You are on page 1of 24

Numerical analysis of a nonlocal Volterra integro-

differential equation
Yan Zhu
Shandong University
Tiantian Dong
Qilu University of Technology
Xiangcheng Zheng
Shandong University
Yiqun Li
University of South Carolina
Sihao Jia
Shandong University
Xu Guo (  guoxu@sdu.edu.cn )
Shandong University

Research Article

Keywords: nonlocal Volterra integro-differential equation, peridynamic differential operator, nite element
method, quadrature rule, error estimate

Posted Date: February 13th, 2024

DOI: https://doi.org/10.21203/rs.3.rs-3939627/v1

License:   This work is licensed under a Creative Commons Attribution 4.0 International License.
Read Full License

Additional Declarations: No competing interests reported.


Numerical analysis of a nonlocal Volterra
integro-differential equation
Yan Zhu1 , Tiantian Dong2,3 , Xiangcheng Zheng4 , Yiqun Li5 ,
Sihao Jia1 , Xu Guo1*
1* Geotechnical and Structural Engineering Center, Shandong University,
Jinan, 250061, China.
2 School of Mathematics and Statistics, Qilu University of Technology

(Shandong Academy of Sciences), Jinan, 250353, China.


3 Shandong Artificial Intelligence Institute, Qilu University of

Technology (Shandong Academy of Sciences), Jinan, 250014, China.


4 School of Mathematics, Shandong University, Jinan, 250100, China.
5 Department of Mathematics, University of South Carolina, Columbia,

SC 29208, USA.

*Corresponding author(s). E-mail(s): guoxu@sdu.edu.cn;


Contributing authors: zhuyan0077@126.com;
10431220893@stu.qlu.edu.cn; xzheng@sdu.edu.cn; yiqunl@email.sc.edu;
202335090@mail.sdu.edu.cn;

Abstract
We study the numerical approximation to a nonlocal Volterra integro-differential
equation, in which the integral term is the convolution product of a positive-
definite kernel and a nonlocal peridynamic differential operator. Compared with
the classical differential operators, the nonlocal peridynamic differential operators
describe, e.g., discontinuities, and have domonstrated more widespread applica-
tions. The equation is discretized in space by the Galerkin finite element method,
and we accordingly prove its error estimate. We then discretize the equation in
time by the backward Euler method, and a positive quadrature rule is combined
to approximate the convolution term. The convergence rate of the fully-discrete
finite element scheme is proved, and numerical experiments are carried out to
substantiate the theoretical findings.

Keywords: nonlocal Volterra integro-differential equation, peridynamic differential


operator, finite element method, quadrature rule, error estimate

1
1 Introduction
Nonlocal Volterra integro-differential equation (VIDE) is a mathematical model that
plays an important role in science and engineering fields such as biology, physics
and finance. In particular, some phenomena involving materials with memory can be
described by the boundary value problems of nonlocal VIDE in [1–4, 46]. This kind
of equation often appears in the form of convolution in practical application, and has
been widely used in many scientific fields such as wave propagation theory, physical
heat conduction and super-flow theory [30, 43, 44].
The classical Volterra integro-differential models that have been widely investi-
gated arise from the theory of viscoelastic material [8, 29]. An important feature of
the classical Volterra integro-differential equation is that its integral term is the con-
volution of positive definite kernel and elliptic partial differential operator [3], which
is based on the continuity assumption of the function. When a discontinuity (crack)
occurs, the basic mathematical structure of the equation breaks down. Therefore, in
the study of discontinuity, it is of significant importance to find a method that can
better describe the real physical process. The nonlocal peridynamic differential oper-
ator can better deal with material discontinuities [10–13], such as material damage as
well as crack propagation. Combining the nonlocal peridynamic differential operator
with the classical model, we can try to provide a new solution to crack propagation
problems in viscoelastic materials.
It is well known that the definition of derivative of classical differential operator
is local, and the main difference between peridynamic differential operator and differ-
ential operator is its nonlocality [5–7, 14, 45]. The advantage of the nonlocal VIDE
is that its integral term is the convolution product of a positive-definite kernel and a
nonlocal peridynamic differential operator. In view of the inadequacies of local theo-
ries, Silling proposes a new nonlocal theory that does not include spatial derivatives,
namely the theory of peridynamic (PD) [15, 16]. Based on the idea of nonlocal action
of PD, a peridynamic differential operator (PDDO) is proposed in [17]. PDDO is a
numerical method based on the nonlocal idea [18, 19], which establishes the connec-
tion between local partial derivatives and nonlocal integrals. The partial differential
equation (PDE) can be reconstructed from local differential form to nonlocal integral
form by using PDDO. The work in [20] develops a weak form of PDDO that solves
the bond association of PDE and eliminates the zero-energy mode by introducing the
near-field range of the bond association [21]. There has been some works on mathemat-
ical and numerical analysis of the nonliner VIDE, and their mathematical properties
have also been studied in [9, 22–26].
We propose and analyze a numerical approximation to the initial-boundary value
problem for nonlocal Volterra integro-differential equation
Z t
ut (x, t) − β (t − s) Lδ u (x, s) ds = f (x, t) , (x, t) ∈ Ω × (0, T ] ;
0
(1)
u (x, t) = 0, (x, t) ∈ BΩ × (0, T ] ; u (x, 0) = v (x) , x ∈ Ω.

Here, u = u(x, t), ut denotes ∂u/∂t, β is a positive-definite kernel function, i.e.,


β ∈ L1,loc (0, T ], and the space L1,loc (0, T ] is a function space with local integrability

2
on the interval (0, T ]. The nonlocal peridynamic differential operator (PDDO) is given
by [27, 28] Z

−Lδ u (x, t) = −2 γ(x, y) u(y, t) − u(x, t) dy, (2)
Ω∪BΩ
where γ(x, y) : Rd × Rd → R0+ be a symmetric kernel function, i.e., γ(x, y) = γ(y, x).
In this paper, we develop the semi-discrete and the fully-discrete finite element
approximations to the model (1) and accordingly prove their error estimate. The rest
of the paper is organized as follows: in Section 2, we introduce the notations and
some preliminaries for the peridynamic differential operator. In Section 3, we discretize
the equation in space by the Galerkin finite element method and in time by the
backward Euler method, and a positive quadrature rule is combined to approximate
the convolution term. In Section 4, we prove the error estimate for the semi-discrete
and the fully-discrete finite element scheme. In Section 5, numerical experiments are
carried out to substantiate the theoretical findings.

