COMMUN. MATH. SCI.

c 2007 International Press

Vol. 5, No. 4, pp. 851–864

ON THE WELL-POSEDNESS OF THE LINEAR PERIDYNAMIC

MODEL AND ITS CONVERGENCE TOWARDS THE NAVIER
EQUATION OF LINEAR ELASTICITY∗
ETIENNE EMMRICH† AND OLAF WECKNER‡

Abstract. The non-local peridynamic theory describes the displacement field of a continuous
body by the initial-value problem for an integro-differential equation that does not include any spatial
derivative. The non-locality is determined by the so-called peridynamic horizon δ which is the radius
of interaction between material points taken into account.
Well-posedness and structural properties of the peridynamic equation of motion are established
for the linear case corresponding to small relative displacements. Moreover the limit behavior as
δ → 0 is studied.

Key words. linear elasticity, non-local theory, peridynamic equation, Navier equation

AMS subject classifications. 35Q72, 74B05, 74B99, 74H10, 74H20, 74H25

1. Introduction
The peridynamic model has been introduced in Silling [33] as a non-local elasticity
theory in integral form that avoids spatial derivatives. It essentially relies upon differ-
ences of the displacement of material points interacting within a prescribed horizon.
Here also lies the possible advantage as the evolution of discontinuities in the displace-
ment or the gradient of the displacement might be inherently described within the
peridynamic model. Typical applications are the autonomous propagation of cracks
of phase boundaries.
Non-local theories taking into account effects of long-range interactions in elastic
materials, and their application to problems of solid and fracture mechanics have
been studied for a long time, cf. the pioneering work by Kröner [24], Eringen [20]
and the references cited therein, the monographs by Kunin [25] and Rogula [32], and
more recently (without being exhaustive) Altan [1, 2], Bažant & Jirásek [5], Chen
et al. [13, 12], Lei et al. [27], Pisano & Fuschi [29], Polizzotto [30, 31], Wang &
Dhaliwal [40, 41].
Different aspects of the peridynamic model such as its theoretical substantiation,
its numerical approximation, and its application have been studied in Bobaru et al. [9,
10], Dayal & Bhattacharya [14], Emmrich & Weckner [16, 17, 18], Gerstle et al. [22,
23], Silling [34], Silling & Askari [36, 35], Silling & Bobaru [37], Silling et al. [38],
Weckner & Abeyaratne [42], Weckner & Emmrich [43].
Here, we establish new results on well-posedness and dissipativity for the linear
peridynamic model. This might be seen as a first step towards the analysis of the
nonlinear model that allows, in contrast to the linear case, also large displacements
as they occur, e.g. for cracks. Existence and uniqueness follow from interpreting
the equation of motion as an abstract ordinary differential equation with a bounded
operator in a Lebesgue space. The exact solution can be represented by Duhamel’s
principle from which then a priori estimates and stability follow. For deriving a
∗ Received: May 15, 2007; accepted (in revised version): August 29, 2007. Communicated by

Claude Le Bris.
† Technische Universität Berlin, Institut für Mathematik, Straße des 17. Juni 136, 10623 Berlin,

Germany ([email protected]).
‡ The Boeing Company, Mathematics & Engineering Analysis, P.O. Box 3707, MC 7L-21, Seattle,

WA 98124-2207, United States ([email protected]).

851
852 WELL-POSEDNESS AND CONVERGENCE OF THE PERIDYNAMIC MODEL

priori estimates uniform in the parameter δ describing the non-locality, an inequality

established in Bourgain et al. [11] relating the H 1 -norm of a function to a double
integral over divided differences of the function is employed. First results on well-
posedness can be found in Emmrich & Weckner [16, 18].
Moreover, this paper is a first attempt to study the relation between the general
linear peridynamic model and the classical Navier equation of linear elasticity. It turns
out that the integral operator in the peridynamic equation of motion, which describes
the spatial interaction, applied on a smooth function becomes in the limit of vanishing
non-locality just the differential operator of the Navier equation. This result can be
proven for a large class of linear material ansatzes. However, the convergence only
takes place in an interior subdomain of the domain under consideration. This also
reflects the fact that there is no place for posing boundary conditions within the
peridynamic model as there are no spatial derivatives.
Besides, we prove an expansion of the peridynamic integral operator in terms of
higher order spatial derivatives in an interior domain. The proof mainly relies upon
a Taylor expansion of the displacement field. From studying particular nonlinear dis-
persion relations for the one-dimensional infinite peridynamic bar, it is already known
that the peridynamic integral operator equals for smooth functions a differential op-
erator of infinite order that is a series of derivatives of even order (cf. Weckner &
Abeyaratne [42]). Again for the infinite peridynamic bar but relying on another ma-
terial model, it has been shown in Emmrich & Weckner [16] that the peridynamic
integral operator becomes for sufficiently smooth functions the differential operator
of the wave equation augmented by a mixed derivative of fourth order (second order
in time and space) that describes lateral contraction.
It is interesting to see that Taylor expansions similar to those arising here have also
been observed in Arndt & Griebel [4] in connection with higher-order gradient theories
for crystalline solids in the one-dimensional case. The relation between atomistic and
continuum models for crystalline solids and corresponding expansions have also been
studied in E & Ming [15]. We also refer to Blanc et al. [6, 7, 8] and Le Bris &
Lions [26] for a discussion of the relation between molecular models and models on a
continuum level.
Throughout this paper, we rely upon the usual notation for Lebesgue- and
Bochner-Lebesgue-spaces as well as spaces of continuously differentiable functions.
By C, we denote a generic positive constant that is independent of δ.
2. The peridynamic equation of motion
The governing equation in the peridynamic theory is the second-order in time
partial integro-differential equation
Z
2
ρ(x)∂t u(x,t) = f (x, x̂,u(x,t),u(x̂,t),t)dx̂ + b(x,t), (x,t) ∈ V × (0,T ), (2.1)
V

in Lagrangian coordinates x, where ρ denotes the mass density that shall be bounded
away from zero with 1/ρ ∈ L∞ (V), u the displacement field of the body that occupies
the volume V ⊆ Rd (d ∈ {1,2,3}), f the pairwise force function that describes the in-
ternal forces, and b an inhomogeneity that collects all external forces per unit volume.
By T > 0, the time under consideration is denoted. The volume V is supposed to be
a bounded domain of class C 0,1 . Equ. (2.1) is supplemented by the initial conditions

u(·,0) = u0 , ∂t u(·,0) = u̇0 . (2.2)

 E. EMMRICH AND O. WECKNER 853

Note that no boundary conditions appear as there are no spatial derivatives.

