Peridynamic Theory of Solid Mechanics

S. A. Silling∗
R. B. Lehoucq†

Sandia National Laboratories

Albuquerque, New Mexico 87185-1322 USA

April 28, 2010

Dedicated to the memory of James K. Knowles

Contents
1 Introduction 4
1.1 Purpose of the peridynamic theory . . . . . . . . . . . . . . . 4
1.2 Summary of the literature . . . . . . . . . . . . . . . . . . . . 5
1.3 Organization of this article . . . . . . . . . . . . . . . . . . . 11

2 Balance laws 13
2.1 Balance of linear momentum . . . . . . . . . . . . . . . . . . 13
2.2 Principle of virtual work . . . . . . . . . . . . . . . . . . . . . 18
2.3 Balance of angular momentum . . . . . . . . . . . . . . . . . 19
2.4 Balance of energy . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.5 Master balance law . . . . . . . . . . . . . . . . . . . . . . . . 24

3 Peridynamic states: notation and properties 28

4 Constitutive modeling 32
4.1 Simple materials . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.2 Kinematics of deformation states . . . . . . . . . . . . . . . . 34
4.3 Directional decomposition of a force state . . . . . . . . . . . 34
4.4 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
∗
Multiscale Dynamic Material Modeling Department, [email protected]
†
Applied Mathematics and Applications Department, [email protected]

1
 4.5 Thermodynamic restrictions on constitutive models . . . . . . 36
4.6 Elastic materials . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.7 Bond-based materials . . . . . . . . . . . . . . . . . . . . . . 39
4.8 Objectivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.9 Isotropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.10 Isotropic elastic solid . . . . . . . . . . . . . . . . . . . . . . . 43
4.11 Peridynamic material derived from a classical material . . . . 44
4.12 Bond-pair materials . . . . . . . . . . . . . . . . . . . . . . . 44
4.13 Example: a bond-pair material in bending . . . . . . . . . . . 48

5 Linear theory 50
5.1 Small displacements . . . . . . . . . . . . . . . . . . . . . . . 50
5.2 Double states . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
5.3 Linearization of an elastic constitutive model . . . . . . . . . 52
5.4 Equations of motion and equilibrium . . . . . . . . . . . . . . 53
5.5 Linear bond-based materials . . . . . . . . . . . . . . . . . . . 54
5.6 Equilibrium in a one dimensional model . . . . . . . . . . . . 56
5.7 Plane waves and dispersion in one dimension . . . . . . . . . 60

6 Relation to other theories 62

6.1 Deformation gradient and the deformation state . . . . . . . . 62
6.2 Peridynamic stress tensor . . . . . . . . . . . . . . . . . . . . 63
6.3 Convergence in the limit of small horizon . . . . . . . . . . . 64
6.4 Elasticity tensor derived from a peridynamic material . . . . 66
6.5 Nonlocal theories . . . . . . . . . . . . . . . . . . . . . . . . . 67

7 Discrete particles as peridynamic bodies 70

7.1 Self-equilibrated subregions . . . . . . . . . . . . . . . . . . . 70
7.2 Linear and angular momentum in self-equilibrated subregions 72
7.3 Peridynamic particles . . . . . . . . . . . . . . . . . . . . . . 73
7.4 Particles as a special case of a continuum . . . . . . . . . . . 76
7.5 Multibody potentials . . . . . . . . . . . . . . . . . . . . . . . 77
7.6 Peridynamic stress due to two discrete particles . . . . . . . . 79
7.7 Average stress due to many discrete particles . . . . . . . . . 81

8 Damage and fracture 87

8.1 Damage as part of a constitutive model . . . . . . . . . . . . 87
8.2 Irreversibility of damage growth . . . . . . . . . . . . . . . . . 88
8.3 Bond breakage . . . . . . . . . . . . . . . . . . . . . . . . . . 88
8.4 Ordinary material models with strong damage dependence . . 90
8.5 Bond-based constitutive models with bond breakage . . . . . 90
8.6 Separable damage models . . . . . . . . . . . . . . . . . . . . 91
8.7 Energy balance in progressive damage . . . . . . . . . . . . . 93

2
 8.8 Relation to the Griffith criterion . . . . . . . . . . . . . . . . 95
8.9 Surface energy . . . . . . . . . . . . . . . . . . . . . . . . . . 96

9 Discussion 98

3
1 Introduction
1.1 Purpose of the peridynamic theory
The peridynamic theory of mechanics attempts to unite the mathematical
modeling of continuous media, cracks, and particles within a single frame-
work. It does this by replacing the partial differential equations of the clas-
sical theory of solid mechanics with integral or integro-differential equations.
These equations are based on a model of internal forces within a body in
which material points interact with each other directly over finite distances.
The classical theory of solid mechanics is based on the assumption of
a continuous distribution of mass within a body. It further assumes that
all internal forces are contact forces [73] that act across zero distance. The
mathematical description of a solid that follows from these assumptions
relies on partial differential equations that additionally assume sufficient
smoothness of the deformation for the PDEs to make sense in either their
strong or weak forms. The classical theory has been demonstrated to pro-
vide a good approximation to the response of real materials down to small
length scales, particularly in single crystals, provided these assumptions are
met [52]. Nevertheless, technology increasingly involves the design and fab-
rication of devices at smaller and smaller length scales, even interatomic
dimensions. Therefore, it is worthwhile to investigate whether the classi-
cal theory can be extended to permit relaxed assumptions of continuity, to
include the modeling of discrete particles such as atoms, and to allow the
explicit modeling of nonlocal forces that are known to strongly influence the
behavior of real materials.
Similar considerations apply to cracks and other discontinuities: the
PDEs of the classical theory do not apply directly on a crack or dislocation
because the deformation is discontinuous on these features. Consequently,
the techniques of fracture mechanics introduce relations that are extraneous
to the basic field equations of the classical theory. For example, linear elas-
tic fracture mechanics (LEFM) considers a crack to evolve according to a
separate constitutive model that predicts, on the basis of nearby conditions,
how fast a crack grows, in what direction, whether it should arrest, branch,
and so on. Although the methods of fracture mechanics provide important
and reliable tools in many applications, it is uncertain to what extent this
approach can meet the future needs of fracture modeling in complex media
under general conditions, particularly at small length scales. Similar con-
siderations apply to certain methods in dislocation dynamics, in which the
motion of a dislocation is determined by a supplemental relation.
Aside from requiring these supplemental constitutive equations for the
growth of defects within LEFM and dislocation dynamics, the classical the-
ory predicts some well-known nonphysical features in the vicinity of these

4
singularities. The unbounded stresses and energy densities predicted by the
classical PDEs are conventionally treated in idealized cases by assuming that
their effect is confined to a small process zone near the crack tip or within
the core of a dislocation [38]. However, the reasoning behind neglecting the
singularities in this way becomes more troublesome as conditions and ge-
ometries become more complex. For example, it is not clear that the energy
within the core of a dislocation is unchanged when it moves near or across
grain boundaries. Any such change in core energy could affect the driving
force on a dislocation.
Molecular dynamics (MD) provides an approach to understanding the
mechanics of materials at the smallest length scales that has met with im-
portant successes in recent years. However, even with the fastest computers,
it is widely recognized that MD cannot model systems of sufficient size to
make it a viable replacement for continuum modeling.
These considerations motivate the development of the peridynamic the-
ory, which attempts to treat the evolution of discontinuities according to the
same field equations as for continuous deformation. The peridynamic theory
also has the goal of treating discrete particles according to the same field
equations as for continuous media. The ability to treat both the nanoscale
and macroscale within the same mathematical system may make the method
an attractive framework in which to develop multiscale and atomistic-to-
continuum methods.

1.2 Summary of the literature

The term “peridynamic” first appeared in [60] and comes from the Greek
roots for near and force. The model proposed in [60] treats internal forces
within a continuous solid as a network of pair interactions similar to springs.
In this respect it is similar to Navier’s theory of solids (see Section 6). In
the peridynamic model, the springs can be nonlinear. The responses of the
springs can depend on their direction in the reference configuration, leading
to anisotropy, and on their length. The maximum distance across which a
pair of material points can interact through a spring is called the horizon,
because a given point cannot “see” past its horizon. The horizon is treated
as a constant material property in [60]. The equation of motion proposed
in the original peridynamic theory is
Z
f u(x0 , t) − u(x, t), x0 − x dVx0 + b(x, t)


ρ(x)ü(x, t) = (1)
H

where x is the position vector in the reference configuration of the body B,

ρ is density, u is displacement, and b is a prescribed body force density. H
is a neighborhood of x with radius δ, where δ is the horizon for the mate-
rial. Constitutive modeling, as proposed in [60], consists of prescribing the

5
pairwise force function f (η, ξ) for all bonds ξ = x0 − x and for all relative
displacements between the bond endpoints, η = u0 − u. f can depend non-
linearly on η, and there is no assumption that the bond forces are zero in the
reference configuration. f has dimensions of force/volume2 . Linearization of
the equation of motion results in an expression that is formally the same as
in Kunin’s nonlocal theory [46, 47] although constitutive modeling and other
aspects are different; a comparison between the two models is discussed in
Section 6.5.
A number of papers have investigated various aspects of the linear peri-
dynamic theory. In [70], the static loading by a body force density of an
infinitely long, homogeneous bar is considered. The resulting solutions, ob-
tained using Fourier transforms, demonstrate interesting features not present
in solutions of the classical equilibrium equation. Among these are oscilla-
tions that decay at points far from where the loading is applied, a result
of the nonlocality in the equations. (The physical significance of these fea-
tures is not yet clear.) Dispersion curves are derived from isotropic material
models in [60], along with a variational formulation and some aspects of
material stability. Zimmermann [83] explored many features of theory, in-
cluding certain aspects of wave motion, material stability, and numerical
solution techniques. Zimmermann also studied energy balance for crack
growth within the theory.
Weckner and Abeyaratne [75] studied the dynamics of a one dimensional
bar and obtained a Green’s function representation of the solution. They
also derived expressions for the evolution of discontinuities in the displace-
ment field. Stable discontinuities (i.e., discontinuities that do not grow un-
boundedly over time) can occur for certain choices of the initial data, even
with well-behaved material properties. For other materials, discontinuities
can grow unboundedly over time, leading to a type of material instability.
Green’s functions for three dimensional unbounded elastic isotropic media
were derived in [77] for both statics and dynamics. This work also presented
a comparison between the local and peridynamic theories for linear elastic
solids.
Alali and Lipton [3], Du and Zhou [18, 19], and Emmrich and Weck-
ner [20, 21] establish various existence and uniqueness results for the linear
peridynamic balance of momentum. These papers also draw equivalences
with the weak solution of the classical equations of linear elasticity, and
show, in a precise sense, the well-posedness of the peridynamic equations
in the limit as the nonlocality vanishes. In particular, the limiting solution
coincides with the conventional weak solution given sufficient regularity of
the boundary data and material properties. Within the context of nonlocal
steady-state diffusion, Gunzburger and Lehoucq [35] introduce a nonlocal
Gauss’s theorem and nonlocal Green identities to establish well-posedness

6
of the nonlocal boundary value problem.
The peridynamic theory as outlined in [60] suffers from significant restric-
tions on the scope of material behavior that can be modeled, in particular
the Poisson ratio is always 1/4 for isotropic materials. This motivated a
rethinking of the whole peridynamic theory. The outcome was a concept
which preserves the idea of bonds carrying forces between pairs of particles.
However, in the new approach, the forces within each bond are not deter-
mined independently of each other. Instead, each bond force depends on the
collective deformation (and possibly the rate of deformation and history) of
all the bonds within the horizon of each endpoint. The resulting modified
theory is called state-based, because the mathematical objects that convey
information about the collective deformation of bonds are called peridynamic
states (see Section 3). The technical discussion in the present article deals
primarily with the state-based theory, although the earlier bond-based theory
is shown to be a special case of this. The state-based theory is discussed in
greater detail in [67], which includes a specific isotropic solid material model
in which any Poisson ratio can be prescribed.
It is also shown in [67] that any elastic constitutive model from the
classical theory can be adapted to the peridynamic theory using a nonlo-
cal approximation to the deformation gradient tensor. Application of this
technique to a strain-hardening plasticity model is demonstrated in [74, 27].
The stress tensor provided by the classical constitutive model is mapped
onto the bond forces in a way consistent with the approximation used for
the deformation gradient (see Section 4.11).
A peridynamic stress tensor (see Section 6.2) was derived in [48], al-
though a similar concept was previously discussed in [83]. The peridynamic
stress tensor has a mechanical interpretation similar to the Piola stress ten-
sor in the classical theory. It provides the force per unit area across any
imaginary internal surface. However, in the peridynamic case, the stress
tensor is nonlocal: the forces involved are the nonlocal forces in bonds that
cross from one side of the surface to the other. The peridynamic operator for
the internal force density can be expressed exactly as the divergence of the
peridynamic stress tensor field. Thus, the peridynamic equation of motion
becomes formally the same as the classical equation.
The convergence of the bond-based peridynamic theory to the equations
of classical elasticity theory was demonstrated by Zimmermann [83], and in
the context of isotropic linear elastic solids by Emmrich and Weckner [21].
Within the state-based framework for constitutive modeling, it was shown in
[68] that if a deformation is classically smooth, then the peridynamic oper-
ator for internal force density approaches the classical operator in the limit
of small horizon (see Section 6.3). The limiting process produces a classi-
cal constitutive model for Piola stress as a limiting case of the peridynamic

7
stress for small horizon. In this sense, the peridynamic theory converges to
the classical theory.
Sears and Lehoucq [58] provide a statistical mechanical foundation for
the peridynamic balance of linear momentum. The nonlocality of force in-
teraction is intrinsic and originates in molecular force interaction that is
nonlocal. This analysis is similar to the landmark work of Irving and Kirk-
wood [39], who had the objective of deriving the classical, rather than peri-
dynamic, field equations from statistical physics. The classical balance of
linear momentum is a consequence of the more general peridynamic balance
when the integral operator is expressed as the divergence of a stress tensor.
In the important special case of a pair-potential, Noll [58, 49] in effect de-
rives the peridynamic balance of linear momentum as an intermediate step
in deriving the classical balance from the principles of statistical mechanics.
Gerstle et al. extended the peridynamic mechanics model to diffusive
processes including heat conduction and migration of species due to high
electrical current density [31]. They applied the combined nonlocal equa-
tions incorporating multiple physical mechanisms, including species diffu-
sion, heat transport, mechanics, and electrical conduction, to a model prob-
lem demonstrating the failure of an electronic component due to electromi-
gration.
Nearly all of the applications of the peridynamic model to date rely on
numerical solutions. A numerical technique for approximating the peridy-
namic field equations was proposed in [61]. This numerical method simply
replaces the volume integral in (1) with a finite sum:
ρi n+1 n n−1

X
u − 2u i + u = f (unj − uni , xj − xi )Vi + bni
h2 i i
j∈H

where i is the node number, n is the time step number, h is the time step
size, and Vi is the volume (in the reference configuration) of node i. This
numerical method is meshless in the sense that there are no geometrical
connections, such as elements, between the discretized nodes. Adaptive
refinement and convergence of the discretized method in one dimension are
discussed in [11].
Damage is incorporated into this numerical method by causing the “bonds”
between interacting nodes to break irreversibly. Although this breakage oc-
curs independently among all the bonds, their failure tends to organize it-
self along two dimensional surfaces that are interpreted as cracks. Cracks
progress autonomously: their advance is determined only by the field equa-
tions and constitutive model at the bond level. There is no supplemental
relation that dictates crack growth. In particular, the stress intensity factor
is not used. Because of the nonlocal nature of the equations, fields near a
crack tip in the numerical results are bounded. A computer solution to one

8
of the Kalthoff-Winkler problems [40], which is regarded in the computa-
tional fracture mechanics community as an important benchmark problem,
is presented in [61]. Additional examples, as well as more details about the
numerical method, are discussed in [65, 76].
Autonomy of crack growth is also demonstrated by Kilic and Madenci
[43], who apply the peridynamic method in a numerical model of cracking
in glass plates. The cracks are driven by a temperature gradient that causes
thermal stresses [82]. In the geometry considered, the crack growth is mostly
stable. In some cases, the cracks curve and branch. The numerical model
reproduces many aspects of the experiments.
Dayal and Bhattacharya [16] developed a peridynamic material model
designed to reproduce martensitic phase transformations. Numerical studies
showed that this model predicts phase boundaries with finite thickness and
detailed structure. These authors further showed that the model uniquely
determines a kinetic relation for the motion of phase boundaries. This result
is analogous to the autonomous growth of cracks: the motion of the defect
is determined by the field equations and the constitutive model.
Finite element discretization techniques for the peridynamic equations
have been proposed by Zimmermann [83] and by Weckner et al. [77]. Macek
[50] demonstrated that standard truss elements available in the Abaqus com-
mercial finite element code can be used to represent peridynamic bonds.
These peridynamic elements can be applied in part of an FE mesh with
standard elements in the remainder of the mesh. The resulting FE model of
the peridynamic equations was applied in [50] to penetration problems. A
finite element formulation was also developed by Chen and Gunzburger [14],
who consider the one dimensional equations for a finite bar. Weckner and
Emmrich investigated certain discretizations of the peridynamic equation of
motion, including Gauss-Hermite quadrature, and applied these to initial
value problems to demonstrate convergence [78, 22].
Among applications of the peridynamic model to real systems, Bobaru
[10, 9] demonstrated the application of a numerical model to small scale
structures, including nanofibers and nanotubes. The nanofiber model is
multiscale in the sense that it involves both short-range forces within a fiber
and long-range van der Waals forces between fibers. The meshless property
of the numerical method, as well as the ability to treat long-range forces,
is helpful in these applications because of the need to generate models of
complex, random structures. Silling and Bobaru [66] additionally applied
the method to the dynamic fracture of brittle-elastic membranes. This study
demonstrated the acceleration of a crack to a limiting growth velocity that
is consistent with the properties of the material.
Small scale numerical applications of the peridynamic equations are
also demonstrated by Agwai, Guven, and Madenci [1, 2] and by Kilic and

9
Madenci [44], who studied cracking and debonding in electronic integrated
circuit packaging. Their model explicitly includes a temperature-dependent
term in the material model for bond forces and so can be applied to damage
driven by thermal stresses.
Concrete, because it is heterogeneous and brittle unless large compressive
confining stress is present, is a good example of a material in which the
standard assumptions of LEFM do not apply, at least on the macroscale.
The process of cracking in concrete tends to occur through the accumulation
of damage over a significant volume before localizing into a discontinuity,
which itself usually follows a complex, three dimensional path. Damage and
its progression to cracking in concrete are often cited as processes in which
nonlocality is important [8, 55]. Gerstle et al. [34, 30, 32, 33] have applied
the peridynamic method to the failure of concrete structures, including the
debonding of reinforcing bar from concrete. This includes development a
micropolar version of the theory, in which rotational degrees of freedom are
included in the computational nodes.
Impact against brittle structures is a natural application for the peridy-
namic model, because cracks grow “autonomously:” fracture nucleation and
evolution occur as an outcome of the material model and equation of motion,
so any number of cracks can grow in any degree of complexity. Peridynamic
analysis of impact is demonstrated in [64, 17].
Application to damage and fracture in composite laminates is discussed
in [7, 80, 81, 6]. It is demonstrated that the strong anisotropy in a uniaxially
reinforced lamina can be reproduced by making the bond response in (1)
dependent on the direction of the bond in the reference configuration. The
anisotropy also applies to damage: the criterion for bond breakage can also
be dependent on bond direction. From this conceptually simple treatment
of anisotropy, the complexity of damage and fracture in composites can be
reproduced to a surprising degree by a homogenized peridynamic model.
Kilic, Agwai, and Madenci [42] developed an innovative numerical model
of a composite lamina that is not homogenized, but instead treats the con-
stituents explicitly within the mesh. This model reproduces the influence of
stacking sequence on damage and progressive failure in laminates.
The peridynamic method was applied by Foster to the interpretation
of experiments on dynamic fracture initiation [26]. This application used
a state-based peridynamic material adapted from a viscoplastic material
model for metals using the technique discussed in Section 4.11. This work
successfully reproduced the effect of loading rate on crack initiation in steel.
The use of the peridynamic theory as multiscale method is currently
in its early stages. Preliminary work is reported in [5]. Solution of the
peridynamic continuum equations within the LAMMPS molecular dynamics
code is described in [56].

10
 Multiscale analysis of a fiber-reinforced composite in the limit of small
fiber diameter is treated by Alali and Lipton [3] for different types of assumed
limiting behavior of the constituent materials and their interfaces. These
authors also investigate the homogenized models resulting from alternative
ways of coupling the peridynamic horizon to the geometrical length scales
naturally present in the material during this limiting process.

1.3 Organization of this article

The purpose of this article is to present an up-to-date, consistent develop-
ment of the peridynamic theory. In Section 2 we develop systematically the
equations for global and local balance of linear momentum, angular momen-
tum, and energy. This leads to a statement of the principle of virtual work,
as well as the peridynamic form of the first law of thermodynamics.
Section 3 contains a discussion of the notation and properties of peri-
dynamic states, which are the mathematical objects used in constitutive
modeling. The term “states” is chosen in analogy with the traditional usage
of this term in thermodynamics: these objects contain descriptions of all
the relevant variables that affect the conditions at a material point in the
body. In the case of the peridynamic model, these variables are the nonlocal
interactions between a point and its neighbors.
The general form of constitutive models is discussed in Section 4, includ-
ing the appropriate notion of elastic materials. Conditions for isotropy and
objectivity are discussed. The Coleman-Noll method for obtaining restric-
tions on constitutive dependencies is applied, revealing a restriction on the
sign of rate-dependent terms. Specific material models are described that
highlight material behavior that the peridynamic model can describe but
the classical theory cannot.
Linearization is treated in Section 5. The linearized peridynamic mate-
rial properties are contained in the modulus state, which is analogous to the
fourth order elasticity tensor in the classical theory. The equation of motion
becomes a linear integro-differential equation in the linearized theory. The
equation of equilibrium is a linear Fredholm integral equation of the second
type.
In Section 6, we compare the peridynamic theory to the classical theory.
The peridynamic stress tensor is defined, and it is shown that under certain
conditions, the peridynamic equation of motion converges to the classical
PDE. A comparison between the peridynamic model and some other nonlo-
cal theories is also presented.
Section 7 demonstrates that a description of discrete particles can be ob-
tained as the limiting case of peridynamic regions of finite volume as their
sizes are shrunk to zero. The resulting description involves forces that are
more general than pair interactions. Then, it is shown that such a collection

11
of “peridynamic particles” can be represented within the peridynamic con-
tinuum equations using generalized functions. In particular, any multibody
potential can be represented exactly in terms of a peridynamic constitutive
model. The peridynamic stress tensor and its volume average are derived
for a system of discrete particles, and it is shown that these averages obey
the peridynamic equation of motion.
Damage and fracture are discussed in Section 8. It is shown how ir-
reversible damage can be included in the peridynamic expression for free
energy in a constitutive model. Damage evolution is treated as part of the
material model. A peridynamic version of the J-integral is derived that
gives the rate of energy dissipation of a moving defect; this is related to the
Griffith criterion for crack growth. An expression for the surface energy of a
crack is derived in terms of the work done on bonds that initially connected
material on one side of a crack to material on the other side.