2 Preliminaries
2.1 Notations
We come up with some preliminaries for later use. For the positive integer m and a
real number 1 ≤ p ≤ ∞, let Lp (Ω) be the space of p-th Lebesgue integrable function
on Ω and H m (Ω) be the Hilbert spaces of L2 (Ω) whose derivative is a function of order
m. Let H m (Ω) := W m,2 (Ω) and H0m (Ω) be the completion of C0m (Ω), where C0m (Ω)
is the space of infinitely differentiable functions with compact support in Ω, and in
H m (Ω) [31–33]. For simplicity, we drop the notation Ω in spaces and corresponding
norms in the rest of the work. For a positive integer r and real-valued functions v,
Z 1/2  X 1/2
2
∥v∥ = ∥v∥L2 = v 2 dx , ∥v∥r = ∥v∥H r = ∥Dα v∥ , (3)
Ω |α|≤r

α1 α
with α = α1 , · · · , αm ), Dα = ∂/∂x1 ) · · · (∂/∂xm ) m , express an arbitrary deriva-
m
X
tive with respect to x of order |α| = αi , so that the sum in (3) contains derivatives
i=1
of all orders not exceeding r [34, 35].

2.2 Nonlocal peridynamic differential operator problem


We set up some of the symbols used throughout the paper and present the relevant
lemmas and results. Before we start to discuss the problem (1), we introduce the strong
form of a nonlocal volume-constrained Poisson problem [28, 36–39]: given b : Ω → R,
find g : Ω ∪ BΩ → R, such that

−Lδ g (x) = b (x) , x ∈ Ω; g(x) = 0, x ∈ BΩ, (4)

3
where Lδ is the nonlocal PDDO, i.e.,
Z
−Lδ g (x, t) = −2 γ(x, y) g(y, t) − g(x, t))dy. (5)
Ω∪BΩ

Let γ(x, y) be supported by a norm-induced ball H(x, δ) of radius δ centered on x,


which has a bounded supported non-negative kernel [40–42], i.e.,

H(x, δ) := supp(γ(x, ·)) = y ∈ Rd : |y − x| ≤ δ . (6)

Let Ω ⊂ Rd be a bounded open domain. Its associated interaction domain is defined as


an external set of points, with the Ω interacting with the internal points (see Figure 1),
i.e., 
BΩ := y ∈ Rd \ Ω : ∃x ∈ Ω such that |y − x| ≤ δ . (7)

Fig. 1 Interaction domain of the small elliptic domain Ω (in white) with its corresponding thickness
δ (in green). Yellow is the two balls of radius δ centered on two points in Ω ∪ BΩ.

For the function w(x) defined by x ∈ Ω ∪ BΩ, the following function spaces are
defined [28]: 
V (Ω ∪ BΩ) := w ∈ L2 (Ω ∪ BΩ) : |||w||| < ∞ ,
and we define the norm
Z Z
2 2 2
|||w||| = |w (y)−w (x)| γ (x, y) dydx + ∥w∥L2 (Ω∪BΩ)
Ω∪BΩ Ω∪BΩ
2
= P (w, w) + ∥w∥L2 (Ω∪BΩ) ,

where P (·, ·) is the bilinear form associated with the nonlocal peridynamic differential
operator. We choose the test function v(x) : Ω ∪ BΩ → R to get
Z Z
P (u (x, t) , v (x)) = [v (y)−v (x)] γ (x, y) [u (y, t)−u (x, t)] dydx. (8)
Ω∪BΩ Ω∪BΩ

4
We also introduce the constrained energy space

V0 (Ω ∪ BΩ) := {w ∈ V (Ω ∪ BΩ) : w|ΩB = 0},

the norm
2
|||w|||0 = P (w, w).
We write the weak form of problem (4),

P (g, ψ) = (b, ψ) , ∀ψ ∈ H01 , (9)

and then the problem of approximation is to find a function gh ∈ Sh ,

P (gh , χ) = (b, χ) , ∀χ ∈ Sh , (10)

let Sh be a family of the finite-dimensional subspaces of H01 (Ω). In applications, h


is typically the maximal length of a finite element under the finite element space
definition.
Notice that, as a result of (9) and (10),

P (gh − g), χ = 0, ∀χ ∈ Sh . (11)

For the one-dimensional case with a uniform partition and the following kernel
function 
 1 , |y − x| ≤ δ
γ(x, y) = δ 3 ,
 0, |y − x| > δ
the following estimate holds.
Lemma 1. [28, Theorem 5.3] Assume we have a uniform grid in one space dimension
and δ = h. Then for g ∈ H 2 , we have

∥g − gh ∥H 1 ≤ Ch ∥g∥H 2 , (12)

p dependent on ∥g∥H 2 , and is independent of h and δ.


where C > 0 is a constant
We write ∥w∥V = P (w, w) for w ∈ V , note that H 1 is continuously embedded
in V , we have
∥w∥V ≤ C ∥w∥H 1 , ∀w ∈ H 1 , (13)
where the constant C is independent of δ. We define the peridynamic projection
(denoted by PD projection for simplicity) Rh : H01 (Ω) → Sh by

P (Rh v, χ) = P (v, χ), ∀ χ ∈ Sh , for v ∈ H01 . (14)

We assume the following approximation property of the PD projection

∥Rh v − v∥H 1 ≤ Ch ∥v∥H 2 , ∀v ∈ H 2 ∩ H01 , (15)

which clearly holds true for the case considered in Lemma 1.

5
3 Discretization scheme
The purpose of this section is to investigate the standard Galerkin finite element
scheme for the approximate solution of the initial boundary value problem of model
(1).

