Journal of the Mechanics and Physics of Solids

49 (2001) 1539 – 1568

www.elsevier.com/locate/jmps

On the characterization of geometrically necessary

dislocations in *nite plasticity
Paolo Cermellia , Morton E. Gurtinb;∗
a Dipartimento
di Matematica, Universita di Torino, I-10123 Torino, Italy
b Department of Mathematics, Carnegie Mellon University, Pittsburgh, PA 15213-3190, USA

Received 5 May 2000; received in revised form 1 December 2000

Abstract
We develop a general theory of geometrically necessary dislocations based on the decompo-
sition F = Fe Fp . The incompatibility of Fe and that of Fp are characterized by a single tensor
G giving the Burgers vector, measured and reckoned per unit area in the microstructural (inter-
mediate) con*guration. We show that G may be expressed in terms of Fp and the referential
curl of Fp , or equivalently in terms of Fe−1 and the spatial curl of Fe−1 . We derive explicit
relations for G in terms of Euler angles for a rigid-plastic material and — without neglecting
elastic strains — for strict plane strain and strict anti-plane shear. We discuss the relationship
between G and the distortion of microstructural planes. We show that kinematics alone yields
a balance law for the transport of geometrically necessary dislocations.
c 2001 Published by
Elsevier Science Ltd.

Keywords: A. Dislocations; B. Crystal plasticity; B. Finite strain

1. Introduction

Modern treatments of *nite plasticity are based on the Kr?oner (1960) — Lee (1969)
decomposition F = Fe Fp of the deformation gradient F = ∇y into structural (elastic)
and plastic components, where x = y(X; t) represents the deformation that carries ma-
terial points X in the reference con*guration into their positions x at time t in the
deformed con*guration. For a single crystal, Fp (X) represents the “local deformation”
of referential line segments to line segments dl = Fp (X) dX in the microstructural
(or lattice) con+guration, a deformation resulting solely from the formation of defects

∗Corresponding author. Fax: +1-412-268-6380.

E-mail addresses: cermelli@dm.unito.it (P. Cermelli), mg0c@andrew.cmu.edu (M.E. Gurtin).

0022-5096/01/$ - see front matter  c 2001 Published by Elsevier Science Ltd.

PII: S 0 0 2 2 - 5 0 9 6 ( 0 0 ) 0 0 0 8 4 - 3
1540 P. Cermelli, M.E. Gurtin / J. Mech. Phys. Solids 49 (2001) 1539 – 1568

such as dislocations; Fe (X) represents the “local deformation” of the segments dl into
segments dx = Fe (X) dl due to stretching and rotation of the lattice.
An important feature of the Kr?oner–Lee decomposition is that, while F is compatible
(the gradient of a vector *eld), Fe and Fp are generally incompatible, a property related
to the formation of dislocations. Such dislocations are termed geometrically necessary,
as they arise solely from the underlying kinematics, and their intrinsic characterization
is basic to general theories of plasticity.
In crystal physics dislocations may be quanti*ed by the Burgers vector, which rep-
resents the closure de*cit of circuits deformed from a perfect lattice, and one may ask
whether in a continuum theory it is possible to characterize such dislocations through a
tensor *eld G that measures the local Burgers vector per unit area. In fact, the problem
is not the absence of such a *eld, but rather the plethora of *elds that have appeared
in the literature. Here our central task is to show that there is but a single measure of
geometrically necessary dislocations consistent with physically motivated requirements.
To place our ideas in context we begin, after *xing notation, with a brief historical
discussion of this question.
With the exception of Subsection 2.1,
∇; Div; and Curl
denote the gradient, divergence, and curl with respect to the material point X in the
reference con+guration;
grad; div; and curl
denote the divergence, gradient, and curl with respect to the point x = y(X; t) in the
deformed con+guration; ’˙ denotes the material time derivative (holding X *xed).

1.1. Historical perspective. Critical comments

The early treatments of geometrically necessary dislocations, generally referred to as

GNDs, are based on a representation of the crystalline structure by three independent
director *elds ek (x). These *elds are viewed as the generators, on a microscopic scale,
of a Bravais lattice deformed via a tensor *eld A(x) through a transformation ek (x) =
A(x)ck of a basis ck of a *xed reference lattice L. Defectiveness is then measured,
in the spirit of Burgers, by quantifying the closure failure of in*nitesimal circuits in
this microscopic lattice. This point of view is easily reconcilied with the formulation in
terms of the Kr?oner–Lee decomposition upon identifying the elastic deformation tensor
Fe with A. This approach to defectiveness is apparently due to Kondo (1952, 1955),
who framed his treatment within a diKerential–geometric framework using the theory
of connections with defectiveness characterized by the non-holonomicity of the director
*elds. His measure of GNDs — the torsion of the connection generated by the director
*elds — results in a third-order tensor *eld that equivalently may be expressed as the
second-order tensor *eld
GeKo = J e Fe−1 curl Fe−1 (1.1)
e e
with J = det F . At about the same time, and independently, Nye (1953) used physical
arguments to justify a formula relating the local Burgers vector to the local rotation of
 P. Cermelli, M.E. Gurtin / J. Mech. Phys. Solids 49 (2001) 1539 – 1568 1541

the directors; tacit in Nye’s work is the neglect of elastic strains and the assumption
of in*nitesimal rotations, and his resulting defect measure is

curl We

with We the skew tensor that represents in*nitesimal rotations of the lattice. Based
on Nye’s ideas, Bilby et al. (1955), Eshelby (1956), Fox (1966), and Acharya and
Bassani (2000), respectively, proposed tensor *elds essentially equivalent to

curl Fe Fe−T ; curl Fe ; Fe−1 curl Fe−1 ; and ReT curl Fe−1 (1.2)

as measures of GNDs in *nite deformations.

An approach essentially diKerent from those described above is that of Kr?oner
(1960), who, based on the Kr?oner–Lee decomposition, introduces the defect measures

GeKr = curl Fe−1 FeT and GpKr = Curl Fp : (1.3)

Kr?oner then goes on to develop a complete theory within the context of in*nitesimal
displacements, where GeKr = GpKr . This is apparently the *rst instance in which the
problem of the equivalence of elastic and plastic defect measures is addressed.
The variety of measures described above — each proposed as a representation of
GNDs — begs the question as to which measure(s), if any, have an intrinsic physical
meaning. Requirements, not mutually independent, that one might consider as reason-
able characterizations of such a measure are that:
(i) G should measure the local Burgers vector in the microstructural con*guration,
per unit area in that con*guration;
(ii) G should, at any point, be expressible in terms of the +eld Fp in a neighborhood
of the point, since, by *at, Fp characterizes the defect structure near the point in
question;
(iii) G should be invariant under superposed compatible elastic deformations and also
under compatible local changes in reference con*guration, since these — being
compatible — should not result in an intrinsic change in the distribution of GNDs
near any point.
Note that conditions (ii) and (iii) trivially render G frame-indiKerent; of (1.2), only
(1:2)3; 4 are frame-indiKerent. It is clear from the work of Teodosiu (1970, 1982) and
Davini (1986) that the measure GeKo satis*es (i) and (iii), but (ii) would seem prob-
lematic. On the other hand, GpKr trivially satis*es (ii), but violates (i). The remaining
measures listed in (1.2) violate (i) and (iii). Teodosiu (1970, 1982) noted that GeT Ko n
represents the Burgers vector — measured in the reference lattice — for an in*nitesimal
circuit in L enclosing a surface element with unit normal n. Davini (1986) showed
that GeKo is invariant under superposed compatible elastic deformations (cf. Davini and
Parry, 1989; Cermelli and Sellers, 2000). Finally, Noll (1967) was led to GeKo in his
theory of materially uniform simple bodies with inhomogeneities, GpKr was used by
Naghdi and Srinivasa (1993), and GeKr was used by D lużewski (1996).
1542 P. Cermelli, M.E. Gurtin / J. Mech. Phys. Solids 49 (2001) 1539 – 1568

1.2. The geometric dislocation tensor G

Our main result is the existence of a tensorial measure of GNDs, the geometric
dislocation tensor G, consistent with requirements (i) – (iii) stated above. In physical
terms,

GT n gives the local Burgers vector in the microstructural con+guration — per

unit area in that con+guration — for those dislocation lines piercing the plane
with unit normal n.

We refer to GT n as the Burgers vector for . What is most important, G may be

expressed in a form
1 p
G= F Curl Fp ; (1.4)
Jp
that depends only on the plastic part Fp of the deformation gradient and equivalently
in Kondo’s form
G = Je Fe−1 curl Fe−1 ; (1.5)
that depends only on the *eld Fe . Because we do not require that det Fp = 1, our
discussion allows for the formation of voids and the interaction of voids with other
defects.
The importance of the alternative plastic and elastic representations of G is that
(i) For situations in which lattice strains are small, GNDs would seem amenable to
experimental study through measurement of lattice rotations (cf. Sun et al., 1998,
2000); the relation for G in terms of Fe (in this case a rotation) allows for its
experimental study.
(ii) In developing a constitutive theory that allows for GNDs, it would seem advanta-
geous to use the representation for G in terms of Fp , which characterizes defects,
leaving Fe to describe the stretching and rotation of the lattice. The importance of
the relation for G in terms of Fp is underlined by a relation we derive giving Ġ
for a single crystal as a function of Fp , Curl Fp , the slips, and the slip gradients.
Given a *xed unit vector n, consider the microstructural plane normal to n. We
show that n · G(x)n — the normal component of the Burgers vector for — is related
to the distortion of . Precisely, we show that n · Gn = 0 everywhere if and only if
is undistorted; that is, if and only if convects to a family of smooth surfaces in
the deformed con*guration. The *eld n · Gn might therefore be useful as a constitutive
quantity related to hardening due to the formation of GNDs. This *eld, which we term
the “distortion modulus”, accounts for the normal component of the Burgers vector of
those dislocations impinging transversally on . When is a slip plane, these are
traditionally termed “forest dislocations” and are thought to be responsible for stage II
hardening (Kuhlmann-Wilsdorf, 1989).
When elastic strains are negligible, so that Fe is a rotation Re , GNDs are amenable
to experimental study through the measurement of lattice rotations as described by
 P. Cermelli, M.E. Gurtin / J. Mech. Phys. Solids 49 (2001) 1539 – 1568 1543

Re . Material scientists typically describe rotations by means of Euler angles via a

decomposition of the form

Re = Q3 Q2 Q1 ; Qi (x) = Q(#i (x); ci )

in which the ci are constant unit vectors and Q(#i (x); ci ) represents a counterclockwise
rotation about ci through an angle #i (x). As one of our main results we show that the
geometric dislocation tensor GO = Re GReT referred to the deformed con*guration has
the simple form
3


GO = {cOi ⊗ grad #i − (cOi · grad #i )1}
i=1

with cOi being the vectors ci convected to the deformed con*guration.