12
2 Balance laws
We derive the peridynamic balances of linear and angular momentum in a
more systematic way than has previously appeared in the literature [67].
We then postulate the global balance of energy for a subregion in a peridy-
namic body, which leads to the local balance of energy. The energy balance
involves both heat transport and mechanical power. The global energy bal-
ance introduces the absorbed power and supplied power for a subregion. An
important result is that the internal energy defined in terms of these powers
is an additive quantity, leading to a meaningful definition of internal energy
density.
The balances of linear momentum, angular momentum, and energy are
shown to adhere to a canonical structure, which we call the master balance
law. This law expresses the rate of change of any additive quantity within
a subregion as the sum of interactions between points inside and outside
of the subregion, plus a source term. These interactions appear within
the integrand of an integral operator in the master balance law, and the
antisymmetry of this integrand plays a crucial role. This antisymmetry
allows the integral operator to be written as the integral of the divergence
of a nonlocal flux. (An analogous master balance law also exists in the
classical theory.)

2.1 Balance of linear momentum

Let B be the reference configuration of a closed, bounded body with reference
mass density ρ. Let y(·, ·) be a motion of B, so y(x, t) is the position at
time t ≥ 0 of a material point x ∈ B. The deformed image of B under y is
denoted Bt (Figure 1). Define the velocity field by
v(x, t) = ẏ(x, t) ∀x ∈ B, t ≥ 0.
Let b be the external body force density field. Let L(x, t) be the force per
unit volume at time t on x due to interactions with other points in the body.
The force vector on a subregion P ⊂ B is given by
Z
(L + b) dV ,
P

in which the integration is performed in the reference configuration. Apply-

ing Newton’s second law to this subregion,
Z Z Z
d
ρẏ dV = ρÿ dV = (L + b) dV , (2)
dt P P P

hence, by localization, the equation of motion in terms of L is

ρ(x)ÿ(x, t) = L(x, t) + b(x, t) ∀x ∈ B, t ≥ 0. (3)

13
Newton’s second law applied to B requires that
Z Z
d
ρẏ dV = b dV . (4)
dt B B

Setting P = B in (2) and comparing the result with (4) shows that L must
be self-equilibrated:
Z
L(x, t) dVx = 0 ∀t ≥ 0.
B

Now let f (·, ·, ·) be a vector-valued function such that

Z
L(x, t) = f (x0 , x, t) dVx0 ∀x ∈ B, t ≥ 0, (5)
B

and such that f is antisymmetric:

f (x, x0 , t) = −f (x0 , x, t) ∀ x, x0 ∈ B, t ≥ 0. (6)

For a given L, such an f can always be found; an example is

1
f (x0 , x, t) = (L(x, t) − L(x0 , t)) (7)
V
where V is the volume of B in the reference configuration. The function
f , which plays a fundamental role in the peridynamic theory, is called the
dual force density. It has dimensions of force per unit volume squared. In
general, the vectors f (x0 , x, t) and f (x, x0 , t) are not parallel to the vector
y(x0 , t) − y(x, t). The particular choice of f given in (7) is not very useful in
practice; it is given only to demonstrate that for a given L, an f satisfying (5)
and (6) always exists. In applications, f is determined by the deformation
through the constitutive model.
The antisymmetry of f stated in (6) allows the balance of linear momen-
tum on a subregion P ⊂ B to be expressed in a form in which f connects
only points in the interior of P to points in its exterior. To see this, note
that (6) implies Z Z
f (x0 , x, t) dVx0 dVx = 0. (8)
P P
Therefore, from (2), (5), and (8),
Z Z Z Z
d 0
ρẏ(x, t) dV = f (x , x, t) dVx0 dVx + b(x, t) dVx . (9)
dt P P B\P P

The following converse is also true: if (9) holds for all subregions P ⊂ B,
then (6) holds. To see this, choose any two subregions N ⊂ B and N 0 ⊂ B
such that N ∩ N 0 = ∅ (Figure 3). Also define R = B \ (N ∪ N 0 ). Since

14
 Figure 1: Peridynamic body and its motion y.

Figure 2: Dual force density f between two points has contributions from
the bond force density t at both points.

15
B \ N = N 0 + R and B \ N 0 = N + R, it follows that for any f , whether
antisymmetric or not,
"Z Z Z Z
+
N B\N N0 B\N 0
Z Z Z Z Z Z 
− − − f (x0 , x) dVx0 dVx = 0. (10)
N N0 N0 N N ∪N 0 R

It follows from the linearity of the integral operator that

Z Z Z 
+ − (ρÿ(x, t) − b(x, t)) dVx = 0,
N N0 N ∪N 0

hence, from (9),

"Z Z Z Z Z Z #
+ − f (x0 , x) dVx0 dVx = 0.
N B\N N0 B\N 0 N ∪N 0 R

Subtracting this from (10) yields

Z Z Z Z 
+ f (x0 , x) dVx0 dVx = 0.
N N0 N0 N

Since this equation must hold for arbitrary disjoint N and N 0 , localization
results in (6). Thus, the balance of linear momentum (9) implies that f
possesses the antisymmetry (6).
It is convenient, but not entirely accurate, to think of f (x0 , x, t) as phys-
ically representing the force vector (per unit volume squared) that x0 exerts
on x. The reason this interpretation is not accurate is that there is not
necessarily a direct physical connection between x0 and x that gives rise to
the force. For example, if L is given, the particular f given by (7) would
generate L regardless of whether each x0 and x have any direct mechanical
interaction, such as a spring connecting the two points.
For a given f field satisfying (5) and (6), let t(·, ·, ·) denote a vector-
valued function such that

f (x0 , x, t) = t(x0 , x, t) − t(x, x0 , t) ∀ x, x0 ∈ B, t ≥ 0. (11)

Such a t function can always be found; an example is given by

f (x0 , x, t)
t(x0 , x, t) = ∀ x, x0 ∈ B, t ≥ 0.
2
The function t is called the bond force density and is the basic quantity
produced by a constitutive model in the peridynamic theory (Figure 2). Like
f , the bond force density has dimensions of force per unit volume squared.

16
 Figure 3: Antisymmetry of f .

In order to simplify notation, set

t = t(x0 , x, t), t0 = t(x, x0 , t),
f = f (x0 , x, t), f 0 = f (x, x0 , t)
y = y(x, t), y0 = y(x0 , t),
(12)
ρ = ρ(x), b = b(x, t),
L = L(x, t),
dV = dVx , dV 0 = dVx0 .
From (5) and (11), the force density is given by
Z
t − t0 dV 0 .


L= (13)
B

From (9) and (11), the global balance of linear momentum for any subregion
P ⊂ B is
Z Z Z Z
d 0

0
ρẏ dV = t − t dV dV + b dV . (14)
dt P P B\P P

From (3) and (5), the local balance of linear momentum is

Z
ρÿ = f dV 0 + b ∀x ∈ B, t ≥ 0 (15)
B

17
or equivalently, using (13),
Z
t − t0 dV 0 + b


ρÿ = ∀x ∈ B, t ≥ 0. (16)
B

The local balance of linear momentum is also called the equation of motion.
By setting ÿ = 0 in (16), the equilibrium equation is found to be
Z
t − t0 dV 0 + b = 0


∀x ∈ B.
B

The double integral in (14) represents a nonlocal flux of linear momentum

through the boundary of P. This term is analogous to the contact force on
a subregion in the classical, local theory. Equation (14) is an example of
nonlocal balance principles whose structure is discussed in Section 2.5.

2.2 Principle of virtual work

Boundary and initial conditions can be incorporated into the balance of
linear momentum (16) by formulating a variational problem [51]. Let B ∗ ⊂
B have a nonzero volume. B ∗ consists of the points where the motion is
prescribed. Let w(·, ·) be a motion of B, and use the abbreviated notation

w = w(x, t), w0 = w(x0 , t).

The principle of virtual work is stated as follows:

Z Z Z Z
0 0
ρÿ · w dV + t · (w − w) dV dV = b · w dV (17)
B B B B

for all motions w such that

w=0 on B ∗ . (18)

We now demonstrate that the principle of virtual work implies the balance
of linear momentum. Using the change of variables x ↔ x0 leads to the
identity
Z Z Z Z
0 0
t · (w − w) dV dV = − (t − t0 ) · w dV 0 dV . (19)
B B B B

Inserting (19) into (17) results in

Z  Z 
ρÿ − (t − t0 ) dV 0 − b · w dV = 0.
B B

Since this must hold for any choice of w satisfying (18), it follows that
Z
ρÿ = (t − t0 ) dV 0 + b on B \ B ∗ .
B

18
 This leads to the initial boundary-value problem for the balance of linear
momentum (16)
 Z


 ρÿ = (t − t0 ) dV 0 + b on B \ B ∗ ,
B

∗ (20)

 y=y on B ∗ ,
on B \ B ∗ ,

ẏ(·, 0) = v0 (·)


where y∗ and v0 are prescribed functions. Conversely, working backwards

shows that any solution of the initial boundary-value problem (20) also
satisfies the principle of virtual work (PVW) statement (17).

2.3 Balance of angular momentum

Let B be a closed, bounded body, and as before, let P ⊂ B be a subregion.
The angular momentum in P with respect to an arbitrary reference point
y0 is defined by Z
A(P) = (y − y0 ) × ρẏ dV . (21)
P
This definition asserts that there are no hidden variables or degrees of free-
dom other than velocity that contain angular momentum. Since ẏ × ρẏ = 0,
(21) implies Z
Ȧ(P) = (y − y0 ) × ρÿ dV .
P
From this and (3),
Z
Ȧ(P) = (y − y0 ) × (L + b) dV . (22)
P

Global balance of angular momentum on B requires that the rate of change

of total angular momentum equal the total moment due to external forces:
Z
Ȧ(B) = (y − y0 ) × b dV . (23)
B

This equation asserts that there are no external moments other than those
arising from b. Comparing the last two equations and setting P = B places
a restriction on L: Z
(y − y0 ) × L dV = 0, (24)
B
which means that the moments generated by internal forces must be self-
equilibrated. Conversely, (22) and (24) imply (23).
Suppose the bond force density field t is such that
Z
(y0 − y) × t dV 0 = 0 ∀x ∈ B, t ≥ 0. (25)
B

19
A bond force density field satisfying (25) will be called nonpolar. This name
is chosen to contrast the present situation with “micropolar” continuum
theories that permit a nonzero moment to be exerted on material points:
the definition (25) asserts that the net moment about y(x, t) exerted by
t(·, x) vanishes. Micropolar theory has been proposed, for example, as a
way of modeling granular flow [41]. A micropolar peridynamic model has
been proposed [30] but is beyond the scope of the present article.
If t is nonpolar, then the global balance of angular momentum on B
necessarily holds. To see this, compute the left hand side of (24) using (5),
(8), and (11):
Z Z Z
(y − y0 ) × L dV = (y − y0 ) × f dV 0 dV
B ZB ZB Z Z
0
= y × f dV dV − y0 × f dV 0 dV
ZB ZB B B

= y × (t − t0 ) dV 0 dV .
B B

Using the change of variables x ↔ x0 to eliminate the t0 term and using (25)
leads to
Z Z Z
(y − y0 ) × L dV = (y − y0 ) × t dV 0 dV = 0,
B B B

so (24) holds. As discussed above, this implies that the global balance of
angular momentum on B (23) holds.
Next, we further investigate the balance of angular momentum on subre-
gions and use the results to derive the local balance of angular momentum.
Assume that t is nonpolar, let P ⊂ B be a subregion, and let y0 = 0. From
(5), (11) and (22),
Z Z Z
0 0
Ȧ(P) = y × (t − t ) dV dV + y × b dV .
P B P

Add the expression

y0 × t − y0 × t
to the integrand in the double integral. Rearranging yields
Z Z
Ȧ(P) = (y − y0 ) × t dV 0 dV
P B Z Z Z
+ (y0 × t − y × t0 ) dV 0 dV + y × b dV . (26)
P B P

20
Since the bond force densities are nonpolar, by (25), the first term on the
right hand side vanishes. Also, the integrand in the second term is antisym-
metric in x and x0 , therefore
Z Z
(y0 × t − y × t0 ) dV 0 dV = 0.
P P

So, (26) implies

Z Z Z
0 0 0
Ȧ(P) = (y × t − y × t ) dV dV + y × b dV ,
P B\P P

or, recalling (21),

Z Z Z Z
d 0 0 0
y × ρẏ dV = (y × t − y × t ) dV dV + y × b dV , (27)
dt P P B\P P

which holds for any P ⊂ B. (22) and (27) are equivalent statements of the
global balance of angular momentum for a subregion under the assumption
of nonpolar bond force densities. The structure of (27) is similar to that
of (14) in that the two terms on the right hand side represent nonlocal flux
and source rate. The underlying structure of balance principles of this type
is discussed further in Section 2.5.
By localizing (27), a form of the local balance of angular momentum is
obtained:
Z
y × ρÿ = (y0 × t − y × t0 ) dV 0 + y × b ∀x ∈ B, t ≥ 0.
B

This equation is equivalent to (25).

2.4 Balance of energy

Let q(x0 , x, t) denote the rate of heat transport, per unit volume squared,
from x0 to x. It is required that q be antisymmetric:

q(x, x0 , t) = −q(x0 , x, t) ∀x, x0 ∈ B, t ≥ 0. (28)

Nonlocal heat transport is assumed here for consistency with the mechanical
model, although the subsequent development of the energy balance could
be repeated with a local heat model. Nonlocality is important in radiative
heat transport. In the limit of small interaction distances, nonlocal heat
conduction is physically the same as the local model.
Let r(x, t) denote the heat source rate at x. The rate at which heat is
supplied to a subregion P ⊂ B is given by
Z Z Z
0
Q(P) = q dV dV + r dV , (29)
P B\P P

21
where the abbreviation q = q(x0 , x, t) is used. Taking the scalar product of
both sides of the balance of linear momentum (16) with the velocity v and
integrating over P results in
ρv · v
Z Z Z Z
d
t − t0 · v dV 0 dV +


dV = b · v dV . (30)
dt P 2 P B P

The identity

t − t0 · v = t · v0 − t0 · v − t · v0 − v

implies that for all P ⊂ B,

Z Z
t − t0 · v dV 0 dV


P B Z Z Z Z
0 0 0
t · v0 − v dV 0 dV ,



= t · v − t · v dV dV −
ZP ZB PZ BZ

t · v0 − t0 · v dV 0 dV − t · v0 − v dV 0 dV ,



=
P B\P P B
(31)

where the antisymmetry of the dual power density defined by

pd (x0 , x) = t · v0 − t0 · v (32)

was used in the last step. Using (31), we may rewrite (30) as the power
balance
K̇(P) + Wabs (P) = Wsup (P) (33)
where the kinetic energy in P is defined by
ρv · v
Z
K(P) = dV ,
P 2

the power absorbed by P is defined by

Z Z
t · v0 − v dV 0 dV ,


Wabs (P) = (34)
P B

and the power supplied to P is defined by

Z Z Z
0 0 0


Wsup (P) = t · v − t · v dV dV + b · v dV .
P B\P P

We postulate the following global form of the first law of thermodynamics:

Ė(P) + K̇(P) = Wsup (P) + Q(P) (35)

22
where E(P) is the internal energy in P. Subtracting (33) from (35) results
in
Ė(P) = Wabs (P) + Q(P). (36)
This result asserts that the rate of change of internal energy is the sum of
the absorbed power and the rate of heat supplied.
Using (28), it follows from the definitions (29) and (34) that both Wabs
and Q are additive quantities, i.e., for P1 , P2 ⊂ B where P1 ∩ P2 = ∅,

Wabs (P1 ∪ P2 ) = Wabs (P1 ) + Wabs (P2 ), (37)

Q(P1 ∪ P2 ) = Q(P1 ) + Q(P2 ). (38)

Therefore, by (36), the internal energy E is also additive. It follows that

there exists a scalar quantity ε(x, t) called the internal energy density such
that Z
E(P) = ε dV . (39)
P
From (29), (34), (36), and (39),
Z Z Z Z Z Z
0 0 0


ε̇ dV = t · v − v dV dV + q dV dV + r dV . (40)
P P B P B\P P

By (28), Z Z Z Z
0
q dV dV = q dV 0 dV .
P B\P P B

From this and (40),

Z  Z Z 
0 0 0


−ε̇ + t · v − v dV + q dV + r dV = 0.
P B B

Since this must hold for any P ⊂ B, localization leads to the local statement
of the first law of thermodynamics:

ε̇ = pabs + h + r. (41)

where the local heat transport rate at x is defined by

Z
h= q dV 0
B

and the absorbed power density at x is defined by

Z
t · v0 − v dV 0 .


pabs = (42)
B

pabs is the analogue of the stress power in the classical theory.

23
 It is worthwhile to contrast the peridynamic power balance developed
in this section with earlier approaches that lead to nonadditive definitions
of internal energy. The key difference lies in our usage of the peridynamic
quantities absorbed and supplied power, rather than the traditional ideas of
internal and external power that appear in literature on the thermodynamics
of nonlocal media. To see this, define the internal and external power by
Z Z
Wint (P) = f · v dV 0 dV ,
P
Z Z P Z
0
Wext (P) = f · v dV dV + b · v dV .
P B\P P

Wint (P) consists of the rate of work done on material points in P by inter-
actions with other points in P. Wext (P) represents the work done by all
other interactions, including body forces. These quantities are related to
Wabs and Wsup via
Z Z
t · v0 − v dV 0 dV


Wabs (P) = −Wint (P) +
P B\P
Z Z
t · v0 − v dV 0 dV .


Wsup (P) = Wext (P) +
P B\P

Inserting the above expressions for the absorbed and supplied power replaces
(33) with the following alternate statement of the power balance:
K̇(P) − Wint (P) = Wext (P).
However, Gurtin and Williams [36] demonstrate that Wint and Wext are not
additive quantities, in the sense of (38), leading to their conclusion that there
is no additive notion of the internal energy density analogous to (36). The
antisymmetretry of the dual power density pd defined in (32) is also necessary
for the additivity of the absorbed and supplied power expenditures. As the
next section demonstrates, additivity and antisymmetry are intrinsic to well
formulated nonlocal balance laws.

2.5 Master balance law

The global balances of linear momentum (14), angular momentum (27),
and energy (35) over any subregion P ⊂ B possess the following canonical
structure: Z Z Z
Ė(P) = D dV 0 dV + s dV , (43)
P B\P P

where D(·, ·) : B × B → Rd and s(·) : B → Rd .

Here, d = 1 if E is scalar
valued or d = 3 if it is vector valued. It is assumed that D is antisymmetric:
D(x0 , x) = −D(x, x0 ) ∀x, x0 ∈ B. (44)

24
(In the remaining discussion, the abbreviations D = D(x0 , x) and s = s(x)
are used, and t does not appear explicitly.) The balance (43) states that
the rate of change of the extensive quantity E(P) is composed of two terms.
The first term represents interactions between P its exterior. The second
term represents external sources. The functions D and s are called the dual
interaction density and the source rate respectively. (44) implies
Z Z Z Z
0
D dV dV = D dV 0 dV ,
P B\P P B

hence (43) may be rewritten as

Z Z Z
0
Ė(P) = D dV dV + s dV . (45)
P B P

From (45), it is immediate that

Ė(P1 ∪ P2 ) = Ė(P1 ) + Ė(P2 ),

where P1 and P2 are any two disjoint subregions of B. This establishes that
E is additive. It follows that there exists a density function e on B such that
Z
E(P) = e dV
P

for any subregion P ⊂ B. Inserting this expression into (45), localization

leads to the local balance
Z
ė = D dV 0 + s. (46)
B

Table 1 lists the dual interaction densities and source rates for the three
nonlocal balances previously introduced.
We now demonstrate that the master balance law (43) can be written
in a more traditional form, i.e., the first term on the right hand side of
(43) corresponds to a nonlocal flux acting on the boundary of P. This is
accomplished by invoking two lemmas due to Noll [53, 49], and crucially
depends upon the antisymmetry of D.
Suppose that the dual interaction density D is antisymmetric and con-
tinuously differentiable, and that

|D(x, x0 )| 6 K|x − x0 |−` x ∈ B, x0 ∈ R3 \ B,

for positive constants K and ` < 3. Then Noll’s lemma I provides a closed

25
 Balance (Eq.) e D s

Linear momentum (14) ρẏ t − t0 b

Angular momentum (27) y × ρẏ y0 × t − y × t0 y×b

ρẏ · ẏ
Energy (35) ε+ q + t · ẏ0 − t0 · ẏ r + b · ẏ
2

Table 1: Global balance principles. y denotes the motion of the body B,

and P ⊂ B.

form expression for a tensor of order one1 or two,

Z Z 1 
1
T(x) = − D (x + λz, x − (1 − λ)z) dλ ⊗ z dV , (47)
2 R3 0

such that Z
∇·T= D dV 0 .
B
Noll’s lemma II then implies that
Z Z Z Z Z
0
∇ · T dV = D dV dV = D dV 0 dV , (48)
P P B P B\P

where the antisymmetry of the dual density D is invoked for the second
equality. Using the divergence theorem, this allows us to rewrite the master
balance law (43) in the familiar form2
Z Z
Ė(P) = Tn dV + s dV .
∂P P

Localization then gives the counterpart of (46) as

ė = ∇ · T + s.

