ 0 ; Je := det(Fe ) =
. (216)
ρ
Also, if the material follows an isotropic elastic constitutive relation, then the Cauchy stress (σ) is an isotropic
function of the left stretch (Ve ). Hence the yield function may alternatively be expressed as
f (σ; T, qj ) ≤ 0 . (217)
This function is also invariant with respect to rigid body rotations and is an isotropic function of σ.
Recall that the plastic rate of deformation in the current configuration can be written as
On the yield surface, the plastic rate of deformation is zero. Hence the arbitrariness of σ implies that on the yield
surface
ϕ0 = ϕ1 = ϕ2 = 0 . (219)
Also, if we consider the alternative expression for dp from equation (204):
1 qb0 1 qb0
1 1
2
s − tr s2 1 2
s − tr s2 1
dp = − 1 + (ϕ1 − 2 p ϕ2 ) s + ϕ2 =− 1 + ξ1 s + ξ2 (220)
3 ρI 3 3 ρI 3
qb0 = ξ1 = ξ2 = 0 . (221)
If one of the internal variables is a second order tensor, then frame indifference of the yield function is more
difficult to achieve in general. More details can be found in [SW01].
It is common in plasticity theory for the rate of plastic deformation dp to be derived from a flow potential φ such
that
∂φ
dp = . (222)
∂σ
The principle of maximum plastic dissipation is has been used by some researchers to prove the existence of a
plastic potential. However, the existence of such a potential does not have the same theoretical standing as the free
energy and internal energy potentials.
27
7.8.1 Isotropic material with scalar internal variables
For an isotropic material with scalar internal variables, the plastic flow potential can be assumed to have the form
φ ≡ φ(p, J2 , J3 ; T, q0 , qj ) (223)
where p is the pressure and J2 , J3 are invariants of the deviatoric stress s defined as
1 1 1
J2 := − (tr (s))2 − tr s2 = tr s2 = s : s
2 2 2 (224)
1
J3 := det(s) = tr s3 .
3
Then,
∂φ ∂φ ∂p ∂φ ∂J2 ∂φ ∂J3
dp = = + + . (225)
∂σ ∂p ∂σ ∂J2 ∂σ ∂J3 ∂σ
Now, using the expressions for the derivatives of the invariants of a second-order tensor (see [TN92], p. 26,
equation 9.11 and Appendix, item 16) and noting that tr (s) = 0, we have
1 ∂p 1
p = − tr (σ) =⇒ =− 1
3 ∂σ 3
1 ∂J2
J2 = tr s2
=⇒ =s (226)
2 ∂σ
∂J3 1
= s2 − tr s2 1 .
J3 = det(s) =⇒
∂σ 2
Therefore (225) can be written as
1 ∂φ ∂φ ∂φ 2 1 2
dp = − 1+ s+ s − tr s 1 . (227)
3 ∂p ∂J2 ∂J3 2
Hence,
∂φ
tr (dp ) = − (228)
∂p
and
1 ∂φ ∂φ 2 1 2
ηp = dp − tr (dp ) = s+ s − tr s 1 . (229)
3 ∂J2 ∂J3 2
Clearly
∂φ 1 ∂φ
ηp = + 1 (230)
∂σ 3 ∂p
implies that the usual assumption that ηp = ∂φ/∂σ only holds when ∂φ/∂p = 0.
