P .I .C . M .

– 2018
Rio de Janeiro, Vol. 4 (3559–3588)

AN INVITATION TO NONLOCAL MODELING, ANALYSIS

AND COMPUTATION
Q D (杜强)

Abstract
This lecture serves as an invitation to further studies on nonlocal models, their
mathematics, computation, and applications. We sample our recent attempts in the
development of a systematic mathematical framework for nonlocal models, includ-
ing basic elements of nonlocal vector calculus, well-posedness of nonlocal variational
problems, coupling to local models, convergence and compatibility of numerical ap-
proximations, and applications to nonlocal mechanics and diffusion. We also draw
connections with traditional models and other relevant mathematical subjects.

1 Introduction
Nonlocal phenomena are ubiquitous in nature but their effective modeling and simula-
tions can be difficult. In early mathematical and scientific inquiries, making local approx-
imations has been a dominant strategy. Over centuries, popular continuum models are
presented as partial differential equations (PDEs) that are expressed by local information
in an infinitesimal neighborhood and are derived, in their origins, for smooth quantities.
Entering into the digital age, there have been growing interests and capabilities in the
modeling of complex processes that exhibit singularities/anomalies and involve nonlocal
interactions. Nonlocal continuum models, fueled by the advent in computing technology,
have the potential to be alternatives to local PDE models in many applications, although
there are many new challenges for mathematicians and computational scientists to tackle.
While mathematical analysis and numerical solution of local PDEs are well estab-
lished branches of mathematics, the development of rigorous theoretical and computa-
tional framework for nonlocal models, relatively speaking, is still a nascent field. This
Supported in part by the U.S. NSF grants DMS–1719699, AFOSR FA9550-14-1-0073 MURI Center for Ma-
terial Failure Prediction Through Peridynamics, OSD/ARO/MURI W911NF-15-1-0562 on Fractional PDEs for
Conservation Laws and Beyond: Theory, Numerics and Applications, and the Columbia University.
MSC2010: primary 45P05; secondary 65R20, 35R09, 74G15, 76M25, 47G10.
Keywords: Nonlocal operators, nonlocal function spaces, nonlocal vector calculus, numerical methods,
asymptotic compatibility, heterogeneous localization, nonlocal in time dynamics.

3559
 3560 QIANG DU (杜强)

lecture serves as an invitation to further studies on this emerging subject. We refer to

Du [n.d.], an NSF-CBMS monograph, for a review on the historical development, recent
progress and connections with other mathematical subjects such as multiscale analysis,
calculus of variations, functional analysis, fractional PDEs, differential geometry, graph,
data and image analysis, deep learning, as well as various applications. Instead of a brief
survey, we present here samples of our recent attempts to develop a systematic mathemati-
cal framework for nonlocal models, including some basic building blocks, algorithms and
applications. In particular, our discussions are centered around nonlocal models with a
finite range of interactions typically characterized by a horizon parameter ı. Their local
(ı ! 0) and global (ı ! 1) limits offer natural links to local and fractional PDEs and
their discretization are also tied with graph operators, point clouds and discrete networks.
A few questions on nonlocal modeling, analysis and computation are addressed here: how
do nonlocal models compare with local and discrete models and how are they connected
with each other? what are the ingredients of nonlocal vector calculus? how to develop
robust discretization of nonlocal models that are asymptotically compatible with their lo-
cal limit? how to get well defined trace maps in larger nonlocal function spaces to couple
nonlocal and local models? and, how to explain the crossover of diffusion regimes using
nonlocal in time dynamics? It is our intention to demonstrate that studies on nonlocal
modeling not only provoke the discovery of new mathematics to guide practical model-
ing efforts, but also provide new perspectives to understand traditional models and new
insight into their connections.

2 Modeling choices and emergence of nonlocal modeling

Mathematical models have various types, e.g., discrete or continuum, and deterministic
or stochastic. Historically, influenced by the great minds like Newton, Leibniz, Maxwell
and others, most popular continuum models are those given by PDEs whose simple close-
form or approximate solutions have often been utilized. As more recent human endeavors,
nevertheless, computer simulations have made discrete models equally prominent.
We consider some simple continuum and discrete equations as illustrations. Let u =
u(x) be a function to be determined on a domain (an interval) Ω  R. The differential
equation
d 2u
L0 u(x) = (x) = f (x; u(x)); 8 x 2 Ω
dx 2
with a prescribed function f = f (x; u), represents a local continuum model: it only in-
volves the value and a few derivatives of the solution at any single point x. By introducing
a set of grid points fxj g in Ω, equally spaced with a grid spacing h and the standard 2nd
order center difference operator Lh = Dh2 on the grid. We then have a discrete difference
 NONLOCAL MODELS 3561

model
u(xj + h) 2u(xj ) + u(xj h)
Lh u(xj ) = Dh2 u(xj ) = = f (xj ; u(xj )) 8xj
h2
that, as h ! 0, approximates the local continuum model. In comparison, we consider

(1) Lı u(x) = f (x; u(x)); x 2 Ω;

which is a nonlocal model Du [2015, 2017b] and Du and X. Tian [2015] with a nonlocal
operator Lı defined, for a prescribed nonlocal interaction kernel ! ı associated with a
given horizon parameter ı > 0, by
ˆ ı
u(x + s) 2u(x) + u(x s)
(2) Lı u (x) = ! ı (s)ds:
0 s2
The model (1) is generically nonlocal, particularly if the support of ! ı extends beyond
the origin. It, at any x, involves function values of u at not only x but possibly its ı-
neighborhood. With ! ı = ! ı (s) a probability density function, Lı can be interpreted as
a continuum average (integral) of the difference operator Ds2 over a continuum of scales
s 2 [0; ı]. This interpretation has various implications as discussed below.
First, differential and discrete equations are special cases of nonlocal equations: let
d2 2
! ı (s) be the Dirac-delta measure at either s = 0 or h, we get L0 = dx 2 or Lh = Dh

respectively, showing the generality of nonlocal continuum models. A better illustration

is via a limiting process, e.g., for smooth u, small ı, and ! ı going to the Dirac-delta at
s = 0, we have
ˆ ı
d 2u d 4u d 2u
Lı u (x) = (x) ! ı (s)ds + c2 ı 2 (x) +


 = L0 u(x)
dx 2 0 dx 4 dx 2
showing that nonlocal models may resemble their local continuum limit for smooth quan-
tities of interests (QoI), while encoding richer information for QoIs with singularity.
In addition, with a special class of fractional power-law kernel ! ı (s) = c˛;1 jsj1 2˛ for
0 < ˛ < 1 and ı = 1, Lı leads to a fractional differential operator Bucur and Valdinoci
[2016], Caffarelli and Silvestre [2007], Nochetto, Otárola, and Salgado [2016], Vázquez
[2017], and West [2016]:
ˆ 1  ˛
d2
L1 u (x) = c˛;1 Ds2 u(x)jsj1 2˛ ds = u(x):
0 dx 2

One may draw further connections from the Fourier symbols of these operators Du [n.d.]
and Du and K. Zhou [2011].
 3562 QIANG DU (杜强)

Nonlocal models and operators have many variations and extensions. For example,
one may define a nonlocal jump (diffusion) operator for a particle density u = u(x),
ˆ
Lı u (x) = (ˇ(x; y)u(y) ˇ 0 (y; x)u(x)) dy ;

with ˇ = ˇ(x; y) and ˇ 0 = ˇ 0 (y; x) the jumping rates. We can recover (2) if ˇ(x; y) =
ˇ 0 (y; x) = jx yj 2 ! ı (jx yj), and make connections with stochastic processes Du,
Huang, and Lehoucq [2014].
Other extensions include systems for vector and tensor fields such as nonlocal models
of mechanics. A representative example is the peridynamic theory Silling [2000] which at-
tempts to offer a unified treatment of balance laws on and off materials discontinuities, see
Bobaru, Foster, Geubelle, and Silling [2017] for reviews on various aspects of peridynam-
ics. We briefly describe a simple linear small strain state-based peridynamic model here.
Let Ω be either Rd or a bounded domain in Rd with Lipshitz boundary, and Ω = Ω [ ΩI
where ΩI is an interaction domain. Let u = u(x; t ) = y(x; t ) x denote the displacement
field at the point x 2 Ω and time t so that y = x + u gives the deformed position, the
peridynamic equation of motion can be expressed by
ut t (x; t ) = Lı u(x; t ) + b(x; t ); 8x 2 Ω; t > 0;
where  is the constant density, b = b(x; t ) the body force, and Lı u the interaction force
derived from the variation of the nonlocal strain energy. Under a small strain assumption,
for any x; x0 = x + , the linearized total strain and dilatational strain are given by
ˆ

(3) s(u)(x0 ; x) := e()
and dı (u)(x) := !ı (x0 ; x)s(u)(x0 ; x)d x0 :
jj
where e() = /jj,  = u(x+) u(x) and the kernel !ı has its support over a spherical
neighborhood jx0 xj < ı (with ı being the horizon parameter) and is normalized by
ˆ
!ı (x0 ; x)d x0 = 1:

The linearized deviatoric strain is denoted by Sı (u)(x0 ; x) := s(u)(x0 ; x) dı (u)(x).