In view of the balance of the angular momentum of the mass-free bond be-
tween x and x̂, the pairwise force function f always points from the current po-
sition x + u(x,t) of the point x to the current position x̂ + u(x̂,t) of the point x̂.
Moreover, the invariance against a rigid body motion, the assumption of no explicit
time dependence, and Newton’s third law lead to f (x, x̂,u, û,t) = f (x, x̂, û − u) with
f (x̂,x,−η) = −f (x, x̂,η) for all x, x̂,η := û − u. It is typical for the peridynamic
model to require

f (x, x̂,η) = 0 if |x̂ − x| ≥ δ, (2.3)

where δ > 0 is the so-called peridynamic horizon of interaction. Here and in the
sequel, | · | denotes the Euclidean vector or spectral matrix norm. This cut-off in
the reference configuration becomes, however, rather questionable in the presence of
large deformations where it might be necessary to redefine the cut-off in the actual
or present configuration.
If the material is microelastic in the sense of Silling [34] then there exists a pairwise
micropotential w such that f (x, x̂,η) = ∇η w(x, x̂,η). In order to emphasise explicitly
the dependence on the dimension d and the horizon δ, we shall use in the following
the subscript d,δ. One of the simplest nonlinear models that has been suggested is
the proportional microelastic material model with
x̂ + û − x − u |x̂ + û − x − u| − |x̂ − x|
f d,δ (x, x̂, û − u) = cd,δ s , s := , (2.4)
|x̂ + û − x − u| |x̂ − x|
where s denotes the bond stretch that is the relative change of the length of a bond.
The constant of proportionality cd,δ is to be determined in such a way that the defor-
mation energy density of a homogeneous body under isotropic expansion arising from
the peridynamic model coincides with the energy density known from the classical
linear elasticity theory, i.e. η = s(x̂ − x) with some s > 0, and
1
Z
wd,δ (x, x̂,s(x̂ − x))dx̂ ≡ eel,d , (2.5)
2
B(x;δ)

where
9Ks2 12Ks2 9Ks2
eel,1 = , eel,2 = , eel,3 =
10 5 2
is the elastic energy density from the classical linear elasticity theory for the one-,
two-, and three-dimensional cases, respectively. Here, K = E/(3(1 − 2ν)) denotes the
bulk modulus, ν the Poisson ratio, and E the Young modulus. Moreover, we denote
by B(x;δ) ⊂ Rd the open ball of radius δ with respect to | · | and with center x ∈ Rd .
The peridynamic model as considered so far is restricted to the Poisson ratio ν = 1/4.
With ν = 1/4, one easily obtains (cf. Emmrich & Weckner [18])
18K 72K 18K
c1,δ = , c2,δ = , c3,δ = . (2.6)
5δ 2 5πδ 3 πδ 4
We should mention here that the pairwise force function of the proportional micro-
elastic material model fulfills the scaling property

f d,εδ (εx,εx̂,εη) = ε−(d+1) f d,δ (x, x̂,η), ε > 0.

854 WELL-POSEDNESS AND CONVERGENCE OF THE PERIDYNAMIC MODEL

A first-order approximation justifies for small relative displacements η the general

linear ansatz

f (x, x̂,η) = f0 (x, x̂) + C(x, x̂) · η

with the stiffness tensor (or micromodulus function) C = C(x, x̂) and f0 denoting
forces in the reference configuration. Without loss of generality, we may assume
f0 ≡ 0, since otherwise f0 can be incorporated into the right-hand side b. In general,
the stiffness tensor C is neither definite nor dependent on the length |x̂ − x| only.
However, C has to be symmetric with respect to its arguments as well as with respect
to its tensor structure such that C(x̂,x) = C(x, x̂) and C(x, x̂)T = C(x, x̂). The cor-
responding micropotential is then given by w(x, x̂,η) = η · C(x, x̂) · η/2. In view of
(2.3), we shall require C(x̂,x) = 0 if |x̂ − x| ≥ δ.
The stiffness tensor can be shown to read as

C(x, x̂) = λd,δ (|x̂ − x|)(x̂ − x) ⊗ (x̂ − x) (2.7)

for a linear microelastic material (cf. also Silling [33]). The measurable function
λd,δ : R+
0 → R with λd,δ (r) = 0 for r ≥ δ determines the specific material model and
depends on d and δ. Note that |C(x, x̂)| = |λd,δ (|x̂ − x|)||x̂ − x|2 .
The linear peridynamic equation of motion (2.1) now reads as
Z
ρ(x)∂t2 u(x,t) = λd,δ (|x̂ − x|)(x̂ − x) ⊗ (x̂ − x) · (u(x̂,t) − u(x,t)) dx̂ + b(x,t)
V∩B(x;δ)

=: (Ld,δ u)(x,t) + b(x,t), (x,t) ∈ V × (0,T ). (2.8)