From equation (204) we have
1 qb0
2 1 2
dp = − 1 + ξ1 s + ξ2 s − tr s 1 . (231)
3 ρI 3
Comparing expressions (225) and (231) we see that
∂φ qb0 ∂φ ∂φ
= ; = ξi ; = ξ2 . (232)
∂p ρI ∂J2 ∂J3
28
We can normalize equation (231) to get
! !2
1 qb0 √ s s 1
dp = − 1 + ξ1 ( s : s) √ + ξ2 (s : s) √ − 1 (233)
3 ρI s:s s:s 3
or,
! !2
1 qb0 p s s 1
dp = − 1 + ξ1 ( 2 J2 ) √ + ξ2 (2 J2 ) √ − 1 (234)
3 ρI 2 J2 2 J2 3
or,
!2
1 qb0 p s s 2
dp = − 1 + ξ1 J2 √ + ξ2 J2 √ − 1 (235)
3 ρI J2 J2 3
Define, p
Γ1 := 2 ξ1 J2 ; Γ2 := 3 ξ2 J2 . (236)
Then " #
1 qb0 Γ1 s Γ2 s2 2
dp = − 1+ √ + − 1 . (237)
3 ρI 2 J2 3 J2 3
Hence, the derivatives of the flow potential are also required to be related to the functions Γ1 and Γ2 vis
∂φ qb0 ∂φ Γ1 ∂φ Γ2
= ; = ξi = √ ; = ξ2 = . (238)
∂p ρI ∂J2 2 J2 ∂J3 3 J2
8 Appendix
1. If the internal energy and the Gibbs free energy have the functional dependence
g = g(SI , T, q0 , qj ) and e = e(SI , T, q0 , qj )
show that
1 ∂e
» –
∂η ∂Ee
= − q0−1 SI : . (239)
∂T T ∂T ∂T
Recall that (for j = 1 . . . n)
Therefore,
∂g ∂e ∂η ∂Ee
=− +η+T + q0−1 SI : . (241)
∂T ∂T ∂T ∂T
Now,
∂g
=η. (242)
∂T
Hence,
∂e ∂η ∂Ee
0=− +T + q0−1 SI : (243)
∂T ∂T ∂T
or,
1 ∂e
» –
∂η ∂Ee
= − q0−1 SI : . (244)
∂T T ∂T ∂T
29
2. Show that if
e = e(Ee , T, q0 , qj ) and η = η(Ee , T, q0 , qj ) (245)
then
∂η 1 ∂e
= . (246)
∂T T ∂T
Recall that the Helmholtz free energy is given by
Therefore,
∂ψ ∂e ∂η
= −η−T . (248)
∂T ∂T ∂T
Now
∂ψ
= −η . (249)
∂T
Hence we have
∂e ∂η
0= −T (250)
∂T ∂T
or,
∂η 1 ∂e
= . (251)
∂T T ∂T
show that
∂η ∂SI
= q0−1 . (253)
∂Ee ∂T
Recall that
∂ψ
= −η . (254)
∂T
Therefore,
∂2ψ
„ «
∂ ∂ψ ∂η
= =− . (255)
∂Ee ∂T ∂T ∂Ee ∂Ee
Now
∂ψ
= q0−1 SI . (256)
∂Ee
Hence,
∂η ∂SI
= −q0−1 . (257)
∂Ee ∂T
4. For thermoplastic materials, show that the specific heats are related by the relation
„ «
∂SI ∂Ee
Cp − Cv = q0−1 SI − T : . (258)
∂T ∂T
The specific heats at constant strain and constant stress are defined as
∂e(Ee , T, q0 , qj ) ∂e(SI , T, q0 , qj )
Cv := and Cp := . (259)
∂T ∂T
From the previous results in this Appendix, we have
∂e(Ee , T, q0 , qj ∂η(Ee , T, q0 , qj )
=T (260)
∂T ∂T
and
∂e(SI , T, q0 , qj ) ∂η(SI , T, q0 , qj ) ∂Ee
=T + q0−1 SI : . (261)
∂T ∂T ∂T
Therefore,
∂η(SI , T, q0 , qj ) ∂Ee ∂η(Ee , T, q0 , qj )
Cp − Cv = T + q0−1 SI : −T . (262)
∂T ∂T ∂T
30
Now, note that
η = η(Ee , T, q0 , qj ) = η(SI , T, q0 , qj ) . (263)
Therefore we can write
∂η(SI , T, q0 , qj ) ∂η(Ee , T, q0 , qj ) ∂Ee ∂η(Ee , T, q0 , qj )
= : + . (264)
∂T ∂Ee ∂T ∂T
Hence,
∂η(Ee , T, q0 , qj ) ∂Ee ∂η(Ee , T, q0 , qj ) ∂Ee ∂η(Ee , T, q0 , qj )
Cp − Cv = T : +T + q0−1 SI : −T (265)
∂Ee ∂T ∂T ∂T ∂T
or, „ «
∂η(Ee , T, q0 , qj ) ∂Ee
Cp − Cv = T + q0−1 SI : . (266)
∂Ee ∂T
Since
∂η ∂SI
= q0−1
∂Ee ∂T
we then have
„ «
∂SI ∂Ee
Cp − Cv = q0−1 −T + SI : . (267)
∂T ∂T
5. Show that
ρ
» “ ”–
1 ” 1“
σ:d= SI : FeT · Ḟe + ḞeT · Fe + FeT · Fe · Ḟp · Fp−1 + Fp−T · ḞpT · FeT · Fe . (268)
ρI 2 2
Recall that “ ” “ ” “ ” “ ”
tr AT · B = tr A · B T = tr B T · A = tr B · AT = A : B = B : A . (269)
Then, from equations (71) and (68) we have
“ ”
σ : d = tr σ T · d = tr (σ · d)
ρ “ ”
= tr (Fe · SI · FeT ) · (Ḟe · Fe−1 + Fe−T · ḞeT + Fe · Ḟp · Fp−1 · Fe−1 + Fe−T · Fp−T · ḞpT · FeT )
2 ρI
(270)
ρ h “ ” “ ” “ ”
= tr Fe · SI · FeT · Ḟe · Fe−1 + tr Fe · SI · ḞeT + tr Fe · SI · FeT · Fe · Ḟp · Fp−1 · Fe−1
2 ρI
“ ”i
+ tr Fe · SI · Fp−T · ḞpT · FeT ) .