3.1 Discretization in space


We consider the one-dimensional case, using the Galerkin finite element method. For
any t ≥ 0, finite dimensional Sh ⊂ H01 (Ω), we take the test function χ ∈ H01 (Ω), where
the Ω ⊂ R be a bounded open domain. The weak form of the original problem can be
written as:
Z t
(ut , χ) + β(t − s)P (u (s) , χ)ds = (f, χ) , ∀χ ∈ H01 (Ω), t ≥ 0, (16)
0

we say that a function u = u(x, t) is a weak solution of equation (1) if u(·, 0) = v.


For any t ≥ 0, write the discrete forms of the problem uh : [0, +∞) → Sh by
Z t
(uh,t , χ) + β(t − s)P (uh (s), χ)ds = (f, χ), ∀χ ∈ Sh , t ≥ 0, uh (0) = vh , (17)
0

where vh is an appropriate approximation of v in Sh .

3.2 Discretization in time


We shall now consider the discretization in time of the above spatial semi-discrete
problem (17). We introduce a time step k, set tn = nk, where n is a non-negative
integer, and let U n ∈ Sh be the approximation of uh (tn ) where the time derivative in
(17) adopts the backward difference quotient, i.e., ∂t U n = (U n − U n−1 )/k.
n
 X
Set Dn φ(x, t) = ωnj φj , as the discrete quadrature formula,
j=0

n Z
 X tn
Dn φ(x, t) = ωnj φj ≈ β(tn − s)φ(x, s)ds, with φj = φ(x, tj ). (18)
j=0 0

Replace the integral in (1) with a finite sum, and then we get the fully-discrete
problem

(∂¯t U n , χ) + Dn (P (U, χ)) = (fn , χ), for χ ∈ Sh , n ≥ 1, U 0 = vh . (19)

We introduce the quadrature error


Z tn
 
ϵn φ(x, t) = Dn φ(x, t) − β(tn − s)φ(x, s)ds, (20)
0

6
and we then introduce the global quadrature error which will be use in the following
estimate
XN
 
ϵ̃N φ(x, t) = k ϵn φ(x, t) 1 , for tN ≤ T, (21)
n=1
1
when φ ∈ C([0, T ], H (Ω)).
We assume that all functions that occur are real-valued and satisfy
Z T Z t
β (t − s) ϕ(s)dsϕ(t)dt ≥ 0, ∀T > 0, ϕ ∈ C[0, T ]. (22)
0 0

The kernel β is smooth for t ≥ 0 and satisfies

β ∈ L1 (0, T ), and (−1)j β (j) (t) ≥ 0 for t > 0, j = 0, 1, 2. (23)

Before estimating the error between the discrete solution U N and the exact solution
u, we first prove a stability result for the completely discrete problem. As we mentioned
in (22) where we discussed stability, naturally, for the quadratic forms analogous to
T
the quadratic integral in (22), and for any ∀Φ = (Φ0 , . . . , ΦN ) , we have

N
X N X
X n
QN (Φ) = k Dn (Φ)Φn = k ωnj Φj Φn ≥ 0. (24)
n=1 n=1 j=0

We shall call such a quadrature formula positive. We note that, in general, if some
ωn0 are non-zero, then (24) cannot hold because QN (Φ) lacks a quadratic term in Φ0 .
Therefore, in our first result we may assume that ωn0 = 0 for n ≥ 1.
We find that ωnn > 0 is a necessary condition for Dn to be positive when n ≥ 1.
This shows that the matrix coefficient of U n in (19) contains a term in the stiffness
matrix corresponding to P .
Take the value φ(x, tj ) in (tj−1 , tj ], and thus use
n Z n
 X tj X
Dn φ(x, t) = β(tn − s)φ(x, tj )ds = κn−j φ(x, tj ), (25)
j=1 tj−1 j=1

where Z tj+1
κj = β(s)ds. (26)
tj

Lemma 2. Assume that β ∈ L1 (0, T ) and satisfies


 (23). Then the integration rule
(25) and (26) is positive see [3, Lemma 4.7] , we have

N
X Z tN

k ϵn φ ≤ C1 k ∥φt ∥ds, for tN ≤ T, (27)
n=1 0

where C1 > 0 is a constant.

7
Theorem 1. Under the assumption of Lemma 2, we have

N
X Z tN
 
ϵ̃N φ = k ϵn φ 1
≤ CT k ∥φt ∥1 ds. (28)
n=1 0


Proof. The sequence {κj }j=0 is convex since, due to the convexity of β(t),
Z tj+1
κj+2 − 2κj+1 + κj = {β (s + 2k) − 2β (s + k) + β(s)} ds ≥ 0. (29)
tj

The κj are also bounded, and the quadrature formula is positive in the case of a
smooth β.
We note that
N
X N
X N
X
  
k ϵn φ(x, t) 1
≤k ϵn φ(x, t) +k ∂x ϵn φ(x, t) = ε1 + ε2 ,
n=1 n=1 n=1

because of formula (20), we have


 
∂x ϵn φ(x, t) = ϵn φx (x, t) ,

the error estimate for ε1 is obtained from Lemma 2, and now we just need to prove ε2 .
Next, according the definition of κj we arrive at
n Z
 X tj
ϵn φx (x, t) = β(tn − s) {φx (x, tj ) − φx (x, s)} ds, (30)
j=1 tj−1

so that
n Z
X tj Z tj

ϵn φx (x, t) ≤ β(tn − s) |φx,t (x, σ)| dσds
j=1 tj−1 tj−1
n Z (31)
X tj
≤ κn−j |φx,t (x, σ)| dσ.
j=1 tj−1