Noll [54] calls T a reacher. This terminology draws a distinction with the
abstract notion of a contactor corresponding to a surface interaction. For
1
The integrand is understood as
Z 1
z D (x + λz, x − (1 − λ)z) dλ
0

for an order one tensor, or flux vector, T.

2
When the tensor T is of order one, then Tn is understood to be T · n.

26
instance, when the interaction is a force, a contactor is a contact stress
associated with the classical continuum notion of contact force.
The conclusion of Noll’s lemma II given by (48) implies that
Z
Tn dV = 0,
∂B

and equivalently expresses that the sum of the internal interactions in the
body is zero.
As shown in Section 2.1 for the case D = f , the second equality in
(48) implies the antisymmetry of D that was assumed in (44). Lehoucq
and Silling [48] provide an expression (see (120) below) for the peridynamic
stress tensor in terms of the bond force density. This expression is derivable
from (47) with D = f .

27
Figure 4: The family H contains the relative position vectors (bonds) con-
necting x to points such as x0 within a distance δ of x.

3 Peridynamic states: notation and properties

The remainder of this paper largely involves mappings from pairs of points
(x, x0 ) to some quantity. As an aid to keeping track of these mappings, it
is convenient to introduce objects called “peridynamic states.” Consider a
body B. Let δ be a positive number, called the horizon. For a given x ∈ B,
let Hx be the neighborhood of radius δ with center x (Figure 4). Define the
family of x by

H = ξ ∈ (R3 \ 0)  (ξ + x) ∈ (Hx ∩ B) .
 

A vector ξ ∈ H is called a bond connected to x. H differs from Hx in that

the former is centered at 0 and contains bonds, while the latter is centered
at x and contains position vectors of material points.
A peridynamic state Ah · i is a function on H. The angle brackets h · i
enclose the bond vector; parentheses and square brackets will be used later
to indicated dependencies of the state on other quantities. A state need not
be a differentiable or continuous function of the bonds in H.
If the value Ahξi is a scalar, then A is a scalar state. The set of all scalar
states is denoted S. Two special scalar states are the zero state and the

28
unity state defined respectively by

0hξi = 0, 1hξi = 1 ∀ξ ∈ H.

If the value of Ahξi is a vector, then A is a vector state. The set of all vector
states is denoted V. Two special vector states are the null vector state and
the identity state defined by

0hξi = 0, Xhξi = ξ ∀ξ ∈ H (49)

where 0 is the null vector.

An example of a scalar state is given by

ahξi = 3c · ξ ∀ξ ∈ H,

where c is a constant vector. An example of a vector state is given by

Ahξi = ξ + c ∀ξ ∈ H.

Another useful kind of state, called a double state, maps pairs of bonds
ξ, ζ ∈ H into second order tensors, and is written Ahξ, ζi. The set of all
double states is denoted D.
In the following, a and b are scalar states, A and B are vector states, and
V is a vector. Some elementary operations on states are defined as follows,
for any ξ ∈ H:

(a + b)hξi = ahξi + bhξi, (A + B)hξi = Ahξi + Bhξi

(ab)hξi = ahξibhξi, (aB)hξi = ahξiBhξi

(A · B)hξi = Ahξi · Bhξi, (A ⊗ B)hξi = (Ahξi) ⊗ (Bhξi)


(A ◦ B)hξi = A Bhξi , (A · V)hξi = (Ahξi) · V
where the symbol · indicates the usual scalar product of two vectors in R3
and ⊗ denotes the dyadic (tensor) product of two vectors. Also define a
scalar state |A| by
|A|hξi = |Ahξi| (50)
and the dot products
Z Z
a•b= ahξibhξi dVξ , A•B= Ahξi · Bhξi dVξ (51)
H H

where, once again, the symbol · denotes the scalar product of two vectors in
R3 . The norm of a scalar state or a vector state is defined by
√ p
||a|| = a • a, ||A|| = A • A. (52)

29
Most of the constitutive models in peridynamics involve functions of states,
and it is helpful to define a notion of derivatives of such functions. If ψ(·) :
S → R is a function of a scalar state, its Fréchet derivative ∇ψ, if it exists,
is defined by
ψ(A + a) = ψ(A) + ∇ψ(A) • a + o(||a||) (53)
for all scalar states A and a. ∇ψ is a scalar state.
If Ψ(·) : V → R is a function of a vector state, its Fréchet derivative ∇Ψ,
if it exists, is similarly defined by

Ψ(A + a) = Ψ(A) + ∇Ψ(A) • a + o(||a||) (54)

for all vector states A and a. ∇Ψ is a vector state.

For functions of more than one state, for example Ψ(A, B), the Fréchet
derivatives with respect to the two arguments will be denoted ΨA and ΨB
respectively. The notation ∂/∂A denotes the derivative of a function with
respect to A, if the argument depends either directly or indirectly on A.
For example, if f (·) : R → R, then
∂
f (ψ(A)) = ∇φ(A), φ(A) := f (ψ(A)).
∂A
In this case, it is easily shown from (53) that the following chain rule applies:
∂
f (ψ(A)) = f 0 (ψ(A))∇ψ(A).
∂A
where f 0 denotes the first derivative of f .
The operations on states such as the dot product defined above occur
repeatedly in manipulations, but their use does not restrict the physics that
can be modeled. Note that S, V, and D are infinite dimensional linear
vector spaces (assuming that H contains an infinite number of bonds), but
this does not preclude the modeling of nonlinear behavior. For example, the
discussion of constitutive modeling in Section 4 below deals with nonlinear
functions of states.
A state field is a state valued function of position in B and possibly time.
These dependencies are written in square brackets:

A[x, t]

for any x ∈ H and t ≥ 0. An example of a scalar state field is given by

a[x, t]hξi = |ξ + x|t ∀ξ ∈ H, x ∈ B, t ≥ 0.

Finally, the dependence of a state valued function of other quantities is

written in parentheses, for example

A(B).

30
An example of a state valued function of a vector state is given by

a(B) = |B|3 ,

i.e., using the definition (50),

a(B)hξi = |Bhξi|3 ∀ξ ∈ H, x ∈ B, t ≥ 0.

A vector state is analogous to a second order tensor in the classical

theory, because it maps vectors (bonds) into vectors. However, the mapping
performed by a vector state is not necessarily a linear transformation on
the bond vectors, i.e., Ahξi is not necessarily a linear function of ξ. The
additional notation described above is needed because of this nonlinearity
and nonlocality.
The mappings defined by states provide the fundamental objects on
which constitutive models operate in the nonlocal setting of peridynamics.
In the classical theory, a constitutive model for a simple material specifies
a tensor (stress) as a function of another tensor (deformation gradient).
In the peridynamic theory, a constitutive model instead provides a vector
state (called the force state) as a function of another vector state (called the
deformation state). The way this works is discussed in the next section.

31
4 Constitutive modeling
The discussion in Section 2 introduced the bond force density field t without
specifying how this t is determined in a particular motion. This determi-
nation is provided by the constitutive model, also called the material model,
which contains all information about the response of a particular mate-
rial. In the peridynamic theory, the constitutive model supplies t(x0 , x, t) in
terms of the deformation at any given time, the history of deformation, and
any other physically relevant quantities. This discussion does not include
damage, which is the subject of Section 8.
The state that maps bonds connected to x into their deformed images is
called the deformation state and denoted Y[x, t]. Angle brackets are used
to indicate a bond that this state operates on. For a motion y, at any t ≥ 0,

Y[x, t]hx0 − xi = y(x0 , t) − y(x, t) (55)

for any x ∈ B and any x0 ∈ B such that x0 − x ∈ H (Figure 5). The values
of any t(x0 , x, t) are given by the force state T:

t(x0 , x, t) = T[x, t]hx0 − xi. (56)

With this definition, the absorbed power density defined in (42) takes the
form
pabs = T • Ẏ (57)
where the dot product is defined in the previous section. Recall that this
absorbed power density is the peridynamic analogue of the stress power
σ · Ḟ, where σ is the Piola stress tensor and F = ∂y/∂x is the deformation
gradient tensor.
In terms of the force state, the equation of motion (16) has the form
Z  
ρ(x)ÿ(x, t) = T[x, t]hx0 − xi − T[x0 , t]hx − x0 i dVx0 + b(x, t) (58)
B

for all x ∈ B, t ≥ 0. The equilibrium equation is then

Z  
T[x]hx0 − xi − T[x0 ]hx − x0 i dVx0 + b(x) = 0
B

for all x ∈ B.

4.1 Simple materials

The constitutive model determines the force state at any x and t. For a
simple material and a homogeneous body, the force state depends only on
the deformation state:
T[x, t] = T̂(Y[x, t])

32
Figure 5: The deformation state Y[x, t] maps each bond in the family of x
to its deformed image.

33
where T̂(·) : V → V is a function whose value is a force state. Suppressing
from the notation the dependence on x and t,

T = T̂(Y) (59)

which is analogous to the Piola stress in a simple material in the classical

theory, σ = σ̂(F). If the body is heterogeneous, an explicit dependence on
x is included:
T = T̂(Y, x).
If the material is rate dependent, the constitutive model would additionally
depend on the time derivative of the deformation state:

T = T̂(Y, Ẏ, x).

4.2 Kinematics of deformation states

The deformation state defined in (55) provides a mapping from each bond
ξ in the family of x to its deformed image Yhξi. It is assumed that at any
t ≥ 0, y(·, t) is invertible:

x1 6= x2 =⇒ y(x1 , t) 6= y(x2 , t) ∀x1 , x2 ∈ B.

This assumption implies

6 0
Yhξi = ∀ξ ∈ H.

Otherwise, there are essentially no kinematical restrictions on Y. All of the

following are allowed:
• Nondifferentiability (as might occur near an inclusion or a phase bound-
ary).

• Discontinuities (such as a crack).

• Voids and other defects.

However, not all these allowable features would appear, or be capable of
appearing, in a given application.

4.3 Directional decomposition of a force state

As discussed in Section 2.3, bond force densities are assumed to be non-
polar, as defined in (25). This provides an admissibility condition on the
constitutive model. In terms of the force state, the condition for nonpolarity
is written as Z
Yhξi × T̂(Y)hξi dVξ = 0 ∀Y ∈ V. (60)
H

34
This requirement means that the force state at x exerts no net moment on
a small volume surrounding B \ x.
For any deformation state Y, define the direction state by
Y
M= (61)
|Y|

(see (50) for notation). Using the abbreviation T = T̂(Y), define the
collinear and orthogonal parts of the force state by

Tk = (M ⊗ M)T, T⊥ = T − Tk . (62)

Thus, for any ξ ∈ H,

Tk hξi = (Mhξi · Thξi)Mhξi (63)

which is parallel to the deformed bond. Similarly, T⊥ hξi is orthogonal to

the deformed bond. From (61) and (63),
Z
Yhξi × Tk hξi dVξ = 0
H

regardless of constitutive model. From this and the second of (62), the
condition for nonpolarity (60) is equivalent to
Z
Yhξi × T⊥ hξi dVξ = 0.
H

The constitutive model T̂ is called ordinary if, for all Y ∈ V,

Tk = T (64)

where T = T̂(Y). Otherwise, the constitutive model is nonordinary. From

(60) and (64), evidently all ordinary constitutive models are nonpolar. (The
converse of this is not true.)

4.4 Examples
An example of a simple peridynamic material model is given by

Y
T̂(Y) = a |Y| − |X| M, M= ∀Y ∈ V,
|Y|

where a is a constant. Writing this out in detail,

Yhξi
Thξi = a |Yhξi| − |ξ| ∀Y ∈ V,
|Yhξi|

35
for any bond ξ ∈ H. In this material, the magnitude of the bond force
density vector t is proportional to the bond extension (change in length of
the bond). The direction is parallel to the deformed bond. In this exam-
ple, the bonds respond independently of each other: Thξi depends only on
Yhξi. Materials with this property are called bond-based and are discussed
in Section 4.7.
A much larger class of materials incorporates the collective response of
bonds. This means that the force density in each bond depends not only on
its own deformation, but also on the deformation of other bonds. A simple
example is given by


Thξi = a |Yhξi| − |Yh − ξi| Mhξi.

In this material, the bond force density for any bond ξ is proportional to
the difference in deformed length between itself and the bond opposite to
ξ. (Note that in general Yh − ξi 6= −Yhξi, since the two bonds ξ and −ξ
can deform independently of each other.) This material is an example of a
bond-pair model, discussed in Section 4.12.
The mean elongation of all the bonds in a family is defined by
Z Z
1

ē = |Yhξi| − |ξ| dVξ , VH = dV.
VH H H

A material model in which the magnitudes of forces in the bonds are identical
to each other and depend only on the mean elongation is provided by

T = aēM.

In Section 4.10, the mean elongation in the bonds (weighted by scalar state)
is used to define a nonlocal volume change. This provides a way to char-
acterize an isotropic solid using the conventional bulk modulus and shear
modulus.

4.5 Thermodynamic restrictions on constitutive models

In this section it is shown that the force state can be related to a free energy
function, which is subject to certain restrictions due to the second law of
thermodynamics. The first law of thermodynamics asserts the equivalence
of mechanical energy and heat energy. At any point x ∈ B, the local form
of the first law (41) with the absorbed power density given by (57) takes the
form
ε̇ = T • Ẏ + h + r (65)
where ε is the internal energy density, h is the rate of heat transfer due
to interaction with other points in B, and r is a prescribed source rate (all
these quantities are per unit volume in the reference configuration).

36
 The second law of thermodynamics is expressed by the Clausius-Duhem
inequality:
θη̇ ≥ r + h (66)
where θ is the absolute temperature and η is the entropy density. Now define
the free energy density by
ψ = ε − θη. (67)
Following Coleman and Noll [15], certain restrictions on the constitutive
response will now be derived. Taking the time derivative of (67) leads to

ψ̇ = ε̇ − θ̇η − θη̇.

From this and (65), it follows that

ψ̇ = T • Ẏ + h + r − θ̇η − θη̇. (68)

Combining this expression with (66), the variables ε, η̇, and r are eliminated
to yield
T • Ẏ − θ̇η − ψ̇ ≥ 0. (69)
Now assume that ψ and η have the following dependencies:

ψ = ψ(Y, Ẏ, θ), η = η(Y, Ẏ, θ),

hence ψ̇ involves the Fréchet derivatives of ψ with respect to Y and Ẏ,

which are denoted ψY and ψẎ respectively:

ψ̇ = ψY • Ẏ + ψẎ • Ÿ + ψθ θ̇,

with a similar expression for η̇. Combining these with (69) leads to
   
T − ψY • Ẏ − ψẎ • Ÿ − ψθ + η θ̇ ≥ 0. (70)

The method of Coleman and Noll assumes that, in the present case of peri-
dynamics, the quantities Ẏ, Ÿ, and θ̇ can, in principle, be varied indepen-
dently. The inequality (70) must hold for all such choices. This results in
the following conclusions:

η = −ψθ , ψẎ = 0.

The first of these is a standard relation in thermodynamics. The second

states that the free energy is independent of Ẏ. Next, following Fried’s
development [29] for the thermodynamics of discrete particles, decompose
the force state into parts that are independent of and dependent on Ẏ
respectively:
T(Y, Ẏ, θ) = Te (Y, θ) + Td (Y, Ẏ, θ) (71)

37
where the superscript e stands for “equilibrium” and d stands for “dissipa-
tive.” Then, setting θ̇ = 0 and Ÿ = 0 in (70) and using (71),
 
Te (Y, θ) − ψY (Y, θ) • Ẏ + Td (Y, Ẏ, θ) • Ẏ ≥ 0

where the terms that are independent of Ẏ have been grouped together.
The conclusions are therefore

Te (Y, θ) = ψY (Y, θ) (72)

and
Td (Y, Ẏ, θ) • Ẏ ≥ 0. (73)
Equation (73) is the dissipation inequality for rate-dependent materials in
peridynamics, and it must hold for all choices of Ẏ. It states that the
rate-dependent part of the constitutive model must dissipate energy at a
nonnegative rate. Interestingly, (73) does not imply that

Td hξi · Ẏhξi ≥ 0 ∀ξ ∈ H.

In other words, there can be some bonds that “generate energy” provided
there are other bonds that dissipate at least this much energy. A version of
the dissipation inequality for materials undergoing damage will be discussed
in Section 8.2.

4.6 Elastic materials

If the free energy density depends only on Y, the material is called elastic,
and by convention the free energy density is called the strain energy density
and denoted W = Ŵ (Y). Then by (72),

Ẇ = T • Ẏ (74)

for any Y and

T̂ = ŴY .
Since Ŵ is a function of only one variable, this can also be written as

T̂ = ∇Ŵ . (75)

For a body composed of an elastic material (not necessarily homogeneous),

by setting w = ẏ in the principle of virtual work expression (17) and using
(56) and (74), it follows that for an elastic material,

ρẏ · ẏ
Z Z Z
d d
dV + W dV = b · ẏ dV .
dt B 2 dt B B

38
Thus, as in the classical theory, work performed on an elastic peridynamic
body by external loads is converted into a combination of kinetic energy and
recoverable strain energy.
A mechanical interpretation of the Fréchet derivative of Ŵ in an elastic
material is as follows. Suppose the family is deformed, then held fixed.
Choose a single bond ξ, surrounded by a small volume dV . While continuing
to hold all other bonds fixed, increment the position of the small volume by a
small vector . If the material is elastic, then there is a vector t, independent
of , such that the resulting change in W is given by

dW = t ·  dV .

The value of this vector is t = Thξi. An elastic material model can be either
ordinary or nonordinary: elasticity does not require that Thξi k Yhξi.

4.7 Bond-based materials

Suppose that each bond has its own constitutive relation, independent of
the others. Then there is a function t̂(·, ·) on R3 × H such that

Thξi = t̂(Yhξi, ξ) (76)

for all Y ∈ V and all ξ ∈ H. Such a material model is called bond-based.

The requirement of nonpolarity (60) implies that any bond-based mate-
rial model is ordinary. To see this, suppose that it is nonordinary. Then, by
definition, there is some deformation state Y0 and some bond ξ0 such that

c := Y0 hξ0 i × t0 6= 0, t0 = t̂(Y0 hξ0 i, ξ0 ).

Start with this Y0 and let all other bonds except ξ0 be held fixed while
ξ0 is further deformed. (Strictly speaking, we are deforming the material
point x + ξ0 , while holding all other material points fixed, where x is the
point whose constitutive model is under consideration.) Because (60) must
continue to hold during this process, any choice of Yhξ0 i leaves Yhξ0 i ×
t̂(Yhξ0 i, ξ0 ) unchanged, i.e.,

z × t̂(z, ξ0 ) = c (77)

for any vector z = Yhξ0 i. One such choice is

z = αc

where α is a nonzero scalar with the appropriate dimensions for this expres-
sion to make sense. Then by (77),

αc × t̂(αz, ξ0 ) = c.

39
This can only hold if c = 0, proving that the material model is ordinary.3
In an elastic bond-based body, there is a scalar-valued function ŵ(p, ξ)
called the bond potential, where p is a vector, such that
Z
Ŵ (Y) = ŵ(Yhξi, ξ) dVξ , t̂(Yhξi, ξ) = ŵp (Yhξi, ξ). (78)
H

Note that the first argument of ŵ in this integrand is a vector, not a vector
state. ŵp denotes the partial derivative with respect to this argument.
Recall the result proved above that any bond-based material model is
ordinary. An implication of this result for elastic bond-based materials is
that ŵ(p, ξ) can depend on p only through |p|, i.e., through the deformed
length of the bond. To confirm this, choose a deformed bond vector p
and consider a rotation of this vector at some angular velocity ω. Then
dp/dt = ω × p. Therefore

d dp
ŵ(p, ξ) = ŵp (p, ξ) · = ŵp (p, ξ) · (ω × p).
dt dt
Since the material is ordinary, there is some scalar β, with appropriate
dimensions, such that
ŵp (p, ξ) = βp.
Combining the last two equations,
d
ŵ(p, ξ) = βp · (ω × p).
dt
Since, for any vector ω, p ⊥ (ω × p), it follows that

d
ŵ(p, ξ) = 0.
dt
This proves that ŵ(p, ξ) is unchanged by a rigid rotation of p; therefore, ŵ
depends on p only through |p|. So, we can write, for an elastic bond-based
material model,

ŵ(p, ξ) = w(e, ξ), e = |p| − |ξ|

for some function w. Then, by the first of (78),

Z
Ŵ (Y) = w(ehξi, ξ) dVξ ,
H
3
The discussion of this result in [60] is flawed because it treats only pairs of material
points in isolation from all other material points, neglecting the possibility that these other
points could somehow cancel out a couple between the pair.

40
where e is the scalar extension state, defined by

e = |Y| − |X| or ehξi = |Yhξi| − |ξ| ∀ξ ∈ H.

Let the partial derivative of w(e, ξ) with respect to e be denoted we (e, ξ).
By the second of (78) and the chain rule,
Yhξi
t̂(Yhξi, ξ) = we (ehξi, ξ)M, M= . (79)
|Yhξi|
If the body is homogeneous and composed of bond-based material, it is
sometimes convenient to consider each bond as the fundamental object for
purposes of constitutive modeling: set
Z
1
w̆ e, x0 , x dVx0 , e = |y(x0 ) − y(x)| − |x0 − x|


W (x) =
2 Hx
where Hx is the neighborhood of x with radius equal to the horizon, and

w̆(e, x0 , x) = 2w(e, x0 − x).