1.3. Strict plane strain and strict anti-plane shear

We consider generalizations of plane strain and anti-plane shear de*ned by the re-
quirement that in each case the tensor *elds Fp and Fe have a form identical to that
of F; we use the adjective strict to denote these generalizations.
For strict plane strain G is a “pure edge tensor”

G = e ⊗ g; g⊥e

with e being the out-of-plane normal. A consequence of this relation is that microstruc-
tural planes parallel to e are undistorted. If the material is rigid plastic, then, for #e
the rotation angle corresponding to Re , g may be expressed in the simple form

g = ReT grad #e :

Thus g rotated to the deformed con*guration is normal to surfaces #e = constant.

For strict anti-plane shear with e the unit vector that de*nes the shear axis, Fp =
1 + e ⊗
p , with
p orthogonal to e, and the geometric dislocation tensor is a “pure
screw tensor”

G = e ⊗ h; h = curl
p ; he:

Microstructural planes perpendicular to e are therefore distorted at any point at which

curl
p = 0.

1.4. Dynamics

Kinematics alone yields a balance law for the transport of GNDs. Consider a pre-
scribed microstructural plane with unit normal l and a prescribed unit vector b,
which need bear no relation to l. Then (l; b) = l · Gb is the component in the di-
rection b of the Burgers vector, per unit area, on . We show that any such (signed)
1544 P. Cermelli, M.E. Gurtin / J. Mech. Phys. Solids 49 (2001) 1539 – 1568

density 1  = (l; b) evolves in accord with the balance law

˙ = −Div qR + R
with dislocation Pux qR and dislocation supply R de*ned by
qR = Fp−1 (l × (LpT b)); R = l · Lp Gb + b · Lp Gl;
p
where Lp = Ḟ Fp−1 . This balance is one of our main results; one of its more interesting
consequences is that plastic Pow, as characterized by non-vanishing values of Lp , is
always associated with a Pux of dislocations, whether or not G vanishes. Here we
emphasize that G — and hence the density  — accounts only for GNDs, as it measures
the net dislocation density associated with macroscopic incompatibility.
Finally, we derive expressions for single crystals showing the explicit relationship
of Ġ to the slips (microshear rates) and their gradients.

2. Preliminaries

2.1. Mappings of surfaces

Let h be a smooth one-to-one mapping of a region D1 of R3 onto another such

region D2 , with H = ∇h; det H ¿ 0. Further, let Si ⊂ Di (i = 1; 2) be smooth surfaces
with S2 = h(S1 ), and with Si oriented by a smooth unit-normal *eld ni . Then, modulo
a change in the sign of n2 ,
H−T n1
n2 = ; (2.1)
|H−T n1 |
where H−T = (HT )−1 . The surface jacobian of h as a mapping of S1 onto S2 is given
by — = (det H)|H−T n1 |, so that, for f a scalar *eld and T a tensor *eld,
   
f dA2 = f— dA1 ; Tn2 dA2 = (det H)TH−T n1 dA1 (2.2)
S2 S1 S2 S1

with dAi the area measure on Si . (More precisely, if the domain of T is D1 , then the
integral over D2 should involve the composition T ◦ h; similarly for f.) We refer to
the vector measures ni dAi (i = 1; 2) as surface elements with unit normal ni and area
dAi . Since the *elds f and T are arbitrary, we view (2.2), formally, as asserting that
the surface element n1 dA1 on S1 is mapped by h into the surface element
n2 dA2 = (det H)H−T n1 dA1 (2.3)

1 An alternative approach to the modeling of dislocations has been suggested by Nye (1953). Noting that
the Burgers vector of a single dislocation in a crystal must belong to a well-de*ned crystallographically
determined set, dislocations are accounted for by assigning independent densities of screw and edge type,
each corresponding to dislocations with a given Burgers vector in this set. These elementary dislocations
may be combined to form the tensor G, but G does not uniquely characterize the elementary densities.
For instance, in the “fcc-deconstruction” (Sun et al., 1998, 2000) there are 18 elementary densities (cf.
footnote 2), but only 9 independent components of G, a diQculty overcome by a minimization technique
that furnishes a lower bound for the total density.
 P. Cermelli, M.E. Gurtin / J. Mech. Phys. Solids 49 (2001) 1539 – 1568 1545

on S2 (with area element dA1 mapped into dA2 = — dA1 and n1 mapped into n2 via
(2.1)).

2.2. The curl operator. Stokes theorem. The skew tensor w×

This subsection concerns tensor analysis in R3 with ∇ and curl the underlying
gradient and curl operators.
The curl of a tensor *eld T is the tensor *eld de*ned by
(curl T)c = curl(TT c) for all constant vectors c:
Let  be the boundary curve of a smooth surface S oriented by a continuous unit
normal *eld n, with the boundary curve  oriented in a manner consistent with Stokes’
theorem for smooth vector *elds f:
 
f · dx = (curl f) · n dA: (2.4)
 S

We then have Stokes’ theorem for a tensor +eld T:

 
T dx = (curl T)T n dA: (2.5)
 S

The veri*cation of (2.5) is immediate: simply apply (2.4) with f = TT c and c constant.
When convenient we use the standard notation of cartesian tensor analysis — in-
cluding summation convention — with respect to the basis e1 = (1; 0; 0), e2 = (0; 1; 0),
e3 = (0; 0; 1). In particular, the component form of curl T is given by
@Tjs
(curl T)ij = irs
@xr
with irs the alternating symbol. (A cautionary note: for some authors the curl of T is
the transpose of our curl T:)
For e a unit vector, the tensor
P(e) = 1 − e ⊗ e (2.6)
is the projection onto the plane perpendicular to e, while e⊥ denotes the plane per-
pendicular to e.
Given any vector w, w× is the skew tensor de*ned by
(w×)c = w × c for all vectors c;
in components (w×)ij = irj wr . Then
(w×)(u×) = u ⊗ w − (u · w)1; (2.7)
so that, for |w| = 1, (w×)(w×) = −P(w). Given any skew tensor W, there is a unique
vector w, called the axial vector of W, such that
W=w×:
1546 P. Cermelli, M.E. Gurtin / J. Mech. Phys. Solids 49 (2001) 1539 – 1568

Let  be a scalar *eld, let f and u be vector *elds, let S and T be tensor *elds with
T constant, and let a be a constant vector. The following identities will be useful:

curl ∇f = 0;
curl(TS) = (curl S)TT ;
curl(T) = (∇×)TT ;
a · curl f = (a×) · ∇f;
curl(u ⊗ f) = (curl f) ⊗ u − (f×)(∇u)T ;
curl(f×) = (div f)1 − ∇f: (2.8)

Moreover, if rather than being a constant, T() is a function of , then

 T
dT
curl(TS) = (curl S)TT + (∇×) S : (2.9)
d
To establish these identities note *rst that, for c a constant vector,
(curl ∇f)c = curl{(∇f)T c} = curl∇(f · c) = 0;
{curl(TS)}c = curl{ST (TT )c)} = (curl S)TT c;
{curl(T)}c = curl( TT c) = (∇×)TT c;
which veri*es (2:8)1−3 . Identity (2:8)4 follows from the de*nitions of curl and a×.
Identity (2:8)5 follows from the computation
{curl(u ⊗ f)}c = curl {(c · u)f} = (c · u)(curl f) − f × ∇(c · u)
= {(curl f) ⊗ u − (f×)(∇u)T }c;
while (2:8)6 is a consequence of the relation {curl(f×)}c = curl(c × f) and standard
vector identities.
Assume next that T = T(). The curl of TS is the curl holding T *xed plus the curl
holding S *xed. Denoting the latter by curlT , we may use (2:8)2 to conclude that
curl(TS) = (curl S)TT + curlT (TS):
Further, for c a constant vector,
 T
dT
{curlT (TS)}c = curlT {(TS)T c} = (∇×) S c;
d
so that curlT (TS) = (∇×)((dT=d)S)T ; which yields (2.9).
The tensor e× satis*es e × =P(e)(e×)P(e). Thus, since e× is skew and P(e) sym-
metric, we have the identity
(e×) · S = (e×) · { P(e)S P(e)} = (e×) · skw{P(e)SP(e)} (2.10)
for any tensor S. Here and, in what follows, we de*ne the symmetric and skew parts
of a tensor S by
sym S = 12 (S + ST ); skw S = 12 (S − ST ):
 P. Cermelli, M.E. Gurtin / J. Mech. Phys. Solids 49 (2001) 1539 – 1568 1547

Finally, since (det T)ijk = pqr Tip Tjq Tkr ; it follows that, for T invertible,
T(b×)TT = (det T)(T−T b) × : (2.11)