Because
N
X Z tN −j+1
κn−j = β(s)ds ≤ ∥β∥L1 (0,T ) , for tN ≤ T, (32)
n=j 0

applying Minkowski’s inequality, summing n and then reversing the summing order,
we get (N )Z
N
X XN X tj

k ϵn φx (x, t) ≤ k κn−j ∥φx,t (x, s)∥ ds, (33)
n=1 j=1 n=j tj−1

8
for any tN ≤ T , we end up with

N
X Z tN

k ϵn φx (x, t) ≤ C2 k ∥φx,t (x, s)∥ ds. (34)
n=1 0

√ 1
Applying the inequality (a + b) ≤ 2(a2 + b2 ) 2 , the Theorem 1 is obtained,
Z tN Z tN Z tN
∥φt (x, s)∥ ds + ∥φx,t (x, s)∥ ds = (∥φt (x, s)∥ + ∥φx,t (x, s)∥)ds
0 0 0
√ Z tN 2 2 1
≤ 2 (∥φt (x, s)∥ + ∥φx,t (x, s)∥ ) 2 ds
0
√ Z tN
= 2 ∥φt (x, s)∥1 ds,
0

Z tN Z tN Z tN
C1 k ∥φt (x, s)∥ ds + C2 k ∥φx,t (x, s)∥ ds ≤ CT k ∥φt (x, s)∥1 ds. (35)
0 0 0

4 Error estimate
Error estimates will be derived for both the semi-discrete and fully-discrete solutions.

4.1 Semi-discrete finite element approximation


The error between the finite element function uh and the exact solution u is estimated
below.
Theorem 2. Assume that β is a positive-definite kernel, where Ph represents the
orthogonal projection of v on the Sh with respect to the inner product in L2 . Under
the appropriate regularity assumptions, for the solutions of equations (17) and (1),
 Z t 
∥uh (t) − u (t)∥1 ≤ C ∥vh − v∥1 + Ch ∥v∥2 + ∥ut ∥2 ds , for t ≥ 0. (36)
0

Proof. Write the error as the sum of the two terms

uh (t) − u(t) = (uh (t) − Rh u(t)) + (Rh u(t) − u(t)) = θ(t) + ρ(t). (37)

According to (15), we have at once:


 Z t 
∥ρ(t)∥1 = ∥u(t) − Rh u(t)∥1 ≤ Ch ∥u (t)∥2 ≤ Ch ∥v∥2 + ∥ut ∥2 ds , (38)
0

and it remains in the range of θ(t).

9
Let Ah : Sh → Sh be a positive definite discrete operator of P defined as

P (ψ, χ) = (Ah ψ, χ) , ∀ψ, χ ∈ Sh . (39)

By our definitions
Z t
(θt , χ) + β (t − s) P (θ (s) , χ) ds
0
Z t
= (uh,t , χ) + β (t − s) P (uh (s) , χ) ds
0
Z t (40)
− (Rh ut , χ) − β (t − s) P (Rh u (s) , χ) ds
0
Z t
= (f, χ) − (Rh ut , χ) − β (t − s) P (Rh u (s) , χ) ds = (ut − Rh ut , χ),
0

or Z t
(θt , χ) + β (t − s) P (θ (s) , χ) ds = −(ρt , χ), ∀χ ∈ Sh , t > 0, (41)
0
the definition of PD projection is utilized in the last step of the above (40).
Since θ ∈ Sh , we may choose χ = Ah θ(t) in (41) and conclude
Z t
P (θt , θ(t)) + β (t − s) (Ah θ (s) , Ah θ(t)) ds = − (ρt , Ah θ (t)) = −P (Ph ρt , θ (t)),
0

1 d
noting that P (θt , θ(t)) = P (θ, θ), and then integrating over [0, T ],
2 dt
Z T Z t
2
∥θ (T )∥1 + β(t − s)(Ah θ (s) , Ah θ(t))dsdt
0 0
( Z )
T
2
≤C ∥θ (0)∥1 + ∥Ph ρt ∥1 ∥θ∥1 dt .
0

It is easy to know that the dual integral is non-negative according to (22), which can
be obtained
( Z T )
∥θ (T )∥1 ≤ C ∥θ (0)∥1 + ∥Ph ρt ∥1 dt , ∀T > 0, (42)
0

and

∥θ (0)∥1 = ∥vh − Rh v∥1 ≤ ∥Rh v − v∥1 + ∥vh − v∥1 , ∥Rh v − v∥1 ≤ Ch ∥v∥2 .

10
Hence, using also the assumed boundedness of Ph in H 1 (Ω),
( Z )
T
∥θ(T )∥1 ≤ C ∥vh − v∥1 + Ch ∥v∥2 + ∥ρt ∥1 dt , for T ≥ 0. (43)
0

In summary, this leads to Theorem 2.

4.2 Fully-discrete finite element approximation


We are now ready to present our stability result for (19).
Lemma 3. Under the assumption of Theorem 2, Dn is positive, then the solution of
(19) satisfies
XN
U N 1 ≤ C ∥vh ∥1 + Ck ∥fn ∥1 . (44)
n=1

Proof. With Ah defined by (39), (Ah ψ, χ) = P  (ψ, χ), ∀ψ, χ ∈ Sh . So that, setting
χ = Ah U n in (19) and noting that Dn P (U, χ) = P Dn (U ), χ , we have
  
∂¯t U n , Ah U n + Dn P (U, Ah U n ) = fn , Ah U n ,
  
P ∂¯t U n , U n + P Dn (U ), Ah U n = P fn , U n , (45)
put into ∂¯t U = (U − U
n n n−1
)/k,

P (U n , U n ) − P (U n−1 , U n ) + kP (Dn (U ), Ah U n )
(46)
= kP (fn , U n ) = kP (Ph fn , U n ).