Then, the small strain quadratic last energy density functional is given by
ˆ
2
(4) Wı (x; f; g) = jdı (u)(x)j + !ı (x+; x) js(u)(x + ; x) dı (u)(x)j2 d 
Ω

where  represents the peridynamic bulk modulus and  the peridynamic shear modulus.
For  = , we get a nondimensionalized bond-based peridynamic energy density
Mengesha and Du [2014] and Silling [2000]
ˆ ˇ ˇ
ˇ  u(x + ) u(x) ˇ2
(5) Wı (x; f; g) = !ı (x + ; x) ˇˇ
ˇ d ;
jj jj ˇ
 NONLOCAL MODELS 3563

For a scalar function u = u(x), we get a simple one dimensional energy density
ˆ
ju(y) u(x)j2
Wı (x; fug) = ! ı (jy xj) dy ;
jy xj2
associated with the nonlocal operator in (1), if a translation invariant and even kernel ! ı
is adopted. This special case has often served as a benchmark problem for peridynam-
ics, even though in most practical applications, peridynamic models do take on nonlinear
vector forms to account for complex interactions Du, Tao, and X. Tian [2017].

3 Nonlocal vector calculus and nonlocal variational problems

We introduce the theory through an example, accompanied by some general discus-
sions.

A model equation. The systematic development of the nonlocal vector calculus was
originated from the study of peridynamics Du, Gunzburger, Lehoucq, and K. Zhou [2013].
Let us consider a time-independent linear bond-based peridynamic model associated with
the strain energy (5) given by
(6) ˆ  
u(x + ) u(x)
Lı u(x) = 2 !ı (x + ; x)
e() e()d  = b(x) ; 8 x 2 Ω;
Ω jj2
where u is a displacement field and b is a body force. Intuitively, (6) describes the force
balance in a continuum body of linear springs, with the spring force aligned with the
undeformed bond direction between any pair of points x and x0 = x + . This gives a
nonlocal analog of classical linear elasticity model with a particular Poisson ratio, yet it
does not, at the first sight, share the same elegant form of linear elasticity. Nonlocal vector
calculus can make the connections between local and nonlocal models more transparent.

Examples of nonlocal operators. Let us introduce some nonlocal operators as illustra-

tive examples. First, we define a nonlocal two-point gradient operator G for any v : Rd !
Rm such that G v : Ω  Ω ! Rd n is a two-point second-order tensor field given by,
v(x0 ) v(x) x0 x
(7) (G v)(x0 ; x) = e(x0 x)˝ where e(x0 x) = 0 ; 8 x0 ; x 2 Ω  :
jx0 xj jx xj
There are two cases of particular interests, namely, n = 1 and n = d . In the latter case,
we also define a nonlocal two-point divergence operator D by
v(x0 ) v(x)
(8) (Dv)(x0 ; x) = e(x0 x)
= Tr(G v)(x; x0 ) ; for n = d:
jx0 xj
 3564 QIANG DU (杜强)

For peridynamics, (Dv)(x0 ; x) corresponds to the linearized strain described in (3).

Next, we define a nonlocal two-point dual divergence operator D  acting on any two-
point scalar field Ψ : Ω  Ω ! R such that D  Ψ becomes a vector field given by,
ˆ

e(x0 x) 0
(9) (D  Ψ)(x) = Ψ(x; x0 ) + Ψ(x0 ; x) d x ; 8 x 2 Ω :
Ω jx0 xj

We may interpret D  and D as adjoint operators to each other in the sense that
ˆ ˆ ˆ

(10) v(x)
(D Ψ)(x)d x = (Dv)(x0 ; x)Ψ(x0 ; x)d x0 d x
Ω Ω Ω

for all v and Ψ that make integrals in (10) well defined. The duality may also be written
more canonically as (v; D  Ψ)Ω = (Dv; Ψ)Ω Ω where (
;
)Ω and (
;
)Ω Ω de-
note L2 inner products for vector and scalar fields in their respective domains of definition.
Similarly, we can define a nonlocal two-point dual gradient operator G  acting on any two-
point vector field Ψ : Ω  Ω ! Rd by the duality that (v; G  Ψ)Ω = (G v; Ψ)Ω Ω
for G given by (7).
Some basic elements of nonlocal vector calculus are listed in Table 2 in comparison
with the local counterpart. Discussions on concepts like the nonlocal flux and further
justifications on labeling G and D as two-point gradient and divergence can be found in
Du [n.d.] and Du, Gunzburger, Lehoucq, and K. Zhou [2013].

Newton’s vector calculus , Nonlocal vector calculus

Differential operators, local flux , Nonlocal operators, nonlocal flux
Green’s identity, integration by parts Nonlocal Green’s identity (duality)
ˆ ˆ ¨
,
u
∆v v
∆u = u
@n v v
@n u u
D  (Dv) v
D  (Du) = 0
Ω @Ω Ω Ω

Table 1: Elements of vector calculus: local versus nonlocal.

Reformulation of nonlocal models. Let the kernel in (6) !ı = !ı (x0 ; x) = !ı (x; x0 )

be symmetric. We consider D  Ψ with Ψ(x0 ; x) = !ı (x0 ; x)(Du)(x0 ; x). This leads to

Lı u = D  (!ı Du) = b

a concise reformation of (6) that starts to resemble, in appearance, the PDE form of clas-
sical elasticity, with local differential (gradient and divergent) operators replaced by their
nonlocal counterparts. Analogously, a scalar nonlocal diffusion equation for a translation
 NONLOCAL MODELS 3565

invariant !ı , i.e., !ı (x0 ; x) = ! ı (x0 x) = ! ı (x x0 ), and its reformulation can be given

by
(11) ˆ
v(x + ) 2v(x) + v(x )
Lı v(x) = ! ı () d  = f (x) , G  (! ı G v) = f:
Ω jj2
similar to the one-dimensional version (2). Moreover, not only nonlocal models can be
nicely reformulated like classical PDEs, their mathematical theory may also be developed
in a similar fashion along with interesting new twists Du, Gunzburger, Lehoucq, and K.
Zhou [2013, 2012] and Mengesha and Du [2014, 2015].

Variational problems. Let Ω  Rd be a bounded domain with Lipshitz boundary and

ΩI be where constraints on the solution are imposed. We consider an energy functional
¨
1
(12) Eı (u) := juj2Sı (b; u)Ω with juj2Sı := ! ı (x0 x) (Du(x0 ; x))2 d x0 d x
2 2 2
Ω Ω

for a prescribed body force b = b(x) 2 L2 (Ωı )d and a kernel ! ı = ! ı () satisfying
8
< ! ı ()  0 is radially ˆsymmetric, Bı (0)  supp(! ı )  Bı (0)  Rd
(13)
: for 0 <  < 1, and ! ı () d  = 1 :
Bı (0)

Let S2ı be the set of u 2 L2 (Ω )d with kuk2 ı = kuk2L2 (Ω )d + juj2 ı finite, which is a
S2 S2
separable Hilbert space with an inner product induced by the norm k
kSı Mengesha and
2
Du [2014]. For a weakly closed subspace V  L2 (Ω )d that has no nontrivial affine maps
with skew-symmetric gradients, we let Vc;ı = S2ı \ V . One can establish a compactness
result on Vc;ı Mengesha and Du [2014] and Mengesha [2012]:
Lemma 3.1. For a bounded sequence fun g2Vc;ı , limn!1 jun jSı = 0 gives
2
kun kL2 (Ω ) ! 0.
This leads to a nonlocal Poincaré inequality and the coercivity of the energy functional.
Proposition 3.2 (Nonlocal Poincaré). There exists a positive constant C such that
kukL2 (Ω )d  C jujSı ; 8 u 2 Vc;ı :
2

The well-posedness of the variational problem then follows Mengesha and Du [2014].
Moreover, one can get a uniform Poincaré constant, independent of ı as ı ! 0, if the
nonlocal interaction kernels behave like a Diract-delta sequence. More specifically, they
satisfy that
ˆ
(14) lim ! ı ()d  = 0; 8  > 0:
ı!0 jj>
 3566 QIANG DU (杜强)

The assumption is particularly true for a rescaled kernel ! ı () = ı d !(/ı) Du [n.d.]
and Mengesha and Du [2014].
Note that the above line of analysis can be carried out by extending similar results
for the scalar function spaces originated from the celebrated work Bourgain, Brezis, and
Mironescu [2001] and further studied in Ponce [2004]. One complication for vector fields
is that the energy seminorm only uses a projected difference Du instead of the total dif-
ference, see Mengesha [2012] and Mengesha and Du [2014, 2015, 2016] for detailed dis-
cussions.
The nonlocal Poincaré inequality and energy coercivity then imply a well-posed vari-
ational formulation of the nonlocal model through minimizing Eı (u) over u 2 Vc;ı . The
weak form of the Euler–Lagrange equation is given by

Bı (u; v) := (! ı Du; Dv)Ω Ω = (u; b)Ω ; 8 v 2 Vc;ı :

We note a special case with Vc;ı = S0;ı , the closure of C01 (Ω)d in S2ı with all of its
elements satisfying u(x) = 0 on ΩI = Ωı = fx 2 Rd nΩ j dist(x; Ω) < ıg, corresponding
to a problem with a homogeneous nonlocal Dirichlet constraint on a ı-layer around Ω, see
Figure 1.

ΩI = Ωı u = 0 on @Ω
Lı u = D  (! ı D)u = b; in Ω;
(15) ı ı!0 L0 u = b
u = 0; in ΩI = Ωı : x0
Ω x
Ω

Figure 1: A nonlocal constrained value problem and its local PDE limit

Furthermore, under the assumption (14), we can show that as ı ! 0, the solution
of (15), denoted by uı , converges in L2 (Ω) to the solution u0 2 H01 (Ω) of the equation
L0 u = (d +2) 1 (∆u+2r(r
u)) = b in Ω Mengesha and Du [2014], thus compatible
with linear elasticity.