3. Basic kinematics

3.1. Macroscopic kinematics. Plastic strain. Structural deformation tensor

Consider a body BR identi*ed with the open region of space it occupies in a *xed
reference con*guration, and assume that, in this con*guration, BR is homogeneous,
although possibly defective.
Let X denote an arbitrary material point of BR . Assume that the body is evolving,
but *x the time and suppress it in what follows. A motion of BR (at that time) is then
a smooth one-to-one mapping
x = y(X)
with deformation gradient
F = ∇y (3.1)
consistent with J = det F ¿ 0:
The conceptual hypothesis underlying the theory is that there is a “microscopic
structure” — a lattice in the case of a single crystal — with respect to which micro-
scopic kinematical hypotheses can be framed. Speci*cally, the theory is based on the
decomposition
F = F e Fp ; (3.2)
p
where F , the plastic strain, represents the distortion of material elements due to the
formation of defects in the microscopic structure, while Fe , the structural deformation
tensor, represents stretching and rotation of this structure. Unlike F, the tensors Fp
and Fe do not generally correspond to deformations (i.e., are not gradients of vector
*elds), but because Fp and Fe are invertible, we may view these tensors as deforma-
tions of in*nitesimal neighborhoods. We use the term microstructural con+guration
— or lattice con+guration in the case of a single crystal — for the collection of in-
*nitesimal con*gurations obtained by applying Fp locally to reference increments dX
or, equivalently, by applying Fe−1 locally to increments dx. By (3.2),
J = Je Jp ; Je = det Fe ; Jp = det Fp (3.3)
and we assume, without loss in generality, that Je ¿ 0 and Jp ¿ 0. For speci*city, we
consider the microstructural *elds Fe ; Fp ; Je , and Jp as functions of X on BR .
The tensor *eld Lp de*ned by the relation
p
Ḟ = Lp Fp (3.4)
represents the plastic strain rate, measured in the microstructural con*guration. A
p
standard identity then yields the analogous relation J˙ = Jp tr Lp : Particular microscopic
structures are often characterized by restrictions on the form of the tensor Lp .
1548 P. Cermelli, M.E. Gurtin / J. Mech. Phys. Solids 49 (2001) 1539 – 1568

The polar decompositions

Fp = Rp Up = Vp Rp ;
F e = R e Ue = V e R e (3.5)
de*ne the plastic and elastic rotations Rp and Re , the plastic and elastic right-stretch
tensors Up and Ue , and the plastic and elastic left-stretch tensors Vp and Ve . A case
of special interest corresponds to situations in which the microscopic structure may
be treated as rigid, so that the sole source of local deformation is due to the Pow of
defects. Such materials — referred to as rigid-plastic — are de*ned by the restriction
Ue = Ve = 1 (so that Fe = Re ; Je = 1): (3.6)

3.2. Convection of geometric quantities

We use the term convect to indicate the manner in which geometrical objects
“deform” during the motion. Thus the reference body BR convects to the deformed
body
def
BO = y(BR );
an oriented surface SR with unit normal nR convects to the oriented surface SO = y(SR )
with unit normal
F−T nR
nO = (3.7)
|F−T nR |
(cf. (2.1)); the surface element nR dAR on SR convects to the surface element nO d AO on
SO de*ned by
nO d AO = J F−T nR dAR (3.8)
Consistent with this, we use the following notation for a geometrical object {: : :}
(such as a unit normal *eld) that may be described relative to some or all of the
underlying con*gurations:
(a) {: : :}R denotes its representation in the reference con*guration;
(b) {: : :} (no embellishments) denotes its representation in the microstructural con*g-
uration;
(c) {: : :} denotes its representation in the deformed con*guration.
We use this scheme even when the quantity has no representation in the microstructural
con*guration (e.g., BR is the reference body and BO is the deformed body), but in each
case the quantities will be consistent with our use of the term “convect” in the sense
that {: : :} convects from {: : :}R and convects to {: : :}, and so forth.
The ambient space of the microstructural con*guration consists of a collection of
copies of R3 , one copy L(X) for each material point X. L(X) should be viewed
as the ambient space into which an in*nitesimal neighborhood of X is carried by the
linear transformation Fp (X) — or to which an in*nitesimal neighborhood of x=y(X) is
carried backwards by Fe−1 (x). (Throughout, we use abbreviations such as Fe−1 =(Fe )−1
 P. Cermelli, M.E. Gurtin / J. Mech. Phys. Solids 49 (2001) 1539 – 1568 1549

and Fp−T = (Fp )−T :) The operation of integration is physically meaningless on L(X),
as integration is not local, but the notion of a surface element n dA (with unit normal n
and area dA) does have meaning, as n dA is local. Thus, bearing in mind the paragraph
containing (2.3), we formally stipulate that a unit normal nR and surface element
nR dAR at X convect to the unit normal and surface element
Fp−T nR
n= and n dA = Jp Fp−T nR dAR (3.9)
|Fp−T nR |
in L(X), and that n and n dA convect to the unit normal and surface element
Fe−T n
nO = and nO d AO = J e Fe−T n dA (3.10)
|Fe−T n|
at x in the deformed con*guration. Since F = Fe Fp , (3:9)2 and (3:10)2 are consistent
with (3:8). It suQces to consider relations (3:9)2 and (3:10)2 as formal, as we shall
use them only to make meaningful the notion of a tensorial density measured per unit
area in the lattice con*guration.

3.3. Single crystals. Slip

A microscopic structure of particular importance is a single crystal. In this case we

restrict attention to plastic Pow induced by the motion of dislocations on prescribed
slip systems " = 1; 2; : : : ; A, with each system " de*ned by a slip direction s" and a
slip-plane normal m" , where
s" · m" = 0; |s" |; |m" | = 1; s" ; m" = constant: (3.11)
The plane with normal m" and the line on this plane de*ned by s" then represent the
slip plane and the slip line for ", and the tensor
S " = s " ⊗ m" (3.12)
is referred to as the Schmid tensor for ".
The presumption that Pow take place through slip manifests itself in the requirement
that the evolution of Fp be governed by slips (microshear-rates) " on the individual
slip systems via the Pow rule
A

A


Lp = " (s" ⊗ m" ) = " S" : (3.13)
"=1 "=1
p
By (3:11)1 and (3.13), tr Lp = 0 and a consequence of the diKerential equation J˙ =
p p p p
J tr L is that J = 1 for all time if J = 1 initially.

3.4. Isochoric Fp . Decomposition for Fp non-isochoric

To characterize internal damage due solely to the formation of voids one might
restrict the plastic strain to a dilatation,
Fp = #−1 1; Jp = #−3 (3.14)
1550 P. Cermelli, M.E. Gurtin / J. Mech. Phys. Solids 49 (2001) 1539 – 1568

with # the stretch from the microstructural con*guration to the reference; (3.4) then
yields the Pow rule
Lp = $1; $ = −ln˙#: (3.15)
p
An arbitrary plastic strain F may be decomposed into the product
Fp = #−1 Fp0 ; Jp = #−3 ; (3.16)
of a dilatation #−1 1 and a plastic strain that is isochoric in the sense that det Fp0 =1:
Fp0
Then, by (3.4), the plastic Pow, as represented by Lp , admits the additive decompo-
sition
Lp = $1 + Lp0 ;

into dilatational and isochoric Pows as represented by $1, $ = −ln˙#, and Lp0 = Ḟ0 Fp−1
p
0
with tr Lp0 = 0: This decomposition with Lp0 in the form (3.13) would represent the
interaction of slip with void-formation in single crystals.

3.5. Plane strain. Strict plane strain

Under plane strain the deformation x = y(X) has the component form
xi = yi (X1 ; X2 ) (i = 1; 2); x3 = X3 :
Plane strain results in a deformation gradient F and a velocity gradient L that are
independent of X3 and consistent with
F = P(e)FP(e) + e ⊗ e; L = P(e)LP(e); e ≡ e3 ; (3.17)
so that their component matrices have the respective forms
   
· · 0 · · 0
 · · 0;  · · 0:
0 0 1 0 0 0
Tensor *elds that are independent of X3 and of the form (3:17)1 are termed planar.
Note that this de*nition excludes L, which annihilates vectors parallel to e.
The constraint of plane strain does not generally render Fp and Fe planar tensors.
Indeed, for a rigid-plastic material, the polar decomposition (3.5) yields F = (Re Rp )Up
with Up the square root of FT F; hence Up and Re Rp must be planar, but individually
Re and Rp need not be planar.
A direct consequence of (3.4) is that, if Fp is planar at some time, then Fp and
(hence) Fe are planar if and only if Lp e = LpT e = 0 (so that Lp has the form (3:17)2 ).
We use the term strict plane strain to describe plane strain with Fp and Fe planar
tensor +elds. Under strict plane strain, Fp e = e and FpT e = e, and similarly for Fe (and
F); hence referential planes perpendicular to e convect to planes perpendicular to e
in the microstructural and deformed con+gurations.
For single crystals one can be assured of strict plane strain provided one restricts
attention to planar slip systems; that is, slip systems " that satisfy
s" · e = 0; m" · e = 0; s " × m" = e (3.18)
 P. Cermelli, M.E. Gurtin / J. Mech. Phys. Solids 49 (2001) 1539 – 1568 1551

with slips " independent of X3 . All other slip systems are ignored. There is a large
literature based on this hypothesis. The resulting fully two-dimensional kinematics is
important in constructing simple mathematical models, often based on two slip systems.
(Cf., e.g., Asaro, 1983, pp. 45 – 46, 84 –97 and the references therein; Prantil et al.,
1993. Cf. also Kalinindi and Anand (1993) who discuss plane strain allowing for a
full three-dimensional collection of slip systems.)