Now, simplify P U n−1 , U n ,

 1  1 
P U n−1 , U n ≤ P U n , U n + P U n−1 , U n−1 ,
2 2

therefore, (46) can be simplified as


   
P U n , U n − P U n−1 , U n−1 + 2k Dn (Ah U ), Ah U n = 2kP Ph fn , U n . (47)

N
X
Since QN (Φ) = k Dn (Φ) Φn ≥ 0,
n=1

N
X Z
2k(Dn (Ah U ), Ah U n ) = 2 QN (Ah U )dx ≥ 0, (48)
n=1 Ω

11
and then we have
N
X
2 2
UN 1
≤ C ∥vh ∥1 + Ck ∥Ph fn ∥1 ∥U n ∥1 . (49)
n=1

Now, with M chosen so that U M 1


= max Uj 1
, we have
0≤j≤N

M
X
2 2 2
UN 1
≤ UM 1
≤ C ∥vh ∥1 + Ck ∥Ph fn ∥1 ∥U n ∥1
n=1
(50)
M
X
≤ C ∥vh ∥1 U M 1
+ Ck ∥Ph fn ∥1 U M 1
,
n=1

so we get
M
X
UN 1
≤ UM 1
≤ C ∥vh ∥1 + Ck ∥Ph fn ∥1 . (51)
n=1
Continue to use the properties of Ph ,

N
X
UN 1
≤ C ∥vh ∥1 + Ck ∥fn ∥1 , (52)
n=1

from which the result follows.


We can now state and prove our error estimate for (19).
Theorem 3. Assume that β is positive definite, if Dn is positive and if vh is chosen
so that
∥vh − v∥1 ≤ Ch ∥v∥2 , (53)
then for the solutions of (19) and (1), we have
 Z tn 
∥U n − u(tn )∥1 ≤Ch ∥v∥2 + ∥ut ∥2 ds
0
Z tn Z tn  (54)
+ CT k ∥utt ∥1 ds + ∥−Lδ ut ∥1 ds .
0 0

Proof. Using the peridynamic projection Rh : H01 (Ω) → Sh defined in (15) we write

U n − u(tn ) = U n − Rh u(tn ) + Rh u(tn ) − u(tn ) = θn + ρn . (55)

By (38) we have
 Z tN 
n
∥ρ ∥1 = ∥u(tn )−Rh u(tn )∥1 ≤ Ch ∥u(tn )∥2 ≤ Ch ∥v∥2 + ∥ut ∥2 ds . (56)
0

12
For θn we obtain by our definitions that

∂¯t θn , χ + P (Dn (θ) , χ)
= (∂¯t U n , χ) + P (Dn (U ) , χ) − (∂¯t Rh u(tn ), χ) − P (Dn (Rh u) , χ) (57)
= (fn , χ) − (∂¯t u(tn ), χ) − (∂¯t ρn , χ) − P (Dn (u) , χ) ,

because of the following formula,


Z tn
(fn , χ) = (ut (tn ), χ) + β(tn − s)P (u (s) , χ)ds, (58)
0

we end up with
 
∂¯t θn , χ + P (Dn (θ) , χ) = ut (tn ) − ∂¯t u(tn ), χ
Z tn (59)
+ β(tn − s)P (u (s) , χ)ds − P (Dn (u) , χ) − (∂¯t ρn , χ).
0

Thus, letting

ω n = ω1n + ω2n + ω3n


Z tn 
 
¯
= (ut (tn ) − ∂t u(tn ) + β(tn − s) − Lδ u(s) ds − Dn (−Lδ u) − ∂¯t ρn ,
0

we have 
∂¯t θn , χ + P (Dn (θ) , χ) = (ω n , χ) , (60)
and applying Lemma 3,

N
X
θN 1
≤ C θ0 1
+ Ck ∥ω n ∥1 . (61)
n=1

Here,
θ0 1
= ∥vh − Rh v∥1 ≤ ∥vh − v∥1 + ∥Rh v − v∥1 ≤ Ch ∥v∥2 . (62)
Further,
N
X N
X N
X N
X
k ∥ω n ∥1 = k ∥ω1n ∥1 + k ∥ω2n ∥1 + k ∥ω3n ∥1 . (63)
n=1 n=1 n=1 n=1
The following deals with the above three items separately,

u(tj ) − u(tj−1 )
ω1j + ut (tj ) = ,
k
Z tj
kω1j = u(tj ) − u(tj−1 ) − kut (tj ) = − (s − tj−1 )utt (s)ds, (64)
tj−1

13
and so, summation,

N
X N
X
k ∥ω1n ∥1 = ∥u(tn ) − u(tn−1 ) − kut (tn )∥1
n=1 n=1
N
X Z tn
= (s − tj−1 )utt (s)ds
n=1 tn−1
1 (65)
N Z
X tn
≤ Ck ∥utt (s)∥1 ds
n=1 tn−1
Z tN
= Ck ∥utt (s)∥1 ds.
0

N
X
Now deal with the middle, e
ϵN (φ) = k ∥ϵn (φ)∥1 ,
n=1

N
X N
X
Ck ∥ω2n ∥1 = Ck ∥ϵn (−Lδ u)∥1 ≈ C e
ϵN (−Lδ u) . (66)
n=1 n=1

Finally, the third item is dealt with, ρn = Rh u(tn ) − u(tn ),

 u(tn ) − u(tn−1 )
ω3n = ∂¯t ρn = ∂¯t Rh u(tn ) − u(tn ) = (Rh − I)∂¯t u(tn ) = (Rh − I) ,
k
N
X N
X Z tn N
X Z tn
k ∥ω3n ∥1 = (Rh − I) ut (s)ds = Rh ut (s) − ut (s)ds
n=1 n=1 tn−1 n=1 tn−1
1 1
N Z
X tn N Z
X tn
≤ ∥Rh ut (s) − ut (s)∥1 ds ≤ Ch ∥ut ∥2 ds (67)
n=1 tn−1 n=1 tn−1
Z tN
= Ch ∥ut ∥2 ds.
0
Combined with Theorem 1 mentioned earlier, we get the final error estimate, for any
tn ≤ T ,  Z tn 
n
∥U − u(tn )∥1 ≤Ch ∥v∥2 + ∥ut ∥2 ds
0
Z tn Z tn 
+CT k ∥utt ∥1 ds + ∥−Lδ ut ∥1 ds .
0 0

14
5 Numerical experiments
In this section, we evaluate the spatial and temporal convergence properties of the
finite element schemes for the nonlocal Volterra integro-differential equation (1) respec-
tively. We will evaluate the accuracy of numerical solutions and test their convergence
properties in one-dimensional problems. Here are two simple examples to verify the
results.