Elements of mathematical foundation of nonlocal models. Without going into details,

we summarize some basic elements in Table 2. For brevity, the illustration is devoted to
the case of nonlocal diffusion model and its local limit, with K denoting a generic 2nd
order positive definite coefficient tensor, and the same notation (
;
) for L2 inner products
of scalar, vector and tensor fields over their respective domains.
 NONLOCAL MODELS 3567

Local variational problems , Nonlocal variational problems

Local energy: (ru; ru) , Nonlocal energy: (! ı G u; G u)
Sobolev space H (Ω) 1
, Nonlocal function space S2ı
 (
Local balance (PDE): Nonlocal balance:
,
r
(Kru) = f G  (! ı G u) = f

Boundary conditions on @Ω , Volumetric constraints on ΩI

 
Local weak forms: Nonlocal weak forms:
,
(Kru; ru) = (f; v); 8 v (! ı G u; G v) = (f; v); 8 v

Classical Poincaré: kukL2  ckrukL2 , Nonlocal Poincaré: kukL2  cjujSı

2

Table 2: Elements of variational problems: local versus nonlocal.

Other variants of nonlocal operators and nonlocal calculus. As part of the nonlocal
vector calculus, there are other on possible variants to the nonlocal operators introduced
here, e.g., the one-point nonlocal divergence and nonlocal dual gradient given by
ˆ ˆ
v(x0 ) + v(x) 0
D v(x) = ı (x0 x)(Dv)(x0 ; x)d x0 ; G v(x) = ı (x0 x)e(x0 x) ˝ dx
Ω Ω jx0 xj

for an averaging kernel ı . With ı approaching a Dirac-delta measure at the origin

as ı ! 0, D and G recover the conventional local divergence and gradient opertors
Du, Gunzburger, Lehoucq, and K. Zhou [2013] and Mengesha and Du [2016]. They also
form a duality pair and have been used for robust nonlocal gradient recovery Du, Tao, X.
Tian, and J. Yang [2016]. Their use in the so-called correspondence peridynamic materials
models could be problematic but more clarifications have been given recently Du and X.
Tian [n.d.]. Moreover, these one-point operators are needed to reformulate more general
state-based peridynamics Du, Gunzburger, Lehoucq, and K. Zhou [2013] and Mengesha
and Du [2015, 2016] where the equation of motion is often expressed by Silling [2010],
Silling, Epton, Weckner, Xu, and Askari [2007], and Silling and Lehoucq [2010]
ˆ
ut t = fT[x; u]hx0 xi T[x0 ; u]hx x0 igd x0

with T[x; u]hx0 xi and T[x0 ; u]hx x0 i denoting the peridynamic force states. In fact,
D u can be used to represent the linear dilational strain in (3). Thus, we once again see
that the study of nonlocal models of mechanics further enriches the mathematical theory
of nonlocal operators and makes nonlocal vector calculus highly relevant to applications.
 3568 QIANG DU (杜强)

4 Numerical discretization of nonlocal models

There are many ways to discretizef nonlocal models Du [2017a], such as mesh-free
Bessa, Foster, Belytschko, and Liu [2014], Parks, Seleson, Plimpton, Silling, and Lehoucq
[2011], and Parks, Littlewood, Mitchell, and Silling [2012], quadrature based difference
or collocation Seleson, Du, and Parks [2016], H. Tian, H. Wang, and W. Wang [2013],
and X. Zhang, Gunzburger, and Ju [2016a,b], finite element Tao, X. Tian, and Du [2017],
H. Tian, Ju, and Du [2017], and K. Zhou and Du [2010] and spectral methods Du and J.
Yang [2017]. In particular, finite difference, finite element and collocation schemes in one
dimension were considered in X. Tian and Du [2013], including comparisons and analysis
of the differences and similarities. Discontinuous Galerkin approximations have also been
discussed, including conforming DG Chen and Gunzburger [2011] and Ren, C. Wu, and
Askari [2017] nonconforming DG X. Tian and Du [2015] and local DG Du, Ju, and Lu
[2018].
Since nonlocal models are developed as alternatives when conventional continuum
PDEs can neither capture the underlying physics nor have meaningful mathematical so-
lutions, we need to place greater emphasis on verification and validation of results from
the more tortuous simulations. A common practice for code verification is to consider the
case where the nonlocal models can lead to a physically valid and mathematically well-
defined local limit on the continuum level and to check if one can numerically reproduce
solutions of the local limit by solving nonlocal models with the same given data. Such
popular benchmark tests may produce surprising results as discussed here.

Asymptotical compatibility. Addressing the consistency on both continuum and dis-

crete levels and ensuring algorithmic robustness have been crucial issues for modeling and
code development efforts, especially for a theory like peridynamics that is developed to
capture highly complex physical phenomena. In the context of nonlocal models and their
local limits, the issues on various convergent paths are illustrated in the diagram shown
in Figure 2 X. Tian and Du [2014] (with smaller discretization parameter h representing
finer resolution).
The paths along the diagram edges are for taking limit in one of the parameters while
keeping the other fixed:  shows the convergence of solutions of nonlocal continuum
models to their local limit as ı ! 0, which has been established for various linear and
nonlinear problems; | is a subject of numerical PDE; } assures a convergent discretiza-
tion to nonlocal problem by design; ~ is more intriguing, as it is not clear whether the local
limit of numerical schemes for nonlocal problems would remain an effective scheme for
the local limit of the continuum model. An affirmative answer would lead to a nice com-
mutative diagram, or asymptotic compatibility (AC) X. Tian and Du [ibid.], one can follow
 NONLOCAL MODELS 3569

~
Discrete uhı ı!0
uh0 Discrete
Nonlocal Local
h!

h!0
h!0
} ı! 0 |
0
Continuum ı!0 Continuum
Nonlocal uı u0 Local PDE

Figure 2: A diagram of possible paths between uı , uhı , uh0 and u0 via various limits.

either the paths through those marked with } and  or ones marked with ~ and | to get
the convergence of uhı to u0 .
AC schemes offer robust and convergent discrete approximations to parameterized
problems and preserve the correct limiting behavior. While the variational characteriza-
tion and framework are distinctive, they are reminicent in spirit to other studies of conver-
gent approximations in the limiting regimes, see for example Arnold and Brezzi [1997],
Guermond and Kanschat [2010], and Jin [1999].

Getting wrong solution from a convergent numerical scheme. To motivate the AC

schemes, we consider a 1d linear nonlocal problem Lı uı (x) = b(x) for x 2 (0; 1),
where Lı is given by (2) with a special kernel, i.e.,
ˆ ˆ
3 ı
3 ı 2 2
(16) Lı u (x) = 3 u(x + s) 2u(x) + u(x s) ds = 3 h Dh u(x)dh ;
ı 0 ı 0
We impose the contraint that uı (x) = u0 (x) for x 2 ( ı; 0) [ (1; 1 + ı) where u0
00
solves the local limiting problem u0 (x) = b(x) in R. On the continuum level, we have
uı ! u0 as ı ! 0 in the appropriate function spaces, as desired. For (16), we may obtain
a discrete system if we replace the continuum difference Lı by discrete finite differences
through suitable quadrature approximations (leading to the quadrature based finite differ-
ence discretization as named in X. Tian and Du [2013]). For example, following Du and
X. Tian [2015] and X. Tian and Du [2013], we consider a scheme for (16) obtained from
a Riemann sum quadrature: for 1  i  N = 1/h, ı = rh,
r
3h X 2
(17) Lhı ui = (D u)i = b(xi ) ;
ı 3 m=1 mh

where fui g are approximations of fu(xi )g at nodal points fxi = ihgN +r

i = r . For any given
ı > 0, we can show the convergence of the discretization as h ! 0 for any given ı by
combining both stability with consistency estimates X. Tian and Du [2013]. However, by
 3570 QIANG DU (杜强)

considering a special case with r = 1 in (17), we end up with a scheme 3(Dh2 u)i = bi ,
which converges to the differential equation 3u00 (x) = b(x) as h = ı ! 0, but not
to the correct local limit. In other words, if we set h and ı to zero proportionally, the
numerical solution of the discrete scheme for the nonlocal problem yields a convergent
approximation to a wrong local limit associated with, unfortunately, a consistently over-
estimated elastic constant!
The possibility of numerical approximations converging to a wrong solution is alarm-
ing; if without prior knowledge, such convergence might be mistakenly used to verify or
disapprove numerical simulation, and we see the risks involved due to the wrong local lim-
its produced by discrete solutions to nonlocal models. Although illustrated via a simple
example here, it has been shown to be a generic feature of discretizations represented by
(17) and other schemes such as the piecewise constant Galerkin finite element approxima-
tions, for scalar nonlocal diffusion models and general state-based peridynamic systems
Du and X. Tian [2015] and X. Tian and Du [2013, 2014].

Robust discretization via AC schemes. On a positive note, the complications due to

the use of discrete schemes like (17) can be resolved through other means. For example,
it is proposed in X. Tian and Du [2013] that an alternative formulation works much more
2
robustly by suitably adjusting the weights for the second order differences fDmh ug so
that the elastic constant always maintain its correct constant value 1, independently of r!
Hence, as shown in X. Tian and Du [ibid.], we have a scheme that is convergent to the
nonlocal model for any fixed ı as h ! 0 and to the correct local limit whenever ı ! 0
and h ! 0 simultaneously, regardless how the two parameters are coupled. Moreover,
for a fixed h, it recovers the standard different scheme for the correct local limit models
as ı ! 0. Thus, we have a robust numerical approximation that is free from the risk of
going to the wrong continuum solution. Naturally, it is interesting to characterize how
such schemes can be constructed in general.