A short calculation shows that Ld,δ v vanishes if v is a vector field such that

v(x) = A · x + b, AT = −A, (2.9)

for some skew-symmetric matrix A ∈ Rd×d and some vector b ∈ R. For small displace-
ments as in the linear case, such a vector field can be seen as describing rigid body
motions. In the three-dimensional case, we have v(x) = a ∧ x + b for some a,b ∈ R3
(cf. also Temam & Miranville [39, Sect. 1.4] and Oleinik et al. [28]).
For the rest of the paper, let
Zδ Zδ
k+d+1 |·|
ℓk,d,δ := λd,δ (r)r dr , k = 0,1,2,... , ℓd,δ := |λd,δ (r)|rd+1 dr .
0 0

By virtue of (2.5), we require

s2
Z
ξ · λd,δ (|ξ|)ξ ⊗ ξ · ξ dξ = eel,d ,
4
B(0;δ)

which yields a first assumption on λd,δ , namely,

4eel,d 9K 24K 9K
ℓ2,d,δ = , i.e., ℓ2,1,δ = , ℓ2,2,δ = , ℓ2,3,δ = , (2.10)
s2 σd 5 5π 2π
where σd comes from the integration with respect to the spherical angle coordinates
such that σ1 = 2, σ2 = 2π, σ3 = 4π.
 E. EMMRICH AND O. WECKNER 855

Well-posedness of the linear peridynamic equation of motion has been studied by

the authors in [43] for the unbounded one-dimensional case and in [18] for the bounded
two- and three-dimensional case in a framework with Lp (V)d with sufficiently large
p. The proofs of the two statements below follow essentially the same lines as in [18]
and shall be omitted here.
|·|
Proposition 2.1. If ℓd,δ < ∞ then the integral operator Ld,δ defined by (2.8) is a
linear mapping in L∞ (V)d with
|·|
kLd,δ kL(L∞ (V)d ) ≤ 2ℓd,δ σd . (2.11)

|·|
Theorem 2.2. If ℓd,δ < ∞ then there exists for every u0 , u̇0 ∈ L∞ (V)d , b ∈
L (0,T ;L (V) ) a unique mild solution u ∈ C 1 ([0,T ];L∞ (V)d ) to the initial-value
1 ∞ d

problem (2.8), (2.2) that satisfies the a priori estimate

kukC 1 ([0,T ];L∞ (V)d ) ≤ Cd,δ ku0 kL∞ (V)d + ku̇0 kL∞ (V)d + kbkL1 (0,T ;L∞ (V)d ) (2.12)

with
 q 
|·| |·|
Cd,δ = max(1,T,2T ℓd,δ σd )cosh T 2ℓd,δ σd . (2.13)

If b ∈ C([0,T ];L∞ (V)d ) then u ∈ C 2 ([0,T ];L∞ (V)d ).

The stability estimate (2.12) with (2.13) immediately follows from the represen-
tation of the solution by means of Duhamel’s principle (cf. again [18]),
−1
p  p p 
u(t) = cos t −Ld,δ u0 + −Ld,δ sin t −Ld,δ u̇0
Z t
−1
p  p 
+ −Ld,δ sin (t − s) −Ld,δ b(s)ds (2.14)
0

with the formal notation

∞ ∞
p  X t2n n −1
p  X t2n+1
Lnd,δ .
p
cos t −Ld,δ := L , −Ld,δ sin t −Ld,δ :=
n=0
(2n)! d,δ n=0
(2n + 1)!

The series are uniformly convergent on any compact time interval.

It should be noted that, due to (2.10), the quantities ℓ2,d,δ are independent of δ
and thus, because of (2.11) and
|·|
0 < ℓ2,d,δ ≤ δ 2 ℓd,δ , (2.15)

the family of operators Ld,δ : L∞ (V)d → L∞ (V)d may not be uniformly bounded as
δ → 0. Reasoning in the same way shows that also the stability constant Cd,δ from
(2.13) behaves badly as δ → 0. Furthermore, we can prove the following results.
|·|
Proposition 2.3. If ℓd,δ < ∞ then the integral operator Ld,δ defined by (2.8) is a
linear mapping in L2 (V)d with
|·|
kLd,δ kL(L2 (V)d ) ≤ 2ℓd,δ σd . (2.16)
856 WELL-POSEDNESS AND CONVERGENCE OF THE PERIDYNAMIC MODEL

Moreover, Ld,δ : L2 (V)d → L2 (V)d is self-adjoint, i.e.,

(Ld,δ v,w)L2 (V)d = (v,Ld,δ w)L2 (V)d ∀v,w ∈ L2 (V)d .

If λd,δ is nonnegative then Ld,δ : L2 (V)d → L2 (V)d is also dissipative, i.e.

(Ld,δ v,v)L2 (V)d ≤ 0 ∀v ∈ L2 (V)d ,

where (Ld,δ v,v)L2 (V)d = 0 if and only if v is of type (2.9) or λd,δ vanishes identically.
Proof. We immediately find
Z
|·|
|(Ld,δ v)(x)| ≤ |λd,δ (|x̂ − x|)||x̂ − x|2 |v(x̂)|dx̂ + ℓd,δ σd |v(x)|.
V

Denoting the integral on the right-hand side by I(x), an application of the Cauchy-
Schwarz inequality yields
Z
|·|
|I(x)|2 ≤ ℓd,δ σd |λd,δ (|x̂ − x|)||x̂ − x|2 |v(x̂)|2 dx̂.
V

Integration then gives

Z Z Z 
|·| |·|
|I(x)|2 dx ≤ ℓd,δ σd |λd,δ (|x̂ − x|)||x̂ − x|2 dx |v(x̂)|2 dx̂ ≤ (ℓd,δ σd )2 kvk2L2 (V)d ,
V V V

from which the first assertion follows.