Since “ ” “ ”
tr AT · B = tr B · AT (271)
we have “ ” “ ” “ ”
tr [Fe · SI · FeT ] · [Ḟe · Fe−1 ] = tr [Ḟe · Fe−1 ] · [Fe · SI · FeT ] = tr Ḟe · SI · FeT
“ ” (272)
= tr SI · FeT · Ḟe = SI : (FeT · Ḟe ) .
Also, “ ” “ ”
tr Fe · SI · ḞeT = tr SI · ḞeT · Fe = SI : (ḞeT · Fe ) . (273)
Similarly,
“ ” “ ”
tr [Fe · SI · FeT ] · [Fe · Ḟp · Fp−1 · Fe−1 ] = tr [Fe · Ḟp · Fp−1 · Fe−1 ] · [Fe · SI · FeT ]
“ ”
= tr Fe · Ḟp · Fp−1 · SI · FeT (274)
“ ”
= tr SI · FeT · Fe · Ḟp · Fp−1 = SI : (FeT · Fe · Ḟp · Fp−1 ) .
And, “ ” “ ”
tr Fe · SI · Fp−T · ḞpT · FeT ) = tr Fp−T · ḞpT · FeT · Fe · SI = SI : (Fp−T · ḞpT · FeT · Fe ) . (275)
Therefore,
ρ
» “ ”–
1 T T
” 1“
T −1 −T T T
σ:d= SI : Fe · Ḟe + Ḟe · Fe + Fe · Fe · Ḟp · Fp + Fp · Ḟp · Fe · Fe . (276)
ρI 2 2
31
6. Show that
1hb b ˙ ˙
i
Fe · F p · Fbp−1 · Fbe−1 + (Fbe · Fb p · Fbp−1 · Fbe−1 )T
ηp = (277)
2
The volumetric/distortional split of the deformation gradient is given by
where Fb is the distortional component of the deformation gradient. Clearly, we can perform the same decomposition for the elastic
and plastic parts of the deformation gradient, i.e.,
where
Je := det(Fe ) ; det(Fbe ) = 1 ; Jp := det(Fp ) ; det(Fbp ) = 1 . (280)
Since
J = det(F ) = det(Fe · Fp ) = det(Fe ) det(Fp ) = Je Jp . (281)
we can check that the product of the two decompositions gives us the original decomposition of f back, i.e.,
F = Fe · Fp = (Je1/3 Fbe ) · (Jp1/3 Fbp ) = J 1/3 Fbe · Fbp = J 1/3 Fb =⇒ Fb = Fbe · Fbp (282)
and
det(F ) = det(Fe · Fp ) = det(Je1/3 Fbe ) det(Jp1/3 Fbp ) = Je det(Fbe ) Jp det(Fbp ) = Je Jp = J . (283)
Recall that
1h i
dp =Fe · Ḟp · Fp−1 · Fe−1 + (Fe · Ḟp · Fp−1 · Fe−1 )T . (284)
2
We want to express the quantities Fe and Fp in the expression in terms of their volumetric and distortional components. Let us
consider the rate quantities first. We thus have
d 1/3 b ˙ 1 dJp b ˙
Fp = Jp1/3 Fbp =⇒ Ḟp = (Jp ) Fp + Jp1/3 Fb p = Jp−2/3 Fp + Jp1/3 Fb p . (285)
dt 3 dt
Now, from (88) and (92), we have
dJp ρ̇I
= Jp tr (dp ) = −Jp . (286)
dt ρI
Therefore, " #
1 1/3 ρ̇I b ˙
1/3 b 1/3 ρ̇I ˙
Ḟp = − Jp Fp + Jp F p = Jp − Fbp + Fb p . (287)
3 ρI 3 ρI
Also,
Fp−1 = Jp−1/3 Fbp−1 . (288)
Then " #
ρ̇I ˙ ρ̇I ˙
Ḟp · Fp−1 = − Fbp + Fb p · Fbp−1 = − 1 + Fb p · Fbp−1 . (289)
3 ρI 3 ρI
This leads to the next step
" #
“ ” ρ̇I ˙b
“ ”
Fe · Ḟp · Fp−1 · Fe−1 1/3 b
= Je Fe · − 1 + F p · Fbp −1
· Je−1/3 Fbe−1
3 ρI
" #
ρ̇I ˙ −1
= Fe · −
b 1 + F p · Fp
b b · Fbe−1 (290)
3 ρI
ρ̇I ˙
=− 1 + Fbe · Fb p · Fbp−1 · Fbe−1 .