Quadrature based finite difference AC scheme. Approximations for multidimensional

scalar nonlocal diffusion equations have been developed Du, Tao, X. Tian, and J. Yang
[n.d.], which are not only AC but also preserve the discrete maximum principle. We con-
sider a set of nodes (grid points) fxj g of a uniform Cartesian mesh with a mesh size h and a
multi-index j corresponding to xj = hj. It is natural to approximate the nonlocal operator
in (11) by
ˆ  
u(xi + z) 2u(xi ) + u(xi z)
(18) Lı u(xi )  Ih W (z)! ı (z)d z;
jzj2 W (z)
where Ih represents the piecewise d -multi-linear interpolation operator in z associated
with the uniform Cartesian mesh fxj = hjg, but the key that is crucial for the AC property
 NONLOCAL MODELS 3571

and the discrete maximum principle is to choose a properly defined nonnegative weight
W = W (z). The choice adopted in Du, Tao, X. Tian, and J. Yang [ibid.] corresponds
to W (z) = 1/jzj1 where jzj1 denotes the `1 norm in Rd . This particular weight makes
the quadrature exact for all quadratic functions. One can then show, through a series of
technical calculations, that the resulting numerical solution converges to the solution of
the nonlocal model on the order of O(h2 ) for a fixed ı > 0, and converges to that of
the local limit model on the order of O(ı 2 + h2 ) as both h; ı ! 0 simultaneously, thus
demonstrating the AC property.

AC finite element approximations. For multidimensional systems, one can extend, as

in X. Tian and Du [2014], to more general abstract settings using conforming Galerkin
finite element (FE) methods on unstructured meshes. In particular, the concept and theory
of asymptotically compatible schemes are introduced for general parametrized variational
problems. A special application is to pave a way for identifying robust approximations to
linear nonlocal models that are guaranteed to be consistent in the local limit. Specifically,
we have the following theorem that agrees with numerical experiments reported in the
literature Bobaru, M. Yang, Alves, Silling, Askari, and Xu [2009] and X. Tian and Du
[2013].

Theorem 4.1. Let uı be the solution of (15) and uı;h be the conforming Galerkin FE
approximation on a regular quasi-uniform mesh with meshing parameter h. If the FE
space Vı;h contains all continuous piecewise linear elements, then kuı;h u0 kL2 (Ω) ! 0
as ı ! 0 and h ! 0. If in addition, the FE subspace is given by a conforming FE space of
the local limit PDE model with zero extension outside Ω with u0;h being the FE solution,
then on each fixed mesh, kuı;h u0;h kL2 ! 0 as ı ! 0. On the other hand, if Vı;h is the
piecewise constant space and conforming for (15), then kuı;h u0 kL2 ! 0 if h = o(ı)
as ı ! 0.

The above theorem, proved under minimal solution regularity, remains valid for nonlo-
cal diffusion and state-based peridynamic models. The same framework of AC schemes
can establish the convergence of numerical approximation to linear fractional diffusion
equations (that correspond to ı = 1) via the approximation of a nonlocal diffusion model
with a finite horizon X. Tian, Du, and Gunzburger [2016]. For example, consider a scalar
fractional diffusion model, for ˛ 2 (0; 1),
ˆ
u(x) u(x0 ) 0
( ∆)˛ u = f; on Ω; u = 0; on Rd nΩ; ( ∆)˛ u(x) = Cd;˛ dx ;
Rd jx x0 jd +2˛

and Cd;˛ is a positive constant dependent on d and ˛. We have that X. Tian, Du, and
Gunzburger [ibid.],
 3572 QIANG DU (杜强)

Theorem 4.2. Let uı be the solution of the above fractional diffusion model with the
integral truncated to a spherical neighborhood of radius ı > 0. Let uhı be a conforming
Galerkin FE approximation with the discretization parameter h, then kuhı uı kH ˛ ! 0
as h ! 0 for any given ı and kuhı u1 kH ˛ ! 0 as ı ! 1 and h ! 0.

We note that studies of AC schemes have been extended to nonconforming DG FE X.

Tian and Du [2015], local DG FE Du, Ju, and Lu [2018], spectral approximation Du and
J. Yang [2016] and nonlocal gradient recoveries Du, Tao, X. Tian, and J. Yang [2016].
There were also extensions to nonlinear nonlocal models Du and Huang [2017] and Du
and J. Yang [2016].

5 Nonlocal and local coupling

Nonlocal models can be effective alternatives to local models by accommodating sin-
gular solutions, which makes nonlocal models particularly useful to subjects like fracture
mechanics. Yet treating nonlocality in simulations may incur more computation. Thus,
exploring localization and effective coupling of nonlocal and local models can be helpful
in practice. Nevertheless, nonlocal models, unlike local PDEs, generically do not employ
local boundary or interface conditions imposed on a co-dimension-1 surface, hence mo-
tivating the development of different approaches for local-nonlocal coupling Li and Lu
[2017] and Du, Tao, and X. Tian [2018].

Heterogeneous localization. A particular mathematical quest for a coupled local and

nonlocal model is through heterogeneous localization, as initiated in X. Tian and Du
[2017].
The aim is to characterize subspaces of L2 (Ω), denoted by S(Ω), that are significantly
larger than H 1 (Ω) and have a continuous trace map into H 1/2 (@Ω). One such example
is defined as the completion of C 1 (Ω) with respect to the nonlocal norm for a kernel ı ,
  12
kukS(Ω) = kuk2L2 (Ω) + juj2S(Ω) ; with
ˆ ˆ
(u(y) u(x))2
juj2S(Ω) = ı (x; y) d yd x:
Ω Ω\Bı (x) jy xj2

The main findings of X. Tian and Du [ibid.] are that the trace map exists and is contin-
uous on a nonlocal function space S(Ω) if the radius of the support of ı , i.e., the horizon,
is heterogeneously localized as x ! @Ω. By considering such a class of kernels, the study
departs from many existing works, such as Bourgain, Brezis, and Mironescu [2001], cor-
responding to typical translate-invariant kernels. In X. Tian and Du [2017], the class of
 NONLOCAL MODELS 3573

kernels under consideration is given by

 
1 jy xj
(19) (x; y) = ˆ

jı(x)jd ı(x)

where ˆ = ˆ(s) is a non-increasing nonnegative function defined for s 2 (0; 1) with a

finite d 1 moment. The heterogeneously defined horizon ı = ı(x) approaches zero
when x ! Γ  @Ω. A simple choice taken in X. Tian and Du [ibid.] is ı(x) =  dist(x; Γ)
for  2 (0; 1].
The following proposition has been established in X. Tian and Du [ibid.], which is of
independent interests by showing the continuous imbedding of classical Sobolev space
H 1 (Ω) in the new heterogeneously localized nonlocal space S(Ω). The result generalizes
a well-known result of Bourgain, Brezis, and Mironescu [2001] for the case with a constant
horizon and translation invariant kernel.

Proposition 5.1. For the kernel in (19) and the horizon ı(x) = dist(x; Γ) with  2
(0; 1), H 1 (Ω) is continuously imbedded in S(Ω) and for any u 2 H 1 (Ω), kukS(Ω) 
C kukH 1 (Ω) where the constant C = C (Ω) is independent of  for  small.

New trace theorems. A key observation proved in X. Tian and Du [2017] is that, with
heterogeneously vanishing interaction neighborhood when x ! @Ω, we expect a well
defined continuous trace map from the nonlocal space S(Ω), which is larger than H 1 (Ω),
to H 1/2 (@Ω).

Theorem 5.2 (General trace theorem). Assume that Ω is a bounded simply connected
Lipschitz domain in Rd (d  2) and Γ = @Ω, for a kernel in (19) and the heterogeneously
defined horizon given by ı(x) =  dist(x; Γ) for  2 (0; 1]. there exists a constant C
depending only on Ω such that the trace map T for Γ satisfies kT ukH 12 (Γ)  C kukS(Ω) ,
for any u 2 S(Ω).

By Proposition 5.1, we see that the above trace theorem is indeed a refinement of the
classical trace theorem in the space H 1 (Ω), with the latter being a simple consequence.

An illustrative example with a simple kernel on a stripe domain. A complete proof

of the trace Theorem 5.2 is presented in X. Tian and Du [ibid.]. To help understanding
what the result conveys and how it compares with other relevant works, it is suggestive to
consider a special case.
For Ω and Γ, we take a special stripe domain Ω = (0; r)  Rd 1 and a portion of its
boundary Γ = f0g  Rd 1 for a constant r > 0, see equation (20) and Figure 3.
 3574 QIANG DU (杜强)

(20)8 x̃
ˆ
ˆ (0;1) (jy xj)jy xj2 r
ˆ
< (x; y) = ;
jı(x)jd +2
ˆ where ı(x) = dist(x; Γ) = x1 ;
ˆ
Γ = f0g  Rd 1
:̂ Ω=(0; r)Rd 1
8 x = (x1 ; x̃); x̃ 2 Rd 1 :
Figure 3: Nonlocal kernel and depiction of the stripe geometry.