4. Burgers vector. The geometric dislocation tensor G

By (2:8)1 , Curl Fp = 0 when the plastic strain is compatible (the gradient of a vector
*eld); Curl Fp therefore provides a measure of the incompatibility of the plastic strain.
By Stokes theorem, for @SR the boundary curve of a smooth surface SR in the reference
body,
 
bp (@SR ) ≡ Fp dX = (Curl Fp )T nR dAR : (4.1)
@SR SR

For a single crystal the microscopic structure is a lattice and, since the vector
(Curl Fp )T nR lies in the lattice con+guration, one might associate (Curl Fp )T nR dAR
with the Burgers vector corresponding to the boundary curve of a surface element with
normal nR , but that would be incorrect, as the surface element nR dAR lies in the ref-
erence con+guration rather than in the lattice con*guration. This is easily recti*ed. By
(3:9)2 ; n dA = Jp Fp−T nR dAR is the surface element in the lattice con*guration from
which nR dAR convects; thus, formally,
1
(Curl Fp )T nR dAR = (Curl Fp )T FpT n dA (4.2)
Jp
with n dA the surface element in the lattice con*guration, so that (1=Jp )(Fp Curl Fp )T n dA
is the local Burgers vector corresponding to the “boundary curve” of the surface ele-
ment n dA in the lattice con+guration. Thus, for
1 p
Gp = F Curl Fp ; (4.3)
Jp
GpT n provides a measure — based on the plastic strain — of the (local) Burgers vector
in the lattice con*guration — per unit area in that con*guration — for the plane
with unit normal n.
On the other hand, let SO be a smooth surface in the deformed body and consider
the line integral
 
e O
b (@S) ≡ F e−1
dx = (curl Fe−1 )T nO d AO (4.4)
@SO SO

(Fe−1
= (F ) ). As before, curl Fe−1 = 0 when Fe is compatible; curl Fe−1 therefore
e −1

provides a measure of the local incompatibility of the structural deformation. Arguing

as in the steps leading to (4.3), we may use (3:10)2 , again formally, to conclude that
(curl Fe−1 )T nO d AO = J e (curl Fe−1 )T Fe−T n dA:
1552 P. Cermelli, M.E. Gurtin / J. Mech. Phys. Solids 49 (2001) 1539 – 1568

Thus, by (4.4), for

Ge = J e Fe−1 curl Fe−1 ;
GeT n provides a measure — based on the elastic deformation — of the Burgers vector,
per unit area in the lattice con*guration, for the plane with unit normal n. The *elds
GpT n and GeT n purportedly characterize the same Burgers vector. To reconcile this,
note that, if SO convects from SR , then
  
p p −1
F dX = F F F dX = Fe−1 dx;
@SR @SR @SO
O = bp (@SR ). Therefore, by (3.8),
so that be (@S)
  
(Curl Fp )T nR dAR = (curl Fe−1 )T nO d AO = (J (curl Fe−1 )T F−T nR ) dAR ;
SR SO SR

and, as SO is arbitrary, (Curl F ) = J (curl F ) F :

p T e−1 T −T

Finally, since F = Fe Fp and J = J e Jp , we arrive at our main result: The tensor +elds
G and Ge coincide. We refer to
p

1
G = p Fp Curl Fp = J e Fe−1 curl Fe−1 (4.5)
J
as the geometric dislocation tensor. Identity (4.5), our main result, is central to most
of what follows. Reiterating, GT n represents the (local) Burgers vector in the mi-
crostructural con*guration — per unit area in that con*guration — for the plane
with unit normal n; or in more physical terms, the local Burgers vector, per unit area,
for those dislocation lines piercing . Here and in what follows, we refer to GT n as
the Burgers vector even when the microstructural con+guration does not represent a
single crystal.
An important consequence of (4.5) is that arguments regarding G are invariant under
the replacement of the *eld Fp by the *eld Fe−1 provided operations in the reference
con*guration are replaced by analogous operations in the deformed con*guration.

5. Canonical decompositions of the geometric dislocation tensor

Because of its tensorial nature, G bears some comparison to the tensors of strain
and stress. An in*nitesimal strain tensor may be written as a sum of simple extensions
in mutually perpendicular directions, or equivalently as a purely volumetric strain plus
simple shears on three mutually perpendicular planes. Decompositions of this type
apply also to G, but with diKerent physical interpretation. The building blocks of
such decompositions form the content of the following de*nitions; in these de*nitions
attention is focused on a given material point, and l is a unit vector.
(a) G is a pure edge-tensor if
G = l ⊗ g with g perpendicular to l; (5.1)
so that g, the Burgers vector for l⊥ , is parallel to the plane l⊥ . In this case g is
termed the principal Burgers vector and l is the line direction.
 P. Cermelli, M.E. Gurtin / J. Mech. Phys. Solids 49 (2001) 1539 – 1568 1553

(b) G is a pure screw-tensor if

G=l⊗h with h parallel to l; (5.2)
T ⊥ ⊥
so that G l = h, the Burgers vector for the plane l , is perpendicular to l . In this
case h is termed the principal Burgers vector and l is the line direction.
(c) G is an axial edge-tensor if
G = ×;
so that, given any n, the Burgers vector GT n is always parallel to the plane n⊥ .
(d) G is an isotropic screw-tensor if
G = ’1;
so that, given any n, the Burgers vector GT n is always perpendicular to the plane
n⊥ .
As we shall see, G is a pure edge-tensor for strict plane strain and a pure screw-tensor
for strict anti-plane shear, and in each case the line direction is normal to the cross-
sectional plane of the body.
Our next result gives canonical decompositions of G into “pure tensors” of the
above form. Given any material point, the geometric dislocation tensor G may be
decomposed:
(i) into an isotropic screw-tensor plus a sum of three pure edge-tensors with respect
to mutually orthogonal planes; speci+cally, there are a scalar ’, an orthonormal
basis {l1 ; l2 ; l3 }, and vectors {g1 ; g2 ; g3 } with gi perpendicular to li such that
3


G = ’1 + li ⊗ g i ; (5.3)
i=1
(ii) into an axial edge-tensor plus a sum of three pure screw-tensors with mutually
orthogonal Burgers vectors; speci+cally, there are a vector , an orthonormal
basis {l1 ; l2 ; l3 }, and vectors {h1 ; h2 ; h3 } with hi parallel to li such that
3


G = (×) + li ⊗ hi : (5.4)
i=1

(The orthonormal basis {l1 ; l2 ; l3 } in (i) is generally diKerent from that in (ii).)
The veri*cation of these decompositions is not diQcult. Consequences of standard
results are that G may be written in the form G = ’1 + G0 with tr G0 = 0. Further,
there are scalars {.1 ; .2 ; .3 } and an orthonormal basis {l1 ; l2 ; l3 } such that sym G0 =
2 sym(.3 N12 + .2 N31 + .1 N23 ) with Nij = li ⊗ lj . (This is simply the decomposition of
a deviatoric “strain-tensor” into simple shears on mutually orthogonal planes; cf., e.g.,
Gurtin, 1972, p. 36.) Further, given this decomposition, there are scalars {/1 ; /2 ; /3 }
such that skw G0 = −2 skw(/3 N12 + /2 N31 + /1 N23 ). Thus, letting g1 = (.3 − /3 )l2 +
(.2 + /2 )l3 ; g2 = : : : ; g3 = : : : (with g2 and g3 obtained by cyclic permutation of the
3
indices), we *nd that G0 = i=1 li ⊗gi ; which implies (5.3), since G0 =sym G0 +skw G0 .
Finally, there are an orthonormal basis {l1 ; l2 ; l3 }, scalars {#1 ; #2 ; #3 }, and a vector 
3
such that sym G = i=1 #i li ⊗ li with skw G = ×, which yields (5.4).
1554 P. Cermelli, M.E. Gurtin / J. Mech. Phys. Solids 49 (2001) 1539 – 1568

Consequences of this argument are the identities

3

3


tr G = 3’; |G0 |2 = |gi |2 ; |skw G|2 = 2||2 ; |sym G|2 = |hi |2 : (5.5)
i=1 i=1

By (3.14), for the special case in which Fp is a dilatation, G is an axial edge-tensor:

G = −(∇# ×): (5.6)
More generally, as a consequence of (2.8) we have a second canonical decomposition
of G: Consider the decomposition Fp = #−1 Fp0 ; det Fp0 = 1, of Fp into isochoric and
dilatational parts (cf. (3.16)). Then
G = #G0 − Fp0 (∇# ×) FpT p p
0 ; G0 = F0 Curl F0 ; (5.7)
rendering G0 the geometric dislocation tensor for the isochoric part of Fp .

6. G described relative to the reference and deformed con2gurations

G has a referential counterpart that may be obtained by transforming the vector

GT n dA to the reference con*guration by premultiplication by Fp−1 and then converting
n dA to the referential surface element nR dAR using (3:9)2 ; the result is
GR = Jp Fp−1 GFp−T = (Curl Fp )Fp−T ; (6.1)
GTR nR gives the Burgers vector transported back to — and measured per unit area in
— the reference con*guration, nR dAR being the relevant surface element. Similarly,
transforming GT n dA to the deformed con*guration by premultiplication by Fe and
then converting n dA to the deformed surface element nO d AO using (3:10)2 results in the
tensor *eld
1
GO = e Fe GFeT = (curl Fe−1 )FeT ; (6.2)
J
T
GO nO gives the Burgers vector convected to — and measured per unit area in — the
deformed con*guration, nO d AO being the relevant surface element. GR and GO represent
the geometric dislocation tensor G referred, respectively, to the reference and deformed
con*gurations. Note that GO = GeKr (cf. (1.3)).
A homogeneous transformation of the microstructural con+guration is de*ned by
a constant invertible tensor M together with the transformations:
Fp∗ = MFp ; Fe∗ = Fe M−1 (6.3)
(so that F∗ = F). Under such a transformation

1 p∗
G∗ = J e∗ (Fe∗ )−1 curl((Fe∗ )−1 ) = F Curl Fp∗ ;
Jp∗
∗
G∗R = (Curl Fp∗ )(Fp∗ )−T ; GO = curl((Fe∗ )−1 )(Fe∗ )T : (6.4)
 P. Cermelli, M.E. Gurtin / J. Mech. Phys. Solids 49 (2001) 1539 – 1568 1555