A straightforward calculation shows
1
ZZ


(Ld,δ v,w)L2 (V)d = − λd,δ (|x̂ − x|) (v(x̂) − v(x)) · (x̂ − x) ×
2
VV


(w(x̂) − w(x)) · (x̂ − x) dx̂dx (2.17)

which symmetry and thus self-adjointness of Ld,δ : L2 (V)d → L2 (V)d implies. In the
case v = w, this also proves dissipativity if λd,δ is nonnegative. Applying [39, Propo-
sition 1.2] shows that (v(x̂) − v(x)) · (x̂ − x) vanishes identically if and only if v is of
type (2.9). This proves the last assertion.
By standard arguments for abstract differential equations with a linear bounded
operator and Duhamel’s principle (cf. e.g. Emmrich [19, Ch. 7]), we can prove the
following theorem.
|·|
Theorem 2.4. If ℓd,δ < ∞ then there exists for every u0 , u̇0 ∈ L2 (V)d , b ∈
L1 (0,T ;L2 (V)d ) a unique mild solution u ∈ C 1 ([0,T ];L2 (V)d ) to the initial-value prob-
lem (2.8), (2.2) that satisfies the a priori estimate (2.12), (2.13) with L∞ (V)d being
replaced by L2 (V)d . If b ∈ C([0,T ];L2 (V)d ) then u ∈ C 2 ([0,T ];L2 (V)d ).
The solution is again given by (2.14), Ld,δ : L2 (V)d → L2 (V)d is the generator of a
strongly continuous cosine function on L2 (V)d , and the operator describing the corre-
sponding one-dimensional system is the generator of a strongly continuous semigroup
on L2 (V)d × L2 (V)d (cf. also Arendt et al. [3, Theorem 3.14.7, Corolarry 3.14.9] and
Fattorini [21]).
 E. EMMRICH AND O. WECKNER 857

As in Proposition 2.1 and Theorem 2.2, we cannot conclude uniform boundedness

with respect to δ. This problem can be overcome as follows. As usual, we denote by
| · |H 1 (V)d the standard semi-norm of H 1 (V)d .
Lemma 2.1. Assume v ∈ H 1 (V) and let φ ∈ L1 (Rd ), φ ≥ 0. Then

|v(x̂) − v(x)|2
ZZ
φ(x̂ − x) dx̂dx ≤ C kφkL1 (Rd ) |v|2H 1 (V) ,
|x̂ − x|2
VV

where C depends only on V.

The lemma is a special case of Bourgain et al. [11, Theorem 1].
|·|
Proposition 2.5. Let ℓd,δ < ∞. If λd,δ is nonnegative and fulfills (2.10) then Ld,δ
is a linear mapping on H 1 (V)d with

(Ld,δ v,w)L2 (V)d ≤ Cℓ2,d,δ |v|H 1 (V)d |w|H 1 (V)d ∀v,w ∈ H 1 (V)d ,

where C depends on the domain and its dimension but is independent of δ.

Proof. Let φ(ξ) := λd,δ (|ξ|)|ξ|4 (ξ ∈ Rd ). From (2.17), we conclude with the
Cauchy-Schwarz inequality

(Ld,δ v,w)L2 (V)d ≤

1/2 ZZ 1/2
|v(x̂) − v(x)|2 |w(x̂) − w(x)|2
ZZ
1
φ(x̂ − x) dx̂dx φ(x̂ − x) dx̂dx .
2 |x̂ − x|2 |x̂ − x|2
VV VV

Lemma 2.1 due to Bourgain et al. [11] yields

(Ld,δ v,w)L2 (V)d ≤ C kφkL1 (Rd ) |v|H 1 (V)d |w|H 1 (V)d

with C only depending on V and d. The assertion now follows in view of

Z
kφkL1 (Rd ) = λd,δ (|ξ|)|ξ|4 dξ = ℓ2,d,δ σd .
B(0;δ)

The proposition above shows in particular that

kLd,δ kL(H 1 (V)d ,(H 1 (V)d )∗ ) ≤ C ℓ2,d,δ .

Because of (2.10), the family of linear bounded operators Ld,δ : H 1 (V)d → (H 1 (V)d )∗
is thus uniformly bounded with respect to δ.
As we can infer from (2.17) and the proof above, Ld,δ might also be considered as a
linear bounded operator from W 1,∞ (V)d in its dual with kLd,δ kL(W 1,∞ (V)d ,(W 1,∞ (V)d )∗ )
being bounded independently of δ.
An immediate consequence of the preceding propositions is the following theorem.
|·|
Theorem 2.6. Let ℓd,δ < ∞ and assume that λd,δ is nonnegative and fulfills (2.10).
The unique solution u ∈ C 2 ([0,T ];L2 (V)d ) to the initial-value problem (2.8), (2.2) with
u0 ∈ H 1 (V)d , u̇0 ∈ L2 (V)d , and b ∈ C([0,T ];L2 (V)d ) then satisfies the a priori estimate


kukC 1 ([0,T ];L2 (V)d ) ≤ C ku0 kH 1 (V)d + ku̇0 kL2 (V)d + kbkL2 (0,T ;L2 (V)d ) (2.18)
858 WELL-POSEDNESS AND CONVERGENCE OF THE PERIDYNAMIC MODEL

with C being independent of δ.