This case serves as not only a helpful step towards proving the more general trace
Theorem 5.2 but also an illustrative example on its own. Indeed, this special nonlocal
(semi)-norm is
ˆ ˆ
(u(y) u(x))2
(21) juj2S(Ω) = d yd x :
Ω Ω\fjy xj<jx1 jg jx1 j2+d

Clearly, the denominator x1 penalizes the spatial variation only at x1 = 0, thus S(Ω)
contains all functions in L2 (Ω̃) (and possibly discontinuous) for any domain Ω̃ with its
closure being a compact subset of Ω. Hence, functions in S(Ω) are generally not expected
to have regularity better than L2 (Ω0 ) over any strict subdomain Ω0 . Yet, as elucidated in
X. Tian and Du [2017], due to the horizon localization at the boundary, the penalization
of spatial variations provides enough regularity for the functions in S(Ω) to have well-
defined traces just on the boundary itself. Intuitively, this is a natural consequence of
the localization of nonlocal interactions on the boundary. In contrast, a standard norm
associated with fractional Sobolev space is defined by
ˆ ˆ
2 (u(y) u(x))2
jujH ˛ (Ω) = d yd x :
Ω Ω jy xj2˛+d
The regularity of the functions is effected by the denominator which vanishes at x = y.
We now state the special trace theorem, see X. Tian and Du [ibid.] for a complete proof.

Theorem 5.3 (Special trace theorem). For Ω = (0; r)  Rd 1 and Γ = f0g  Rd 1

, there
exists a constant C depends only on d such that for any u 2 C 1 (Ω̄) \ S(Ω),
 
kukL2 (Γ)  C r 1/2 kukL2 (Ω) + r 1/2 jujS(Ω) ; for d  1;
 
jujH 1/2 (Γ)  C r 1 kukL2 (Ω) + jujS(Ω) ; for d  2:

where the nonlocal semi-norm of S(Ω) is as given in (20).

Coupled local and nonlocal models. We use Ω and Ω+ to denote two open domains in
Rd that satisfy Ω \Ω+ = Γ, a co-dimension-1 interface, and Ω to denote their union. We
 NONLOCAL MODELS 3575

consider the coupling of a local model on Ω with a nonlocal model on Ω+ , see Figure 4.
Let S(Ω+ ) be the nonlocal space with heterogeneous localization on the boundary. By the
trace theorem, we define the energy (solution) space and the test function space as

W(Ω) = fu 2 H 1 (Ω ) \ S(Ω+ ) j u = u+ on Γg

W0 (Ω) = fu 2 W(Ω) j u = 0 on @Ωg;

where fu˙ (x)g denotes the traces of u defined from Ω˙ respectively. From Proposi-
tion 5.1, we have the space H 1 (Ω) continuously imbedded in W(Ω) and H01 (Ω) also
continuously imbedded in W0 (Ω). For u 2 W(Ω), its norm is defined as kukW(Ω) =
kukH 1 (Ω ) +kukS(Ω+ ) . For g 2 H 1/2 (@Ω) and f 2 L2 (Ω), we have a coupled nonlocal-
to-local model (22).

(22) ˚
min 12 juj2H 1 (Ω ) + 12 juj2S(Ω+ ) (f; u)Ω ; ∆u = f Lu = f
1 u 2 S(Ω+ )
subject to u 2 W(Ω) and uj@Ω = g: u 2 H (Ω )
Γ
Figure 4: Variational formulation of a coupled local-nonlocal model.

Well-posedness of the coupled model. For (22) to be well-posed, the coercivity of the
energy functional is the key, which is consequence of a Poincaré inequality on W0 (Ω).
The latter can be established in a similar fashion as that on the nonlocal space with the
constant horizon (and the local Sobolev space H01 (Ω) as well). We thus have

Proposition 5.4. The coupled variational problem (22) has a unique minimizer u 2
W0 (Ω).

The seamless coupling of the nonlocal and local model means that one could use the
same numerical discretization to solve the coupled problems if the heterogenous localiza-
tion of horizon can be handled effectively. Indeed, this is where we can circle back to
utilize the concept of robust asymptotically compatible schemes X. Tian [2017], Du, Tao,
and X. Tian [2018], and X. Tian and Du [2014].

6 Nonlocal in time dynamics

Spatial nonlocality is often accompanied by temporal correlations and memory effects.
The latter involves nonlocality in time. Let us note first that a major difference in time and
 3576 QIANG DU (杜强)

space nonlocality is perhaps the generic time irreversibility. While a local time derivative
may be defined by an infinitesimal change either backward to the history or forward to the
future, it is more natural to view nonlocal time derivative as only dependent on past history.
Thus, it is of much interests to reconsider the basic operators of the nonlocal vector calculus
to accommodate the nonlocal interactions that are not symmetric. Of course, the issue of
symmetry does not only pertain to changes in time. In earlier works, nonlocal gradients
of the upwind type, variants of the operators given in Section 3, have been utilized in the
modeling of convective effects H. Tian, Ju, and Du [2017] and in the nonlocal formulation
of conservation laws Du and Huang [2017] and Du, Huang, and LeFloch [2017]. They
have also been used to perform nonlocal gradient recovery Du, Tao, X. Tian, and J. Yang
[2016]. The first rigorous treatment of a nonlocal in time dynamics with a finite memory
span, in the spirit of nonlocal vector calculus, was given in Du, J. Yang, and Z. Zhou
[2017], which we follow here.

Nonlocal time derivative and nonlocal-in-time dynamics. We take the operator

ˆ ı
u(t) u(t s)
(Gı u)(t ) = lim ı (s) ds; for t > 0;
!0  s
as the nonlocal time derivative for a nonnegative density kernel ı that is supported in the
interval [0; ı). This leads to the study of an abstract nonlocal-in-time dynamics:

(23) Gı u + Au = f; 8 t 2 ΩT = (0; T )  R+ ; u(t ) = g(t); 8 t 2 ( ı; 0)  R :

for a linear operator A in an abstract space, together with some nonlocal initial (historical)
data g = g(t ). We recall a well-posedness result for (23) corresponding to A = ∆ on a
bounded spatial domain Ω with a homogeneous Dirichlet boundary condition Du, J. Yang,
and Z. Zhou [ibid.].
Theorem 6.1. For f 2 L2 (0; T ; H 1 (Ω)), the problem (23) for A = ∆ on Ω with the
homogeneous Dirichlet boundary condition and g(x; t )  0 has a unique weak solution
u 2 L2 (0; T ; H01 (Ω)). Moreover, there is a constant c, independent of ı, f and u, such
that. kukL2 (0;T ;H01 (Ω)) + kGı ukL2 (0;T ;H 1 (Ω))  ckf kL2 (0;T ;H 1 (Ω)) .
The nonlocal-in-time diffusion equation may be related to fractional in time sub-diffusion
equations like @˛t u ∆u = 0 for ˛ 2 (0; 1) Du, J. Yang, and Z. Zhou [2017], Metzler
and Klafter [2004], and Sokolov [2012] by taking some special memory kernels Allen,
Caffarelli, and Vasseur [2016]. Such equations have often been used to describe the con-
tinuous time random walk (CTRW) of particles in heterogeneous media, where trapping
events occur. In particular, particles get repeatedly immobilized in the environment for a
trapping time drawn from the waiting time PDF that has a heavy tail Metzler and Klafter
 NONLOCAL MODELS 3577

[2004]. In general though, (23) provides a new class of models, due to the finite memory
span, that serves to bridge anomalous and normal diffusion, with the latter being the limit
as ı ! 0. Indeed, the model (23) can also be related to a trapping model, see Du [n.d.],
Du, J. Yang, and Z. Zhou [2017], and Du and Z. Zhou [2018] for more detailed studies.

Crossover of diffusion regimes. Diffusions in heterogeneous media have important

implications in many applications. Using single particle tracking, recent studies have
revealed many examples of anomalous diffusion, such as sub-diffusion with a slower
spreading process in more constricted environment Berkowitz, Klafter, Metzler, and Scher
[2002], He, Song, Su, Geng, Ackerson, Peng, and Tong [2016], and Jeon, Monne, Ja-
vanainen, and Metzler [2012]. Meanwhile, the origins and models of anomalous diffusion
might differ significantly Korabel and Barkai [2010], McKinley, Yao, and Forest [2009],
and Sokolov [2012]. On one hand, new experimental standards have been called for Sax-
ton [2012]. On the other hand, there are needs for in-depth studies of mathematical models,
many of which are non-conventional and non-local Du, Huang, and Lehoucq [2014], Du,
Gunzburger, Lehoucq, and K. Zhou [2012], and Sokolov [2012].
Motivated by recent experimental reports on the crossover between initial transient
sub-diffusion and long time normal diffusion in various settings He, Song, Su, Geng,
Ackerson, Peng, and Tong [2016], the simple dynamic equation (23) with A = ∆ pro-
vides an effective description of the diffusion process encompassing these regimes Du
and Z. Zhou [2018]. For model (23), the memory effect dominates initially, but as time
goes on, the fixed memory span becomes less significant over the long life history. As
a result, the transition from sub-diffusion to normal diffusion occurs naturally. This phe-
nomenon can be illustrated by considering the mean square displacement (MSD) m(t)
which can be explicitly computed Du and Z. Zhou [ibid.]. In Figure 5, we plot a solution
of Gı m(t) = 2, i.e., the mean square displacement of the nonlocal solution for f  0
and ı (s) = (1 ˛)ı ˛ 1 s ˛ with ˛ = 0:2 and ı = 0:5. The result again illustrates the
analytically suggested transition from the early fractional anomalous diffusion regime to
the later standard diffusion regime. This ”transition” or ”crossover” behavior have been
seen in many applications, e.g. diffusions in lipid bilayer systems of varying chemical
compositions Jeon, Monne, Javanainen, and Metzler [2012, Fig.2], and lateral motion of
the acetylcholine receptors on live muscle cell membranes He, Song, Su, Geng, Ackerson,
Peng, and Tong [2016, Figs.3, 4].