This change allows the resulting “bond-based theory” to be developed with-

out using the formalism of states. The bond-based theory is the subject of
[60], in which w̆ is called the micropotential and the material model is called
microelastic. Because the bond-based theory was developed earlier than the
state-based theory, and because its constitutive models do not require the
additional complexity of Fréchet derivatives, the vast majority of applica-
tions of peridynamics have been performed within the bond-based theory.
However, as noted in Section 1.2, the bond-based theory suffers from severe
limitations on the material response it can reproduce, notably the restric-
tion on the Poisson ratio ν = 1/4 for isotropic microelastic solids. It is
demonstrated in Section 4.10 below that this restriction is removed in the
state-based theory.

4.8 Objectivity
As in the classical theory, invariance of a strain energy density function in the
peridynamic theory with respect to rigid rotation following a deformation
leads to a notion of material frame indifference, or objectivity. Let O+ denote
the set of all proper orthogonal tensors. For any Q ∈ O+ and any A ∈ V,
let QA be the vector state defined by

(QA)hξi = Q(Ahξi) ∀ξ ∈ H

and similarly define the state AQ by

(AQ)hξi = AhQξi ∀ξ ∈ H.

41
Consider an elastic material such that
Ŵ (QY) = Ŵ (Y) ∀Q ∈ O+ , Y ∈ V. (80)
Let Q be fixed. Consider any Y ∈ V and a small increment δY ∈ V. From
(54), (75), and (80), neglecting terms of higher order than δY,
T̂(QY) • δ(QY) = T̂(Y) • δY.
Since T is vector valued, by the properties of the transpose of a tensor,
QT T̂(QY) • δY = T̂(Y) • δY.

Since this must hold for every small δY, and since QT = Q−1 , it follows
that (80) implies
T̂(QY) = QT̂(Y) ∀Q ∈ O+ , Y ∈ V. (81)
Any simple material model, whether elastic or not, that satisfies (81) is
called objective. Objectivity can be assumed as an admissibility requirement
for any material model in the absence of some externally dictated special
direction in space, such as an electric field. It is easily shown [63] that an
objective elastic material necessarily satisfies the condition for nonpolarity
(60).

4.9 Isotropy
Consider an elastic material model with the property that
Ŵ (YQ) = Ŵ (Y) ∀Q ∈ O+ , Y ∈ V. (82)
Proceeding as in the previous section, choose any Q ∈ O+ and any Y ∈ V,
then consider a small increment δY ∈ V. From (54), (75), and (82),
T̂(Y) • δY = T̂(YQ) • δ(YQ)
Z
= T̂(YQ)hξi · δYhQξi dVξ
ZH
= T̂(YQ)hQ−1 ξ 0 i · δYhξ 0 i dVξ0
H
= T̂(YQ)Q−1 • δY

where the change of variable ξ 0 = Qξ has been used. Since this result must
hold for every δY, it follows that (82) implies
T̂(YQ) = T̂(Y)Q ∀Q ∈ O+ , Y ∈ V. (83)
Any material model, whether elastic or not, satisfying (83) is called isotropic.
If the material model is isotropic, then the force state is invariant with
respect to pre-rotations applied before the stretch.

42
4.10 Isotropic elastic solid
A peridynamic material model for a constitutively linear isotropic elastic
solid was proposed in Section 15 of [67]. A nonlocal dilatation is defined by
3
ϑ= (ωx) • e, m = (ωx) • x (84)
m
where ω is the scalar influence state, which serves as a weighting function,
and the scalar extension state is defined by

e = |Y| − x, x = |X|.

It can be shown [67] that for any choice of ω, if the deformation is small and
homogeneous, ϑ defined in (84) equals the trace of the classical linear strain
tensor. (The coefficient 3/m in (84) is chosen so that this is true.)
Define an elastic material in which the strain energy density contains two
terms representing the contribution of the volume change and of everything
else in the deformation state, respectively:

kϑ2 α
Ŵ (Y) = + (ωed ) • ed (85)
2 2
where k and α are constants and
ϑx
ed := e − ei , ei := .
3
The scalar state ei is called the isotropic part of the extension state, and ed
is called the deviatoric part. The isotropic part contains length changes of
bonds due to isotropic expansion of the family. The deviatoric part contains
the remainder of the length changes, which may be due to shear or to other
types of deformation within the family. After evaluating the applicable
Fréchet derivatives [67], the force state is given by
 
3kϑ d Y
T̂(Y) = ωx + αωe M, M= .
m |Y|

Since the bond force densities are parallel to the deformed bonds, this is an
ordinary material model. This material model is constitutively linear in the
sense that the force state depends linearly on the extension state. However,
it does not assume linear kinematics as will be assumed in the linearized
peridynamic theory discussed below in Section 5. For small, homogeneous
deformations, the strain energy density in the peridynamic material model
(85) equals that of an isotropic linear elastic solid in the classical theory
provided k is the bulk modulus for the material and α = 15µ/m, where µ is
the shear modulus [67].

43
4.11 Peridynamic material derived from a classical material
Suppose a material model from the classical theory is given in the following
form:
∂y
σ = σ̂(F), F=
∂x
where σ is the Piola stress tensor, σ̂ is a function, and F is the deformation
gradient tensor. A peridynamic material model can be derived from this as
follows [67, 74, 27]. (An alternative approach making use of the principle
of virtual work can also be used [51].) A nonlocal approximation to the
deformation gradient tensor is defined by
Z 
F̄ = ωhξiYhξi ⊗ ξ dVξ K−1
H

where ω is the scalar influence state and K is the symmetric positive definite
shape tensor defined by
Z
K= ωhξiξ ⊗ ξ dVξ .
H

The force state is determined by mapping the resulting σ back onto the
bonds as follows:
T̂(Y) = ω σ̂(F̄)K−1 X.
The peridynamic stress tensor (see Section 6.2) corresponding to this peri-
dynamic material model equals σ̂(F) in the special case of homogeneous
deformation of a homogeneous body.

4.12 Bond-pair materials

Let w be a scalar-valued function of four vectors:

w(p, q, r, s)

with partial derivatives with respect to the first two arguments denoted by

wp (p, q, r, s), wq (p, q, r, s).

Suppose an elastic material has its strain energy density function given by
Z
Ŵ (Y) = w(Yhξi, Yhχ(ξ)i, ξ, χ(ξ)) dVξ (86)
H

where χ(·) : H → H is a continuously differentiable and invertible function.

(Note that the four arguments of w in the integrand are vectors, not vector

44
states, because Y is evaluated at the specific bonds ξ and χ(ξ).) Let χ−1
be the inverse mapping of χ:

ζ = χ(ξ) ⇐⇒ ξ = χ−1 (ζ).

Let the Jacobian determinants of the forward and inverse mappings be de-
fined by
   
J(ξ) = det grad χ(ξ), J −1 (ζ) = det grad χ−1 (ζ).
   

Mechanically, the Ŵ defined in (86) sums up energies due to interactions

between pairs of bonds ξ and χ(ξ). Such a material is called a bond-pair
material (Figure 6).
To determine the associated force state, the Fréchet derivative of this Ŵ
is evaluated as follows. Consider an increment in the deformation state δY.
Then, from (86),
Z h


δ Ŵ = wp Yhξi, Yhχ(ξ)i, ξ, χ(ξ) · δYhξi
H

i
+ wq Yhξi, Yhχ(ξ)i, ξ, χ(ξ) · δYhχ(ξ)i dVξ .

Now use the change of variables ζ = χ(ξ) in the wq term to obtain

Z


δ Ŵ = wp Yhξi, Yhχ(ξ)i, ξ, χ(ξ) · δYhξi dVξ
H Z
wq Yhχ−1 (ζ)i, Yhζi, χ−1 (ζ), ζ · δYhζi J −1 (ζ) dVζ .


+
H

In the second integral, replace the dummy variable of integration ζ by ξ:

Z h


δ Ŵ = wp Yhξi, Yhχ(ξ)i, ξ, χ(ξ)
H
i
+ wq Yhχ−1 (ξ)i, Yhξi, χ−1 (ξ), ξ J −1 (ξ) · δYhξi dVξ .

Comparing this result with (75), the force state can be read off:


Thξi = ∇Ŵ hξi = wp Yhξi, Yhχ(ξ)i, ξ, χ(ξ)
+ wq Yhχ−1 (ξ)i, Yhξi, χ−1 (ξ), ξ J −1 (ξ). (87)

Bond-based materials are a special case of bond pair materials with χhξi = ξ
for all ξ.

45
Figure 6: In a bond-pair material, the bond force density in each bond ξ is
determined by its own deformation and that of another bond χ(ξ).

46
Figure 7: Top: Unit vectors m(p, q) and m(q, p). Bottom: Peridynamic
beam based on a bond-pair material. The forces tend to restore the relative
bond angles to their initial value, which in this case is π.

47
4.13 Example: a bond-pair material in bending
Consider the bond-pair material defined by (86) with
c
w(p, q, r, s) = (θ − θ0 )2 , (88)
4
p·q r·s
θ = cos−1 , θ0 = cos−1
|p||q| |r||s|
where c is a constant and both θ and θ0 are in the interval [0, π]. (χ will be
defined later.) θ is the angle between the deformed bonds p and q, while θ0
is the angle between the undeformed bonds r and s. Mechanically, if c > 0,
this material resists changes in the angle between the bonds r and s. The
elastic material model defined by (86) and (88) is objective, because it does
not refer to any special direction in space. Using (88) and the chain rule to
obtain wp , one finds, for θ 6= 0,
 
∂w ∂θ c(θ − θ0 ) −1 ∂ cos θ
wp = =
∂θ ∂p 2 sin θ ∂p
 
c(θ0 − θ) 1 |q| cos θ
= q− p .
2 sin θ |p||q| |p|

A more suggestive form of this expression is

c(θ0 − θ)
wp = m(p, q) (89)
2|p|

where 
 0   if θ = 0,
m(p, q) = 1 q p
− cos θ , if θ 6= 0.
|q| |p|

sin θ
Geometrically, m(p, q) is the unit vector normal to p that is coplanar with
p and q such that q · m(p, q) ≥ 0 (Figure 7). Similarly,

c(θ0 − θ)
wq = m(q, p). (90)
2|q|

To define the pairing of bonds, take

χ(ξ) = −ξ ∀ξ ∈ H, (91)

hence J = J −1 = 1. Then, by (87), (89), and (90),

Thξi = wp Yhξi, Yh − ξi, ξ, −ξ + wq Yh − ξi, Yhξi, −ξ, ξ
c(π − θ)
= m(Yhξi, Yh − ξi)
|Yhξi|

48
where
Yhξi · Yh − ξi
θ = cos−1 , 0 ≤ θ ≤ π.
|Yhξi||Yh − ξi|
Note that Thξi ⊥ Yhξi (see Figure 7).
This material does not offer resistance to any homogeneous deformation.
The strain energy density W changes only in response to nonhomogeneous
deformations. This response is an aspect of nonlocality, because the finite-
ness of the bond lengths is what gives rise to the angle changes that result
in changes in strain energy density.
An application of this material model is the bending of a beam. The
strain energy increases according to deformations of the beam involving
curvature. This can be thought of as a nonlocal version of an Euler beam.
However, in the traditional treatment of an Euler beam, a new PDE is in-
troduced, reflecting the resistance to curvature. This fourth order PDE is
virtually unrelated to the second order PDEs of the classical theory of elas-
ticity. In contrast, in the peridynamic beam model proposed here, the fun-
damental equation of motion is unchanged from the basic three dimensional
peridynamic equation of motion. The peridynamic beam model simply uses
a particular choice of material model, which is the bond-pair model with the
choice of χ given in (91).

49
5 Linear theory
Like the linear classical theory, the linear peridynamic theory concerns small
deformations. However, the applicable notion of smallness is different in the
peridynamic theory, because it does not restrict the deformation gradient,
and even allows discontinuities. Under this assumption of smallness, the
peridynamic equation of equilibrium reduces to a linear integral equation.
Linearization of the bond-based peridynamic theory is discussed in [60,
83]. The discussion below pertains to the more general state-based theory
and largely follows [63]. See Section 1.2 for a summary of work to date
making use of the linear theory.

5.1 Small displacements

Let B be a body with horizon δ. Consider a time-independent deformation
y0 , which may be large. (The role of y0 in the linearization will become
clear in Section 5.3 below.) Let u be a displacement field superposed on y0 ,
and define a vector state field by
U[x, t]hq − xi = u(q, t) − u(x, t), ∀x ∈ B, (q − x) ∈ H. (92)
The displacement field is said to be small if
`δ (93)
where
`= sup |u(q, t) − u(x, t)|. (94)
|q−x|≤δ

(Strictly speaking, (93) is actually a condition on the relative displacements,

rather than the displacements themselves.) This idea of a small displacement
field is a nonlocal analogue of the concept in classical linear theory that
|grad u|  1. The peridynamic definition of a small displacement field (93)
does not restrict rigid translations of a body, but it does restrict rigid body
rotations to small angles. More importantly, it allows for possible small
discontinuities in u, a key difference from the classical linear theory.
Recalling the definition of the norm of a vector state (52), for a small
displacement field, (93) and (94) imply that
||U|| = O(`).

5.2 Double states

In the linear peridynamic theory, the analogue of the classical fourth order
elasticity tensor is a double state (see definition following (49)). Before de-
veloping the linear theory further, it is helpful to introduce some properties
of double states.

50
 If D is a double state, then for every pair of bonds ξ and ζ in H, the value
of Dhξ, ζi is a second-order tensor. The set of all double states is denoted
D. In the following, D and E are double states, while A and B are vector
states. Define the vector state D • A by
Z
(D • A)hξi = Dhξ, ζi Ahζi dVζ ∀ξ ∈ H
H

where the integrand is the product of a second order tensor with a vector;
the component form is Dij Aj . The adjoint of D is a double state defined by

D† hξ, ζi = DT hζ, ξi ∀ξ, ζ ∈ H.

where the superscript T indicates the tensor transpose. Note that the order
of the bonds is switched when taking the adjoint in addition to taking the
tensor transpose. D is self-adjoint if

D† = D.

Also define the vector state A • D by

A • D = D† • A.

For any vector states A and B, the following identity holds:

B • D† • A = A • D • B.

If S(·) : V → V is Fréchet differentiable, then

S(A + a) = S(A) + ∇S(A) • a + o(||a||) ∀A, a ∈ V (95)

where ∇S(A) is a double state. As before, in the case of a function of

two states, such as S(A, B), the Fréchet derivative with respect to each is
denoted SA or SB .
If Ψ(·) : V → R, then the second Fréchet derivative of Ψ, if it exists, is a
double state defined by

∇∇Ψ = ∇(∇Ψ) on V.

The following list summarizes, omitting some details, three important results
that are proved in [63]:
(i) If Ψ(·, ·) : V ×V → R and Ψ is twice continuously Fréchet differentiable,
then the order of differentiation in the mixed second Fréchet derivatives
of Ψ is interchangeable, i.e.,

(ΨA )B (A, B) = (ΨB )A (A, B) ∀A, B ∈ V.

51
 (ii) If Ψ(·) : V → R and Ψ is twice continuously Fréchet differentiable,
then
(∇∇Ψ)† = ∇∇Ψ on V.

(iii) If S(·) : V → V and S is continuously Fréchet differentiable, then

(∇S)† = ∇S on V (96)

if and only if there exists a twice continuously Fréchet differentiable

function Ψ(·) : V → R such that

S = ∇Ψ on V.

(This result is analogous to Poincaré’s theorem in vector calculus.)

5.3 Linearization of an elastic constitutive model

Let the force state for a body B be given by T̂(Y), where T̂ is the constitutive
model. Suppose that T̂ is Fréchet differentiable, and denote its Fréchet
derivative by T̂Y . Consider an equilibrated deformation y0 corresponding
to a time-independent external body force density field b0 , and define

Y0 [x]hq − xi = y0 (q) − y0 (x) ∀x, q ∈ B

and
T0 [x] = T̂(Y0 [x], x) ∀x ∈ B.
Define a double state field called the modulus state field by

K[x] = T̂Y (Y0 [x], x) ∀x ∈ B. (97)

Let u be a small displacement field superposed on y0 , and define the dis-

placement state field by (92). Define the linearized constitutive model by

T[x, t] = T0 [x] + K[x] • U[x, t] ∀x ∈ B, t ≥ 0. (98)

By (95), the linearized model differs from the full model by a term of order
o(||U||). If the material is elastic, then by definition

T̂(Y, x) = ŴY (Y, x) ∀Y ∈ V, x ∈ B

where Ŵ is the strain energy density function. Therefore, by (97),

K[x] = ŴYY (Y0 [x], x) ∀x ∈ B, (99)

and by (96),
K† = K on B. (100)

52
Furthermore, by result (iii) of Section 5.2, (100) implies that the material
is elastic.
The fact that the modulus state K[x] depends on Y0 in (99) represents
a key advantage of the peridynamic approach over other nonlocal models,
because it encompasses the coupling between the large deformation state
Y0 and subsequent small motions of the body. This includes the coupling
between a large bond force density T0 hξi and small rotation of the bond. A
striking example of the importance of these rotations is given in Example 4
of [63], in which large compressive forces in bonds couple to subsequent
rotations to result in material instability.
For an elastic material, the force state in the linearized model (98) can be
obtained from the Fréchet derivative of the following strain energy density
function:
lin 1
T = ŴU (U, x), Ŵ lin (U, x) = T0 [x] • U + U • K[x] • U.
2

5.4 Equations of motion and equilibrium

Continuing under the assumptions of the previous section, the peridynamic
equation of motion (58) under a body force density field b̂ is
Z  
ρ(x)ÿ(x, t) = T[x, t]hp − xi − T[p, t]hx − pi dVp + b̂(x, t)
B

for all x ∈ B, t ≥ 0. Since, by assumption, Y0 is equilibrated under b0 ,

Z  
T0 [x]hp − xi − T0 [p]hx − pi dVp + b0 (x) = 0 ∀x ∈ B.
B
Subtracting the last two equations and using (98) leads to
Z 
ρ(x)ü(x, t) = (K[x] • U[x, t])hp − xi
B

− (K[p] • U[p, t])hx − pi dVp + b(x, t) (101)
where
b(x, t) = b̂(x, t) − b0 (x) = 0 ∀x ∈ B, t ≥ 0.
Using (92) in (101) and simplifying results in
Z


ρ(x)ü(x, t) = C0 (x, q) u(q, t) − u(x, t) dVq + b(x, t) (102)
B
for all x ∈ B, t ≥ 0, where C0 is the tensor valued function defined by
Z 
C0 (x, q) = K[x]hp − x, q − xi
B

− K[p]hx − p, q − pi + K[q]hx − q, p − qi dVp (103)

53
for all x, q ∈ B. (Recall from (51) that the dot products in (101) contain
volume integrals.) If the material is elastic, then by (100) and (103),

CT0 (q, x) = C0 (x, q) ∀x, q ∈ B. (104)

Setting the acceleration term to zero in (102) yields the linearized equation
of equilibrium:
Z


C0 (x, q) u(q) − u(x) dVq + b(x) = 0 (105)
B

for all x ∈ B. This is a Fredholm linear integral equation of the second kind.
C0 is called the micromodulus tensor field.

5.5 Linear bond-based materials

If the material is bond-based as well as elastic, recall from (76) and (79)
that there is a bond potential function w(e, ξ, x) such that

T̂(Y, x)hξi = we (ehξi, ξ, x)Mhξi ∀ξ ∈ H (106)

where
Yhξi
ehξi = |Yhξi| − |ξ|, Mhξi = . (107)
|Yhξi|
(x is included among the arguments in w(e, ξ, x) to account for heterogene-
ity.) To evaluate the modulus state field K using (97), the Fréchet derivative
of T̂ is found as follows. For a small increment δY in the deformation state,
according to (106) and the chain rule,

δ T̂ = wee Mδe + we δM.

From (107),

δY
δe = M · δY, δM = 1 − M ⊗ M ,
|Y|

hence

δY
δ T̂ = wee (M ⊗ M) · δY + we 1 − M ⊗ M .
|Y|
From this, (95), and (97),

K[x]hξ, ζi = wee (e0 hξi, ξ, x)M0 ⊗ M0

1 − M0 ⊗ M0

0
+ we (e hξi, ξ, x) ∆(ζ − ξ). (108)
|Y0 hξi|

54
where
Y0 hξi
e0 hξi = |Y0 hξi| − |ξ|, M0 =
|Y0 hξi|
and ∆ is the three dimensional Dirac delta function. The K given by (108)
is clearly self-adjoint. Because of (106), the second term on the right hand
side of (108) is non-null only if T0 is non-null. This second term represents
the change in direction of T0 in response to an incremental change dY.
The first term on the right hand side of (106) represents the change in the
magnitude of the bond force density.
In the special case of linearization near the reference configuration and
of zero bond force densities in this configuration, (108) specializes to

ξ⊗ξ
K[x]hξ, ζi = wee (0, ξ, x)∆(ζ − ξ) . (109)
|ξ|2

In this case, (103) and (109) imply, after evaluating the integral,

  (q − x) ⊗ (q − x)
C0 (x, q) = wee (0, q − x, x) + wee (0, x − q, q)
|q − x|2
(x − p) ⊗ (x − p)
Z
− ∆(q − x) wee (0, x − p, p) dVp . (110)
B |x − p|2

From (110), it follows that in addition to the symmetry (104) that always
holds for linearized elastic material models, the following symmetries also
hold if the material is bond-based:

CT0 (x, q) = C0 (x, q), C0 (q, x) = C0 (x, q) ∀x, q ∈ B.

When (110) is substituted into the linearized equation of motion (102), the
term involving ∆(q−x) integrates to 0, and the resulting equation of motion
for this body is
Z


ρ(x)ü(x, t) = C(x, q) u(q, t) − u(x, t) dVq + b(x, t) (111)
B

for all x ∈ B, t ≥ 0, where

  (q − x) ⊗ (q − x)
C(x, q) = wee (0, q − x, x) + wee (0, x − q, q) .
|q − x|2

If the horizon δ is constant throughout B, then the region of integration in

(111) can be replaced by the neighborhood of radius δ centered at x. This
smaller region of integration can be used because wee (0, ξ, x) = 0 whenever
|ξ| > δ. The smaller region cannot be used in the more general expression
(102).