By (2:8)2 ; Curl(MFp )=(Curl Fp )MT and curl((Fe M−1 )−1 )=(curl Fe−1 )MT ; we there-
fore have the following transformation law for the geometric dislocation tensor under
a homogeneous transformation of the microstructural con*guration:
G∗ = (det M)−1 MGMT : (6.5)
∗
On the other hand, G∗R O O
= GR and G = G, so that the geometric dislocation tensor —
when referred to the reference or deformed con*guration — is invariant.
For a rigid-plastic material the local relation between the lattice and the deformed
con*guration is a rotation Re , and the lattice as it would appear to an observer is
simply the lattice as framed in the microstructural con*guration rotated via Re . The
tensor
GO = Re GReT = (curl ReT )ReT (6.6)
therefore represents the true Burgers vector; that is, the Burgers vector as seen by
an observer. Measurements of lattice rotations are generally made with respect to the
orientation of the lattice at a particular point x0 in the deformed con*guration. With
this in mind, let Re0 denote the lattice rotation at x0 , so that
Qe = Re ReT
0

represents the rotation from x0 to x. Further, choose M = Re0 in transformation (6.3)

∗
and write GO 0 = GO . Then GO 0 represents the true Burgers vector as reckoned by an
experimenter who measures lattice rotations with respect to x0 , and, by (6:3)2 ,
GO 0 = (curl QeT )QeT :
Thus, when convenient, we may, at any given time, identify the lattice con*guration
with the lattice at a particular point x0 in the deformed con*guration. Consistent with
this one replaces Re by Qe = Re ReT p e p
0 and F by R0 F .

7. Rigid-plastic materials

7.1. Euler-angle representation of G

In situations for which elastic strains are negligible, geometrically necessary disloca-
tions are amenable to experimental study through the measurement of lattice rotations.
Further, since lattice orientation is often characterized through the use of Euler angles,
it would seem important to have at hand a simple relation for G in terms of gradients
of Euler angles. This is the central objective of this section.
The discussion of rotations is greatly simpli*ed using the exponential of a tensor A
as de*ned by the power series

∞
1 k
exp(A) = A ; (7.1)
k!
k=0
a de*nition that renders Z(#)=exp(#A) the unique solution of the initial-value problem
dZ
= AZ; Z(0) = 1: (7.2)
d#
1556 P. Cermelli, M.E. Gurtin / J. Mech. Phys. Solids 49 (2001) 1539 – 1568

If W is a skew tensor, then R = exp(W) is a rotation, and any such rotation R may
be written in this form. In particular, given a unit vector e and a scalar #,
def
Q(#; e) = exp(# e×) (7.3)
represents the rotation about e whose counterclockwise angle of rotation is #. Fix e
and consider (7.3) as a function Q(#) = Q(#; e). Then
dQ dQT
= (e×) Q; = −(e×) QT : (7.4)
d# d#
The *rst identity follows from (7.2). To establish the second, note that, since Q is a
rotation about e; (e×)QT = QT (e×). Thus, since e× is skew, (7:4)1 implies (7:4)2 .
A representation of a rotation R in terms of Euler angles is a decomposition of R
into the product of three rotations
R = Q(#3 ; c3 ) Q(#2 ; c2 ) Q(#1 ; c1 ) (7.5)
of angles #i about unit vectors ci . The standard Euler-angle representation is obtained
by choosing c1 = c3 = k and c2 = i, with (i; j; k) an orthogonal basis; in this case the
angles are generally denoted by #3 = ’1 ; #2 = ; #1 = ’2 , so that
R = Q(’1 ; k) Q( ; i) Q(’2 ; k); ’1 ; ’2 ∈ [0; 2
); ∈ [0;
]: (7.6)
Any rotation may be represented in the standard form (7.6). (Cf., e.g., Synge (1960)
and Kocks et al. (1998) for discussions of this and other rotation-representations com-
mon in mechanics and materials science.)
It is most convenient to work with the geometric dislocation tensor
GO = Re GReT = (curl ReT )ReT
(cf. (6.6)) referred to the deformed con*guration. Consider the Euler-angle decompo-
sition
Re = Q3 Q2 Q1 ; Qi (x) = Q(#i (x); ci );
where the ci are constant unit vectors (with c1 microstructural). Before proceeding with
O note that since c1 is the axis of the initial rotation Q1 ; Q3 Q2 Q1 c1 =
the calculation of G,
Q3 Q2 c1 represents c1 convected to the deformed con*guration; and, since the rotation
Q2 follows Q1 ; Q3 Q2 c2 = Q3 c2 represents the convected value of c2 . The unit vectors
cO1 = Q3 Q2 c1 ; cO2 = Q3 c2 ; cO3 = c3
therefore represent the vectors ci convected to the deformed con*guration.
O note that, by (2.9), (2.11), and (7.4), for B a rotation,
To calculate G,
curl(QTi BT ) = (curl BT )Qi + (grad #i ×)B(ci ×)Qi ;
= (curl BT )Qi + (grad #i ×){(Bci )×}BQi : (7.7)
Applying (7.7) with i = 1 and B = Q3 Q2 yields
curl ReT = {curl(QT2 QT3 )}Q1 + (grad #1 ×){(Q3 Q2 c1 )×}Q3 Q2 Q1 :
 P. Cermelli, M.E. Gurtin / J. Mech. Phys. Solids 49 (2001) 1539 – 1568 1557

Two more applications of (7.7), *rst to curl(QT2 QT3 ) and then to curl(QT3 ), leads to the
expression
curl ReT = {(grad #3 ×)(Oc3 ×) + (grad #2 ×)(Oc2 ×) + (grad #1 ×)(Oc1 ×)}Re :
Thus, in view of (2.7), when represented in terms of Euler angles, the geometric
dislocation tensor — referred to the deformed con+guration — has the form
3


GO = {Oci ⊗ grad #i − (Oci · grad #i )1}: (7.8)
i=1

In using the standard Euler-angle decomposition for Re to compute G, one should

bear in mind that for = 0; Re does not uniquely determine the angles ’1 and ’2 , a
de*ciency that could result in unbounded values of grad ’1 and grad ’2 near points at
which = 0.

7.2. In+nitesimal rotations. Nye’s relation

The power series expansion (7.1) of Re = exp(×),  = #e, yields the formal ap-
proximation
Re ∼ 1 + We ; We = × (7.9)
e
for small rotation-angles #. The tensor *eld W represents the in+nitesimal rotation
and its axial vector  gives the in+nitesimal vector angle of rotation. Consistent with
(7.9), we approximate the geometric dislocation tensor G = ReT curl ReT by the tensor
*eld Ginf = −curl We = −curl(×): Since curl(p×) = (div p)1 − grad p, this yields Nye’s
relation (1953)
Ginf = N − (tr N)1; N = grad : (7.10)

8. When are microstructural planes undistorted?

A given microstructural plane — which in the case of a single crystal may or

may not correspond to a crystalline plane — is represented by its unit normal n. By
(3:9)1 and (3.10), n, which is constant, convects from a +eld nR (X) of unit normals
in the reference con*guration and to a +eld nO (x) of unit normals in the deformed
con*guration, where
#R nR = FpT n; #R = |FpT n|;
On = Fe−T n; #O = |Fe−T n|
#O (8.1)
an interesting question is whether the *eld nO , say, represents a unit normal *eld for
a family of smooth surfaces in the deformed con*guration, either globally or locally.
When this is so, the microstructural plane may be termed undistorted. It is the
purpose of this section to show that — in sense to be made precise — the scalar +eld
n · Gn measures the distortion of the microstructural plane .
Let denote a +xed plane in the microstructural con*guration, with n its unit
normal. We say that is globally undistorted if the unit-normal *eld nO (x) to which
1558 P. Cermelli, M.E. Gurtin / J. Mech. Phys. Solids 49 (2001) 1539 – 1568

n convects is a normal *eld for a family of smooth surfaces in the deformed body B; O
O O
that is, if, given any x in B: (i) x is contained in a single surface S of the family;
and (ii) nO (x) is normal to SO at x. This notion could equally well have been phrased
with respect to the reference con*guration: there is a family of smooth surfaces in BO
for which nO is a normal *eld if and only if there is a family of smooth surfaces in BR
for which nR is a normal *eld.
We now show that: a necessary and suAcient condition that be globally undis-
torted is that n · Gn vanish at each point of the body.
To establish this assertion, note *rst that, by (8.1),
(FpT n) · Curl(FpT n) = #R nR · [#R Curl nR + (∇#R ) × nR ] = #R2 nR · Curl nR
and
2
(Fe−T n) · curl(Fe−T n) = #O nO · curl nO :
Therefore (4.5) yields the interesting relations
2 #R2
n · Gn = J e #O nO · curl nO = nR · Curl nR : (8.2)
Jp
Thus a necessary and suAcient condition that n · Gn vanish at a given point is that
nO · curl nO = 0 (or equivalently, nR · Curl nR = 0). Roughly speaking, nO · curl nO represents
the twist of the vector *eld nO about itself, and similarly for nR · Curl nR . Note that for
a rigid-plastic material,
n · Gn = nO · curl nO :
Finally, a classical theorem of Frobenius (Choquet-Bruhat and DeWitt-Morette, 1977,
p. 235) asserts that a necessary and suQcient condition that nO (x) be a normal *eld for
a family of smooth surfaces in BO is that nO · curl nO = 0 everywhere in B;
O thus, by (8.2),
is globally undistorted if and only if n · Gn ≡ 0.
By (5.6), n · Gn vanishes identically in dilatational Pows; hence in such Pows mi-
crostructural planes are globally undistorted.
One would generally not expect n · Gn to vanish everywhere, and we therefore ask
whether there is a local analog of the foregoing result appropriate to behavior in an
in*nitesimal neighborhood of a given point. With this in mind, choose a point x0 in the
deformed con*guration and consider the underlying *elds as functions of the position
x = y(X) in B. O Let denote a *xed plane in the microstructural con*guration, with
n its unit normal. We say that is locally undistorted at x0 if there is a smooth,
oriented surface MO in BO through x0 such that nO and the unit normal *eld mO for MO
coincide to *rst-order near x0 : given any curve z( ) on MO through x0 with z(0) = x0 ,



d d
On(z(0)) = m(z(0));
O On(z( )) = O
m(z( )) : (8.3)
d =0 d =0

It then follows that: a necessary and suAcient condition that be locally undistorted
at a given point is that n · Gn vanish at that point.
The proof of this assertion, which is technical and beyond the scope of the present
paper, is given by Cermelli and Gurtin (2000, Section 13:1).
 P. Cermelli, M.E. Gurtin / J. Mech. Phys. Solids 49 (2001) 1539 – 1568 1559