Proof. According to Theorem 2.4, there is a unique solution u ∈ C 2 ([0,T ];L2 (V)d ).
For proving (2.18), we firstly observe that for all v ∈ C 1 ([0,T ]; L2 (V)d )

1 d
(v(t),v(t))L2 (V)d = (∂t v(t),v(t))L2 (V)d ;
2 dt
an analogous relation holds true for the second time derivative if v ∈ C 2 ([0,T ];L2 (V)d ).
Testing (2.8) by ∂t u then leads (with the symmetry of Ld,δ ) to

ρ d
k∂t u(t)k2L2 (V)d = (Ld,δ u(t),∂t u(t))L2 (V)d + (b(t),∂t u(t))L2 (V)d
2 dt
1 d
≤ (Ld,δ u(t),u(t))L2 (V)d + kb(t)kL2 (V)d k∂t u(t)kL2 (V)d .
2 dt
With Young’s inequality, integration, dissipativity, and Gronwall’s lemma, we come
up with
 Zt 
k∂t u(t)k2L2 (V)d ≤C − (Ld,δ u0 ,u0 )L2 (V)d + k∂t u0 k2L2 (V)d + kb(s)k2L2 (V)d ds ,
0

where C depends on 1/ρ ∈ L∞ (V) and T only. The assertion follows with Proposi-
tion 2.5 and since
Zt
ku(t)kL2 (V)d ≤ ku0 kL2 (V)d + k∂t u(s)kL2 (V)d ds.
0

A main advantage of (2.18) is the fact that the stability constant now is indepen-
dent of δ and thus furnishes uniform boundedness of a sequence of solutions uδ as
δ → 0.
So far, our existence results were based upon abstract differential equations gov-
erned by a linear bounded operator that maps a Banach space into itself. However,
by virtue of Proposition 2.5 in combination with Proposition 2.3, we may also think
of applying the theory of weak solutions to second-order evolution equations based
upon the Gelfand triple H 1 (V)d ⊂ L2 (V)d ⊂ (H 1 (V)d )∗ (cf. e.g. Wloka [44, Ch. V]).
This would allow to consider the peridynamic equation of motion (2.8) in the same
functional analytic setting as the Navier equation of linear elasticity theory. Unfortu-
nately, the operator −Ld,δ : H 1 (V)d → (H 1 (V)d )∗ has not yet been proven to satisfy a
Gårding inequality.
3. The limit behaviour for vanishing non-locality
In the following, we consider δ ∈ Λ where Λ ⊂ R+ is a null sequence with maxΛ < δ0
for some δ0 > 0. We set

V0 := {z ∈ V : dist(z,∂V) > δ0 }.

|·|
Corollary 3.1. Let ℓd,δ < ∞ and assume that λd,δ is nonnegative and fulfills (2.10).
The sequence {uδ }δ∈Λ of solutions to the initial-value problem (2.8), (2.2) with u0 ∈
 E. EMMRICH AND O. WECKNER 859

H 1 (V)d , u̇0 ∈ L2 (V)d , b ∈ C([0,T ];L2 (V)d ) and the sequence of its time derivatives then
possesses a subsequence that converges weakly ∗ in L∞ (0,T ;L2 (V)d ) as δ → 0 (δ ∈ Λ).

Proof. The result is a direct consequence of Theorem 2.6 applying the Banach-
Alaoglu theorem (upon noting that L∞ (0,T ;L2 (V)d ) is the dual of the separable
Banach space L1 (0,T ;L2 (V)d )).
It remains open for future work to study the properties of the weak∗ limit of
{uδ }δ∈Λ . Also the following observations give rise to further study.
The classical Navier equation of linear elasticity reads for (x,t) ∈ V × (0,T ) as
ρ(x)∂t2 u(x,t) = µ∆u(x,t) + (λ + µ) grad divu(x,t) + b(x,t) =: (Lu)(x,t) + b(x,t)
(3.1)
with the Lamé coefficients µ = E/(2(1 + ν)) and λ = νE/((1 − 2ν)(1 + ν)), cf. e.g.
Temam & Miranville [39]. Note that for the Poisson ratio ν = 1/4, we find µ = λ =
2E/5 = 3K/5. The Navier equation is supplemented by initial conditions (2.2) and
appropriate boundary conditions.
|·|
Theorem 3.2. Let ℓd,δ < ∞ for all δ ∈ Λ and assume (2.10). If v ∈ C 2 (V)d and λd,δ
is nonnegative, then
kLd,δ v − LvkL∞ (V0 )d → 0 as δ→0 (δ ∈ Λ). (3.2)

If v ∈ C m (V)d (m ≥ 2) then the expansion

[m/2]
(2n)
X
(Ld,δ v)(x) = (Lv)(x) + (Ld,δ v)(x) + Rm;d,δ (v;x)
n=2

(2n)
holds for all x ∈ V0 and δ ∈ Λ. Here, Ld,δ is a differential operator of or-
der 2n given by (3.3), (3.4), and (3.5), respectively, and the remainder satisfies
|·| |·|
kRm;d,δ (v;x)kL∞ (V0 )d = o(δ m ℓd,δ ) with ℓd,δ ≥ Cδ −2 . If, in addition, λd,δ is nonneg-
ative then
(2n)
kLd,δ vkL∞ (V0 )d ≤ C δ 2(n−1) ℓ2,d,δ kvkC 2n (V0 )d , kRm;d,δ (v;x)kL∞ (V0 )d = o(δ m−2 )

with ℓ2,d,δ being independent of δ and given by (2.10).