55
5.6 Equilibrium in a one dimensional model
This section describes application of the linear theory to an infinitely long
bar under static loading by a body force density field b. It is shown that
the problem can be analyzed using the Fourier transform, and a Green’s
function solution is derived. Consider a homogeneous bar with constant
cross-sectional area A, infinitely long in both directions, oriented along the
x-axis. The material model is bond-based and linear elastic with horizon
δ. The transverse dimensions of the bar are much smaller than δ. Define a
coordinate system in which x1 is the axial direction. Let

x = x1 , u = u1 , b = b1 ,

C(q − x) = AC11 (x, q) ∀x, q ∈ R3 .

The equilibrium equation (105) simplifies to
Z ∞


C(q − x) u(q) − u(x) dq + b(x) = 0, x ∈ R. (112)
−∞

By (104),
C(−ξ) = C(ξ) ∀ξ ∈ R3 .
As shown in [75], this C is related to the Young’s modulus E that would be
measured in the static extension of a long bar by
Z ∞
E= ξ 2 C(ξ) dξ.
0

Let v ∗ denote the Fourier transform of any function v(x):

Z ∞
∗
v (κ) = e−iκx v(x) dx,
−∞

with inverse transform given by

Z ∞
1
v(x) = eiκx v ∗ (κ) dκ.
2π −∞

The Fourier variable κ physically represents the wave number, κ = 2π/λ,

where λ is the wavelength. Taking the Fourier transform of (112) leads to
Z ∞Z ∞ h
i
e−iκx C(q − x) u(q) − u(x) dq dx + b∗ (x) = 0.
−∞ −∞

By the convolution theorem, this implies

C ∗ (κ)u∗ (κ) − P u∗ (κ) + b∗ (κ) = 0

56
where Z ∞
P = C(ξ) dξ = C ∗ (0).
−∞
The transformed displacement can therefore be written formally as

b∗ (κ)
u∗ (κ) = , (113)
M (κ)

where
M (κ) = P − C ∗ (κ). (114)
After inversion of the transform, the displacement is given by
Z ∞ Z ∞
1 1 b∗ (κ)
u(x) = eiκx u∗ (κ) dκ = eiκx dκ
2π −∞ 2π −∞ M (κ)
Z ∞Z ∞
1 b(z)
= eiκ(x−z) dz dκ
2π −∞ −∞ M (κ)
Z ∞
= g(x, z)b(z) dz (115)
−∞

where g is a Green’s function given by

∞
eiκ(x−z)
Z
1
g(x, z) = dκ.
2π −∞ M (κ)

Some particular cases of static solutions to this one dimensional problem

are discussed in [70] and [75], the latter of which also derives a dynamic
Green’s function. An example taken from [70] is shown in Figure 8. In this
example, two opposite point loads of magnitude b0 are applied at the points
x = ±a, thus
b(x) = b0 (∆(x − a) − ∆(x + a))
where ∆ is the one dimensional Dirac delta function. The peridynamic
material model has a micromodulus function given by

3E/δ 2 , |ξ| ≤ δ

C(ξ) =
0, otherwise

where E is the Young’s modulus. The length scale in the material model
(the horizon) and the length scale in the loading are related, in this example,
by the arbitrary choice
a = δ/4.
The peridynamic solution contains delta functions, indicated by the vertical
arrows in the figure, located at ±a. The jumps in displacement shown in the
figure at ±3a and ±5a are a result of the discontinuities in C(ξ) at ξ = ±δ;

57
Figure 8: Peridynamic and classical displacement fields for two opposite
point loads applied at x = ±a, where a = δ/4.

they would not appear if a continuous C were used. The solution also in-
cludes oscillations that decay with distance from the points of application of
the loads. As shown in [70] analytically, each successive oscillation contains
a discontinuity in a higher derivative of u than the one before it. The peri-
dynamic result approaches the classical displacement field asymptotically at
large distances.
If the length scale in the material model, i.e., the horizon, is reduced so
that δ  a while holding E constant, then the peridynamic displacement
field approaches the classical field at all points except x = ±a. At these
two points, the delta functions persist regardless of δ (see [70] for details).
The convergence in the limit of small horizon is consistent with the results
of Section 6.3. The analysis in that section does not apply at the points
x = ±a because the deformation is not smooth there; thus the peridynamic
operator is not expected to converge to the classical operator.
For a given body force density field b, if a solution u to the equilibrium
equation (112) exists, then the function u0 defined by

u0 = u + uh

is also a solution to (112), where uh is any solution to the following homo-

58
geneous equilibrium equation:
Z ∞


Luh := C(q − x) uh (q) − uh (x) dq = 0, x ∈ R, t ≥ 0. (116)
−∞

where L is a linear operator. For any C, one such uh is provided by uh ≡

c, where c is any constant. (A similar statement is true of the classical
theory: any rigid translation of an equilibrium solution is also an equilibrium
solution.)
Other homogeneous solutions may also exist for certain choices of C. The
set of all these uh functions is the null space of L. If M (κ) has a nonzero
root κ0 , then it is easily confirmed that the function

uh (x) = eiκ0 x

is a solution to the homogeneous equation (116). For such a material, the

transformed displacement u∗ given by (113) does not exist if b∗ (κ0 ) 6= 0.
In other words, there is no deformation of the bar that can equilibrate an
applied body force density that has a nonzero Fourier component at κ0 . In
this case, (115) may not apply.
Materials having a micromodulus C such that M has a nonzero root are
not typical well-behaved materials. For example, if C is nonnegative and
continuous on R, it follows from (114) that M has only the root κ = 0. (It
is too restrictive to assume that C is strictly positive, since this excludes
some materials that have physically reasonable behavior.)
An example of a material for which M does have nonzero roots has its
micromodulus given by

C(ξ) = ∆(ξ − α) + ∆(ξ + α)

where α is a positive constant that is similar to the interatomic spacing in

a discrete lattice model. For this material, by (114),

M (κ) = 2(1 − cos ακ),

which has an infinite number of roots κ0 = 2πn/α, where n is any integer.

If the loading has nonzero Fourier components at these roots, i.e., if
 
∗ 2πn
b 6= 0
α

for some integer n, then a solution to the equilibrium equation (112) fails
to exist. However, if b contains no Fourier components κ outside the set
(−2π/α, 0) ∪ (0, 2π/α), then the solution exists and is given by (115). This
means that as α → 0, there is a larger and larger interval on the κ axis

59
for which the peridynamic model behaves, for purposes of existence and
uniqueness, like the classical theory. This observation is consistent with the
general rule that in the limit of small material length scale, the peridynamic
theory behaves similarly to the classical theory, provided certain smoothness
conditions are met. (In the classical theory, only the Fourier component
κ = 0 is excluded from b, because it represents loading that is not self-
equilibrated.)
In summary, the existence of solutions to the equilibrium problem (112)
depends on both the material properties and the loading. A solution u for
well-behaved materials (no nonzero roots of M ) is given by (115), provided
b∗ (0) = 0. For any such solution u, u + c is also a solution for any constant
c. If M has nonzero roots, the existence of solutions depends on the Fourier
spectrum of the loading.

5.7 Plane waves and dispersion in one dimension

The equation of motion corresponding to (112) for the infinite, homogeneous
bar is given by
Z ∞


ρü(x, t) = C(q − x) u(q, t) − u(x, t) dq + b(x, t), x ∈ R, t ≥ 0. (117)
−∞

To investigate plane waves in the bar with wave number κ and angular
frequency ω, assume a motion of the form

u(x, t) = ei(κx−ωt)

with b ≡ 0. Substituting this expession into (117) results in

Z ∞
2 i(κx−ωt) −iωt
C(q − x) eiκq − eiκx dq


−ρω e =e
−∞

which implies the condition

ρω 2 = M (κ)

where M is given by (114). This provides the following dispersion relation:

s
M (κ)
ω(κ) = ± .
ρ

The corresponding phase velocity is given by

s
ω(κ) 1 M (κ)
c(κ) = =± .
κ κ ρ

60
Figure 9: Dispersion curves for stable and unstable peridynamic materials.

Evidently, stable waves exist for a given κ only if M (κ) > 0. Figure 9
illustrates possible dispersion curves for a peridynamic material exhibiting
stable waves at all wave numbers, and also a material that is unstable for
some wave numbers due to “imaginary wave speeds.” The peridynamic
dispersion curves have the same slope at the origin as the classical dispersion
line if their Young’s moduli are the same.

61
6 Relation to other theories
In this section we identify quantities and principles that allow comparison
between the peridynamic and the classical theories, as well as some non-
local theories. This comparison would not be complete without discussing
Navier’s theory of solids (see [71, 72]). Navier conceived of a continuum as a
smoothed out distribution of masses that interact with each other through a
central potential. This model treated the relative motion of each such pair
according to a first order approximation based on the displacement gradient
components. In this sense, Navier’s model was a local theory, even though
conceptually it involved long-range forces. Because of the assumption of
central potentials, isotropic materials were restricted to a Poisson ratio of
1/4 in Navier’s theory. The local theory of Cauchy that was introduced
after Navier’s does not suffer from this restriction, and this is the classical
continuum theory that continues to enjoy wide usage and acceptance today.
In the next section, we compare the kinematics of this classical theory with
those in the peridynamic theory.

6.1 Deformation gradient and the deformation state

Recall the definition of the deformation state (55) at a point x ∈ B, where
B is a body with horizon δ:

Y[x]hξi = y(x + ξ) − y(x) ∀ξ ∈ H. (118)

(In this discussion, the time variable will not be included explicitly.) If the
deformation is continuously differentiable on B, then the first two terms of
a Taylor expansion yield

y(x + ξ) = y(x) + F(x)ξ + O(|ξ|2 ) ∀ξ ∈ H

where the deformation gradient tensor field F is defined by

∂y
F(x) = (x) ∀x ∈ B.
∂x
Comparing the last two equations,

Y[x]hξi = F(x)ξ + O(|ξ|2 ) ∀ξ ∈ H, x ∈ B. (119)

For a given continuously differentiable deformation, since |ξ| ≤ δ, it follows

that
Y[x] = F(x)X + O(δ 2 ) ∀x ∈ B
where X is the identity state defined in (49). In this sense, conceptually,
the deformation gradient approximates the deformation state. The key dis-
tinctions are as follows:

62
 • The deformation state is nonlocal in that it explicitly relates the de-
formation of material points separated by finite distances, while the
deformation gradient is local.

• The deformation state can be evaluated even if the deformation is not

differentable or continuous.

• The deformation gradient tensor maps small spheres into ellipsoids.

(This is implied by the polar decomposition theorem.) The deforma-
tion state can describe more complex kinematics.

6.2 Peridynamic stress tensor

Recall from Section 2.5 that, as a consequence of Noll’s lemma I [53, 49], the
flux of linear momentum through a surface, i.e., the force per unit area, can
be expressed as the divergence of a tensor field. This tensor field is called
the peridynamic stress tensor field, denoted ν. This can be expressed in
terms of the primitive quantity t as
Z Z ∞Z ∞
ν(x) = (y + z)2 t(x + ym, x − zm) ⊗ m dz dy dΩm (120)
U 0 0

where U is the unit sphere and dΩm is a differential element of solid angle
in the direction of the unit vector m. The integrand can be alternatively
expressed in terms of the force state by substituting

t(x + ym, x − zm) = T[x − zm]h(y + z)mi. (121)

The peridynamic equation of motion in terms of the peridynamic stress

tensor is formally identical to the classical equation of motion:

ρÿ = div ν + b on B, t ≥ 0, (122)

in other words
Z
(t − t0 ) dV 0 = div ν on B, t ≥ 0. (123)
B

The peridynamic stress tensor is similar to the Piola stress in the classical
theory in that it provides the net force per unit area through a closed surface
∂P with unit normal n: Z
FP = νn dA.
∂P
However, the forces described by ν are nonlocal; they represent direct in-
teraction between points such as x + ym in the exterior of ∂P with points
such as x − zm in the interior.

63
6.3 Convergence in the limit of small horizon
Under the assumption of a continuously differentiable deformation, the ap-
proximation (119) becomes more accurate as the horizon δ is reduced, be-
cause |ξ| ≤ δ. It is also reasonable to expect that, since t in (121) is non-null
only when y + z ≤ δ, the (nonlocal) peridynamic stress tensor should ap-
proach the (local) Piola stress tensor as δ is reduced.
To further investigate the convergence of the peridynamic equations to
the classical equations, it is necessary to specify what it means for a peridy-
namic constitutive model to change horizon. Restricting attention to elastic
materials, one way to do this is to require that for a homogeneous deforma-
tion of a homogeneous body, the strain energy density should be invariant
as the horizon changes. To make this precise, suppose a material model Ŵ1
is given with horizon δ1 . For any horizon δ > 0, define s = δ/δ1 . Let Hs
be the family with horizon δ, and let Vs be the set of vector states on Hs .
Define an elastic material model by


Ŵs (Ys ) = Ŵ1 Es (Ys ) ∀ Ys ∈ Vs (124)

where Es (·) : Vs → V1 is defined by

Ys hsξi
Es (Ys )hξi = , ∀ ξ ∈ H1 . (125)
s
Geometrically, Es rescales the length of bonds ξ ∈ Hs to the original family
H1 . To confirm that the new material model defined in (124) possesses
the required invariance under rescaling, let a homogeneous deformation of a
large body B be defined by

y(x) = F0 x + c ∀x∈B

where F0 is a constant tensor, det F0 > 0, and c is a constant vector. If

x ∈ B is sufficiently far from the boundary of the body that its family does
not include any points on this boundary, then from (118),

Ys hξi = F0 ξ ∀ ξ ∈ Hs . (126)

Then for any ξ ∈ V1 , from (125) and (126),

Ys hsξi F0 sξ
Es (Ys )hξi = = = F0 ξ.
s s
So, if the deformation is homogeneous, Es (Ys ) is independent of s. There-
fore, under this assumption, Ŵs (Ys ) defined through (124) is also indepen-
dent of s. This proves that this rescaled material model is invariant under
changes in δ if the deformation is homogeneous.

64
 Now let T̂s denote the constitutive model for the force state derived from
Ŵs :
T̂s (Ys ) = ∇Ŵs (Ys ) ∀Ys ∈ Vs .
It is easily shown [68] that this force state scales with s as follows:

T̂s (Ys )hξi = s−4 T̂1 Es (Ys ) hξ/si

∀ξ ∈ Hs . (127)

Let νs denote the peridynamic stress tensor obtained from its definition
(120) for this T̂s using (121):
Z Z ∞Z ∞
νs (x) = (y + z)2 T̂s (Ys )[x − zm]h(y + z)mi ⊗ m dz dy dΩm .
U 0 0

Returning to the case of a given continuously differentiable deformation with

deformation gradient tensor field F, it can be shown [68] that νs approaches
a limit given by

νs (x) → σ(F(x)) as s → 0 ∀ x ∈ B

where σ is the function defined by

Z
σ(F) = T̂1 (FX)hξi ⊗ ξ dVξ ∀ F ∈ L+ (128)
H1

where L+ is the set of all second order tensors with positive determinant.
σ is called the collapsed stress tensor because it represents the limit, under
the present assumptions, of the peridynamic stress tensor for the horizon
collapsing to zero. It can further be shown [68] that

div νs (x) → div σ(F(x)) as s → 0, ∀x ∈ B, (129)

and that
∂Ω
σ(F) = (F) ∀F ∈ L+
∂F
where Ω is defined by

Ω(F) = Ŵ1 (FX) ∀F ∈ L+ .

Furthermore, σ and Ω inherit properties from the peridynamic material with

horizon δ1 characteristic of a Piola stress tensor:

• Objectivity: Let O+ be the set of all proper orthogonal tensors. Then

Ŵ1 (QY) = Ŵ1 (Y) ∀Y ∈ V1 , Q ∈ O+

=⇒ Ω(QF) = Ω(F) ∀F ∈ L+ , Q ∈ O+ .

65
 • Isotropy:

Ŵ1 (YQ) = Ŵ1 (Y) ∀Y ∈ V1 , Q ∈ O+

=⇒ Ω(FQ) = Ω(F) ∀F ∈ L+ , Q ∈ O+ .

• Balance of angular momentum:

Z
Yhξi × T̂1 (Y)hξi dVξ = 0 ∀Y ∈ V1
H1
=⇒ σ(F)FT = Fσ T (F) ∀F ∈ L+ .

(129) means that if a given deformation is twice continuously differentiable,

we can compare the acceleration fields ÿs computed by the peridynamic
theory, for a material with horizon sδ1 , to those of the classical theory ÿ0 :

ÿs → ÿ0 as s → 0

where Z h i
ÿs = Ts [x]hx0 − xi − Ts [x0 ]hx − x0 i dVx0 + b
H
and
ÿ0 = div σ + b.
In these equations, Ts and σ are derived from the same peridynamic mate-
rial model with horizon δ1 through (127) and (128) respectively.
In this sense, the peridynamic theory converges to the classical theory
in the limit of small horizon. Stated differently, the PDEs of the classical
theory are obtainable from the peridynamic equations as a limiting case.

6.4 Elasticity tensor derived from a peridynamic material

Recall from Section 6.3 that a classical material model can be derived from a
peridynamic model by requiring that the two produce the same stress tensor
for all homogeneous deformations. The classical material model is given by
(128), which provides a Piola stress tensor σ(F), where F is the deformation
gradient tensor.
By specializing this approach to linear peridynamic material models, a
fourth order elasticity tensor can be derived. To do this, substitute the
linear constitutive model (98) into (128) and assume T0 = 0. Consider a
displacement gradient tensor H, not necessarily symmetric, where |H|  1.
By setting F = 1 + H, the following expression for the components of stress
is obtained [63]:
Z Z 
σij = K ik hξ, ζiξj ζl dVζ dVξ Hkl .
H H

66
The classical constitutive model for linear elasticity is

σij = Cijkl Hkl

where Cijkl is the fourth order elasticity tensor. Since H is arbitrary, com-
paring the last two equations leads to the conclusion that
Z Z
Cijkl = K ik hξ, ζiξj ζl dVζ dVξ .
H H

Thus, a classical linear elastic material model has been obtained from a
peridynamic linear elastic model. The two models give the same stress tensor
for homogeneous deformations of a homogeneous body. They are expected
to disagree for nonhomogeneous deformations, and the classical model is not
applicable at all if a discontinuity in the deformation is present.

6.5 Nonlocal theories

In this section, the peridynamic model is compared with other strongly non-
local theories, i.e., theories in which points separated from each other by a
finite distance interact directly. Also included is a comparison with higher
order gradient models, which are weakly nonlocal because they contain a
length scale in the constitutive model but do not explicitly include inter-
actions across finite distances. The literature on such nonlocal models is
large, and only a few representative models are discussed here to illustrate
the main similarities and differences.
Strongly nonlocal theories have been proposed as a way to gain insight
into the role of the finiteness of the interaction distance between atoms,
particularly microstructure in crystals. A pioneering example of such a
nonlocal theory is Kröner’s [45] , which added nonlocal terms, in the form
of an integral operator, to the local equation of motion for a body B:
Z
ρ(x)üi (x, t) = Cijkl uk,lj + Φik (x0 − x)uk (x0 ) dVx0 + bi (x, t)
B

where Cijkl is the fourth order elasticity tensor, and Φik is a function rep-
resenting the effect of long-range interactions. Kröner’s nonlocal model is
linear because the integral is a linear operator. Because Kröner’s equation
of motion, like the classical equation of motion, involves the second partial
derivatives of displacement, it does not lend itself to the study of phenomena
involving discontinuities.
Perhaps the most widely known nonlocal elasticity theory is that of Erin-
gen [24]. In its simplest form [23], its basic equations for an isotropic solid
can be expressed as

ρ(x)üi (x, t) = tij,j (x, t) + bi (x, t)

67
 Z
tij (x, t) = α(|x0 − x|)σij (x0 , t) dVx0
B
σij (x, t) = λδij ukk (x, t) + 2µui,j (x, t)
where λ and µ are the usual Lamé moduli and α is a weighting function.
Thus, in this version of Eringen’s model, a nonlocal stress tensor tij is eval-
uated from the weighted volume average of the local stress tensor σij . Like
Kröner’s model (and the classical local model), the form of Eringen’s equa-
tions prevents it from achieving the goal of peridynamics, which is to apply
the same field equations on or off of discontinuities. (However, Eringen,
Speziale, and Kim [25] successfully treat the problem of a crack in a nonlo-
cal elastic medium by, essentially, representing the crack as a zero-traction
boundary condition. The solution to this problem demonstrates the absence
of unbounded stress fields near the tip of a crack represented in this way
within a nonlocal continuum.)
Kunin [46, 47] developed a nonlocal model in which internal forces are
expressed directly in terms of the displacement:
Z
ρ(x)üi (x, t) = Φik (x0 − x)uk (x0 ) dVx0 + bi (x, t).
B

Formally, this expression is the same as the linearized equation of motion in

the peridynamic model, (102). However, in the peridynamic equation, the
kernel is derived by linearization of a material model in the nonlinear theory,
and therefore allows the micromodulus function C0 in (102) to be obtained
in an unambiguous way. For example, linearization of a peridynamic model
of an isotropic fluid uniquely determines the appropriate C0 function [63].
Deriving the linearized material model as described in Section 5.3 also
preserves the coupling between the forces within a large deformation state
Y0 and a superposed small displacement field. These potentially large forces
have a major effect on the properties of plane waves, and therefore on ma-
terial stability [63]. For example, in a typical crystal, there are large forces
between the atoms regardless of the deformation. If the crystal has zero
stress at a point, then the net force across any plane through that point is
zero. But this does not change the fact that the forces are present; they
merely cancel each other out. The coupling between these forces and su-
perposed displacements cannot be neglected in a theory aimed at providing
insight into the mechanics of microstructures.
A higher order approximation to the deformation state may be obtained
by adding an additional term to the Taylor expansion in (119):

Y[x]hξi = F(x)ξ + (grad F(x))(ξ ⊗ ξ) + O(|ξ|3 ). (130)

where (grad F(x)) is a third order tensor representing a “strain gradient”

term. If a constitutive model explicitly involves this new term, for example

68
if the free energy has the form

ψ(F, grad F),

the resulting mathematical description of the system is a higher order gradi-

ent theory. Theories of this type implicitly involve a length scale because the
dimensions of the two arguments of ψ differ by length. If the deformation has
continuous third derivatives in space, then (130) asymptotically represents
a better approximation to Y then (119). However, if there is a discontinuity
in the deformation such as a crack, then a higher order gradient model is no
more applicable than the classical local model.
Nevertheless, Seleson et al. [59] have shown that a higher order gradient
approximation can be used as an intermediate step in upscaling a molecu-
lar dynamics model to peridynamics. To do this, these authors construct
separate higher order gradient models from both an assumed peridynamic
constitutive model and from the discrete system using an inner expansion
technique [4]. The parameters in the peridynamic model are determined by
matching the coefficients between these two higher order gradient expres-
sions. In effect, this provides a peridynamic model for the discrete system
that is demonstrated to accurately reproduce key features of the molecular
dynamics model, including wave dispersion.