9. Strict plane strain

9.1. Principal Burgers vector g

We now discuss the form of the geometric dislocation tensor for strict plane strain
as de*ned in Section 3.5. We begin with two useful identities. Let T be a planar tensor
*eld. Then
Curl T = e ⊗ q; q = (Curl T)T e (9.1)
with q a planar vector *eld. Moreover,
T(e×) = det T(e×)T−T : (9.2)
T
To verify (9.1), note that, for c a constant vector, T c = p + c3 e with p planar; thus,
since Curl p is parallel to e, P(e)(Curl T)c = P(e)Curl(TT c) = P(e) Curl p = 0: But c is
arbitrary, thus P(e) Curl T=0 and, since P(e)=1−e⊗e, Curl T=e⊗(Curl T)T e=e⊗q.
Further, q · e = e · (Curl T)e = e · Curl(TT e) = e · Curl e = 0 and q is planar. This
proves the *rst assertion. Since T is planar, the second, (9.2), follows from (2.11):
T(e×) = det T {(T−T e)×}T−T = det T(e×)T−T :
Under strict plane strain the tensor G simpli*es considerably. By (9.1),
 T
1 p p 1 p p T 1 p
G = p F Curl F = p (F e) ⊗ (Curl F ) e = e ⊗ Curl F e;
J J Jp
or equivalently,
G = Je Fe−1 curl Fe−1 = e ⊗ (J e curl Fe−1 )T e:
The geometric dislocation tensor is therefore a pure edge-tensor
G=e⊗g (9.3)
with principal Burgers vector g planar and of the form
1
g = Je (curl Fe−1 )T e = p (Curl Fp )T e: (9.4)
J
(A vector *eld u is planar if u is independent of X3 (or equivalently, x3 ) and satis*es
u·e=0, so that Curl u is parallel to e.) Thus, g represents the local Burgers vector in the
microstructural con*guration — per unit area in that con*guration — for cross-sectional
surface elements.
The distortion modulus now has the simple form n · Gn = (g · n)(e · n); we may
therefore conclude from the results of Section 8 that: a microstructural plane parallel
to e is globally undistorted; a microstructural plane parallel to g at some point is
there locally undistorted.
In view of (6.1) and (6.2), the geometric dislocation tensor is given by
GR = e ⊗ gR ; gR = Jp Fp−1 g; (9.5)
when referred to the reference con*guration, and by
1
GO = e ⊗ gO ; gO = e Fe g; (9.6)
J
when referred to the deformed con*guration.
1560 P. Cermelli, M.E. Gurtin / J. Mech. Phys. Solids 49 (2001) 1539 – 1568

The rotation and stretch tensors de*ned by the polar decomposition (3.5) of Fe are
planar tensor *elds; the lattice rotation Re may therefore be expressed in the form
Re = Re (#e ) (9.7)
with #e an angle that measures the lattice rotation in a counterclockwise direction about
e. Applying identities (2.9) and (7:4)3 to Curl Fe−1 with Fe = Ve Re ,
 T
eT
e−1 e−1 e dR e−1
curl F = (curl V )R + (grad #e ×) V
d#e
= (curl Ve−1 )Re − (grad #e ×)[(e×)Fe−1 ]T ;
so that, by (9.2) with T = Fe−1 ,
g = Je {ReT (curl Ve−1 )T e − (e×)Fe−1 (e × grad #e )}
= Je ReT (curl Ve−1 )T e − (e×)(e×)FeT grad #e :
Further, Fe−1 grad #e is planar and, for u planar, (e×)(e×)u = −u; thus
g = J e ReT (curl Ve−1 )T e + FeT grad #e : (9.8)

9.2. Rigid-plastic materials

By (9.6),
gO = Re g (9.9)
represents the true Burgers vector; that is, the Burgers vector for cross-sectional planes
as would be observed by an experimenter (cf. the remark containing (6.6)). Since
Ve = 1, curl Ve−1 = 0; relations (9.8) and (9.9) reduce to
g = ReT grad #e and gO = grad #e ; (9.10)
e
gO is hence normal to surfaces # = constant in the deformed con*guration. Experiments
on the uniaxial compression of single crystals (cf., e.g., Schwartz et al., 1999) exhibit
regions of nearly constant lattice orientation separated by thin layers. Relation (9:10)2
shows that — granted strict plane strain — such layers are necessarily accompanied by
geometrically necessary dislocations and, when suQciently thin, should have Burgers
vector approximately normal to the layer.
Under strict plane strain the underlying *elds may be considered as functions of
their position (x1 ; x2 ) in the cross-sectional plane P. Consider, for the moment, a
single crystal. Then sO" = Re s" and mO " = Re m" and for planar slip systems the family
of deformed slip planes for " may be identi*ed with the family of (smooth) deformed
slip curves in P tangent to the *eld sO" . By (9:10)2 , s" · g = sO" · grad #e represents a
curvature +eld for the deformed slip curves. Similarly, m" · g = mO " · grad #e represents
a curvature *eld for the family of smooth curves tangent to mO " .
Returning to general rigid-plastic materials, ReT grad #e = Fp−T ∇#e , since F = Re Fp
and FT grad #e = ∇#e , and this yields the alternative relation
g = Fp−T ∇#e : (9.11)
 P. Cermelli, M.E. Gurtin / J. Mech. Phys. Solids 49 (2001) 1539 – 1568 1561

In many situations of interest the structural strains are small in the sense that Ve ≈ 1,
so that, formally, Ve−1 ≈ 1; and curl Ve−1 ≈ 0: In this case the foregoing relations
hold to within the same approximation.

10. Strict anti-plane shear

Under anti-plane shear the deformation x = y(X) is given by

x1 = X1 ; x2 = X2 ; x3 = X3 + u(x1 ; x2 ) (10.1)
with u the displacement in the direction e ≡ e3 , and results in a deformation gradient
F of the form
F = 1 + e ⊗
;
·e=0
with
independent of X3 , and with
= ∇u. We require, in addition, that the anti-plane
shear be strict in the sense that
Fp = 1 + e ⊗
p ; Fe = 1 + e ⊗
e
with
p and
e normal to e. Then J e = 1; Jp = 1; and, in addition, the decomposition
F = Fe Fp yields
=
e +
p : Since
= ∇u, curl
= 0; thus curl
p = −curl
e : (By (10.1)
and the fact that the underlying *elds are independent of X3 , the operators “curl” and
“Curl” are here interchangeable.)
Next, by (2:8)5 , curl Fp =(curl
p )⊗e. Moreover, since
·e=0, curl
p and curl
e are
parallel to e; thus curl Fp = e ⊗ curl
p with
p · curl
p = 0. Thus, since G = Fp curl Fp ,
the geometric dislocation tensor is a pure screw-tensor
G=e⊗h (10.2)
with principal Burgers vector
h = curl
p = −curl
e (10.3)
parallel to e. Thus microstructural planes perpendicular to e are distorted at any
point at which curl
p = 0.

11. Transport of geometrically necessary dislocations

11.1. Dislocation densities. Balance law for dislocations

We now show that kinematics alone yields a balance law for the transport of geomet-
rically necessary dislocations. Within a continuum theory edge and screw dislocations
are characterized by the pure edge- and screw-tensors (5.1) and (5.2) and hence by
dyads of the form

l⊥b; edge;
l ⊗ b::: (11.1)
l = b; screw
1562 P. Cermelli, M.E. Gurtin / J. Mech. Phys. Solids 49 (2001) 1539 – 1568

with l and b microstructural unit vectors. 2 We refer to each such pair d = (l; b) as a
dislocation dyad with Burgers direction b and line direction l. The scalar *eld
def
 = (l; b) = l · Gb = (l ⊗ b) · G (11.2)
then represents the tensor G resolved on the dislocation system d. We refer to (l; l)
as a screw density, to (l; b) — with b perpendicular to l — as an edge density. Note
that these densities are in units of length per unit area and are signed.
The vector *elds
lR = FpT l; bR = FpT b
represent l and b transported back to the reference as normals, although not as unit
normals. Since
(Curl Fp )b = Curl bR ; (Curl Fp )l = Curl lR ; (11.3)
if Jp = 1, then (4.5) leads to the simple expression  = lR · Curl bR ; hence
˙ = l̇R · Curl bR + lR · Curl ḃR = l̇R · Curl bR + ḃR · Curl lR − Div(lR × ḃR ):
By (3.4), l̇R = FpT LpT l and ḃR = FpT LpT b, so that, in view of (11.3),
l̇R · Curl bR + ḃR · Curl lR = l · Lp Gb + b · Lp Gl:
Further, we may use (2.11) to conclude that (FpT l) × =Fp−1 (l×)Fp−T ; and hence that
lR × ḃR = {(FpT l)×}FpT LpT b = Fp−1 (l×)LpT b:
We therefore have the balance law for dislocations: Granted Jp = 1, the dislocation
density  = (l; b) has the simple form  = lR · Curl bR and evolves according to
˙ = −Div qR + R (11.4)
with qR = qR (l; b) the dislocation Cux and R = R (l; b) the dislocation supply de+ned
by
qR = Fp−1 (l × (LpT b)); R = 2(sym l ⊗ b) · (Lp G): (11.5)
As is clear from the proof, the Pux and supply may be written alternatively as
qR = lR × ḃR ; R = l̇R · Curl bR + ḃR · Curl lR :
Further, for a single crystal, we may use (3.13) and (4.5) to conclude that
A

A


qR = " (b · s" ) Fp−1 (l × m" ); R = 2(sym l ⊗ b) · " (S" G):
"=1 "=1

2 For crystalline materials there are natural families of such dyads associated with the underlying slip
systems (Sun et al., 1998, 2000). For example, for the {1 1 1}1 1 0 slip systems in fcc crystals a canonical
set of dyads consists of 6 pure screw directions with line directions in the 1 1 0 directions and 12 pure
edge directions whose line directions lie in 1 1 2 directions and whose Burgers vectors lie along 1 1 0.
But it would also seem that the screw dyads corresponding to the slip plane normals are important, as the
corresponding screw densities are the distortion moduli for the slip planes.
 P. Cermelli, M.E. Gurtin / J. Mech. Phys. Solids 49 (2001) 1539 – 1568 1563

Thus slip systems " with s" · l = 0 do not contribute to temporal changes in the screw
density (l; l); in particular, there is no contribution to changes in (m2 ; m2 ) from slip
on 2.