7 Discussion and conclusion

Nonlocal models, arguably more general than their local or discrete analogs, are de-
signed to account for nonlocal interactions explicitly and to remain valid for complex
 3578 QIANG DU (杜强)

Figure 5: The MSD plot for (23) and the MSD curve He, Song, Su, Geng, Ackerson,
Peng, and Tong [2016, Fig. 3] showing the crossover from sub-diffusion to normal
diffusion for immobile AChRs.

systems involving possibly singular solutions. They have the potential to be alternatives
and bridges to existing local continuum and discrete models. Their increasing popular-
ity in applications makes the development of a systematic/axiomatic mathematical frame-
work for nonlocal models necessary and timely. This work attempts to answer a few
questions on nonlocal modeling, analysis and computation, particularly for models in-
volving a finite-range nonlocal interactions and vector fields, To invite further studies
on the subject, it might be more enticing to identify some issues worthy further investi-
gation and to explore connections with other relevant topics. This is the purpose here,
but before we proceed, we note that there are already many texts and online resources
devoted to nonlocal models (scalar fractional equations in particular, see for example
more recent books Bucur and Valdinoci [2016], Vázquez [2017], and West [2016] and
http://www.ma.utexas.edu/mediawiki/index.php/Starting_page). We also re-
fer to Du [n.d.] for more details and references on topics discussed below.

Nonlocal exterior calculus and geometry. While an analogy has been drawn between
traditional local calculus and the nonlocal vector calculus involving nonlocal operators
and fluxes, nonlocal integration by parts and nonlocal conservation laws, the nonlocal
framework still needs to be updated or revamped. For example, a geometrically intrin-
sic framework for nonlocal exterior calculus and nonlocal forms on manifolds is not yet
available. It would be of interests to develop nonlocal geometric structures that are more
general than both discrete complexes and smooth Riemannian manifolds. In connection
with such investigations, there are relevant studies on metric spaces Burago, Ivanov, and
Kurylev [2014] and Fefferman, Ivanov, Kurylev, Lassas, and Narayanan [2015], Laplace-
Beltrami Belkin and Niyogi [2008] and Lévy [2006], and combinatorial Hodge theory
with scalar nonlocal forms Bartholdi, Schick, N. Smale, and S. Smale [2012]. We also
made attempts like Le [2012] to introduce nonlocal vector forms, though more coherent
constructions are desired. Given the close relations between local continuum models of
 NONLOCAL MODELS 3579

mechanics and differential geometry, one expects to find deep and intrinsic connections
between nonlocal mechanics and geometry.

Nonlocal models, kernel methods, graph and data. Discrete, graph, network models
and various kernel based methods in statistics often exhibit nonlocality. Exploring their
continuum limits and localization can offer fundamental insights. In this direction, we
mention some works related to graph Laplacians, diffusion maps, spectral clustering and
so on Coifman and Lafon [2006], Singer and H.-T. Wu [2017], Spielman [2010], Trillos
and Slepčev [2016], and van Gennip and A. L. Bertozzi [2012]. These subjects are also
connected with the geometric analysis already mentioned and applications such as image
and data analysis and learning Buades, Coll, and Morel [2010], Gilboa and Osher [2008],
and Lou, X. Zhang, Osher, and A. Bertozzi [2010]. For instance, one can find, for appli-
cations to image analysis, the notion of nonlocal means Buades, Coll, and Morel [2010]
and nonlocal (NL) gradient operator Gilboa and Osher [2008] together with a graph di-
vergence all defined for scalar fields. Indeed, there have been much works on nonlocal
calculus for scalar quantities, see Du [n.d.] for more detailed comparisons.

Nonlocal function spaces, variational problems and dynamic systems. While there
have been a vast amount of studies on nonlocal functional spaces, related variational prob-
lems and dynamic systems, such as Ambrosio, De Philippis, and Martinazzi [2011], Bour-
gain, Brezis, and Mironescu [2001], Bucur and Valdinoci [2016], Caffarelli and Silvestre
[2007], Silvestre [2014], and West [2016], the majority of them have focused on scalar
quantities of interests and are often associated with fractional differential operators, frac-
tional calculus, fractional Sobolev spaces and fractional PDEs having global interactions.
On the other hand, motivated by applications in mechanics, our recent works can serve as a
starting point of further investigations on nonlocal functional analysis of vector and tensor
fields and systems of nonlocal models. For example, one may consider nonlocal exten-
sions to the variational theory of nonlinear elasticity Ball [2010] and use them to develop
better connections with atomistic modeling. One may further consider nonlocal spaces
that can account for anisotropies and heterogeneities in both state and configuration vari-
ables. Extensions of the new trace theorems on heterogeneously localized nonlocal spaces
to various vector field forms are also topics of more subsequent research. For instance,
one may investigate possible nonlocal generalization of the trace theorems on the normal
component of vector fields in the H (div) space Buffa and Ciarlet [2001]. Moreover, there
are also interesting questions related to nonlocal models of fluid mechanics, including the
nonlocal Navier-Stokes equations involving fractional order derivatives Constantin and
Vicol [2012] and more recently analyzed nonlocal analogs of the linear incompressible
Stokes equation as presented in the following forms, together with a comparison with
 3580 QIANG DU (杜强)

their classical form in the local limit:

  
Lı u + Gı p= b; Lı u + Gı p = b; ∆u + rp= b;
(24) or and
Dı u = 0; Dı u ı 2 L̂ı p= 0; r
u = 0;

where Lı and L̂ı are nonlocal diffusion operators, Gı and Dı are one-point nonlocal
gradient and divergence operators, similar to ones described in Section 3. There are surely
more questions about the extensions to time-dependent and nonlinear systems.

Nonlocal, multiscale and stochastic modeling. Nonlocality arises naturally from model
reductions and has appeared (either knowingly or implicitly) in many early works (such
as the Mori-Zwanzig formalism Chorin, Hald, and Kupferman [2002]). Nonlocal model-
ing could play more prominent roles in multiscale and stochastic modeling, ranging from
bridging atomistic and continuum models, to data-driver model reductions of dynamic
systems. There are also strong connections of nonlocal models with hydrodynamic de-
scriptions of collective behavior and flocking hydrodynamics Motsch and Tadmor [2014]
and Shvydkoy and Tadmor [2017]. Exploring nonlocal models in diffusion and dispersal
processes has also received much attention Fuentes, Kuperman, and Kenkre [2003], Kao,
Lou, and Shen [2010], and Massaccesi and Valdinoci [2017], with the resulting nonlo-
cal models having strong ties with stochastic processes, particularly, non-Gaussian and
non-Markovian behaviors Kumagai [2014] and Zaburdaev, Denisov, and Klafter [2015].
Stochastic nonlocal modeling is certainly an interesting subject on its own. In addition,
inverse problems related to nonlocal models are also essential research subjects of both
theoretical and practical interests and they can also be connected with various design and
control problems.

Nonlocal modeling, numerical analysis and simulation. Numerical simulations of

nonlocal models bring new computational challenges, from discretization to efficient solvers.
To elevate the added cost associated with nonlocal interactions, it is of interests to explore
a whole host of strategies, including local and nonlocal coupling Li and Lu [2017] and
Du, Tao, and X. Tian [2018], adaptive grids Du, L. Tian, and Zhao [2013], multigrid and
fast solvers Du and Z. Zhou [2017] and H. Wang and H. Tian [2012], some of them are
less examined than others and most of topics remain to be further studied. The subject is
naturally linked to sparse and low rank approximations that would allow one to explore
the nonlocal structure to achieve efficient evaluation of nonlocal interactions as well as
the solution of associated algebraic systems. Scalable algorithms via domain decomposi-
tion or other strategies that can particularly handle the information exchange (communi-
cations between processors) involving nonlocal interactions are interesting and important
research questions. Let us also mention that nonlocal models can also become effective
 NONLOCAL MODELS 3581

tools to analyze numerical schemes that were initially developed to solve local PDEs. For
example, to understand the interplay between the smoothing length and the particle spac-
ing in the context of smoothed particle hydrodynamics Gingold and Monaghan [1977]
and Monaghan [2005], nonlocal continuum systems (24) can help providing a rigorous
mathematical foundation for improving the stability and robustness of the discretization
Du and X. Tian [2017]. Another example is concerned with discretization schemes for
multidimensional local diffusion equations through the nonlocal integral formulation Du,
Tao, X. Tian, and J. Yang [n.d.] and Nochetto and W. Zhang [2017], a topic linked with
approximations of fully nonlinear elliptic equations such as the Monge-Ampére. An open
question there is whether or not there are discretization schemes on unstructured meshes
which can preserve the discrete maximum principles and are asymptotically compatible
for general anisotropic and heterogeneous diffusion equations.

Thinking nonlocally, acting locally. The pushes for nonlocal modeling come from sev-
eral fronts. Foremost, the development of nonlocal models is driven by the interests in
studying singular/anomalous/stochastic/multiscale behavior of complex systems where
nonlocal models can potentially unify and bridge different models. Nowadays, the im-
minent growth of nonlocal modeling may also be attributed to the inescapable presence
of nonlocality in the daily human experience. The emergence of augmented reality, infor-
mation technology and data science as well as intelligent computing has been fueling the
popularity of nonlocal modeling as the world is getting more than ever remotely and non-
locally networked together. With extreme computing capabilities beyond doing simple
analytical approximations, we could be ready to tackle nonlocal interactions directly. Yet,
despite the huge lift in computing power, exploring simple representations and closure
relations via local, sparse, low rank or low dimensional approximations is still of great
theoretical interest and practical significance. We thus conclude by saying that promoting
the role of nonlocal modeling is to not only argue for the need to think nonlocally and to
retain nonlocal features wherever necessary, but also point out the importance in utilize
local models wherever feasible, hence to act locally, as our goal is to have the efficiency
and robustness of mathematical modeling and numerical simulations while maintaining
their generality and predicability.