69
7 Discrete particles as peridynamic bodies
In this section it is shown that the ODEs describing the motion of discrete
particles can be obtained as the limiting case of the motion of mutually
interacting peridynamic bodies of finite volume, as their volume is reduced
to zero. This limiting process could not be carried out within the classical
theory of continuum mechanics, because nonlocality is a fundamental as-
pect of the system. The resulting interactions between the discrete particles
have the same basic structure as the peridynamic continuum equations. It is
further shown that these particles can be represented within the continuum
equations with a mass density field and constitutive model that use gener-
alized functions. This allows any multibody potential to be represented as
a peridynamic material model.
Averaging the mass density and the interactions between particles results
in a conventional (smooth) continuum. The averaged peridynamic stress
tensor over a collection of particles provides, in effect, a Piola stress tensor
field that appears in the classical (PDE) equation of motion. This result
achieves the goal of deriving a classical stress tensor field from a set of
particles interacting through an arbitrary multibody potential.

7.1 Self-equilibrated subregions

Let a body B be defined by
N
[
B= Pi ,
i=1

where the Pi are N disjoint bounded subregions. Assume that the nonpo-
larity condition (25) holds on B. For each subregion, define its mass and
center of mass by
Z Z
1
Mi = ρ(x) dVx , yi (t) = ρ(x)y(x, t) dVx . (131)
Pi Mi Pi

Recall the abbreviated notation defined in (12). Define the net force and
that Pj exerts on Pi by
Z Z
Fij = f dV 0 dV (132)
Pi Pj

and the net external body force on Pi by

Z
ext
Fi = b dV . (133)
Pi

70
Within each Pi , let ri denoted the deformed position vector relative to the
deformed center of mass:

ri (x, t) = y(x, t) − yi (t), ∀x ∈ Pi . (134)

The net moment about yi that Pj exerts on Pi is found from

Z Z
τ ij = ri × f dV 0 dV . (135)
Pi Pj

where ri = ri (x, t). The net external moment on Pi about yi is given by

Z
ext
τi = ri × b dV . (136)
Pi

The condition
Fii = 0, i = 1, . . . , N (137)
is always satisfied because of (6) and (132). If the subregions are separated
by empty space and are small in size relative to the distances between them,
it is reasonable to require on physical grounds that

τ ii = 0, i = 1, . . . , N. (138)

A system in which (138) holds will be called self-equilibrated. This condition

does not follow from the antisymmetry of f , except in the special case of an
ordinary material (see Section 4.3).
An alternate form of the condition (138) may be derived as follows. By
(11), (132), (134), (135), and (137),
Z Z Z Z
0
τ ii = ri × f dV dV = yi × Fii + ri × f dV 0 dV
ZPi ZPi Pi Pi
0 0
= (yi + ri ) × (t − t ) dV dV
ZPi ZPi
= y × (t − t0 ) dV 0 dV
ZPi ZPi
= (y0 − y) × t dV 0 dV .
Pi Pi

where the change of dummy variable of integration x ↔ x0 has been used in

the last step. Since the material is assumed to be nonpolar, (25) holds. So,
the last expression implies
Z Z
τ ii = (y0 − y) × t dV 0 dV .
Pi B\P i

71
Therefore, an alternate form of the condition for self-equilibration (138) is
Z Z
(y0 − y) × t dV 0 dV = 0, i = 1, . . . , N. (139)
Pi B\P i

We emphasize that self-equilibration does not hold in general; it reflects the

special physical situation of small bodies separated by large empty distances.

7.2 Linear and angular momentum in self-equilibrated sub-

regions
From the local balance of linear momentum (15) integrated over Pi ,
Z Z Z Z
0
ρÿ dV = f dV dV + b dV .
Pi Pi B Pi

In view of (131), (132), (133), and (137), this is equivalent to

X
Mi ÿi = Fij + Fext
i , i = 1, . . . , N. (140)
j6=i

Equation (140) is the balance of linear momentum for the subregions ex-
pressed in terms of the centers of mass, the net force between the subregions,
and the net external forces. It holds regardless of whether the subregions
are self-equilibrated.
Now consider the balance of angular momentum about the origin in
subregion Pi :
Z Z Z Z Z
d 0
ρy × ẏ dV = ρy × ÿ dV = y × f dV dV + y × b dV .
dt Pi Pi Pi B Pi

Using (134),
Z
ρ(yi + ri ) × (ÿi + r̈i ) dV
Pi
Z Z Z
= (yi + ri ) × f dV 0 dV + (yi + ri ) × b dV . (141)
Pi B Pi

But from (131) and (134),

Z Z
ρri dV = ρr̈i dV = 0.
Pi Pi

From this, the terms on the left hand side of (141) involving ri × ÿi and
yi × r̈i drop out. Grouping the remaining terms involving yi together and
using (132) and (133) results in
 X  Z Z Z Z
ext 0
yi × Mi ÿi − Fij −Fi + ρri ×r̈i dV = ri ×f dV dV + ri ×b dV .
j6=i Pi Pi B Pi

72
By the balance of linear momentum (140), the term in parentheses vanishes,
so that
Z Z Z Z
0
ρri × r̈i dV = ri × f dV dV + ri × b dV .
Pi Pi B Pi

If the subregions are self-equilibrated, (138) allows this to be rewritten in

the form X
ȧi = τ ij + τ ext
i , i = 1, . . . , N (142)
j6=i

where Z
ai = ρri × ṙi dV
Pi

and τ ij and τ ext

i are defined in (135) and (136) respectively. Since ai is the
angular momentum of Pi about its own deformed center of mass, ai can be
thought of as the angular momentum due to the the “spin” of the subregion.
(142) asserts that if the subregions are self-equilibrated, changes in this spin
are independent of net force on the subregion.

7.3 Peridynamic particles

The next step is to investigate the balance of angular momentum in the
limit of self-equilibrated subregions with zero size. This limiting case rep-
resents peridynamic particles. To derive the properties of these particles,
we adopt an ansatz concerning the nature of the forces during this limiting
process such that the net forces remain fixed. Suppose that each subregion
Pi is bounded by a sphere centered at the center of mass in the reference
configuration xi . It is assumed that there exists a number  > 0 such that
|ri | ≤  for all i and all t ≥ 0, where ri is defined in (134). The sizes of the
Pi in the reference configuration are variable and are parameterized by .
For any  > 0 and any Pi , assume that the bond force densities obey the
following ansatz:
t = Tij ϕij (x0 , x, ) (143)
where the Tij are vectors independent of .
R ϕRij is a non-negative function
on Pi × Pj × R+ such that for any  > 0, Pi Pj ϕij = 1. From (143), it is
immediate that Z Z
Tij = t dV 0 dV . (144)
Pi Pj

73
From (11), (135), and (143), since |r| ≤ , it follows that
Z Z 
 
|τ ij | =  ri × (t − t0 ) dV 0 dV 
 
 Pi Pj 
Z Z
≤ |ri | |Tij ϕij − Tji ϕji | dV 0 dV
Pi Pj
Z Z Z Z
≤ |Tij | ϕij + |Tji | ϕji dV 0 dV
Pi Pj Pi Pj


≤  |Tij | + |Tji | ,

hence
τ ij = O() as  → 0.
Adopting an ansatz for b similar to (143) leads to

τ ext
i = O() as  → 0.

From these results and (142), the conclusion is that

ȧi → 0 as  → 0. (145)

This proves that in the limit  → 0, the angular momentum of a peridynamic

particle about its own deformed center of mass is independent of time.
Next we derive a condition for nonpolarity of the interparticle forces Tij .
Continuing under the ansatz (143), since the Pi are self-equilibrated, (139)
implies
Z Z
0= (y0 − y) × t dV 0 dV
Pi B\P i
Z XZ
= (y0 − y) × t dV 0 dV
Pi j6=i Pj
Z XZ
(yj + r0j ) − (yi + ri ) × t dV 0 dV


=
Pi j6=i Pj
X Z XZ
= (yj − yi ) × Tij + (r0j − ri ) × t dV 0 dV
j6=i Pi j6=i Pj
X
= (yj − yi ) × Tij + O() as →0
j6=i

where r0j = rj (x0 , t). So, in the limit  → 0, the requirement of nonpolarity
(25) reduces to X
(yj − yi ) × Tij = 0 (146)
j6=i

74
for all i, which states that the net moment about yi exerted by the Tij on
the other particles is zero.
Using (11), (132), and (144), it follows that

Fij = Tij − Tji

for all i and j. Then by (140), the balance of linear momentum for peridy-
namic particles may be written as
X
Mi ÿi = (Tij − Tji ) + Fext
i (147)
j6=i

for all i.
We have already investigated the dependence of angular momentum of
peridynamic particles about their own deformed center of mass, with the
result (145). Now we consider the balance of angular momentum of particles
about the origin. To do this, once again take the limit as  → 0 in the global
balance of angular momentum (27), leading to
X
yj × Tij − yi × Tji + yi × Fext


Mi yi × ÿi = i . (148)
j6=i

It is easy to show that this relation holds if the forces are nonpolar. To see
this, note that (147) implies
X 
Mi yi × ÿi = yi × (Tij − Tji ) + Fext
i .
j6=i

Add (146) to this result to obtain (148). Thus, if nonpolarity holds, then
the global balance of angular momentum follows from the balance of linear
momentum.
It is a standard result in textbooks that for pairwise interactions between
particles, in which the force vector Fij is parallel to the relative deformed
position vector yj − yi , the balance of angular momentum follows from the
balance of linear momentum. In the present case of more complex inter-
actions, the additional nonpolarity relation (146) is required. Of course,
this additional relation is trivially satisfied in the special case of pairwise
interactions.
By (56), (59), and (144), it follows that the Tij can be expressed in the
form
Tij = T̂ij y1 , . . . , yN ),
where the T̂ij are suitably defined functions. These functions provide a con-
stitutive model for the forces between peridynamic particles. The functions
T̂ij must satisfy the nonpolarity requirement (146).

75
7.4 Particles as a special case of a continuum
In this section we demonstrate that a collection of discrete peridynamic
particles, together with the forces between them, can be represented within
the framework of the continuum theory using generalized functions. To
do this, let xi denote the reference positions of the particles. Define a
peridynamic body and its bond force densities by
X X
ρ(x) = Mi ∆(x − xi ), b(x, t) = Fext
i (t)∆(x − xi ), (149)
i i
XX
t(x0 , x, t) = Tij (t)∆(x − xi )∆(x0 − xj ) (150)
i j6=i

where ∆ is the three dimensional Dirac delta function (which has units
of volume−1 ). To confirm that this body reproduces the accelerations for
peridynamic particles given by (147), substitute these expressions into the
equation of motion (16):
Z
t(x0 , x, t) − t(x, x0 , t) dV 0 + b(x, t)


ρ(x)ÿ(x, t) =
B

to obtain
X X
Mi ∆(x − xi )ÿ(x, t) = Fext
i (t)∆(x − xi )
i i
Z XX
Tij (t) ∆(x − xi )∆(x0 − xj ) − ∆(x0 − xi )∆(x − xj ) dVx0 .


+
B i j6=i

Taking x to be any of the particle reference positions, and carrying out the
integration using the properties of the delta function, leads to (147).
Expressing the constitutive model (150) at xi in terms of a force state,
set
T[x, t]hξi = t(x + ξ, x, t)
to obtain
XX
T[x, t]hξi = Tij (t)∆(x − xi )∆(ξ − (xj − xi )). (151)
i j6=i

With this constitutive model and (149), the equations for peridynamic par-
ticles are seen to be a special case of the continuum theory.

76
7.5 Multibody potentials
Consider a set of N particles with masses M1 , . . . , MN and current positions
y1 , . . . , yN . Let the potential energy of this set of particles be given by
U (y1 , . . . , yn ), where U is an N -body potential. Assume that U has the
property of translational invariance:

U (y1 + c, . . . , yN + c) = U (y1 , . . . , yN ) (152)

for any constant vector c. Also assume that U satisfies the balance of linear
momentum,
N
X ∂U
(y1 , . . . , yN ) = 0, (153)
∂yi
i=1

and the balance of angular momentum,

N
X ∂U
yi × (y1 , . . . , yN ) = 0. (154)
∂yi
i=1

These two assumptions are equivalent to requiring that each N -tuple of

particles have constant total linear and angular momentum in the absence
of any other interactions. Define arbitrary reference positions of the particles
x1 , . . . , xN independent of time, and let the particle locations be described
by the motion y(x, t), so that

yi = y(xi , t), i = 1, . . . , N. (155)

Let x0 be an arbitrary fixed point called the reference point for this N -
tuple of particles. This reference point may or may not coincide with the
reference positions of any of the particles. Define an elastic nonhomogeneous
peridynamic body by the following mass density and body force density:
N
X N
X
ρ(x) = Mi ∆(x − xi ), b(x, t) = Fext
i (t)∆(x − xi ) (156)
i=1 i=1

and by the following strain energy density function:

Ŵ (Y, x) = ∆(x − x0 )U (y1 , . . . , yN ) (157)

where ∆ is the three dimensional Dirac delta function. To put (157) in a

form whose right hand side depends explicitly on Y, recall the definition of
the deformation state (55), and use the abbreviated notation

y = y(x, t), Y = Y[x, t]. (158)

77
Then, for any x,

yi = y + Yhxi − xi, i = 1, . . . , N.

Now we can rewrite Ŵ in terms of the deformation state:

Ŵ (Y, x) = ∆(x − x0 )U (y1 , . . . , yN )

= ∆(x − x0 )U (y + Yhx1 − xi, . . . , y + YhxN − xi)
= ∆(x − x0 )U (Yhx1 − xi, . . . , YhxN − xi). (159)

In the last step of (159), the translational invariance of U was used as stated
in (152). Using (75), it is easily confirmed that the force state corresponding
to this Ŵ is given by
N
X ∂U
T̂(Y, x)hξi = ∆(x − x0 ) ∆(ξ − (xi − x)).
∂yi
i=1

It can also be confirmed directly that with ρ and b supplied by (156), the
equation of motion (16), evaluated at any x = xi , implies
∂U
Mi ÿi = − + Fext
i , i = 1, . . . , N (160)
∂yi
which shows that the force on particle i due to interactions with other mem-
bers of the N -tuple is −∂U/∂yi .
In subsequent discussion, it will be necessary to have the peridynamic
description of the N -tuple of particles in the form (151). This can be ac-
complished by treating the reference point for the N -tuple, x0 , as a particle
with zero mass. Define
∂U
M0 = 0, T0j = , j = 1, . . . , N (161)
∂yj
and
T00 = Tj0 = Tij = 0, i = 1, . . . , N, j = 1, . . . , N. (162)
With these definitions, the equation of motion for peridynamic particles
(147) implies (160). It also implies the additional relation
N
X ∂U
M0 ÿ0 = =0
∂yj
j=1

which vanishes because of (153). With the definitions (161) and (162),
the requirement for nonpolarity (146) is immediately seen to be implied by
(154).
Several results have been obtained:

78
 • This N -tuple of particles interacting through the multibody potential
U can be described exactly as a peridynamic state-based constitutive
model with a strain energy density function Ŵ (·, x) that is nonzero at
a single arbitrary point x = x0 .

• The bond force densities in the force state at this x0 involve only the
partial derivatives of U .

• These bond force densities generate the correct expression for Newton’s
second law for the particles, (160).

• We did not need to identify the force that each particle exerts on the
other, since the forces T0j only involve the gradient of the N -body
potential.

• Nonpolarity of the bond force densities is necessarily satisfied for any

admissible U .

The primary limitation of the method presented here for treating molec-
ular dynamics through the peridynamic equations is that the N -tuples that
interact through the multibody potentials are defined in the reference config-
uration, so it is assumed that these sets of particles do not change over time.
This would be a good approximation for solids, but not fluids, in which the
sets of particles that interact would evolve over time. However, it may be
possible to extend the peridynamic theory to an Eulerian framework, which
would avoid this problem.

7.6 Peridynamic stress due to two discrete particles

Consider two distinct particles i and j with reference positions xi and xj .
Let the force state field be given by

T[x]hξi = Tij ∆(x − xi )∆(ξ − (xj − xi )). (163)

One or both of these points may be a zero-mass reference point for a multi-
body potential, as discussed in the previous section, or they may both have
positive mass. Recall from (120) that the peridynamic stress tensor field is
given by
Z Z ∞Z ∞
ν(x) = (y + z)2 T[x − zm]h(y + z)mi ⊗ m dz dy dΩm (164)
U 0 0

where U is the unit sphere and dΩm is a differential solid angle in the direc-
tion of the unit vector m. Let L denote the open line segment connecting xi
and xj . Comparing the last two equations, evidently the integrand in (164)

79
takes on nonzero values only if x ∈ L and only if m = ±mij , where mij is
the unit vector defined by
xj − x i
mij = (165)
|xj − xi |

(see Figure 10). From this observation and the form of the integrand in
(164), it follows that ν can be expressed in the form

ν(x) = S(x) ⊗ mij (166)

where S is a vector-valued function that takes on non-null values only on L.

Recall from the properties of the peridynamic stress tensor (123) that
Z


div ν(x) = T[x]hξi − T[x + ξ]h − ξi dVξ .
H

Applying this to (163) leads to

div ν(x) = ∆(x − xi ) − ∆(x − xj ) Tij . (167)

Let R be a sphere centered at xi with radius r < |xj −xi | (so that its bound-
ary ∂R intersects L). By (167) and the properties of the delta function,
Z
div ν(x) dV = Tij .
R

From this, (166), and the divergence theorem,

Z Z
Tij = ν(x)n dA = (S(x) ⊗ mij )n dA
∂R ∂R

where n is the outward-directed unit normal vector to ∂R. But since S is

non-null only on L, it follows that n = mij there. Therefore,
Z  Z
Tij = S(x) dA mij · mij = S(x) dA
∂R ∂R

since mij is a unit vector. Since this must hold for every choice of r such
that 0 < r < |xj − xi |, it follows from (166) that
Z
ν(x) dA = Tij ⊗ mij (168)
P

for every plane P normal to L that intersects L. Thus, ν has the structure
of a two-dimensional Dirac delta function. Another way to state this is as
follows:

80
 Figure 10: Interacting particles at xi and xj .

• ν = 0 on R3 − L.

• For any function γ on R3 , using (168),

Z Z Z Z
γν dV = γν dA ds = Tij ⊗ mij γ ds (169)
R3 L Ps L

where s is path length along L, and Ps is the plane normal to L that

intersects L at s. This relation will be used in the next section in
computing the average stress among many particles.

7.7 Average stress due to many discrete particles

Now consider a system of many particles. These may include zero-mass
reference points for multibody potentials (Figure 11), as discussed in Sec-
tion 7.5. From the results of the previous section, the peridynamic stress
tensor field ν is non-null only on the line segments Lij connecting pairs
of particles that interact with each other. (This is unrelated to whether
the particles interact through a pair potential or a multibody potential.) To
make this stress more useful, an averaging function is now introduced. First,
the mechanical significance of an averaged stress field is investigated.

81
Figure 11: Peridynamic particles interacting through a 5-body potential
with reference point x0 .

82
 Let φ be an averaging function on R3 such that
R
φ = 1. A typical choice
of φ would be
φ(q) = c exp(−|q|2 /a2 ) (170)
where a and c are constants. Let ν be the peridynamic stress tensor field.
Recall the local equation of motion in terms of the stress tensor (122):

ρ(x)ÿ(x, t) = div ν(x, t) + b(x, t). (171)

Evaluate this equation at x + q, multiply both sides by φ(q) and integrate

over R3 :
Z
φ(q)ρ(x + q)ÿ(x + q, t) dVq =
Z Z
φ(q)div ν(x + q, t) dVq + φ(q)b(x + q, t) dVq .

Define the following averaged quantities:

Z
ρ̄(x) = φ(q)ρ(x + q) dVq ,
Z
1
ȳ(x, t) = φ(q)ρ(x + q)y(x + q, t) dVq ,
ρ̄(x)
Z
ν̄(x, t) = φ(q)ν(x + q, t) dVq ,
Z
b̄(x, t) = φ(q)b(x + q, t) dVq . (172)

Note that ȳ is a mass-weighted average deformed position vector. In terms

of these quantities, (171) becomes

¨ (x, t) = div ν̄(x, t) + b̄(x, t).

ρ̄(x)ȳ

This is the peridynamic equation of motion in terms of the averaged quan-

tities.
If the distance between interacting particles, i.e., the horizon, is small
compared to the length scale of the averaging function (such as a in (170)),
then it is a good approximation to assume that φ is constant along any of
the bonds. Neglecting the resulting error term, from (169) and (172), for
any pair of particles i and j,


ν̄(x) = (Tij ⊗ mij ) |xj − xi |φ(xi − x)

Using (165), this can also be written as

ν̄(x) = φ(xi − x)Tij ⊗ (xj − xi ).