3 ρI
Using equation (92) we get
1 ˙
Fe · Ḟp · Fp−1 · Fe−1 = tr (dp ) 1 + Fbe · Fb p · Fbp−1 · Fbe−1 . (291)
3
It follows that
1 ˙
(Fe · Ḟp · Fp−1 · Fe−1 )T = tr (dp ) 1 + (Fbe · Fb p · Fbp−1 · Fbe−1 )T . (292)
3
Therefore, » –
1 1 ˙ 1 ˙
dp = tr (dp ) 1 + Fbe · Fb p · Fbp−1 · Fbe−1 + tr (dp ) 1 + (Fbe · Fb p · Fbp−1 · Fbe−1 )T (293)
2 3 3
32
or,
1 1hb b ˙ ˙
i
dp =tr (dp ) 1 + Fe · F p · Fbp−1 · Fbe−1 + (Fbe · Fb p · Fbp−1 · Fbe−1 )T . (294)
3 2
Comparing the above equation with (95) we see that
1hb b ˙ ˙
i
ηp = Fe · F p · Fbp−1 · Fbe−1 + (Fbe · Fb p · Fbp−1 · Fbe−1 )T (295)
2
This expression contains only distortional terms of the deformation gradients and hence must be distortional too. Therefore, the
standard volumetric/deviatoric split of dp is acceptable.
n
(299)
X ∂g
dψ = −q0−1 Ee : dSI − η dT − dqj + q0−1 Ee : dSI + q0−1 SI : dEe
j=1
∂q j
or,
n
X ∂g
de = − dqj + T dη + q0−1 SI : dEe
j=1
∂q j
n
(300)
X ∂g
dψ = −η dT − dqj + q0−1 SI : dEe .
j=1
∂q j
n
X ∂g
de = q0−1 SI : dEe + T dη − dqj (301)
j=1
∂q j
n
X ∂g
dψ = q0−1 SI : dEe − η dT − dqj .
j=1
∂q j
33
Comparing (301) and (303), we clearly see that this special circumstance leads to equations that are equivalent to assuming that
Q0 = 0.
and
Ḟ = V̇e · Q · Up + Ve · Q̇ · Up + Ve · Q · U̇p ; Ḟ T = UpT · QT · V̇eT + UpT · Q̇T · VeT + U̇pT · QT · VeT . (308)
Also,
V̇e = V̇eT ; U̇p = U̇pT ; (310)
Hence
F −T = Ve−1 · Q · Up−1 ; Ḟ T = Up · QT · V̇e + Up · Q̇T · Ve + U̇p · QT · Ve . (311)
The velocity gradient can then be expressed as
and “ ”
σ : (∇v)T = σ : (Ve−1 · V̇e ) + tr σ · Ve−1 · Q · Q̇T · Ve + σ : (Ve−1 · Q · Up−1 · U̇p · QT · Ve )
(315)
= σ : (Ve−1 · V̇e ) + (Q · Q̇T ) : (Ve · σ · Ve−1 ) + σ : (Ve−1 · Q · Up−1 · U̇p · QT · Ve ) .
Now,
Q · QT = 1 =⇒ Q̇ · QT = −Q · Q̇T . (316)
Hence,
σ : (∇v)T = σ : (Ve−1 · V̇e ) − (Q̇ · QT ) : (Ve · σ · Ve−1 ) + σ : (Ve−1 · Q · Up−1 · U̇p · QT · Ve ) . (317)
Adding the two terms, we get
1h
σ:d= σ : (V̇e · Ve−1 + Ve−1 · V̇e ) + (Q̇ · QT ) : (Ve−1 · σ · Ve − Ve · σ · Ve−1 )+
2 i (318)
σ : (Ve · Q · U̇p · Up−1 · QT · Ve−1 + Ve−1 · Q · Up−1 · U̇p · QT · Ve ) .
34
Since σ and Ve are coaxial, their spectral decompositions (eigenprojections) are
3
X 3
X
σ= σi ni ⊗ ni ; Ve = λj nj ⊗ nj (319)
i=1 j=1
and
3
X 1
Ve−1 = nk ⊗ nk . (320)
λk
k=1
Therefore, ! ! !