Acknowledgments. The author would like to thank the organizing committee of ICM
2018 for the invitation. Some of the more detailed discussions presented here was taken
from the joint publications with Xiaochuan Tian, Tadele Mengesha, Max Gunzburger,
Richard Lehoucq, Kun Zhou, Zhi Zhou and Jiang Yang. The author is grateful to them
and many other students and collaborators on the subject (an extended list of them is given
in Du [n.d.]).
 3582 QIANG DU (杜强)

References
Mark Allen, Luis Caffarelli, and Alexis Vasseur (2016). “A parabolic problem with a frac-
tional time derivative”. Arch. Ration. Mech. Anal. 221.2, pp. 603–630. MR: 3488533
(cit. on p. 3576).
Luigi Ambrosio, Guido De Philippis, and Luca Martinazzi (2011). “Gamma-convergence
of nonlocal perimeter functionals”. Manuscripta Math. 134.3-4, pp. 377–403. MR:
2765717 (cit. on p. 3579).
Douglas N. Arnold and Franco Brezzi (1997). “Locking-free finite element methods for
shells”. Math. Comp. 66.217, pp. 1–14. MR: 1370847 (cit. on p. 3569).
John M Ball (2010). “Progress and puzzles in nonlinear elasticity”. In: Poly-, quasi-and
rank-one convexity in applied mechanics. Springer, pp. 1–15 (cit. on p. 3579).
Laurent Bartholdi, Thomas Schick, Nat Smale, and Steve Smale (2012). “Hodge theory
on metric spaces”. Found. Comput. Math. 12.1. Appendix B by Anthony W. Baker,
pp. 1–48. MR: 2886155 (cit. on p. 3578).
Mikhail Belkin and Partha Niyogi (2008). “Towards a theoretical foundation for Laplacian-
based manifold methods”. J. Comput. System Sci. 74.8, pp. 1289–1308. MR: 2460286
(cit. on p. 3578).
Brian Berkowitz, Joseph Klafter, Ralf Metzler, and Harvey Scher (2002). “Physical pic-
tures of transport in heterogeneous media: Advection-dispersion, random-walk, and
fractional derivative formulations”. Water Resources Research 38.10, pp. 9–1 (cit. on
p. 3577).
M. A. Bessa, J. T. Foster, T. Belytschko, and Wing Kam Liu (2014). “A meshfree unifi-
cation: reproducing kernel peridynamics”. Comput. Mech. 53.6, pp. 1251–1264. MR:
3201938 (cit. on p. 3568).
Florin Bobaru, John T. Foster, Philippe H. Geubelle, and Stewart A. Silling, eds. (2017).
Handbook of peridynamic modeling. Advances in Applied Mathematics. CRC Press,
Boca Raton, FL, pp. xxxviii+548. MR: 3560288 (cit. on p. 3562).
Florin Bobaru, Mijia Yang, Leonardo Frota Alves, Stewart A. Silling, Ebrahim Askari,
and Jifeng Xu (2009). “Convergence, adaptive refinement, and scaling in 1D peridy-
namics”. International Journal for Numerical Methods in Engineering 77.6, pp. 852–
877 (cit. on p. 3571).
Jean Bourgain, Haim Brezis, and Petru Mironescu (2001). “Another look at Sobolev spaces”.
In: Optimal control and partial differential equations. IOS, Amsterdam, pp. 439–455.
MR: 3586796 (cit. on pp. 3566, 3572, 3573, 3579).
A. Buades, B. Coll, and J. M. Morel (2010). “Image denoising methods. A new nonlocal
principle”. SIAM Rev. 52.1. Reprint of “A review of image denoising algorithms, with
a new one” [MR2162865], pp. 113–147. MR: 2608636 (cit. on p. 3579).
 NONLOCAL MODELS 3583

Claudia Bucur and Enrico Valdinoci (2016). Nonlocal diffusion and applications. Vol. 20.
Lecture Notes of the Unione Matematica Italiana. Springer, [Cham]; Unione Matemat-
ica Italiana, Bologna, pp. xii+155. MR: 3469920 (cit. on pp. 3561, 3578, 3579).
A. Buffa and P. Ciarlet Jr. (2001). “On traces for functional spaces related to Maxwell’s
equations. I. An integration by parts formula in Lipschitz polyhedra”. Math. Methods
Appl. Sci. 24.1, pp. 9–30. MR: 1809491 (cit. on p. 3579).
Dmitri Burago, Sergei Ivanov, and Yaroslav Kurylev (2014). “A graph discretization of
the Laplace-Beltrami operator”. J. Spectr. Theory 4.4, pp. 675–714. MR: 3299811 (cit.
on p. 3578).
Luis Caffarelli and Luis Silvestre (2007). “An extension problem related to the fractional
Laplacian”. Comm. Partial Differential Equations 32.7-9, pp. 1245–1260. MR: 2354493
(cit. on pp. 3561, 3579).
X. Chen and Max Gunzburger (2011). “Continuous and discontinuous finite element meth-
ods for a peridynamics model of mechanics”. Comput. Methods Appl. Mech. Engrg.
200.9-12, pp. 1237–1250. MR: 2796157 (cit. on p. 3568).
Alexandre J. Chorin, Ole H. Hald, and Raz Kupferman (2002). “Optimal prediction with
memory”. Phys. D 166.3-4, pp. 239–257. MR: 1915310 (cit. on p. 3580).
Ronald R. Coifman and Stéphane Lafon (2006). “Diffusion maps”. Appl. Comput. Har-
mon. Anal. 21.1, pp. 5–30. MR: 2238665 (cit. on p. 3579).
Peter Constantin and Vlad Vicol (2012). “Nonlinear maximum principles for dissipative
linear nonlocal operators and applications”. Geom. Funct. Anal. 22.5, pp. 1289–1321.
MR: 2989434 (cit. on p. 3579).
Qiang Du (n.d.). Nonlocal modeling, analysis and computation, NSF-CBMS Monograph.
To appear in SIAM (cit. on pp. 3560, 3561, 3564, 3566, 3577–3579, 3581).
– (2015). “From peridynamics to stochastic jump process: illustrations of nonlocal bal-
ance laws and nonlocal calculus framework”. Scientia Sinica Mathematica 45.7, pp. 939–
952 (cit. on p. 3561).
– (2017a). “Local limits and asymptotically compatible discretizations”. In: Handbook
of peridynamic modeling. Adv. Appl. Math. CRC Press, Boca Raton, FL, pp. 87–108.
MR: 3618714 (cit. on p. 3568).
– (2017b). “Nonlocal calculus of variations and well-posedness of peridynamics”. In:
Handbook of peridynamic modeling. Adv. Appl. Math. CRC Press, Boca Raton, FL,
pp. 63–85. MR: 3618713 (cit. on p. 3561).
Qiang Du, Max Gunzburger, Richard B. Lehoucq, and Kun Zhou (2012). “Analysis and
approximation of nonlocal diffusion problems with volume constraints”. SIAM Rev.
54.4, pp. 667–696. MR: 3023366 (cit. on pp. 3565, 3577).
– (2013). “A nonlocal vector calculus, nonlocal volume-constrained problems, and nonlo-
cal balance laws”. Math. Models Methods Appl. Sci. 23.3, pp. 493–540. MR: 3010838
(cit. on pp. 3563–3565, 3567).
 3584 QIANG DU (杜强)

Qiang Du and Zhan Huang (2017). “Numerical solution of a scalar one-dimensional monotonicity-
preserving nonlocal nonlinear conservation law”. J. Math. Res. Appl. 37.1, pp. 1–18.
MR: 3642662 (cit. on pp. 3572, 3576).
Qiang Du, Zhan Huang, and Philippe G. LeFloch (2017). “Nonlocal conservation laws. A
new class of monotonicity-preserving models”. SIAM J. Numer. Anal. 55.5, pp. 2465–
2489. MR: 3715381 (cit. on p. 3576).
Qiang Du, Zhan Huang, and Richard B. Lehoucq (2014). “Nonlocal convection-diffusion
volume-constrained problems and jump processes”. Discrete Contin. Dyn. Syst. Ser. B
19.2, pp. 373–389. MR: 3170190 (cit. on pp. 3562, 3577).
Qiang Du, L. Ju, and J. Lu (2018). “A discontinuous Galerkin method for one-dimensional
time-dependent nonlocal diffusion problems”. Mathematics of Computation (cit. on
pp. 3568, 3572).
Qiang Du, Yunzhe Tao, and Xiaochuan Tian (2017). “A Peridynamic Model of Fracture
Mechanics with Bond-Breaking”. Journal of Elasticity, pp. 1–22 (cit. on p. 3563).
– (2018). “Nonlocal models with heterogeneous localization and their application to seam-
less local-nonlocal coupling”. In preparation (cit. on pp. 3572, 3575, 3580).
Qiang Du, Yunzhe Tao, Xiaochuan Tian, and Jiang Yang (n.d.). “Asymptotically com-
patible numerical approximations of multidimensional nonlocal diffusion models and
nonlocal Green’s functions”. To appear in IMA J. Numerical Analysis (cit. on pp. 3570,
3571, 3581).
– (2016). “Robust a posteriori stress analysis for quadrature collocation approximations
of nonlocal models via nonlocal gradients”. Comput. Methods Appl. Mech. Engrg. 310,
pp. 605–627. MR: 3548574 (cit. on pp. 3567, 3572, 3576).
Qiang Du, Li Tian, and Xuying Zhao (2013). “A convergent adaptive finite element al-
gorithm for nonlocal diffusion and peridynamic models”. SIAM J. Numer. Anal. 51.2,
pp. 1211–1234. MR: 3045653 (cit. on p. 3580).
Qiang Du and Xiaochuan Tian (n.d.). “Stability of nonlocal Dirichlet integrals and im-
plications for peridynamic correspondence material modeling”. To appear in SIAM J.
Applied Mathematics (cit. on p. 3567).
– (2015). “Robust discretization of nonlocal models related to peridynamics”. In: Mesh-
free methods for partial differential equations VII. Vol. 100. Lect. Notes Comput. Sci.
Eng. Springer, Cham, pp. 97–113. MR: 3587379 (cit. on pp. 3561, 3569, 3570).
– (2017). “Mathematics of smoothed particle hydrodynamics, part I: nonlocal Stokes
equation” (cit. on p. 3581).
Qiang Du and Jiang Yang (2016). “Asymptotically compatible Fourier spectral approxima-
tions of nonlocal Allen-Cahn equations”. SIAM J. Numer. Anal. 54.3, pp. 1899–1919.
MR: 3514714 (cit. on p. 3572).
 NONLOCAL MODELS 3585