83
For the system with many particles, this becomes
XX
ν̄(x) = φ(xi − x)Tij ⊗ (xj − xi ). (173)
i j6=i

Since φ has dimensions of 1/volume and Tij has dimensions of force, ν̄

has dimensions of force/area. By (149) and the first of (172), the averaged
density for many particles is given by
X
ρ̄(x) = Mi φ(xi − x).
i

If ν̄ is evaluated in the reference configuration, in which yi = xi , then the

requirement (146) for nonpolarity implies that ν̄ is symmetric. To confirm
this, let a be any vector, and note that (146) implies
X X
φi ξij × Tij = 0, ξij = xj − xi , φi = φ(xi − x).
i j6=i

Then, using the BAC-CAB rule and (173),

X X
0= φi (ξij × Tij ) × a
i j6=i
X X
= φi a × (Tij × ξij )
i j6=i
X X 
= φi Tij (a · ξij ) − ξij (a · Tij )
i j6=i
X X 
= φi (Tij ⊗ ξij )a − (ξij ⊗ Tij )a
i j6=i
 
X X
= 2 skw  φi (Tij ⊗ ξij ) a
i j6=i
 
= 2 skw ν̄(x) a

where “skw” means the skew-symmetric part of a tensor. Since this must
hold for any a, it follows that skw ν̄(x) = 0; hence ν̄ is symmetric.
Thus, nonpolarity of forces implies the symmetry of the averaged peri-
dynamic stress tensor, prior to significant deformation of the system. After
deformation, the stress tensor is no longer symmetric except in special cases.
This is comparable to the asymmetry of the classical Piola stress tensor in
a body undergoing large deformation.
Our expression for ν̄ in (173) performs the averaging in the reference
configuration, because the reference position vector x is the spatial variable.

84
Because of this, the weighting for a particle does not change as the particle
moves. This means that if the particles are highly mobile, as in a gas,
the expression (173) for stress is not very useful, because its status as an
observable quantity depends on the assumption that the particles always
remain close together. In this case, it is more useful to perform the averaging
in the deformed configuration, so that the weighting of each particle varies
as it moves closer to or farther from an observation point. To derive the
effect of this change on the averaged stress, define the momentum density
at any point x in space by
X
p̄ = Mi φi ẏi , φi = φ(yi − x).
i

Define a function Φ by
Φ(z, x) = φ(z − x),
thus X
p̄ = Mi Φ(yi , x)ẏi .
i
Observe that
∂Φ ∂Φ
=− . (174)
∂z ∂x
Evaluating the time derivative of p̄ holding x fixed, while using (174) and
temporarily neglecting body forces, leads to
X h ∂Φ
i
p̄˙ = Mi φi ÿi + (ẏi · (yi , x) ẏi
∂z
i
X h ∂Φ
i
= Mi φi ÿi − (ẏi · (yi , x) ẏi
∂x
i
X h
i
= Mi φi ÿi − div Φ(yi , x)ẏi ⊗ ẏi
i
X h
i
= Mi φi ÿi − div φi ẏi ⊗ ẏi
i
X X
= Mi φi ÿi − div Mi φi ẏi ⊗ ẏi .
i i

We are free to choose any reference configuration that is convenient. In this

case, choose it to be the configuration at any time t. Then the peridynamic
stress tensor is given by (173) with xi = yi , so that
p̄˙ = div ν̄ + div κ + b̄
where the kinetic stress tensor is defined by
X
κ=− Mi φi ẏi ⊗ ẏi .
i

85
The kinetic stress tensor accounts for the transport of momentum due to
the motion of particles into or out of a region that is fixed in space. In
contrast, the peridynamic stress tensor accounts only for the acceleration
of particles. Kinetic stress is the primary contributor to pressure in gases,
in which interactions between particles are weak but velocities, including
thermal velocities, are significant. The idea of kinetic stress is not new and
not particular to the peridynamic model, but it is included here to show
that it can be included in the peridynamic concept of momentum balance
in a straightforward way.
The averaged peridynamic stress given by (173) is similar to Hardy’s
expression for the potential contribution to the stress tensor [37], which is
obtained by averaging a large number of particles interacting through pair
potentials. Hardy also discusses the relation of this stress tensor to the virial
stress. The present approach applies to multibody potentials, while ensuring
that the average stress satisfies balance of angular momentum (in the sense
of producing symmetric ν̄ in the reference configuration, as shown above).

86
8 Damage and fracture
This section presents the peridynamic view of damage and its incorporation
into a material model. This is presented within a thermodynamic framework
that highlights the role of irreversibility of damage. Various damage evolu-
tion laws and their implications are described, the simplest being indepen-
dent bond breakage. The energy balance for moving defects is investigated,
leading to peridynamic expressions for the J-integral, surface energy, and
the Griffith criterion for crack growth.

8.1 Damage as part of a constitutive model

Suppose that a material has a free energy function ψ and entropy function
η that depend not only on the deformation state and temperature, but also
on a scalar state φ called the damage state. We write

ψ(Y, θ, φ), η(Y, θ, φ). (175)

The damage state is special in that it cannot decrease over time, thus

φ̇ ≥ 0, i.e., φ̇hξi ≥ 0 ∀ ξ ∈ H. (176)

It is also assumed that

0 ≤ φ ≤ 1, i.e., 0 ≤ φhξi ≤ 1 ∀ ξ ∈ H. (177)

The damage state is determined by the deformation and by other variables

through a prescribed damage evolution law of the form

φ = D(Y, Ẏ, . . . )

or alternatively in terms of the rate of damage growth:

φ̇ = Ḋ(Y, Ẏ, . . . ).

A material model such that, for any ξ ∈ H,

φhξi = 1 =⇒ Thξi = 0 (178)

is said to have strong damage dependence. All other material models have
weak damage dependence.

87
8.2 Irreversibility of damage growth
Recall the inequality (69) derived from the first and second laws in terms of
free energy:
T • Ẏ − θ̇η − ψ̇ ≥ 0.
We now repeat the Coleman and Noll method [15] used previously in Sec-
tion 4.5 to obtain restrictions on the constitutive response in the presence
of evolving damage. Differentiating the first of (175) with respect to time
yields
ψ̇ = ψY • Ẏ + ψθ θ̇ + ψφ • φ̇, (179)
where ψY and ψφ are the Fréchet derivatives of ψ with respect to Y and φ
respectively, and ψθ = ∂ψ/∂θ. Combining the last two expressions yields
   
T − ψY • Ẏ − η + ψθ θ̇ − ψφ • φ̇ ≥ 0. (180)

As in Section 4.5, assume that Ẏ and θ̇ can be varied independently and

arbitrarily in (180), hence

T = ψY , η = −ψθ . (181)

In view of (176), φ̇ cannot be set arbitrarily in (180), leading to the conclu-

sion
ψφ ≤ 0 , (182)
which is the second law restriction on the dependence of free energy on the
damage. Using (181), (179) takes the form

ψ̇ = T • Ẏ − θ̇η + ψφ • φ̇. (183)

Now assume an adiabatic process, so that h = r = 0. Subtracting (68),

which continues to hold in the presence of damage, from (183) leads to the
conclusion
ψ̇ d
η̇ = , ψ̇ d := −ψφ • φ̇ (184)
θ
where ψ̇ d is the rate of energy dissipation. The first of (184) gives the entropy
production due to damage evolution.

8.3 Bond breakage

Recall the definition of the scalar extension state e,

e = |Y| − |X| (185)

88
where X is the identity state defined in (49). A useful example of a damage
evolution law is given by the following model:

φhξi = D(Y, ξ)hξi = max f (ehξi, ξ) (186)

t

where f (e, ξ) is a nondecreasing function of e, 0 ≤ f ≤ 1, and the maximum

is taken over all times up to t. Assume for simplicity that for a given motion,
e is a nondecreasing function of time. Observing from (185) that

Y
ė = M · Y, M= , (187)
|Y|

differentiating (186) with respect to time yields an equivalent damage evo-

lution law in terms of the rate:

φ̇hξi = Ḋ(Y, Ẏ, ξ) = fe (ehξi, ξ)Mhξi · Ẏhξi (188)

where fe denotes the partial derivative of f with respect to e. It is helpful

to introduce a vector state r defined by

rhξi = fe (ehξi, ξ)Mhξi (189)

so that (188) can be written as

φ̇ = r · Ẏ (190)

provided e is nondecreasing. A specific case of such a damage model is bond

breakage in tension, in which


f (e, ξ) = H e − eb (ξ) (191)

where H is the Heaviside step function and eb (ξ) is the prescribed bond
breakage extension for the bond ξ. In this case


fe (e, ξ) = ∆ e − eb (ξ)

and from (188),

φ̇ = ∆(e − eb )M · Ẏ
Alternatively, the same damage evolution law can be defined in terms of the
rate through (190) with
r = ∆(e − eb )M.
Bond breakage in compression can be treated in a similar way.

89
8.4 Ordinary material models with strong damage depen-
dence
Suppose an elastic material model is defined by

Ŵ (Y) = W 0 (e) (192)

where W 0 is a function and e is the scalar extension state, defined by (185).

Because of (187), the chain rule implies

∇Ŵ (Y) = ∇W 0 (e)M.

Now we will modify the material model (192) to include damage. To do

this, define a free energy density function by

ψ(Y, φ) = W 0 (1 − φ)e .


(193)

The first of (181) and the chain rule provide the following force state:

T = ψY = (1 − φ)t0 M, t0 = ∇W 0 (1 − φ)e .


(194)

Because of the (1 − φ) term in this expression for T, evidently (178) holds

for this material, so it has strong damage dependence. By (193),

ψφ = −t0 e. (195)

So, the second law requirement (182) holds provided

t0 e ≥ 0. (196)

This asserts that the scalar bond force density in each bond has the same
sign as the bond’s scalar extension.

8.5 Bond-based constitutive models with bond breakage

An important special case of the ordinary material with strong damage
dependence is obtained if the bond breakage model for evolution of damage
is combined with a bond-based constitutive model. Following (78), choose
Z
0
W (e) = w(ehξi, ξ) dVξ
H

where w(e, ξ) is the differentiable bond potential function. Let the damage
evolution be described by the bond breakage model in (191). Following the
method of the previous section, modify W 0 to include damage by defining
the following free energy function:
Z
0



ψ(Y, φ) = W (1 − φ)e = w (1 − φhξi)ehξi, ξ dVξ . (197)
H

90
From (194), the force state is given by

T = (1 − φ)t0 M, t0 hξi = we (1 − φhξi)ehξi, ξ .

The Fréchet derivative of ψ with respect to φ is supplied by (195), thus

ψφ hξi = −ehξiwe (1 − φhξi)ehξi, ξ (198)

for all ξ ∈ H. The second law condition is given by (196). The important
distinction between this material and more general ordinary materials is
that here, t0 hξi and φhξi for a given bond ξ are determined independently
of whatever happens in all the other bonds.
It is of interest to compute the dissipated energy at a point x up to time
t. To do this, use the second of (184) and (198), assuming for simplicity
that e is nondecreasing:
Z t Z tZ
ψd = −


ψφ • φ̇ dt = ehξiwe (1 − φhξi)ehξi, ξ φ̇hξi dVξ dt.
0 0 H

Recalling that, by definition, ψφ refers to the derivative of ψ holding Y, and

therefore e, constant, it follows that
Z Z φhξi
d d

ψ =− w (1 − σ)ehξi, ξ dσ dVξ .
H 0 dσ

Combining this with (186) and (191), because φhξi changes discontinuously
when the extension reaches the bond breakage extension eb (ξ), it follows
that Z
d
ψ = wb (ξ)φhξi dVξ
H
where the bond breakage energy is defined by

wb (ξ) = w(eb (ξ), ξ) − w(0, ξ).

In this material, the dissipated energy is simply the integral of the bond
breakage energies over all the broken bonds in the family.
In this material, ψ d depends only on the current value of φ. This is not
true for most other materials; a counterexample is the separable damage
model discussed in the next section.

8.6 Separable damage models

The previous section showed how damage can be introduced into a bond-
based material model simply by modifying the bond potential function with
a term that depends on bond damage, as indicated in (197). The present

91
section deals with incorporation of damage into more general elastic consi-
titutive models. Assume that such an “undamaged” model is provided, and
let W 0 (Y) be its strain energy density function. Let T0 = ∇W 0 . Further
assume that

W 0 (X) = 0 and W 0 (Y) ≥ 0 ∀ Y ∈ V. (199)

Define a free energy density function by

ψ(Y, φ) = Φ(φ)W 0 (Y) (200)

where Z Z
1
Φ(φ) = (1 − φhξi)2 dVξ , V = dVξ . (201)
V H H
Because the damage state and the deformation state appear in separate
terms in (200), this type of damage model will be called separable. After
evaluating the Fréchet derivatives and using the first of (181), the force state
including damage is found to be

Thξi = ψY hξi = Φ(φ)T0 hξi. (202)

This means that all the bonds have their force reduced by the same scalar
Φ(φ). Φ decreases monotonically with time because, by the assumption
(176), φ increases monotonically. From (200) and (201), the Fréchet deriva-
tive of free energy with respect to φ is given by

2W 0 (Y)
ψφ hξi = − (1 − φhξi). (203)
V
The second law requirement (182) is necessarily satisfied because of (177),
(199), and (203). The energy dissipation rate is found from (184) and (203)
to be
2W 0 (Y)
Z
d
ψ̇ = −ψφ • φ̇ = (1 − φhξi)φ̇hξi dVξ
V H
= −W 0 (Y)Φ̇.

The dissipated energy up to time t is therefore given by

Z t Z t
ψ d (t) = ψ̇ d (τ ) dτ = − W 0 (Y[τ ])Φ̇(τ ) dτ.
0 0

The separable damage model results in a material characterization with weak

dependence on damage (see Section 8.1), because (202) does not necessarily
imply that bonds with φhξi = 1 have Thξi = 0. This is in contrast to the
materials discussed in Section 8.4, which exhibit strong damage dependence.

92
Figure 12: Subregion Pt containing points where there is energy dissipation
moves to the right with velocity V.

8.7 Energy balance in progressive damage

Consider a closed, bounded subregion Pt with constant shape that translates
through the reference configuration B with velocity V; thus there is a flux
of material through its boundary ∂Pt . Assume a steady-state motion of the
form
y(x, t) = x + u(x − Vt) (204)
where u is some differentiable function (Figure 12).
Assume that body force, kinetic energy, heat transport, and heat sources
are all negligible. Additionally assume an isothermal process, so that θ̇ = 0.
These assumptions along with the local first law expression (65) and (183)
imply
ψ̇ = ε̇ + ψφ • φ̇. (205)
Recalling the shorthand notation in (12), the global first law (35) in this
case has the form
Z Z Z
ε̇ dV = (t · ẏ0 − t0 · ẏ) dV 0 dV . (206)
Pt Pt B\Pt

Because the motion is steady state, the total time derivative of any inten-
sive quantity over Pt vanishes. The Reynolds transport theorem therefore

93
implies, for the free energy ψ,
Z Z Z
d
ψ dV = ψ̇ dV + ψV · n dA = 0
dt Pt Pt ∂Pt

where n is the outward-directed unit normal to ∂Pt . Using this and (205),
Z Z


ε̇ + ψφ • φ̇ dV + ψV · n dA = 0.
Pt ∂Pt

From this and (206),

Z Z Z Z
(t · ẏ0 − t0 · ẏ) dV 0 dV + ψφ • φ̇ dV + ψV · n dA = 0.
Pt B\Pt Pt ∂Pt

Evaluating ẏ and ẏ0 using (204), therefore

Z Z
t0 · (ux V) − t · (u0x V) dV 0 dV


P B\P
Z Z
+ ψφ • φ̇ dV + ψV · n dA = 0 (207)
P ∂P

where the t subscript has been dropped from Pt and

ux = grad u(x − Vt), u0x = grad u(x0 − Vt).

(Note that (207) has the same structure as the master balance law discussed
in Section 2.5.) Using the second of (184), this result can be expressed in
the form Z
J·V = ψ̇ d dV , ψ̇ d = −ψφ • φ̇ (208)
P
where Z Z Z
uTx t0 (u0x )T t 0


J= − dV dV + ψn dA. (209)
P B\P ∂P

This provides the peridynamic equivalent of the J-integral in the standard

theory [57]. (208) and (209) relate the free energy lost in some dissipa-
tive process to quantities along the surface of a subregion P that contains
the material where the dissipation is occurring. This dissipative region can
be much smaller than P. The required quantities on the surface ∂P can
be evaluated if the deformation is known near this surface. Thus, we can
“measure” the dissipation based on these far-field quantities without know-
ing the details of what happens in the dissipative region. Recall that (176)
and (182), which are consequences of the second law, together with (208),
ensure that
J·V ≥0

94
for any V; in other words the rate of energy dissipation is always nonnega-
tive.
Although it was assumed that u is differentiable (with respect to all
spatial coordinates), it is only the directional derivative ux V in the direction
of motion that is used. Therefore, it is permissible to have discontinuities
parallel to the direction of propagation, as would be the case with a crack. As
in the classical development of the J-integral [57], the assumption of a steady
motion excludes curved or oscillatory cracks. However, the peridynamic
method can nevertheless be applied in these cases, as demonstrated by the
numerical studies summarized in Section 1.2.
Evidently, if there is no dissipation within the closed surface ∂P, then
J = 0. So, P can be deformed to include any amount of additional material
in which there is no dissipation occurring without changing the value of J.
In this sense, J is “path independent.”

8.8 Relation to the Griffith criterion

A good approximation in many solids is to assume that a crack will grow
if a definite amount of energy G, called the critical energy release rate, is
available to it per unit area of crack advance. The critical energy release
rate is often thought of as being consumed in separating atoms to create
new surface area, but it can include other processes as well, such as plastic
work in the vicinity of the crack tip. Plasticity in peridynamic materials,
although not treated in the present article, is discussed in [67]. Like the
rate dependence discussed in Section 4.5 (see (73)), plasticity represents a
mechanism whose energy dissipation rate can be included through terms
similar to ψ̇ d discussed here for damage evolution.
The Griffith concept of crack growth can be related to the peridynamic
model as follows. In (208), suppose φ̇ scales with V. In other words, assume
that the damage model is such that there exists a vector state valued function
r(φ) such that
φ̇ = r • Ẏ.
(189) is an example of such an r. From (204),

Ẏ[x]hx0 − xi = −(u0x − ux )V.

Combining the last two equations,

φ̇ = −(G • r) · V (210)

where the double state field G is defined by

G[x]hξ, ζi = uTx (x + ξ) − uTx (x) ∆(ζ − ξ).

95
for all bonds ξ, ζ ∈ H. From (208) and (210),
Z 
J·V = ψφ • G • r dV · V
P

from which it follows that

Z
J= ψφ • G • r dV . (211)
P

If the direction of propagation is parallel to a unit vector a, then J · a is the

energy dissipation (with units of energy/length). If the body is a plate of
thickness β containing a crack through its thickness, and if P is a cylinder
through the thickness containing the crack tip, then the energy dissipated
per unit crack area is
J·a
Ga = .
β
Under the assumptions of the present analysis, we have arrived at the Griffith
model for crack growth: energy is dissipated at a constant rate per unit crack
area, independent of time and propagation speed. Of course, the integrand
in (211) depends on all the details of the deformation and the material
model, including the damage model.

8.9 Surface energy

The results of the previous section show that, under certain conditions, a
crack growing in a peridynamic solid consumes a fixed amount of energy
per unit area of crack growth. This energy can be computed, in a numerical
model of a growing crack, either by evaluating J directly from (211) or using
the expression (209) that was derived from an energy balance on a moving
subregion containing the crack tip. By carrying out this calculation for
different choices of parameters in the damage evolution law D or Ḋ, these
parameters can be calibrated to experimental data on critical energy release
rate.
In this section, a simpler but approximate procedure is presented for
accomplishing the same thing. This procedure assumes that the energy
consumed by a growing crack equals the work required, per unit crack area,
to separate two halves of a body across a plane (Figure 13). Suppose a
plane A separates a large homogeneous body B into two subsets B+ and
B− , where B− includes A. Consider a small patch on A with area a. Let
P be the cylinder normal to A, extending infinitely in both directions from
it, whose cross-section is this small patch. Also define P+ = P ∩ B+ and
P− = P ∩ B− .

96
 Consider a motion with velocity field v given by

c/2 on B+
v=
−c/2 on B−
where c is a constant vector. Now compute the total energy E absorbed by
P in this motion. By (34),
Z t Z ∞Z Z
0
E= Wabs (P) dt = t · (v0 − v) dV 0 dV dt
0 0 P B

where t = t(x0 , x, t), v = v(x), and v0 = v(x0 ). In this discussion, t serves

as a convenient parameter although dynamics is not considered. Because
the body is homogeneous, the energies absorbed by P+ and by P− must be
equal. Also, bonds that do not cross A do not contribute to the integrand,
since v = v0 for these bonds. With these simplifications,
Z ∞Z Z
E=2 t · c dV 0 dV dt, (212)
0 P− B+

in which the factor of 2 appears because the integral over P+ is not included
explicitly. Assume that the material is characterized by a free energy density
function with damage. By (181),

t = t(x0 , x, t) = ψY (Y, φ)hx0 − xi.

Following [26], assume that the material model is such that each bond con-
sumes a prescribed amount of work (per unit volume squared) w0 as the two
parts of the body are separated out to a large distance:
Z ∞
t · c dt = w0 ∀ x ∈ B− , ∀ x0 ∈ B + .
0

An example of such a material model is one in which each bond breaks at

the time τ (x0 , x) at which the net work done on it up to that time equals
w0 , thus
φ[x, t]hx0 − xi = H(t − τ (x0 , x))
where τ is defined by
Z τ (x0 ,x)
t · c dt = w0 .
0
Also, assume that in such a material model, t = 0 for t > τ . The function
τ is not known in advance, but this does not matter; it can be computed
in a numerical simulation “on the fly.” Under these assumptions about the
material, (212) becomes
Z Z
E = 2w0 φ[x, ∞]hx0 − xi dV 0 dV . (213)
P− B+

97
To evaluate this double integral, observe that φ[x, ∞]hx0 − xi = 1 for bonds
that connect points x ∈ P− to points x0 ∈ B+ . Referring to Figure 14, any
point x on the lower vertical axis with 0 ≤ z ≤ δ is connected to points x0
within the spherical cap H ∩ B+ . Working in a spherical coordinate system
with elevation angle ϕ, azimuthal angle ϑ, and radius ξ, (213) reduces to
2π δ δ cos−1 (z/ξ)
πw0 δ 4 a
Z Z Z Z
E = 2w0 a ξ 2 sin ϕ dϕ dξ dz dϑ = . (214)
0 0 z 0 2

Assuming that this surface energy equals the critical energy release rate G
for the material times the area of the patch a,

G = E/a.