11.2. Properties of the dislocation Cux

The dislocation Pux has several interesting properties. Firstly, the microstructural
Cux
def
q(l; b) = Fp qR (l; b) = (l×)LpT b; (11.6)

which is qR convected to the microstructural con*guration, lies in the plane l⊥ . More-

over, plastic Pow is always associated with a Pux of dislocations — that is, Lp = 0
if and only if qR (l; b) = 0 for some choice of l and b — and this is true even when
G ≡ 0, but in this case % ≡ 0; R ≡ 0, and the Pow is equilibrated: Div qR = 0.
Further, the dislocation Cuxes determine Lp : for {e1 ; e2 ; e3 } an arbitrarily chosen or-
thonormal basis and any choice of i and j, i = j; ei · Lp ej = (ei × ej ) · q(ei ; ei ) and
ei ·Lp ei =(ej ×ei )·q(ej ; ei ); the oK-diagonal components of Lp are therefore determined
by the screw-density Puxes q(ei ; ei ), the diagonal components by the edge-density Puxes
q(ei ; ej ).
Consider strict plane strain and strict anti-plane shear, and for each let e denote
the out-of-plane normal. In both cases the supply vanishes identically. For strict plane
strain with b planar, (b; b) vanishes and qR (b; b) is equilibrated and parallel to e,
while both (e; e) and qR (e; e) vanish. The edge densities of interest, namely (e; b)
with b planar, are associated with a planar lattice Pux q given by LpT b rotated about
e through
=2. For strict anti-plane shear Lp has the form e ⊗  with ⊥e and the
density of interest, (e; e), is associated with the lattice Pux q = e × , which is also
normal to e.
The microstructural Pux q corresponding to G resolved on the dislocation system
d = (l; b) was deduced simply by calculation without regard to physical relevance. We
now show that this Pux may be derived from physical considerations alone, at least for
a single crystal and d of edge type. Consider *rst a single active slip system, so that
Lp = " (s" ⊗ m" ). Our computation of Burgers vectors uses a right-hand screw-rule,
and therefore the natural line direction l" for edge dislocations moving on this slip
system is l" = m" × s" , because the Burgers vector associated with this line direction
gives slip corresponding to material above the slip plane (i.e., in the region into which
m" points) moving in the direction s" relative to material below the plane. The Pux q
of dislocations — as resolved on the system d — due to slip on " should have the
following properties: (i) q should be orthogonal to the dislocation line and hence to
l, since the tangential motion of a dislocation line is irrelevant; (ii) q should lie in
the slip plane and should hence be orthogonal to m" ; (iii) the magnitude of q should
be (" s" ) · b; (iv) for l the natural line direction for slip on ", the Pux should be
(s" · b)s" . These conditions determine q. Indeed, (i) and (ii) imply that q should lie
on the line spanned by the unit vector l × m" and hence, by (iii), should have the
form ± " (s" · b) l × m" . Since l" = m" × s" , condition (iv) requires that we take the
1564 P. Cermelli, M.E. Gurtin / J. Mech. Phys. Solids 49 (2001) 1539 – 1568

positive sign
q = " (s" · b) l × m" = (l×)(" S"T )b:
If we allow for the possibility of all slip systems being active, then we arrive at formula
(11.6) for the dislocation Pux as measured in the microstructural con*guration.

11.3. Relations for Ġ

In discussing geometrically necessary dislocations in single crystals it would seem

useful to have available identities relating Ġ to the slips " and their gradients. By
(4.5),

1˙ p 1 p
Ġ = F Curl Fp + p Fp Curl Ḟ ; (11.7)
Jp J
while (2:8)2 and (3.4) yield
p
Curl Ḟ = (Curl Fp )LpT + CurlLp (Lp Fp ); (11.8)
p p p p p
where CurlLp (L F ) is the Curl of L F holding F *xed. Thus, by (3.4) and since
p
J˙ = Jp tr Lp ,
1
Ġ = −(tr Lp )G + Lp G + GLpT + p Fp CurlLp (Lp Fp ): (11.9)
J
The term CurlLp (Lp Fp ) is, in general, complicated, but it does reduce when at-
tention is restricted to single crystals. In that case Lp has the
explicit form (3.13)
A
and CurlLp (Lp Fp ) comes from computing Curl(Lp Fp ) = " " p
"=1 Curl { S F }
p " p
holding F *xed, or equivalently holding S F *xed for each ". Thus (2:8)3 ,
A
(2.11), and the fact that Jp = 1 yield Fp CurlLp (Lp Fp ) = "=1 Fp (∇" ×)FpT S"T =
A p−T
"=1 {(F ∇" )×} S"T .
Further, since ∇" = FT grad " and F = Fe Fp ,
def
Fp−T ∇" = FeT grad " = ∇L " ; (11.10)
" "
where ∇L  , which is intrinsic to the lattice, represents the gradient of  transported
to the lattice. The evolution equation (11.9) therefore takes the form
A


Ġ = [" {S" G + GS"T } − {m" × ∇L " } ⊗ s" ]: (11.11)
"=1

Thus Ġ is linear in the slips " and the slip-gradients ∇" , with coeAcients dependent
on Fp and Curl Fp .

11.4. Strict plane strain. Relations for ġ

Under strict plane strain Lp e = LpT e = 0 and G = e ⊗ g with e constant; the analog
of (11.9) is therefore
1
ġ = (Lp − (tr Lp )1) g + p [CurlLp (Lp Fp )]T e: (11.12)
J
 P. Cermelli, M.E. Gurtin / J. Mech. Phys. Solids 49 (2001) 1539 – 1568 1565

˙ ˙
Next, in view of (3.4) and the identity T−1 = −T−1 ṪT−1 , it follows that Fp−T =
−LpT Fp−T . Thus for a rigid-plastic material an equivalent expression in terms of the
elastic rotation angle follows upon diKerentiation of (9.11):
e
ġ = −LpT g + ∇L #̇ : (11.13)
The specialization to single crystals may be obtained by direct substitution of G = e ⊗ g
into (11.11):
A


ġ = [(g · m" )" + (s" · ∇L " )]s" : (11.14)
"=1

For a rigid-plastic material, g = ∇L #e and this expressions reduces to

A


ġ = [(m" · ∇L #e )" + (s" · ∇L " )]s" : (11.15)
"=1

12. The invariant description of geometrically necessary dislocations

The invariant characterization of geometrically necessary dislocations requires invari-

ance under arbitrary compatible changes in reference con*guration, since such changes
should not induce additional dislocations of that type. The chief purpose of this section
is to determine what functional dependences on Fp and ∇Fp display this invariance.
Let X0 denote an arbitrarily prescribed material point. By a compatible change in
local reference (at X0 ) we mean a smooth one-to-one mapping Z = Z(X) of a neigh-
borhood of X0 onto an open set in R3 ; the points Z then represent new labels for
material points. We write ∇ ˆ for the gradient with respect to points Z, leaving ∇ to
denote the gradient with respect to X. Writing X=X(Z) for the inverse of the mapping
Z = Z(X), let H = ∇X ˆ and assume that det H ¿ 0.
When expressed relative to the new reference, a motion x = y(X) and its gradient
are given by
x = ŷ(Z) = y(X(Z)); F̂ = ∇ ˆ ŷ = FH: (12.1)
The structural transformation Fe is a linear transformation from the microstructural
con*guration to the deformed con*guration and as such is unrelated to the reference
con*guration. We therefore stipulate that Fe be invariant under local changes in refer-
ence. On the other hand, by (12.1) and the decomposition F = Fe Fp , the plastic strain
p
F̂ relative to the new reference must, for consistency, satisfy
p p
F̂ = Fp H; Jˆ = (det H)Jp : (12.2)
A discussion of the geometric dislocation tensor G requires a transformation law for
ˆ for the curl with respect to Z, the desired transformation law is
Curl Fp . Writing Curl
given by
ˆ F̂p = (det H)H−1 Curl Fp
Curl (12.3)
(cf. Cermelli and Gurtin, 2000, Eq. (2.10)).
1566 P. Cermelli, M.E. Gurtin / J. Mech. Phys. Solids 49 (2001) 1539 – 1568