3 3 3
X 1 X X
Ve−1 · σ · Ve = nk ⊗ nk · σi ni ⊗ ni · λj nj ⊗ nj ; (321)
λk i=1 j=1
k=1
Since (
0 if i 6= j
(ni ⊗ ni ) · (nj ⊗ nj ) = (322)
ni ⊗ ni if i = j
we have
3
X
Ve−1 · σ · Ve = σi ni ⊗ ni = Ve · σ · Ve−1 . (323)
i=1
Hence the stress power takes the form
1h
σ:d= σ : (V̇e · Ve−1 + Ve−1 · V̇e )+
2 i (324)
σ : (Ve · Q · U̇p · Up−1 · QT · Ve−1 + Ve−1 · Q · Up−1 · U̇p · QT · Ve ) .
Since
Ve = Q · Ue · QT (326)
we have
Ve−1 = Q · Ue−1 · QT ; V̇e = Q̇ · Ue · QT + Q · U̇e · QT + Q · Ue · Q̇T . (327)
Therefore,
σ : (V̇e · Ve−1 ) = σ : [(Q̇ · Ue · QT + Q · U̇e · QT + Q · Ue · Q̇T ) · (Q · Ue−1 · QT )]
= σ : [Q̇ · QT + Q · U̇e · Ue−1 · QT + Q · Ue · Q̇T · Q · Ue−1 · QT ]
σ : (Ve−1 · V̇e ) = σ : [(Q · Ue−1 · QT ) · (Q̇ · Ue · QT + Q · U̇e · QT + Q · Ue · Q̇T )]
= σ : [Q · Ue−1 · QT · Q̇ · Ue · QT + Q · Ue−1 · U̇e · QT + Q · Q̇T ]
(328)
σ : (Ve · Q · U̇p · Up−1 · QT · Ve−1 ) = σ : [(Q · Ue · QT ) · (Q · U̇p · Up−1 · QT ) · (Q · Ue−1 · QT )]
= σ : [Q · Ue · U̇p · Up−1 · Ue−1 · QT ]
σ : (Ve−1 · Q · Up−1 · U̇p · QT · Ve ) = σ : [(Q · Ue−1 · QT ) · (Q · Up−1 · U̇p · QT ) · (Q · Ue · QT )]
= σ : [Q · Ue−1 · Up−1 · U̇p · Ue · QT ] .
Since the stress power does not contain any terms containing Q̇, we can ignore these terms to get
1 h
σ:d= σ : (Q · U̇e · Ue−1 · QT + Q · Ue−1 · U̇e · QT )
2 i (329)
+ σ : (Q · Ue · U̇p · Up−1 · Ue−1 · QT + Q · Ue−1 · Up−1 · U̇p · Ue · QT ) .
or,
1 h n o
σ:d= σ : Q · (U̇e · Ue−1 + Ue−1 · U̇e ) · QT
2 n oi (330)
+ σ : Q · (Ue · U̇p · Up−1 · Ue−1 + Ue−1 · Up−1 · U̇p · Ue ) · QT .
35
10. If
Fe = Q · Ue and Fp = Up (331)
show that the equations
1 h i
de = Ḟe · Fe−1 + (Ḟe · Fe−1 )T
2 (332)
1 h i
dp = Fe · Ḟp · Fp−1 · Fe−1 + (Fe · Ḟp · Fp−1 · Fe−1 )T .
2
can be expressed as
1 h i
de = Q · (U̇e · Ue−1 + Ue−1 · U̇e ) · QT
2 (333)
1 h i
dp = Q · (Ue · U̇p · Up−1 · Ue−1 + Ue−1 · Up−1 · U̇pT · Ue ) · QT .
2
Since
Fe = Q · Ue and Fp = Up (334)
we have
Ḟe = Q̇ · Ue + Q · U̇e ; Fe−1 = Ue−1 · QT ; Ḟp = U̇p ; Fp−1 = Up−1 . (335)
Therefore,
σ : (Ḟe · Fe−1 ) = σ : [(Q̇ · Ue + Q · U̇e ) · (Ue−1 · QT )] = σ : (Q̇ · QT + Q · U̇e · Ue−1 · QT )
σ : (Ḟe · Fe−1 )T = σ : (Q · Q̇T + Q · Ue−1 · U̇e · QT )
(336)
σ : (Fe · Ḟp · Fp−1 · Fe−1 ) = σ : (Q · Ue · U̇p · Up−1 · Ue−1 · QT )
σ : (Fe · Ḟp · Fp−1 · Fe−1 )T = σ : (Q · Ue−1 · Up−1 · U̇pT · Ue · QT )
Since σ : d = σ : (de + dp ) does not have any terms containing Q̇, we have
1h i
σ : de = σ : (Q · U̇e · Ue−1 · QT ) + σ : (Q · Ue−1 · U̇e · QT )
2 (337)
1h i
σ : dp = σ : (Q · Ue · U̇p · Up−1 · Ue−1 · QT ) + σ : (Q · Ue−1 · Up−1 · U̇pT · Ue · QT )
2
Therefore,
1 h i
de = Q · (U̇e · Ue−1 + Ue−1 · U̇e ) · QT
2 (338)
1 h i
dp = Q · (Ue · U̇p · Up−1 · Ue−1 + Ue−1 · Up−1 · U̇pT · Ue ) · QT .