Proof. For x ∈ V0 , the closed ball B(x;δ) ∋ x̂ lies in V for all δ ∈ Λ. The smoothness
assumed for v then allows the Taylor expansion
m
X 1
k
v(x̂) − v(x) = (x̂ − x) · ∇ v(x) + r m (v;x, x̂)
k!
k=1

with the remainder

Z1
(1 − θ)m−1
m
v(x + θ(x̂ − x)) − v(x) dθ = o(|x̂ − x|m ).


r m (v;x, x̂) = (x̂ − x) · ∇
(m − 1)!
0

Inserting the Taylor expansion into the definition of Ld,δ (see (2.8)) yields
m
(k)
X
(Ld,δ v)(x) = (Ld,δ v)(x) + Rm;d,δ (v;x),
k=1
860 WELL-POSEDNESS AND CONVERGENCE OF THE PERIDYNAMIC MODEL

where
1
Z
(k)
k
(Ld,δ v)(x) := λd,δ (|x̂ − x|)(x̂ − x) ⊗ (x̂ − x) (x̂ − x) · ∇ v(x)dx̂
k!
B(x;δ)

and
Z
Rm;d,δ (v;x) = λd,δ (|x̂ − x|)(x̂ − x) ⊗ (x̂ − x)r m (v;x, x̂)dx̂.
B(x;δ)

(k)
It follows (Ld,δ v)(x) = 0 if k is odd since then the integrand is an odd function in
x̂ − x.
For the one-dimensional case, we immediately find

0 if k is odd,
(
(k)
(L1,δ v)(x) = 2 (k) (3.3)
ℓk,1,δ v (x) if k is even.
k!
(2)
In particular, (2.10) furnishes (L1,δ v)(x) = 9Kv ′′ (x)/5 = (Lv)(x).
Introducing spherical coordinates, we obtain for the two-dimensional case

k Z2π 
(k)
X 1 cos2 φ cosφ sinφ
(L2,δ v)(x) = ℓk,2,δ cosk−j φ sinj φ dφ ∂ (k−j,j) v(x)
j!(k − j)! cosφ sinφ sin2 φ
j=0 0
(3.4)

and for the three-dimensional case

(k)
(L3,δ v)(x)
Z2π Zπ/2 cos2 φ cos2 ψ cosφ sinφ cos2 ψ cosφ cosψ sinψ
!
X 1
= ℓk,3,δ cosφ sinφ cos2 ψ sin2 φ cos2 ψ sinφ cosψ sinψ
j1 !j2 !j3 ! cosφ cosψ sinψ sinφ cosψ sinψ sin2 ψ
j1 +j2 +j3 =k 0 −π/2

× (cosφ cosψ)j1 (sinφ cosψ)j2 sinj3 ψ cosψ dψdφ ∂ (j1 ,j2 ,j3 ) v(x). (3.5)

Here, we write spatial derivatives using multiindices, ∂ α ≡ ∂xα11 ···∂xαdd for α =

(α1 ,...,αd ). With x1 ,...,xd and v1 ,...,vd , the components of x and v, respectively,
are denoted.
In particular, we find with (2.10)
 
(2) 3K 3∂ (20) v1 + 2∂ (11) v2 + ∂ (02) v1
(L2,δ v)(x) = (x) = (Lv)(x)
5 ∂ (20) v2 + 2∂ (11) v1 + 3∂ (02) v2

as well as
(2)
(L3,δ v)(x)
 (200) 
3∂ v1 + 2∂ (110) v2 + 2∂ (101) v3 + ∂ (020) v1 + ∂ (002) v1
3K  (200)
= ∂ v2 + 2∂ (110) v1 + 3∂ (020) v2 + 2∂ (011) v3 + ∂ (002) v2  (x)
5 ∂ (200)
v3 + 2∂ (101)
v1 + ∂ (020) v3 + 2∂ (011) v2 + 3∂ (002) v3
= (Lv)(x).
 E. EMMRICH AND O. WECKNER 861

This, together with v ∈ C m (V)d and (2.15), proves the expansion asserted. Moreover,
for nonnegative λd,δ , we observe that

0 ≤ ℓk,d,δ ≤ δ k−2 ℓ2,d,δ , k = 2,3,...,m. (3.6)

The uniform convergence (3.2) is a direct consequence of the preceding results

2 d
with
S m =S2 since the second-order derivatives of v ∈ C (V) are uniformly continuous
on δ∈Λ x∈V0 B(x;δ) ⊂ V.
In view of (3.3), the expansion in the one-dimensional case reads for a smooth
function v as
9K ′′ ℓ4,1,δ (4) ℓ6,1,δ (6)
L1,δ v = v + v + v + ···
5 12 360
Some straightforward but tedious calculations of the integrals appearing lead for the
two-dimensional case e.g. to
 
(4) ℓ4,2,δ π 5∂ (40) v1 + 4∂ (31) v2 + 6∂ (22) v1 + 4∂ (13) v2 + ∂ (04) v1
L2,δ v =
192 ∂ (40) v2 + 4∂ (31) v1 + 6∂ (22) v2 + 4∂ (13) v1 + 5∂ (04) v2

and

(6) ℓ6,2,δ π 7∂ (60) v1 + 6∂ (51) v2 + 15∂ (42) v1 + 12∂ (33) v2 +
L2,δ v =
9216 ∂ (60) v2 + 6∂ (51) v1 + 9∂ (42) v2 + 12∂ (33) v1 +

+ 9∂ (24) v1 + 6∂ (15) v2 + ∂ (06) v1
,
+ 15∂ (24) v2 + 6∂ (15) v1 + 7∂ (06) v2

and for the three-dimensional case e.g. to

 (400) (310) (301)
v3 + 6∂ (220) v1 +
(4) ℓ4,3,δ π  5∂ (400) v1 + 4∂ (310) v2 + 4∂
L3,δ v = ∂ v2 + 4∂ v1 + 6∂ (220) v2 + 4∂ (211) v3 +
210 ∂ (400)
v3 + 4∂ (301)
v1 + 2∂ (220) v3 + 4∂ (211) v2 +

+ 6∂ (202) v1 + 4∂ (130) v2 + 4∂ (121) v3 + 4∂ (112) v2 + 4∂ (103) v3 +

+ 2∂ (202) v2 + 4∂ (130) v1 + 4∂ (112) v1 +
(202) (121)
+ 6∂ v3 + 4∂ v1 + 4∂ (103) v1 +