Example 1.
we consider one-dimensional problems discretized on a uniform grid, with the
spatial domain Ω = (0, 1) and the time interval [0, T ] = [0, 0.1]. For a given hori-
zon δ = 1/8, its relevant domain is BΩ = (−δ, 0] ∪ [1, 1 + δ) (see Figure 2). Let the
tα−1
positive-definite kernel β (t) = , 0 < α < 1.
Γ (α)

Fig. 2 One-dimensional domain Ω, and associated boundaries BΩ.

Let the constant kernel function



 3 , |y − x| ≤ δ,
γ(x, y) = 2δ 3 (68)
 0, |y − x| > δ.

Firstly, we take u(x, t) = t2 x(1 − x) for (x, t) ∈ (0, 1) × [0, 0.1] and v(x) = 0 for
x ∈ (0, 1). Then we obtain the source term:



 4tα+2 

 2tx 1 − x +   , x ∈ δ, 1 − δ ,

 α α + 1 α + 2 Γ(α)



  6tα+2

 2tx 1 − x + 3   ×

 δ α α + 1 α + 2 Γ(α)

  
x2 x3 δ2 δ3 2 2

f (x, t) = − + − + + xδ − x δ + xδ , x ∈ 0, δ ,

 2 3 2 3



  6tα+2

 2tx 1 − x + 3   ×

 δ α α+1 α−2 Γ α

  2 

 x x3 δ2 δ3 1 

 − − − − xδ − x2 δ − xδ 2 − , x ∈ 1 − δ, 1 .
2 3 2 3 6

Choose the above source term, then we can use u(x, t) = t2 x(1 − x) for (x, t) ∈
(0, 1) × [0, 0.1] as the exact solutions.

15
Table 1 Convergence of the finite element schemes defined on a spatial
partition for the nonlocal Volterra integro-differental equation, for
α = 0.3, 0.5, 0.7, fixed time uniform grid k = 1 × 10−4 in Example 1.

Nx uh
Nt − u(·, T ) rate
H1
α = 0.3 24 7.95E-04 -
25 3.72E-04 1.10
26 1.82E-04 1.03
27 9.03E-05 1.01
α = 0.5 24 7.46E-04 -
25 3.46E-04 1.03
26 1.81E-04 1.01
27 9.03E-05 1.00
α = 0.7 24 7.29E-04 -
25 3.62E-04 1.01
26 1.80E-04 1.00
27 9.04E-05 1.00

Table 2 Convergence of the finite element schemes defined on a


temporal partition for the nonlocal Volterra integro-differental equation,
for α = 0.3, 0.5, 0.7, fixed space uniform grid h = 1/128 in Example 1.

Nt uh
Nt − u(·, T ) rate
H1
α = 0.3 24 5.11E-04 -
25 2.63E-04 0.96
26 1.33E-04 0.98
27 6.65E-05 1.00
α = 0.5 24 5.92E-04 -
25 3.03E-04 0.96
26 1.54E-04 0.98
27 7.70E-05 1.00
α = 0.7 24 5.99E-04 -
25 3.04E-04 0.98
26 1.53E-04 0.99
27 7.65E-05 1.00

In the numerical experiments, we compute the numerical solutions using a uniform


spatial partition of h = 1/16, 1/32, 1/64, 1/128, respectively, and a uniform temporal
mesh of k = 1 × 10−4 . We list the numerical results in Table 1. In the table, we provide
the H 1 (Ω) norms of the error and convergence rates for a sequence of spatial grid sizes.
We observe from Table 1 that for α = 0.3, 0.5, 0.7, the finite element scheme for the
nonlocal Volterra integro-differential equation (1) has a first-order spatial convergence
rate on a uniform temporal mesh of k = 1 × 10−4 . From Table 2, one can also see
that, in the case for which h = 1/128 is fixed, for α = 0.3, 0.5, 0.7, the finite element
scheme has a first-order temporal convergence rate.

16
Example 2.
Example 1 shows that we use the constant kernel function and obtain first-order
convergence, which is consistent with our theoretical prediction. Then we guess that
it also works for more complex kernel functions. Next, we will use the ration kernel
function to verify our conjecture. Compared with Example 1, only change the form
γ(x, y) to: 
 1
2 |y − x|
, |y − x| ≤ δ,
γ(x, y) = δ (69)

0 , |y − x| > δ.
We take u(x, t) = t2 x(1 − x) for (x, t) ∈ (0, 1) × [0, 0.1] and v(x) = 0 for x ∈ (0, 1).
The source term



 4tα+2 

 2tx 1 − x +   , x ∈ δ, 1 − δ ,

 α α + 1 α + 2 Γ α



  4tα+2
2tx 1 − x +
   ×

 δ2 α α + 1 α + 2 Γ α

 
 x 3x2  δ2 
f (x, t) = x x − 1 ln − δ − + 2δ + 1 x + , x ∈ 0, δ ,

 δ 2 2

  α+2

 4t

 2tx 1 − x + 2   ×

 δ α α+1 α+2 Γ α

  

  1−x 3x2  δ2 1 

 x x − 1 ln +δ− + 2 − 2δ x + − , x ∈ 1 − δ, 1 .
δ 2 2 2

We also choose the above source term, then we can use u(x, t) = t2 x(1 − x) for
(x, t) ∈ (0, 1) × [0, 0.1] as the exact solutions.

Table 3 Convergence of the finite element schemes defined on a spatial


partition for the nonlocal Volterra integro-differental equation, for
α = 0.3, 0.5, 0.7, fixed time uniform grid k = 1 × 10−4 in Example 2.

Nx uh
Nt − u(·, T ) rate
H1
α = 0.3 24 4.24E-04 -
25 1.95E-04 1.12
26 9.28E-05 1.07
27 4.53E-05 1.04
α = 0.5 24 3.82E-04 -
25 1.85E-04 1.05
26 9.09E-05 1.03
27 4.52E-05 1.01
α = 0.7 24 3.67E-04 -
25 1.82E-04 1.01
26 9.04E-05 1.01
27 4.54E-05 0.99

17
Table 4 Convergence of the finite element schemes defined on a
temporal partition for the nonlocal Volterra integro-differental equation,
for α = 0.3, 0.5, 0.7, fixed space uniform grid h = 1/128 in Example 2.