2
11. Show that the stress power can be expressed in the intermediate configuration as
P = Je−1 SI : (De + Dp ) . (339)
where
1 h i
De = U̇e · Ue + Ue · U̇e
2 (340)
1h 2 i
Dp = Ue · U̇p · Up−1 + Up−1 · U̇p · Ue2 .
2
36
Using the definitions of de and dp from (165) we have
1h i
Ue · QT · de · Q · Ue = Ue · (U̇e · Ue−1 + Ue−1 · U̇e ) · Ue
2
1h i
= Ue · U̇e + U̇e · Ue
2 (344)
1h i
T
Ue · Q · dp · Q · Ue = Ue · (Ue · U̇p · Up−1 · Ue−1 + Ue−1 · Up−1 · U̇p · Ue ) · Ue
2
1h 2 i
= Ue · U̇p · Up−1 + Up−1 · U̇p · Ue2 .
2
Recall from equations (74) and (75) that
1 “ T ”
De = Ḟe · Fe + FeT · Ḟe
2 (345)
1“ T ”
Dp = Fe · Fe · Ḟp · Fp−1 + Fp−T · ḞpT · FeT · Fe .
2
Recall that Fe = Q · Ue and Fp = Up which implies that
Ḟe = Q̇ · Ue + Q · U̇e ; ḞeT = Ue · Q̇T + U̇e · QT ; FeT = Ue · QT ; Ḟp = U̇p ; Fp−1 = Up−1 ; Fp−T = Up−1 . (346)
Ḟ = Q̇ · Ue · Up + Q · U̇e · Up + Q · Ue · U̇p and Ḟ T = U̇pT · UeT · QT + UpT · U̇eT · QT + UpT · UeT · Q̇T . (353)
37
Adding the two terms, we get
“ ”
Ḟ T · F + F T · Ḟ = U̇p · Ue2 · Up + Up · Ue2 · U̇p + Up · U̇e · Ue + Ue · U̇e · Up
“ ” (355)
+ Up · Ue · Q̇T · Q · Ue + Ue · QT · Q̇ · Ue · Up .
or,
d “ ”
(U 2 ) = Up · (Up−1 · U̇p · Ue2 + Ue2 · U̇p · Up−1 ) · Up + Up · U̇e · Ue + Ue · U̇e · Up . (357)
dt
Comparing with equation (168) we see that
d
(U 2 ) = Up · (Dp + De ) · Up . (358)
dt
13. Show that the strain rate in the intermediate configuration can be expressed as
! n
ρ̇I X ∂Qj
De = Ėe = ρI S : ṠI − SI + αe Ṫ − q̇j (359)
ρI j=0
∂SI
where
∂2g ∂Ee
S := ; αe := . (360)
∂SI2 ∂T
Now, from (74) we have De = Ėe and from (33) we have
∂g
Ee = ρ I . (361)
∂SI
Hence De is completely determined by the Gibbs free energy functional g. Also
∂2g ρ̇I
„ «
∂g ∂ ∂g
Ėe = ρ̇I + ρI = Ee + ρ I . (362)
∂SI ∂t∂SI ρI ∂SI ∂t
Now,
∂g ∂Ee ∂2g
Ee = ρ I =⇒ = q0 (367)
∂SI ∂SI ∂SI2
and from equation (45) we have
∂η ∂Ee
= q0−1 . (368)
∂SI ∂T
38
Hence we can write
n
∂2g ∂2g
„ «
∂ ∂g ∂Ee X ∂Qj
= 2
: ṠI + q0−1 Ṫ − q0−2 Ee q̇0 − q0−1 SI : q̇ −
2 0
q0−1 q̇j . (369)
∂SI ∂t ∂SI ∂T ∂SI j=0
∂SI
Recall that „ «
∂ ∂g
Ėe = q0−1 Ee q̇0 + q0 . (370)
∂SI ∂t
Plugging the expression from (369) into (370), we get
n
∂2g ∂Ee ∂2g X ∂Qj
Ėe = q0 : Ṡ I + Ṫ − S I : q̇ 0 − q̇j . (371)
∂SI2 ∂T ∂SI2 j=0
∂SI
If we define the fourth-order elastic stiffness tensor and the second-order thermal expansion tensor as
∂2g ∂Ee
S := ; αe := (373)
∂SI2 ∂T
we then have
! n
ρ̇I X ∂Qj
De = Ėe = ρI S : ṠI − SI + αe Ṫ − q̇j . (374)
ρI j=0
∂SI
14. Let S be a second-order tensor and let qb(S) be a scalar valued function of S. Show that qb is an isotropic function of S if and only if
there exists a function qe(I1 , I2 , I3 ) such that
qb(S) = qe(I1 , I2 , I3 ) (375)
where I1 , I2 , I3 are the invariants of S defined as
1 ˆ
(tr (S))2 − tr S 2 ; I3 (S) = det(S) .