+ ∂ (040) v1 + 2∂ (022) v1 + ∂ (004) v1
(040) (031) (022) (013)
+ 5∂ v2 + 4∂ v3 + 6∂ v2 + 4∂ v3 + ∂ (004) v2  .
(040) (031) (022) (013)
+ ∂ v3 + 4∂ v2 + 6∂ v3 + 4∂ v2 + 5∂ (004) v3

Example 3.3. The linearisation of the proportional microelastic material model (2.4)
leads to λd,δ (r) = cd,δ r−3 (r < δ) with cd,δ defined in (2.6). The one-dimensional case
|·|
cannot be treated here as ℓ1,δ = ∞. In the two- and three-dimensional case, the model
|·|
yields a weakly singular integral kernel of convolution type with ℓd,δ = O(δ −2 ). So, all
the results of this and the preceding section apply for the two- and three-dimensional
case.
The coefficients in the expansion of Theorem 3.2 are calculated from
cd,δ
ℓk,d,δ = δ k+d−1 , k = 2,4,..., d ∈ {2,3}.
k+d−1
862 WELL-POSEDNESS AND CONVERGENCE OF THE PERIDYNAMIC MODEL

Unfortunately, in this model, the interaction jumps to zero if r = δ. This can be

avoided by taking λd,δ (r) = c′d,δ r−3 exp(−δ 2 /(δ 2 − r2 )) (r < δ). The new constant of
proportionality c′d,δ is again calculated from (2.10).
It should be noted that Arndt & Griebel [4, Sect. 4] have obtained, in the one-
dimensional case, the quite similar expansion (compare with (3.3), (3.6))
∞
X 2 2(n−1) ′
ε (φ ◦ v)(2n) (x)
n=1
(2n)!

for the evolution equation of an atomistic system. Here, φ is some function describing
the potential, e.g. φ(r) = (r − 1)2 /2 for a chain of linear springs, and ε is a scaling
parameter. For a further study of crystalline solids, cf. also E & Ming [15].
The very interesting question whether the sequence of solutions to the peridy-
namic equation of motion converges in some sense towards a certain solution to the
Navier equation cannot be answered so far. In order to find an answer, it might be
essential to describe the behavior of the peridynamic solution near the boundary of
the domain as one lacks a criterion for posing appropriate boundary conditions for
the Navier equation. More precisely, let u be a sufficiently smooth solution to the
Navier equation subject to suitable initial and boundary conditions and for a given
right-hand side. Then wδ := uδ − u with uδ ∈ C 2 ([0,T ];L2 (V)d ) being a solution to
the peridyamic equation with the same initial values and the same right-hand side
satisfies for (x,t) ∈ V × (0,T ) the equation

ρ(x)∂t2 wδ (x,t) = (Ld,δ wδ )(x,t) + (Ld,δ u)(x,t) − (Lu)(x,t)

with initial conditions wδ (·,0) = ∂t wδ (·,0) = 0. The uniform stability (2.18) furnishes

kwδ kC 1 ([0,T ];L2 (V)d ) ≤ C kLd,δ u − LukL2 (0,T ;L2 (V)d )

with C being independent of δ. With Theorem 3.2, we only have an estimate for
Ld,δ u − Lu in the interior domain V0 at hand. What would be needed is, however, a
characterization of the difference between Ld,δ u and Lu near the boundary ∂V.
Acknowledgement. The authors gratefully acknowledge many discussions with
Henrik Büsing and Stephan Kusche.

REFERENCES

[1] B.S. Altan, Uniqueness of initial-boundary value problems in nonlocal elasticity, Int. J.
Solids Structures, 25(11), 1271–1278, 1989.
[2] B.S. Altan, Uniqueness in nonlocal thermoelasticity, J. Thermal Stresses, 14, 121–128,
1991.
[3] W. Arendt et al., Vector-Valued Laplace Transforms and Cauchy Problems, Birkhäuser,
Basel, 2001.
[4] M. Arndt and M. Griebel, Derivation of higher order gradient continuum models from
atomistic models for crystalline solids, Multiscale Model. Simul., 4(2), 531–562,
2005.
[5] Z.P. Bažant and M. Jirásek, Nonlocal integral formulations of plasticity and damage:
survey and progress, J. Eng. Mech., 128(11), 1119–1149, 2002.
[6] X. Blanc, C. Le Bris and F. Legoll, Analysis of a prototypical multiscale method coupling
atomistic and continuum mechanics, M2AN, 39(4), 797–826, 2005.
 E. EMMRICH AND O. WECKNER 863