Nt uh
Nt − u(·, T ) rate
H1
α = 0.3 24 3.68E-04 -
25 1.84E-04 1.00
26 8.94E-05 1.04
27 4.12E-05 1.12
α = 0.5 24 4.34E-04 -
25 2.19E-04 0.99
26 1.08E-05 1.02
27 5.20E-05 1.06
α = 0.7 24 4.72E-04 -
25 2.38E-04 0.99
26 1.19E-04 1.00
27 5.83E-05 1.02

By observing the Tables 3 and 4, after we complicate the kernel function, the non-
local Volterra integro-differential equation (1) can also conclude that the finite element
scheme has first-order spatial convergence rate and first-order temporal convergence
rate, which again verifies the correctness of the theory.
Finally, Example 1 and Example 2 get about the same convergence rate in both
different cases, and first-order convergence is obtained for all cases considered, which
verify our theory as predicted in previous sections.
With the numerically tested convergence of the finite element approximation for
problem (1), we then conduct numerical experiments to assess the performance of the
(1) for predefined initial values of v. We set v = sin(2πx), and use the spatial mesh
size h = 1/128 and the time step size τ = 1×10−4 in the finite element approximation,
and investigate the different behavior of problem (1) at α = 0.3, 0.5, 0.7. By observing
Figures 3 and 4, under the discretization of the finite element scheme, the solutions of
model (1) under different kernels have roughly the same regularity.

Fig. 3 Left to right: the solution surfaces of the equation (1) with a constant kernel.

18
Fig. 4 Left to right: the solution surfaces of the equation (1) with a ration kernel.

6 Conclusion
This paper presents and analyzes a new nonlocal Volterra integro-differential model
and its effective numerical discretization. In particular, the peridynamic differential
operator is nonlocal. Though it may appear first that the nonlocal model involve
more complex definitions of nonlocal interactions, they have advantages in preserving
important aspects of physical features such as mass conservation and the maximum
principle.
In terms of discretization, we use Galerkin finite element method in space and back-
ward Euler method in time, and get the corresponding error estimates. Two numerical
examples are given to verify the convergence order. We obtain that the finite element
scheme of the nonlocal equation has the convergence rate of first-order space and time.
At the same time, the accuracy is achieved and the characteristics of good nonlocal
model are maintained.

Funding. This work was supported by the National Natural Science Foun-
dation of China (No. 12301555), the National Key R&D Program of China
(No. 2023YFA1008903), the Taishan Scholars Program of Shandong Province (No.
tsqn202306083), the National Natural Science Foundation of China grants 12271303,
the Major Fundamental Research Project of Shandong Province of China (No.
ZR2023ZD33), and the Taishan Scholars Program of Shandong Province of China (No.
tsqn201909044).

Declarations
Conflict of interest. The authors declare no competing interests.

19
References
[1] Maccamy, R. C. A model for one-dimensional, nonlinear viscoelasticity. Quart.
Appl. Math. 35, 21–33 (1977).

[2] Maccamy, R. C. An integro-differential equation with application in heat flow.


Quart. Appl. Math. 35, 1–19 (1977).

[3] Mclean, W. & Thomée, V. Numerical solution of an evolution equation with a


positive-type memory term. J. Aust. Math. Soc. Ser. 35, 23–70 (1993).

[4] Mclean, W., Sloan, I. H. & Thomée, V. Time discretization via laplace transfor-
mation of an integro-differential equation of parabolic type. Numer. Math. 102,
497–522 (2006).

[5] Dorduncu, M., Kutlu, A., Madenci, E. & Rabczuk, T. Nonlocal modeling of bi-
material and modulus graded plates using peridynamic differential operator. Eng.
Comput. 39, 893–909 (2022).

[6] Tian, H., Ju, L. & Du, Q. A conservative nonlocal convection–diffusion model
and asymptotically compatible finite difference discretization. Comput. Methods
Appl. Mech. Eng. 320, 46–67 (2017).

[7] Tian, H., Ju, L. & Du, Q. Nonlocal convection-diffusion problems and finite
element approximationsn. Comput. Methods Appl. Mech. Eng. 289, 60–78 (2015).

[8] Zheng, X. & Wang, H. Discretization and analysis of an optimal control of a


variable-order time-fractional diffusion equation with pointwise constraints. J.
Sci. Comput. 91 (2022).

[9] Ciarlet, P. G. The Finite Element Method for Elliptic Problems (SIAM, America,
1978).

[10] Du, N., Guo, X. & Wang, H. A fast state-based peridynamic numerical model.
Commun. Comput. Phys. 27, 274–291 (2020).

[11] Lu, F., An, Q. & Yu, Y. Nonparametric learning of kernels in nonlocal operators.
J. Peridyn. Nonlocal Model. (2023).

[12] Wang, H. & Tian, H. A fast galerkin method with efficient matrix assembly and
storage for a peridynamic model. J. Comput. Phys. 231, 7730–7738 (2012).

[13] Yin, M., Zhang, E., Yu, Y. & Karniadakis, G. E. Interfacing finite elements
with deep neural operators for fast multiscale modeling of mechanics problems.
Comput. Methods Appl. Mech. Eng. 402 (2022).

[14] Pang, G., D’Elia, M., Parks, M. & Karniadakis, G. E. nPINNs: Nonlocal
physics-informed neural networks for a parametrized nonlocal universal Laplacian

20
operator. Algorithms and applications. J. Comput. Phys. 422 (2020).

[15] Silling, S. A. Reformulation of elasticity theory for discontinuities and long-range


forces. J. Mech. Phys. Solids 48, 175–209 (1998).