` ´˜
I1 (S) = tr (S) ; I2 (S) = (376)
2
The scalar function qb is invariant under rigid body rotations R if
An isotropic function is a function that is invariant under rigid body rotations. We will first show that the three invariants of S are
isotropic functions.
Thus, “ ” “ ”
I1 (R · S · RT ) = tr R · S · RT = tr S · RT · R = tr (S) = I1 (S)
1 h “ ” “ ”i 1 h “ ”i
I2 (R · S · RT ) = (tr R · S · RT )2 − tr (R · S · RT )2 = (tr (S))2 − tr R · S · S · RT
2 2 (378)
1 ˆ 2 ` 2 ´˜
= (tr (S)) − tr S = I2 (S)
2
I3 (R · S · RT ) = det(R · S · RT ) = det(R) det(S) det(RT ) = det(S) = I3 (S) .
Hence I1 , I2 , I3 are isotropic functions of S. Therefore, any function qe(I1 , I2 , I3 ) is also an isotropic function of S.
Conversely, we can establish the requirement in equation (375) by assuming that qb is isotropic and then showing that qb(S) = qb(B)
whenever the invariants of S and B are identical and vice versa.
If the tensor S is symmetric then it is completely characterized by its invariants via the spectral decomposition theorem. Let us
therefore assume that S and B are symmetric. Then we can write
3
X 3
X
S= λi si ⊗ si ; B = λi bi ⊗ bi . (379)
i=1 i=1
39
Also let si = Q · bi where Q is a rotation. Then
Therefore
3
X 3
X
Q · B · QT = λi Q · (bi ⊗ bi ) · QT = λi si ⊗ si ) = S . (381)
i=1 i=1
But qb is isotropic, i.e. qb(Q · B · QT ) = qb(B). Therefore qb(S) = qb(B) and hence the representation (375) holds.
d(σ)
b = ϕ0 (I1 , I2 , I3 ) 1 + ϕ1 (I1 , I2 , I3 ) σ + ϕ2 (I1 , I2 , I3 ) σ 2 . (382)
R · d(σ)
b · RT = d(R
b · σ · RT ) (383)
R · d(σ)
b · RT = ϕ0 (Iσ ) R · RT + ϕ1 (Iσ ) R · σ · RT + ϕ2 (Iσ ) R · σ 2 · RT
= ϕ0 (Iσ ) 1 + ϕ1 (Iσ ) R · RT · B · R · RT + ϕ2 (Iσ ) R · RT · B · R · RT · B · R · RT (384)
2
= ϕ0 (Iσ ) 1 + ϕ1 (Iσ ) B + ϕ2 (Iσ ) B .
Since the invariants of σ are isotropic functions, we have
Iσ = IB . (385)
Therefore,
R · d(σ)
b · RT = ϕ0 (IB ) 1 + ϕ1 (IB ) B + ϕ2 (IB ) B 2 = d(B)
b b · σ · RT ) .
= d(R (386)
This implies that the function d(σ)
b is isotropic.
We prove the converse by using the spectral representation of σ. Special treatments are needed for the cases where eigenvalues are
repeated. Let us examine the case where the eigenvalues are distinct (for the other cases consult [Gur81], p. 234).
Let the spectral representation of σ be
3
X
σ= σi ni ⊗ ni . (387)
i=1
d(σ)
b = α0 (σ) 1 + α1 (σ) σ + α2 (σ) σ 2 (390)
16. Let A be a second order tensor and let I1 , I2 , I3 be its principal invariants. Show that
∂I1 ∂I2 ∂I3
=1 ; = I1 1 − AT ; = (A2 − I1 A + I2 1 )T . (391)
∂A ∂A ∂A
The derivative of a scalar valued function φ(A) of a second order tensor A can be defined via the directional derivative using
˛
∂φ d ˛
:B= φ(A + s B)˛˛ (392)
∂A ds s=0
40
where B is an arbitrary second order tensor.
The invariant I3 is given by
I3 (A) = det(A) . (393)
Therefore, from the definition of the derivative,
˛
∂I3 d ˛
:B= det(A + s B)˛˛
∂A ds s=0
" !#˛
d 1 −1
˛
= det s A 1 +A ·B ˛
˛
ds s ˛ (394)
s=0
" !#˛
d 3 1 −1
˛
= s det(A) det 1 +A ·B ˛ .