[7] X. Blanc, C. Le Bris and P.L. Lions, From molecular models to continuum mechanics,
Arch. Rational Mech. Anal., 164(4), 341–381, 2002.
[8] X. Blanc, C. Le Bris and P.L. Lions, Atomistic to continuum limits for computational
materials science, M2AN, 41(2), 391–426, 2007.
[9] F. Bobaru and S.A. Silling, Peridynamic 3D problems of nanofiber networks and carbon
nanotube-reinforced composites, Materials and Design: Proc. Numiform, American
Institute of Physics, 1565–1570, 2004.
[10] F. Bobaru, S.A. Silling and H. Jiang, Peridynamic fracture and damage modeling of
membranes and nanofiber networks, Proc. XI Int. Conf. Fracture, Turin, 5748, 1–6,
2005.
[11] J. Bourgain, H. Brézis and P. Mironescu, Another look at Sobolev spaces, J.L. Menaldi
et al. (eds.), Optimal Control and Partial Differential Equations, IOS Press, Ams-
terdam, Tokyo, 439–455, 2001.
[12] Y. Chen, J.D. Lee and A. Eskandarian, Atomistic viewpoint of the applicability of mi-
crocontinuum theories, Int. J. Solids Structures, 41(8), 2085–2097, 2004.
[13] Y. Chen, J.D. Lee and A. Eskandarian, Dynamic meshless method applied to nonlocal
crack problems, Theor. Appl. Fracture Mech., 38, 293–300, 2002.
[14] K. Dayal and K. Bhattacharya, Kinetics of phase transformations in the peridynamic
formulation of continuum mechanics, J. Mech. Phys. Solids, 54, 1811–1842, 2006.
[15] W. E and P. Ming, Cauchy-Born rule and the stability of crystalline solids: static
problems, Arch. Rational Mech. Anal., 183(2), 241–297, 2007.
[16] E. Emmrich and O. Weckner, Analysis and numerical approximation of an
integro-differential equation modelling non-local effects in linear elasticity,
Math. Mech. Solids, published online first, DOI: 10.1177/1081286505059748, 2005.
[17] E. Emmrich and O. Weckner, The peridynamic equation and its spatial discretisation,
Math. Model. Anal., 12 (1), 17–27, 2007.
[18] E. Emmrich and O. Weckner, The peridynamic equation of motion in non-local elasticity
theory, C.A. Mota Soares et al. (eds.), III European Conference on Computational
Mechanics, Solids, Structures and Coupled Problems in Engineering (Lisbon, June
2006), Springer, 19, 2006.
[19] E. Emmrich, Gewöhnliche und Operator-Differentialgleichungen, Vieweg, Wiesbaden,
2004.
[20] A.C. Eringen, Vistas of nonlocal continuum physics, Int. J. Eng. Sci., 30 (10), 1551–
1565, 1992.
[21] H.O. Fattorini, Second order linear differential equations in Banach spaces, North-
Holland mathematics studies 108, Elsevier, Amsterdam, 1985.
[22] W. Gerstle and N. Sau, Peridynamic modeling of concrete structures, Proc. 5th
Int. Conf. Fract. Mech. Concr. Struct, Li et al. (eds.), Ia-FRAMCOS, 2, 949–956,
2004.
[23] W. Gerstle, N. Sau and S. Silling, Peridynamic modeling of plain and reinforced concrete
structures, SMiRT 18: 18th Int. Conf. Struct. Mech. React. Technol., Beijing, China,
15, 2005.
[24] E. Kröner, Elasticity theory of materials with long range forces, Int. J. Solids Structures,
3, 731–742, 1967.
[25] I.A. Kunin, Elastic Media with Microstructure, vol. I and II. Springer, Berlin, 1982/83.
[26] C. Le Bris and P.L. Lions, From atoms to crystals: a mathematical journey, Bulletin of
the American Mathematical Society, 42 (3), 291–363, 2005.
[27] Y. Lei, M.I. Friswell and S. Adhikari, A Galerkin method for distributed systems with
non-local damping, Int. J. Solids Structures, 43(11-12), 3381–3400, 2005.
[28] O.A. Oleinik, A.S. Shamaev and G.A. Yosifian, Mathematical Problems in Elasticity
and Homogenization, North-Holland Publishing, Amsterdam, 1992.
[29] A.A. Pisano and P. Fuschi, Closed form solution for a nonlocal elastic bar in tension,
Int. J. Solids Structures, 40 (1), 13–23, 2003.
[30] C. Polizzotto, Nonlocal elasticity and related variational principles, Int. J. Solids Struc-
864 WELL-POSEDNESS AND CONVERGENCE OF THE PERIDYNAMIC MODEL

tures, 38 (42–43), 7359-7380, 2001.

[31] C. Polizzotto, Unified thermodynamic framework for nonlocal/gradient continuum me-
chanics, Eur. J. Mech. A/Solids, 22, 651–668, 2003.
[32] D. Rogula, Nonlocal Theory of Material Media, Springer, Berlin, 1982.
[33] S.A. Silling, Reformulation of elasticity theory for discontinuities and long-range forces,
J. Mech. Phys. Solids, 48 (1), 175–209, 2000.
[34] S.A. Silling, Dynamic fracture modeling with a meshfree peridynamic code, K.J. Bathe
(ed.), in Computational Fluid and Solid Mechanics 2003, Elsevier, Amsterdam,
641–644, 2003.
[35] S.A. Silling and E. Askari, Peridynamic modeling of impact damage, F.J. Moody (ed.),
Amer. Soc. of Mech. Eng., New York, 489, 197–205, 2004.
[36] S.A. Silling and E. Askari, A meshfree method based on the peridynamic model of solid
mechanics, Comput. Struct., 83(17-18), 1526–1535, 2005.
[37] S.A. Silling and F. Bobaru, Peridynamic modeling of membranes and fibers, Int. J. Non-
linear Mech., 40, 395–409, 2005.
[38] S.A. Silling, M. Zimmermann and R. Abeyaratne, Deformation of a peridynamic bar,
J. Elasticity, 73, 173–190, 2003.
[39] R. Temam and A. Miranville, Mathematical Modeling in Continuum Mechanics, Cam-
bridge University Press, 2001.
[40] J. Wang and R.S. Dhaliwal, Uniqueness in generalized nonlocal thermoelasticity,
J. Thermal Stresses, 16, 71–77, 1993.
[41] J. Wang and R.S. Dhaliwal, On some theorems in the nonlocal theory of micropolar
elasticity, Int. J. Solids Structures, 30(10), 1331–1338, 1993.
[42] O. Weckner and R. Abeyaratne, The effect of long-range forces on the dynamics of a
bar, J. Mech. Phys. Solids, 53(3), 705–728, 2005.
[43] O. Weckner and E. Emmrich, Numerical simulation of the dynamics of a nonlocal,
inhomogeneous, infinite bar, J. Comp. Appl. Mech., 6(2), 311–319, 2005.
[44] J. Wloka, Partial Differential Equations, Cambridge University Press, 1987.
[45] M. Zimmermann, A Continuum Theory With Long-Range Forces for Solids, PhD The-
sis, Massachusetts Institute of Technology, Department of Mech. Engineering, 2005.