[16] Yu, Y., Bargos, F. F., You, H., Parks, M. L., Bittencourt, M. L. & Karniadakis,
G. E. A partitioned coupling framework for peridynamics and classical theory:
Analysis and simulations. Comput. Methods Appl. Mech. Eng. 340, 905–931
(2018).

[17] Madenci, E., Barut, A. & Futch, M. Peridynamic differential operator and its
applications. Comput. Methods Appl. Mech. Eng. 304, 408–451 (2016).

[18] Du, Q., Gunzburger, M., Lehoucq, R. B. & Zhou, K. A nonlocal vector calculus,
nonlocal volume-constrained problems, and nonlocal balance laws. Math. Models
Methods Appl. Sci. 23, 93–540 (2013).

[19] Madenci, E., Barut, A. & Dorduncu, M. Peridynamic Differential Operator for
Numerical Analysis (Springer, Berlin, 2019).

[20] Li, Z., Huang, D., Ren, H. & Rabczuk, T. Weak form of bond-associated peridy-
namic differential operator for solving differential equations. Eng. Comput. 39,
3491–3507 (2022).

[21] Chen, H. Bond-associated deformation gradients for peridynamic correspondence


model. Mech. Res. Commun. 90, 34–41 (2018).

[22] Dafermos, C. M. & Nohel, J. A. Energy methods for nonlinear hyperbolic volterra
integro-differential equations. Commun. Partial Differ. Equ. 4, 219–278 (1978).

[23] Londen, S. O. On an integrodifferential volterra equation with a maximal


monotone mapping. J. Differ. Equ. 27, 405–420 (1978).

[24] Maccamy, R. C. & Wong, J. S. W. Stability theorems for some functional


equations. Trans. Amer. Math. Soc. 164, 1–37 (1972).

[25] Qiu, W., Chen, Y., Xiao, X. & Zheng, X. Discrete L1 remainder stability of
first and second order schemes for a volterra integro-differential equation. Math.
Comput. Simul. 219, 12–27 (2024).

[26] Staffans, O. J. On a nonlinear hyperbolic volterra equation. SIAM J. Math. 11,


739–812 (2006).

[27] D’Elia, M., Gulian, M., Olson, H. & Karniadakis, G. E. Towards a unified theory
of fractional and nonlocal vector calculus. Fract. Calc. Appl. Anal. 24, 1301–1355
(2020).

21
[28] Pasetto, M., Shen, Z., D’Elia, M., Tian, X., Teask, N. & Kamensky, D. Efficient
optimization-based quadrature for variational discretization of nonlocal problems.
Comput. Methods Appl. Mech. Eng. 396 (2022).

[29] Guo, X. & Zheng, X. Variable-order time-fractional diffusion equation with


mittag-leffler kernel: regularity analysis and uniqueness of determining variable
order. Z. Angew. Math. Phys. 74, 64 (2023).

[30] Zheng, X. & Wang, H. Analysis and discretization of a variable-order fractional


wave equation. Commun. Nonlinear Sci. Numer. Simul. 104, 106047 (2022).

[31] Adams, R. A. & Fournier, J. F. Sobolev Spaces (Elsevier, San Diego, 2003).

[32] Bonito, A., Borthagray, J. P., Nochetto, R. H. & Otárola, E. Numerical methods
for frantional diffusion. Comput. Vis. Sci. 19, 19–46. (2018).

[33] Bonito, A. & Pasciak, J. E. Numerical approximation of fractional powers of


elliptic operators. Math. Comput. 84, 491–512 (2013).

[34] Evans, L. C. Partial Differential Equations (American Mathematical Society,


America, 1998).

[35] Thomée, V. Galerkin finite element methods for parabolic problems (Springer,
Berlin, 1984).

[36] D’Elia, M., Du, Q., Glusa, C., Gunzburger, M., Tian, X. & Zhou, Z., Numerical
methods for nonlocal and fractional models. Acta Numer. 29, 1–124 (2020).

[37] Du, Q., Gunzburger, M., Lehoucq, R. B. & Zhou, K. Analysis and approximation
of nonlocal diffusion problems with volume constraints. SIAM Rev. 54, 667–696
(2012).

[38] Gunzburger, M. & Lehoucq, R. B. A nonlocal vector calculus with application to


nonlocal boundary value problems. Multiscale Model. Simul. 8, 1581–1598 (2010).

[39] Leng, Y., Tian, X., Trask, N. & Foster, J. T. Asymptotically compatible repro-
ducing kernel collocation and meshfree integration for nonlocal diffusion. SIAM
J. Numer. Anal. 59, 88–118 (2021).

[40] D’Elia, M., Du, Q., Gunzburger, M. & B, L. R. Nonlocal convection-diffusion


problems on bounded domains and finite-range jump processes. Comput. Methods
Appl. Mech .Eng. 17, 707–722 (2017).

[41] Felsinger, M., , Kassmann, M. & Voigt, P. The dirichlet problem for nonlocal
operators. Math. Z. 279, 779–809 (2015).

[42] Mengesha, T. & Du, Q. Analysis of a scalar nonlocal peridynamic model with a
sign changing kernel. Discrete Contin. Dyn. Syst. Ser. B 18, 1415–1437 (2013).

22
[43] Xu, B. B., Gao, X. W. & Cui, M. An efficient and accurate hybrid weak-form
meshless method for transient nonlinear heterogeneous heat conduction problems.
Eng. Comput. 38, 969–984 (2020).

[44] Wang, X., Chen, Y. & Yu, J. Wave propagation in viscoelastic functionally graded
nanoplates: Comparison of the integral and differential nonlocal models. Acta
Mech. Solida Sin. 36, 724–733 (2023).

[45] Wang, J., Yang, Z. & Zhang, J. Stability and convergence analysis of high-order
numerical schemes with dtn-type absorbing boundary conditions for nonlocal
wave equations. IMA J. Appl. Math. 44, 604–632 (2022).

[46] Chen, H., Qiu, W., Zaky, M. A. & Hendy, A. S. A two-grid temporal second-order
scheme for the two-dimensional nonlinear volterra integro-differential equation
with weakly singular kernel. Calcolo 60 (2023).

23

You might also like