˛
ds s ˛
s=0
Recall that we can expand the determinant of a tensor in the form of a characteristic equation in terms of the invariants I1 , I2 , I3
using
det(λ 1 + A) = λ3 + I1 (A) λ2 + I2 (A) λ + I3 (A) . (395)
Using this expansion we can write
" !#˛
∂I3 d 1 −1 1 −1 1 −1
˛
:B= s3 det(A) + I1 (A · B) 2 + I2 (A · B) + I3 (A · B) ˛
˛
∂A ds s 3 s s ˛
s=0
˛
d ˆ
1 + I1 (A−1 · B) s + I2 (A−1 · B) s2 + I3 (A−1 · B) s3 ˛˛
˜ ˛
= det(A) (396)
ds s=0
= det(A) I1 (A−1 · B) .
Recall that the invariant I1 is given by
I1 (A) = tr (A) . (397)
Hence,
∂I3
: B = det(A) tr A−1 · B = det(A) [A−1 ]T : B .
` ´
(398)
∂A
Invoking the arbitrariness of B we then have
∂I3
= det(A) [A−1 ]T . (399)
∂A
In an orthonormal basis the components of A can be written as a matrix A. In that case, the right hand side corresponds the
cofactors of the matrix.
For the derivatives of the other two invariants, let us go back to the characteristic equation
det(λ 1 + A) = λ3 + I1 (A) λ2 + I2 (A) λ + I3 (A) . (400)
Using the same approach as before, we can show that
∂
det(λ 1 + A) = det(λ 1 + A) [(λ 1 + A)−1 ]T . (401)
∂A
Now the left hand side can be expanded as
∂ ∂ ˆ 3
λ + I1 (A) λ2 + I2 (A) λ + I3 (A)
˜
det(λ 1 + A) =
∂A ∂A (402)
∂I1 2 ∂I2 ∂I3
= λ + λ+ .
∂A ∂A ∂A
Hence
∂I1 2 ∂I2 ∂I3
λ + λ+ = det(λ 1 + A) [(λ 1 + A)−1 ]T (403)
∂A ∂A ∂A
or, » –
∂I1 2 ∂I2 ∂I3
(λ 1 + A)T · λ + λ+ = det(λ 1 + A) 1 . (404)
∂A ∂A ∂A
Expanding the right hand side and separating terms on the left hand side gives
» –
∂I1 2 ∂I2 ∂I3
(λ 1 + AT ) · = λ3 + I1 λ2 + I2 λ + I3 1
ˆ ˜
λ + λ+ (405)
∂A ∂A ∂A
41
or,
» –
∂I1 3 ∂I2 2 ∂I3 ∂I1 2 ∂I2 ∂I3
λ 1 + +AT · λ + AT · λ + AT · = λ3 + I1 λ2 + I2 λ + I3 1 .
ˆ ˜
λ + λ + (406)
∂A ∂A ∂A ∂A ∂A ∂A
If we define I0 := 1 and I4 := 0, we can write the above as
» –
∂I1 3 ∂I2 2 ∂I3 ∂I4 ∂I0 3 ∂I1 2 ∂I2 ∂I3
1 +AT · λ +AT · λ +AT · λ+AT · = I0 λ3 + I1 λ2 + I2 λ + I3 1 .
ˆ ˜
λ + λ + λ+
∂A ∂A ∂A ∂A ∂A ∂A ∂A ∂A
(407)
Collecting terms containing various powers of λ, we get
„ « „ «
∂I1 ∂I0 ∂I2 ∂I1
λ3 I0 1 − 1 − AT · + λ2 I1 1 − 1 − AT · +
∂A ∂A ∂A ∂A
„ « „ « (408)
∂I3 ∂I2 ∂I4 ∂I3
λ I2 1 − 1 − AT · + I3 1 − 1 − AT · =0.
∂A ∂A ∂A ∂A
Then, invoking the arbitrariness of λ, we have
∂I1 ∂I0
I0 1 − 1 − AT · =0
∂A ∂A
∂I2 ∂I3 ∂I2
I1 1 − 1 − I2 1 − 1 − AT · =0 (409)
∂A ∂A ∂A
∂I4 ∂I3
I3 1 − 1 − AT · =0.
∂A ∂A
This implies that
References
[Gur81] M. E. Gurtin. An Introduction to Continuum Mechanics. Academic Press, New York, 1981.
[TN92] C. Truesdell and W. Noll. The Non-linear Field Theories of Mechanics. Springer-Verlag, New York, 1992.
[Wri02] T. W. Wright. The Physics and Mathematics of Adiabatic Shear Bands. Cambridge University Press,
Cambridge, UK, 2002.
43