From this and (214), solving for w0 yields

2G
w0 = .
πδ 4
Thus, the critical value of work on a bond for bond breakage has been
determined from the measurable quantity G. This result is independent of
the details of the constitutive model.

9 Discussion
The development of the peridynamic theory presented above has emphasized
the unifying aspect of the theory: the same field equations can be applied
directly to traditional continua, to continua with emerging and propagating
defects, and to discrete particles. Does such a unified treatment have any
advantages over standard methods? One possible benefit is that since all
these regimes satisfy the same field equations, it may not be necessary to
devise coupling methods to connect disparate mathematical systems. For
example, communication between an atomistic and a continuous region does
not require coupling between a set of ODEs describing the particles and
PDEs describing the continuum, since both regimes obey the same integro-
differential equations in peridynamics. Development of such an atomistic-
to-continuum coupling within the peridynamic framework is a current area
of research.
Similarly, because the same field equations within the peridynamic model
apply to points either on or off of a discontinuity, cracks and other defects
grow autonomously. Their nucleation and progression are determined by
the equation of motion and the material model, which may include damage
evolution. It is not necessary to provide a supplemental kinetic relation
that dictates the evolution of cracks, as is needed in traditional fracture

98
Figure 13: Computation of surface energy by the total work absorbed by
P− as it separates from B+ .

99
Figure 14: Computation of surface energy by the total work absorbed by
P− as it separates from B+ .

100
mechanics. Instead, cracks appear and grow spontaneously depending on
conditions.
It is worthwhile to compare the peridynamic approach to fracture against
variational approaches [28, 12, 13] in which the growth of a crack is de-
termined by energy minimization, including contributions from continuous
parts of the body and from energy consumed by the crack. Like traditional
fracture mechanics, this variational approach treats cracks as separate enti-
ties from the continuous parts of the deformation, hence the need to include
their energy consumption through separate terms in the variational state-
ment. Thus, the variational approach to fracture has fundamentally different
purposes and characteristics from the peridynamic approach, which treats
damage only through the material model and does not distinguish between
cracked and continuous parts of the body.
A question of fundamental interest is the extent to which the peridy-
namic approach to damage can qualitatively and quantitatively reproduce
the phenomena of fracture, particularly in complex materials and geome-
tries. A closely related question is how the details of the constitutive model,
including the damage evolution law, influence predicted fracture and failure
of materials. As noted in Section 1.2, there are many encouraging numerical
results available that apply peridynamic modeling to a variety of problems
in fracture and fragmentation, including dynamic fracture. In most cases,
these simulations rely on the simplest possible assumptions about material
response and damage. A more comprehensive approach is needed to learn
what insights can be gained from the peridynamic method for fracture, and
what types of material models lead to the best agreement with experimental
data.
As remarked in Section 3, the peridynamic theory uses vector states,
rather than second order tensors, as the fundamental quantities that a con-
stitutive model deals with. These states are infinite dimensional objects,
unlike tensors, which are 9-dimensional. This suggests that there may be
a potentially larger and richer environment provided by the peridynamic
theory in material modeling. This environment includes, as discussed in
the present article, the ability to treat discontinuities and long-range forces
directly. But there may also be other avenues of material modeling in the
peridynamic theory that remain to be explored. For example, it is demon-
strated in [69] that peridynamics reveals a condition for a particular type of
material instability, interpreted as the nucleation of a crack, that is not nec-
essarily well described by mathematical conditions such as loss of ellipticity
in the classical theory. It is shown in [79] that a peridynamic micromodulus
function in one dimension can be obtained from experimental measurements
of wave dispersion data. Material response within peridynamics, including
its implications for material stability and generation of defects, is an open

101
and promising area of research.
As noted in Section 7.5, the representation of a system of a discrete
particles within the peridynamic theory is at present limited by the inter-
actions are defined in the reference configuration due to the Lagrangian
nature of the method. The generality of the treatment could be improved
by developing an Eulerian version of the model that would allow changing
interactions. This would also permit a number of other applications to be
modeled, notably those involving fluids.
Future development of the peridynamic theory will include multiscale ap-
plications. It may be possible to construct a consistent multiscale method
within peridynamics, that is, a rigorously coupled set of models at different
length scales that all have the same mathematical structure. Such a multi-
scale method has been proposed in [62]. In this approach, a set of reduced,
or coarsened, degrees of freedom is chosen from a detailed linearized model
at the smallest length scale. This detailed description could represent a
linearized molecular dynamics model since, subject to the assumptions in
Section 7.4, discrete particles are a special case of a peridynamic contin-
uum. It is possible to evaluate the coarsened micromodulus function such
that the forces within the coarsened model agree with what would have been
evaluated from the full, detailed model, even though the coarsened model
excludes many of the original degrees of freedom. The resulting coarsened
model has the same mathematical structure as the detailed model, i.e., linear
peridynamics. Therefore, the coarsening process can be repeated over and
over hierarchically, leading to successively more economical computational
models.

Acknowledgments
The authors gratefully acknowledge helpful discussions with Drs. John
Aidun, Abe Askari, Etienne Emmrich, Florin Bobaru, John Foster, Max
Gunzburger, Michael Parks, Mark Sears, Pable Seleson, Olaf Weckner, and
Jifeng Xu. The preparation of this article was supported by the US Depart-
ment of Energy as part of a Laboratory Directed Research and Development
project at Sandia National Laboratories. Sandia is a multiprogram labora-
tory operated by Sandia Corporation, a Lockheed Martin Company, for the
United States Department of Energy’s National Nuclear Security Adminis-
tration under contract DE-AC04-94AL85000.

102
References
[1] A. Agwai, I. Guven, and E. Madenci. Peridynamic theory for failure
prediction in multilayer thin-film structures of electronic packages. In
Proc. 58th Electronic Components and Technology Conference (ECTC),
pages 1614–1619. IEEE, 2008.

[2] A. Agwai, I. Guven, and E. Madenci. Peridynamic theory for im-

pact damage prediction and propagation in electronic packages due to
drop. In Proc. 58th Electronic Components and Technology Conference
(ECTC), pages 1048–1053. IEEE, 2008.

[3] B. Alali and R. Lipton. Multiscale analysis of heterogeneous media

in the peridynamic formulation. Technical Report IMA Preprint Se-
ries 2241, Institute for Mathematics and Its Applications, Minneapolis,
Minnesota, USA, 2009.

[4] M. Arndt and M. Griebel. Derivation of higher order gradient con-

tinuum models from atomistic models for crystalline solids. Multiscale
Modeling and Simulation, 4:531–562, 2005.

[5] E. Askari, F. Bobaru, R. B. Lehoucq, M. L. Parks, S. A. Silling, and

O. Weckner. Peridynamics for multiscale materials modeling. Journal
of Physics: Conference Series, 125(012078), 2008.

[6] E. Askari, K. Nelson, O. Weckner, J. Xu, and S. A. Silling. The design

of a hybrid material for multifunctional performance using advanced
analysis techniques and testing. In Proc. 40th ISTC, Memphis, Ten-
nessee, USA. Society for the Advancement of Material and Process
Engineering, 2008.

[7] E. Askari, J. Xu, and S. Silling. Peridynamic analysis of damage and

failure in composites. In 44th AIAA Aerospace Sciences Meeting and
Exhibit, Reno, Nevada, number AIAA 2006-88. American Institute of
Aeronautics and Astronautics, 2006.

[8] Z. P. Bazant and G. Pijaudier-Cabot. Nonlocal continuum damage,

localization instability and convergence. Journal of Applied Mechanics,
55:287–290, 1988.

[9] F. Bobaru. Influence of van der Waals forces on increasing the strength
and toughness in dynamic fracture of nanofibre networks: a peridy-
namic approach. Modelling and Simulation in Materials Science and
Engineering, 15:397–417, 2007.

103
[10] F. Bobaru and S. A. Silling. Peridynamic 3d models of nanofiber net-
works and carbon nanotube-reinforced composites. In NUMIFORM
2004 - Proceedings of the 8th International Conference on Numerical
Methods in Industrial Forming Processes, AIP Conference Proceedings,
Vol. 712, pages 1565–1570. American Institute of Physics, 2004.

[11] F. Bobaru, M. Yang, L. F. Alves, S. A. Silling, A. Askari, and J. Xu.

Convergence, adaptive refinement, and scaling in 1d peridynamics. In-
ternational Journal for Numerical Methods in Engineering, 77:852–877,
2008.

[12] B. Bourdin, G. A. Francfort, and J.-J. Marigo. Numerical experiments

in revisited brittle fracture. Journal of the Mechanics and Physics of
Solids, 48:797–826, 2000.

[13] B. Bourdin, G. A. Francfort, and J.-J. Marigo. The variational approach

to fracture. Journal of Elasticity, 91:5–148, 2008.

[14] X. Chen and M. Gunzburger. Continuous and discontinuous finite ele-

ment methods for a peridynamics model of mechanics. Technical report,
Florida State University, 2010.

[15] B. D. Coleman and W. Noll. The thermodynamics of elastic materials

with heat conduction and viscosity. Archive for Rational Mechanics
and Analysis, 13:167–178, 1963.

[16] K. Dayal and K. Bhattacharya. Kinetics of phase transformations in

the peridynamic formulation of continuum mechanics. Journal of the
Mechanics and Physics of Solids, 54:1811–1842, 2006.

[17] P. N. Demmie and S. A. Silling. An approach to modeling extreme load-

ing of structures using peridynamics. Journal of Mechanics of Materials
and Structures, 2:1921–1945, 2007.

[18] Q. Du and K. Zhou. Mathematical analysis for the peridynamic nonlo-

cal continuum theory. Mathematical Modelling and Numerical Analysis,
To appear, 2010.

[19] Q. Du and K. Zhou. Mathematical and numerical analysis of linear peri-

dynamic models with nonlocal boundary conditions. Preprint, Penn-
sylvania State University, 2010.

[20] E. Emmrich and O. Weckner. Analysis and numerical approximation

of an integro-differential equation modeling non-local effects in linear
elasticity. Mathematics and Mechanics of Solids, 12:363–384, 2007.

104
[21] E. Emmrich and O. Weckner. On the well-posedness of the linear peri-
dynamic model and its convergence towards the Navier equation of lin-
ear elasticity. Communications in Mathematical Sciences, 5:851–864,
2007.

[22] E. Emmrich and O. Weckner. The peridynamic equation and its spatial
discretisation. Mathematical Modelling and Analysis, 12, 2007.

[23] A. C. Eringen. On differential equations of nonlocal elasticity and solu-

tions of screw dislocation and surface waves. Journal of Applied Physics,
54:4703–4710, 1983.

[24] A. C. Eringen. Nonlocal continuum field theories. Springer-Verlag

NewYork, Inc, 2002.

[25] A. C. Eringen, C. G. Speziale, and B. S. Kim. Crack-tip problem in

non-local elasticity. Journal of the Mechanics and Physics of Solids,
25:339–355, 1977.

[26] J. T. Foster. Dynamic crack initiation toughness: experiments and

peridynamic modeling. Technical Report Ph.D. dissertation, Purdue
University, reprinted in SAND2009-7217, Sandia National Laboratories,
Albuquerque, New Mexico, 2009.

[27] J. T. Foster, S. A. Silling, and W. W. Chen. Viscoplasticity using

peridynamics. International Journal for Numerical Methods In Engi-
neering, 81:1242–1258, 2010.

[28] G. A. Francfort and J.-J. Marigo. Revisiting brittle fracture as an

energy minimization problem. Journal of the Mechanics and Physics
of Solids, 46:1319–1342, 1998.

[29] E. Fried. New insights into classical particle mechanics. Discrete and
Continuous Dynamical Systems, to appear.

[30] W. Gerstle, N. Sau, and S. A. Silling. Peridynamic modeling of concrete

structures. Nuclear Engineering and Design, 237:1250–1258, 2007.

[31] W. Gerstle, S. Silling, D. Read, V. Tewary, and R. Lehoucq. Peri-

dynamic simulation of electromigration. Computers, Materials, and
Continua, 8:75–92, 2008.

[32] W. H. Gerstle, N. Sau, and E. Aguilera. Micropolar peridynamic con-

stitutive model for concrete. In 19th International Conference on Struc-
tural Mechanics in Reactor Technology (SMiRT 19), Toronto, Canada,
number SMIRT19-B02-1, pages 1–8, 2007.

105
[33] W. H. Gerstle, N. Sau, and N. Sakhavand. On peridynamic computa-
tional simulation of concrete structures. Technical Report SP265-11,
American Concrete Institute, 2009.
[34] W. H. Gerstle, N. Sau, and S. A. Silling. Peridynamic modeling of plain
and reinforced concrete structures. In 18th International Conference
on Structural Mechanics in Reactor Technology (SMiRT 18), Beijing,
China, number SMIRT18-B01-2, pages 54–68, 2005.
[35] M. Gunzburger and R. B. Lehoucq. A nonlocal vector calculus with
application to nonlocal boundary value problems. Technical Report
SAND2009-4666J, Sandia National Laboratories, Albuquerque, New
Mexico 87185 and Livermore, California 94550, February 2010.
[36] M. E. Gurtin and W. O. Williams. On the first law of thermodynamics.
Arch. Rat. Mech. Anal., 42:77–92, 1971.
[37] R. J. Hardy. Formulas for determining local properties in molecular-
dynamics simulations: Shock waves. Journal of Chemical Physics,
76:622–628, 1982.
[38] J. P. Hirth and J. Lothe. Theory of Dislocations, 2nd edition, pages
63–64. Wiley, New York, 1982.
[39] J. Irving and J. Kirkwood. The statistical mechanical theory of trans-
port processes. IV. the equations of hydrodynamics. J. Chem. Phys.,
18:817–829, 1950.
[40] J. F. Kalthoff and S. Winkler. Failure mode transition at high rates
of shear loading. In C. Y. Chiem, H. D. Kunze, and L. W. Meyer,
editors, Impact Loading and Dynamic Behavior of Materials, Vol. 1,
pages 185–195. DGM Informationsgesellschaft Verlag, 1988.
[41] K.-I. Kanatani. A micropolar continuum theory for the flow of granular
materials. International Journal of Engineering Science, 17:419–432,
1979.
[42] B. Kilic, A. Agwai, and E. Madenci. Peridynamic theory for progressive
damage prediction in center-cracked composite laminates. Composite
Structures, 90:141–151, 2009.
[43] B. Kilic and E. Madenci. Prediction of crack paths in a quenched glass
plate by using peridynamic theory. International Journal of Fracture,
156:165–177, 2009.
[44] B. Kilic and E. Madenci. IEEE Transactions on Advanced Packaging,
33:97–105, 2010.

106
[45] E. Kröner. Elasticity theory of materials with long range cohesive forces.
International Journal of Solids and Structures, 3:731–742, 1967.

[46] I. A. Kunin. Elastic Media with Microstructure I: One-Dimensional

Models. Springer: Berlin, 1982.

[47] I. A. Kunin. Elastic Media with Microstructure II: Three-Dimensional

Models. Springer: Berlin, 1983.

[48] R. B. Lehoucq and S. A. Silling. Force flux and the peridynamic stress
tensor. Journal of the Mechanics and Physics of Solids, 56:1566–1577,
2008.

[49] R. B. Lehoucq and O. A. von Lilienfeld. “Translation of Walter Noll’s

derivation of the fundamental equations of continuum thermodynamics
from statistical mechanics”. Journal of Elasticity, 2010.

[50] R. W. Macek and S. A. Silling. Peridynamics via finite element analysis.

Finite Elements in Analysis and Design, 43:1169–1178, 2007.

[51] E. Madenci, E. Oterkus, and A. Barut. Peridynamic theory based on

variational principle. Presentation, SIAM Conference on Mathematical
Aspects of Materials Science, Philadelphia, May, 2010.

[52] R. Maranganti and P. Sharma. Length scales at which classical elasticity

breaks down for various materials. Physical Review Letters, 98:195504,
2007.

[53] W. Noll. Die Herleitung der Grundgleichungen der Thermomechanik

der Kontinua aus der statistischen Mechanik. Journal of Rational Me-
chanics and Analysis, 4:627–646, 1955. In German, English translation
available [49].

[54] W. Noll. Thoughts on the concept of stress. Journal of Elasticity, 2010.

In press.

[55] J. Ozbolt and Z. P. Bazant. Numerical smeared fracture analysis: non-

local microcrack interaction approach. International Journal for Nu-
merical Methods in Engineering, 39:635–661, 1996.

[56] M. L. Parks, R. B. Lehoucq, S. J. Plimpton, and S. A. Silling. Imple-

menting peridynamics within a molecular dynamics code. Computer
Physics Communications, 179:777–783, 2008.

[57] J. R. Rice. A path independent integral and the approximate analy-

sis of strain concentration by notches and cracks. Journal of Applied
Mechanics, 35:379–386, 1968.

107
[58] M. P. Sears and R. B. Lehoucq. The statistical mechanical foundations
of peridynamics: I. mass and momentum conservation laws. Technical
Report 2009-0791J, Sandia National Laboratories, Albuquerque, NM
87185, 2009.
[59] P. Seleson, M. L. Parks, M. Gunzburger, and R. B. Lehoucq. Peridy-
namics as an upscaling of molecular dynamics. Multiscale Modeling and
Simulation, 8:204–227, 2009.
[60] S. A. Silling. Reformulation of elasticity theory for discontinuities and
long-range forces. Journal of the Mechanics and Physics of Solids,
48:175–209, 2000.
[61] S. A. Silling. Dynamic fracture modeling with a meshfree peridynamic
code. In K.-J. Bathe, editor, Computational Fluid and Solid Mechanics
2003, pages 641–644. Elsevier, 2003.
[62] S. A. Silling. A coarsening method for linear peridynamics. Technical
Report SAND2010-1880J, Sandia National Laboratories, Albuquerque,
New Mexico 87185 and Livermore, California 94550, March 2010.
[63] S. A. Silling. Linearized theory of peridynamic states. Journal of Elas-
ticity, 99:85–111, 2010.
[64] S. A. Silling and E. Askari. Peridynamic modeling of impact damage.
In PVP-Vol.489, Problems Involving Thermal-Hydraulics, Liquid Slosh-
ing, and Extreme Loads on Structures, San Diego, California, number
PVP2004-3049, pages 197–205. ASME, 2004.
[65] S. A. Silling and E. Askari. A meshfree method based on the peridy-
namic model of solid mechanics. Computers and Structures, 83:1526–
1535, 2005.
[66] S. A. Silling and F. Bobaru. Peridynamic modeling of membranes and
fibers. International Journal of Non-Linear Mechanics, 40:395–409,
2005.
[67] S. A. Silling, M. Epton, O. Weckner, J. Xu, and E. Askari. Peridynamic
states and constitutive modeling. Journal of Elasticity, 88:151–184,
2007.
[68] S. A. Silling and R. B. Lehoucq. Convergence of peridynamics to clas-
sical elasticity theory. Journal of Elasticity, 93:13–37, 2008.
[69] S. A. Silling, O. Weckner, A. Askari, and F. Bobaru. Crack nucleation
in a peridynamic solid. International Journal of Fracture, to appear,
2010.

108
[70] S. A. Silling, M. Zimmermann, and R. Abeyaratne. Deformation of a
peridynamic bar. Journal of Elasticity, 73:173–190, 2003.

[71] S. P. Timoshenko. History of the Strength of Materials, pages 104–107.

McGraw-Hill, New York, 1953.

[72] I. Todhunter. A History of the Theory of Elasticity and of the Strength

of Materials, Vol. 1, pages 138–139, 223–224, 283–284. Dover, New
York, 1960.

[73] C. Truesdell. A First Course in Rational Continuum Mechanics. Vol-

ume I: General Concepts, pages 120–121. Academic Press: New York,
1977.

[74] T. L. Warren, S. A. Silling, A. Askari, O. Weckner, M. A. Epton, and

J. Xu. A non-ordinary state-based peridynamic method to model solid
material deformation and fracture. International Journal of Solids and
Structures, 46:1186–1195, 2009.

[75] O. Weckner and R. Abeyaratne. The effect of long-range forces on the

dynamics of a bar. Journal of the Mechanics and Physics of Solids,
53:705–728, 2005.

[76] O. Weckner, A. Askari, J. Xu, H. Razi, and S. Silling. Damage and

failure analysis based on peridynamics - theory and applications. In 48th
AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and
Materials Conference, Honolulu, Hawaii, USA, number AIAA 2007-
2314. American Institute of Aeronautics and Astronautics, 2007.

[77] O. Weckner, G. Brunk, M. A. Epton, S. A. Silling, and E. Askari.

Green’s functions in non-local three-dimensional linear elasticity. Pro-
ceedings of the Royal Society A, 465:3463–3487, 2009.

[78] O. Weckner and E. Emmrich. Numerical simulation of the dynamics of

a nonlocal, inhomogeneous, infinite bar. Journal of Computational and
Applied Mechanics, 6:311–319, 2005.

[79] O. Weckner and S. A. Silling. Determination of the constitutive model

in peridynamics from experimental dispersion data. International Jour-
nal of Multiscale Computational Engineering, under review.

[80] J. Xu, A. Askari, O. Weckner, H. Razi, and S. Silling. Damage and

failure analysis of composite laminates under biaxial loads. In 48th
AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and
Materials Conference, Honolulu, Hawaii, USA, number AIAA 2007-
2315. American Institute of Aeronautics and Astronautics, 2007.

109
[81] J. Xu, A. Askari, O. Weckner, and S. Silling. Peridynamic analysis of
impact damage in composite laminates. Journal of Aerospace Engineer-
ing, 21:187–194, 2008.

[82] A. Yuse and M. Sano. Transition between crack patterns in quenched

glass plate. Nature, 362:329–331, 1993.

[83] M. Zimmermann. A continuum theory with long-range forces for solids.

Technical Report Ph.D. dissertation, Department of Mechanical Engi-
neering, Massachusetts Institute of Technology, 2005.

110