Consider a change in local reference. Then G = (1=Jp )Fp Curl Fp expressed relative
p
p
ˆ F̂p ; thus, by (12.2) and (12.3), the geometric
to the new reference is Ĝ=(1= Jˆ )F̂ Curl
dislocation tensor is invariant under compatible changes in local reference: Ĝ = G.
A similar argument yields the invariance of G under superposed compatible elastic
deformations (Davini, 1986).
This result begs the question: Are there other +elds — expressible in terms of
Fp and ∇Fp — that are invariant under compatible changes in local reference? To
answer this question, consider a relation
4 = F(Fp ; ∇Fp ) (12.4)
giving the value of a *eld 4 at X0 when Fp and ∇Fp are known at X0 . We say that
F is invariant under compatible changes in local reference if, given any such change,
p p
F(F̂ ; ∇ˆ F̂ ) = F(Fp ; ∇Fp ): (12.5)
Invariance Theorem. A necessary and suAcient condition that F be invariant un-
der compatible changes in local reference is that it reduce to a function K of the
geometric dislocation tensor G:
F(Fp ; ∇Fp ) = K(G): (12.6)

Thus — in contrast to the standard prejudice that constitutive dependences on Fp

are unsound — gradient theories meant to characterize GNDs should allow for a de-
pendence on Fp through its presence in the geometric dislocation tensor G.
SuQciency follows from the invariance of G. To establish necessity, assume that F
is invariant under compatible changes in local reference. Without loss of generality,
take X0 = 0 and restrict attention to local reference changes that map X = 0 to Z = 0.
We *rst show that F must reduce to a function F∗ of Fp and Curl Fp :
F(Fp ; ∇Fp ) = F∗ (Fp ; Curl Fp ): (12.7)
p
Let {e1 ; e2 ; e3 } denote an orthonormal basis. Then F may be written in the form
Fp = ei ⊗ fi , with fi = FpT ei . (Recall our use of summation convention.) On the space
of all smooth vector *elds g there is a one-to-one correspondence between skw ∇g
and Curl g. Thus, since ∇Fp = ei ⊗ ∇fi and (Curl Fp )T = ei ⊗ Curl fi (cf. (2:8)5 ), to
establish (12.7), it suQces to show that
F(Fp ; ei ⊗ ∇fi ) = F(Fp ; ei ⊗ skw∇fi ): (12.8)
p T
Let Si = sym ∇fi (0) and (Curl F ) = ei ⊗ Curl fi , choose gi so that fi · gj = 5ij ,
and consider the mapping X = Z − 12 (Z · Sk Z) gk (0). Since the gradient H(Z) = 1 −
gk (0) ⊗ (Sk Z) of this mapping satis*es H(0) = 1, this mapping de*nes a compatible
change in local reference. From (12.2) it follows that under such a change in reference
f̂i = HT fi . Therefore ∇f̂i (0) = ∇fi (0) + ∇(HT (Z)fi (0))|Z=0 , and, because fi · gj = 5ij ,
we obtain HT (Z)fi (0) = fi (0) − Si Z and ∇f̂i (0) = ∇fi (0) − Si = skw∇fi (0). Thus, since
p
F̂ (0) = Fp (0), (12.5) implies (12.8). Therefore (12.7) is satis*ed.
p
ˆ F̂p ).
To establish (12.6), note that, by (12.5) and (12.7), F∗ (Fp ; Curl Fp )=F∗ (F̂ ; Curl
Consider the homogeneous change in reference with gradient H ≡ Fp−1 (0). In
 P. Cermelli, M.E. Gurtin / J. Mech. Phys. Solids 49 (2001) 1539 – 1568 1567

p
this case, (4:5)1 , (12.1), and (12.3) yield F̂ (0) = 1 and Curl ˆ F̂p (0) = G(0), so that
p p
ˆ F̂ ) = F∗ (1; G). Thus (12.7) reduces to (12.6).
F∗ (F̂ ; Curl
In view of the remark in the last paragraph of Section 4, a relation 4=F(Fe ; grad Fe )
is invariant under local superposed compatible deformations of the deformed con*gu-
ration if and only if it has the form 4 = K(G).
The Invariance Theorem was arrived at independently by Parry and SilhavY X y (1999)
X
and Cermelli and Gurtin (2000). Parry and SilhavYy worked with lattice vectors, but
their result is easily translated to one for the *eld Fe ; their work also develops a general
theory of elastic invariance involving gradients of order higher than one.

Acknowledgements

We acknowledge valuable discussions with Brent Adams, Lallit Anand, Bassem

El-Dasher, Melik Demirel, Anthony Rollett, Shaun Sellers, and Christian Teodosiu.
The support of this research by the Italian MURST project “Modelli matematici per la
scienza dei materiali” and by the Department of Energy and National Science Foun-
dation of the United States is gratefully acknowledged.

References

Acharya, A., Bassani, J.L., 2000. Lattice incompatibility and a gradient theory of crystal plasticity. J. Mech.
Phys. Solids, 48 (8), 1565–1595.
Asaro, R.J., 1983. Micromechanics of crystals and polycrystals. Adv. Appl. Mech. 23, 1–115.
Bilby, B.A., Bullough, R., Smith, E., 1955. Continuous distributions of dislocations: a new application of
the methods of non-Riemannian geometry. Proc. Roy. Soc. London A 231, 263–273.
Cermelli, P., Gurtin, M.E., 2000. On the characterization of geometrically necessary dislocations in *nite
plasticity. Report 00-CNA-005, Department of Mathematical Sciences, Carnegie Mellon University.
Cermelli, P., Sellers, S., 2000. Multi-phase equilibrium of crystalline solids. J. Mech. Phys. Solids 48=4,
765–796.
Choquet-Bruhat, Y., DeWitt-Morette, C., 1977. Analysis, Manifolds and Physics, vol. 1. North-Holland,
Amsterdam.
Davini, C., 1986. A proposal for a continuum theory of defective crystals. Arch. Rational Mech. Anal. 96,
295–317.
Davini, C., Parry, G.P., 1989. On defect preserving deformations in crystals. Int. J. Plasticity 5, 337–369.
Dlużewski, P., 1996. On geometry and continuum thermodynamics of movement of structural defects. Mech.
Mater. 22, 23–41.
Eshelby, J.D., 1956. The continuum theory of lattice defects. In: Seitz, F.D., Turnbull, D. (Eds.), Solid State
Physics, vol. 3. Academic Press, New York, pp. 79–144.
Fox, N., 1966. A continuum theory of dislocations for single crystals. J. Inst. Math. Appl. 2, 285–298.
Gurtin, M., 1972. The linear theory of elasticity. In: Truesdell, C. (Ed.), Mechanics of Solids II, Handbuch
der Physik 6=a2. Springer, Berlin.
Kalinindi, S.R., Anand, L., 1993. Large deformation simple compression of a copper single crystal. Metall.
Trans. 24A, 989–992.
Kocks, U.F., TomYe, C.N., Wenk, H.-R., 1998. Texture and Anisotropy. Cambridge University Press,
Cambridge.
Kondo, K., 1952. On the geometrical and physical foundations of the theory of yielding. Proceedings Japan
National Congress of Applied Mechanics, vol. 2, pp. 41– 47.
1568 P. Cermelli, M.E. Gurtin / J. Mech. Phys. Solids 49 (2001) 1539 – 1568

Kondo, K., 1955. Non-Riemannian geometry of imperfect crystals from a macroscopic viewpoint. In: Kondo,
K. (Ed.), RAAG Memoirs of the Unifying Study of Basic Problems in Engineering and Physical Science
by Means of Geometry, vol. 1. Gakuyusty Bunken Fukin-Kay, Tokyo.
Kr?oner, E., 1960. Allgemeine Kontinuumstheorie der Versetzungen und Eigenspannungen. Arch. Rational
Mech. Anal. 4, 273–334.
Kuhlmann-Wilsdorf, D., 1989. Theory of plastic deformation: properties of low energy dislocation structures.
Mater. Sci. Eng. 113A, 1–41.
Lee, E.H., 1969. Elastic-plastic deformation at *nite strains. J. Appl. Mech. 36, 1–6.
Naghdi, P.M., Srinivasa, A.R., 1993. A dynamical theory of structured solids. I Basic developments. Philos.
Trans. Roy. Soc. London A 345, 425–458.
Noll, W., 1967. Materially uniform simple bodies with inhomogeneities. Arch. Rational Mech. Anal. 27,
1–32.
Nye, J.F., 1953. Some geometrical relations in dislocated solids. Acta Metall. 1, 153–162.
X
Parry, G.P., SilhavY y , M., 1999. Elastic scalar invariants in the theory of defective crystals. Proc. Roy. Soc.
London A 455, 4333–4346.
Prantil, V.C., Jenkins, J.T., Dawson, P.R., 1993. An analysis of texture and plastic spin for planar polycrystals.
J. Mech. Phys. Solids 41, 1357–1382.
Schwartz, A.J., St?olken, J.S., King, W.E., Campbell, W.E., Lassila, D.H., Sun, S., Adams, B.L., 1999.
Analysis of compression behavior of a [011] Ta single crystal with orientation imaging microscopy and
crystal plasticity. Report UCRL-JC-133402, Lawrence Livermore National Laboratory.
Sun, S., Adams, B.L., King, W.E., 2000. Observations of lattice curvature near the interface of a deformed
aluminum bicrystal. Philos. Mag. A 80, 9–25.
Sun, S., Adams, B.L., Shet, C., Saigal, S., King, W., 1998. Mesoscale investigation of the deformation *eld
of an aluminum bicrystal. Scripta Mater. 39, 501–508.
Synge, J.L., 1960. Classical dynamics. In: Fl?ugge, S. (Ed.), Principles of Classical Mechanics and Field
Theory, Handbuch der Physik 3=1. Springer, Berlin.
Teodosiu, C., 1970. A dynamical theory of dislocations and its applications to the theory of the elastic-plastic
continuum. In: Simmons, J.A. et al. (Eds.), Proceedings of Conference on Fundamental Aspects of
Dislocation Theory, Washington 1969, N.B.S. Spec. Publ. 317=2, 837–876.
Teodosiu, C., 1982. Elastic Models of Crystal Defects. Springer, Berlin.