– (2017). “Fast and accurate implementation of Fourier spectral approximations of non-

local diffusion operators and its applications”. J. Comput. Phys. 332, pp. 118–134. MR:
3591174 (cit. on p. 3568).
Qiang Du, Jiang Yang, and Zhi Zhou (2017). “Analysis of a nonlocal-in-time parabolic
equation”. Discrete Contin. Dyn. Syst. Ser. B 22.2, pp. 339–368. MR: 3639119 (cit. on
pp. 3576, 3577).
Qiang Du and Kun Zhou (2011). “Mathematical analysis for the peridynamic nonlocal con-
tinuum theory”. ESAIM Math. Model. Numer. Anal. 45.2, pp. 217–234. MR: 2804637
(cit. on p. 3561).
Qiang Du and Z. Zhou (2018). “A nonlocal-in-time dynamic system for anomalous diffu-
sion”. Preprint (cit. on p. 3577).
Qiang Du and Zhi Zhou (2017). “Multigrid finite element method for nonlocal diffusion
equations with a fractional kernel” (cit. on p. 3580).
Charles Fefferman, Sergei Ivanov, Yaroslav Kurylev, Matti Lassas, and Hariharan Nara-
yanan (Aug. 2015). “Reconstruction and interpolation of manifolds I: The geometric
Whitney problem”. arXiv: 1508.00674 (cit. on p. 3578).
MA Fuentes, MN Kuperman, and VM Kenkre (2003). “Nonlocal interaction effects on
pattern formation in population dynamics”. Physical review letters 91.15, p. 158104
(cit. on p. 3580).
Yves van Gennip and Andrea L. Bertozzi (2012). “Γ-convergence of graph Ginzburg-
Landau functionals”. Adv. Differential Equations 17.11-12, pp. 1115–1180. MR: 3013414
(cit. on p. 3579).
Guy Gilboa and Stanley Osher (2008). “Nonlocal operators with applications to image pro-
cessing”. Multiscale Model. Simul. 7.3, pp. 1005–1028. MR: 2480109 (cit. on p. 3579).
R. Gingold and J. J. Monaghan (1977). “Smoothed particle hydrodynamics: theory and
application to non-spherical stars”. Monthly Notices of the Royal Astronomical Society
181, pp. 375–389 (cit. on p. 3581).
Jean-Luc Guermond and Guido Kanschat (2010). “Asymptotic analysis of upwind dis-
continuous Galerkin approximation of the radiative transport equation in the diffusive
limit”. SIAM J. Numer. Anal. 48.1, pp. 53–78. MR: 2608358 (cit. on p. 3569).
Wei He, Hao Song, Yun Su, Ling Geng, Bruce J Ackerson, HB Peng, and Penger Tong
(2016). “Dynamic heterogeneity and non-Gaussian statistics for acetylcholine receptors
on live cell membrane”. Nature communications 7, p. 11701 (cit. on pp. 3577, 3578).
Jae-Hyung Jeon, Hector Martinez-Seara Monne, Matti Javanainen, and Ralf Metzler (2012).
“Anomalous diffusion of phospholipids and cholesterols in a lipid bilayer and its ori-
gins”. Physical review letters 109.18, p. 188103 (cit. on p. 3577).
Shi Jin (1999). “Efficient asymptotic-preserving (AP) schemes for some multiscale kinetic
equations”. SIAM J. Sci. Comput. 21.2, pp. 441–454. MR: 1718639 (cit. on p. 3569).
 3586 QIANG DU (杜强)

Chiu-Yen Kao, Yuan Lou, and Wenxian Shen (2010). “Random dispersal vs. non-local dis-
persal”. Discrete Contin. Dyn. Syst. 26.2, pp. 551–596. MR: 2556498 (cit. on p. 3580).
Nickolay Korabel and Eli Barkai (2010). “Paradoxes of subdiffusive infiltration in disor-
dered systems”. Physical review letters 104.17, p. 170603 (cit. on p. 3577).
Takashi Kumagai (2014). “Anomalous random walks and diffusions: From fractals to ran-
dom media”. In: Proceedings of ICM. Vol. IV. Citeseer, pp. 75–94 (cit. on p. 3580).
Thinh Duc Le (2012). “Nonlocal Exterior Calculus on Riemannian Manifolds”. PhD the-
sis. Penn State University (cit. on p. 3578).
Bruno Lévy (2006). “Laplace-beltrami eigenfunctions towards an algorithm that” under-
stands” geometry”. In: Shape Modeling and Applications, 2006. SMI 2006. IEEE Inter-
national Conference on. IEEE, pp. 13–21 (cit. on p. 3578).
Xingjie Helen Li and Jianfeng Lu (2017). “Quasi-nonlocal coupling of nonlocal diffu-
sions”. SIAM J. Numer. Anal. 55.5, pp. 2394–2415. MR: 3709890 (cit. on pp. 3572,
3580).
Yifei Lou, Xiaoqun Zhang, Stanley Osher, and Andrea Bertozzi (2010). “Image recov-
ery via nonlocal operators”. J. Sci. Comput. 42.2, pp. 185–197. MR: 2578033 (cit. on
p. 3579).
Annalisa Massaccesi and Enrico Valdinoci (2017). “Is a nonlocal diffusion strategy con-
venient for biological populations in competition?” J. Math. Biol. 74.1-2, pp. 113–147.
MR: 3590678 (cit. on p. 3580).
Scott A McKinley, Lingxing Yao, and M Gregory Forest (2009). “Transient anomalous
diffusion of tracer particles in soft matter”. Journal of Rheology 53.6, pp. 1487–1506
(cit. on p. 3577).
Tadele Mengesha (2012). “Nonlocal Korn-type characterization of Sobolev vector fields”.
Commun. Contemp. Math. 14.4, pp. 1250028, 28. MR: 2965673 (cit. on pp. 3565,
3566).
Tadele Mengesha and Qiang Du (2014). “The bond-based peridynamic system with Dirichlet-
type volume constraint”. Proc. Roy. Soc. Edinburgh Sect. A 144.1, pp. 161–186. MR:
3164542 (cit. on pp. 3562, 3565, 3566).
– (2015). “On the variational limit of a class of nonlocal functionals related to peridy-
namics”. Nonlinearity 28.11, pp. 3999–4035. MR: 3424902 (cit. on pp. 3565–3567).
– (2016). “Characterization of function spaces of vector fields and an application in non-
linear peridynamics”. Nonlinear Anal. 140, pp. 82–111. MR: 3492730 (cit. on pp. 3566,
3567).
Ralf Metzler and Joseph Klafter (2004). “The restaurant at the end of the random walk:
recent developments in the description of anomalous transport by fractional dynamics”.
J. Phys. A 37.31, R161–R208. MR: 2090004 (cit. on p. 3576).
J. J. Monaghan (2005). “Smoothed particle hydrodynamics”. Rep. Progr. Phys. 68.8, pp. 1703–
1759. MR: 2158506 (cit. on p. 3581).
 NONLOCAL MODELS 3587

Sebastien Motsch and Eitan Tadmor (2014). “Heterophilious dynamics enhances consen-
sus”. SIAM Rev. 56.4, pp. 577–621. MR: 3274797 (cit. on p. 3580).
Ricardo H. Nochetto, Enrique Otárola, and Abner J. Salgado (2016). “A PDE approach to
space-time fractional parabolic problems”. SIAM J. Numer. Anal. 54.2, pp. 848–873.
MR: 3478958 (cit. on p. 3561).
Ricardo H Nochetto and Wujun Zhang (2017). “Discrete ABP estimate and convergence
rates for linear elliptic equations in non-divergence form”. Foundations of Computa-
tional Mathematics, pp. 1–57 (cit. on p. 3581).
Michael L. Parks, D. J. Littlewood, J. A. Mitchell, and Stewart A. Silling (2012). Peridigm
users’guide. Tech. rep. (cit. on p. 3568).
Michael L. Parks, Pablo Seleson, Steven J. Plimpton, Stewart A. Silling, and Richard B.
Lehoucq (2011). “Peridynamics with LAMMPS: A User Guide, v0. 3 Beta”. Sandia
Report (2011–8253) (cit. on p. 3568).
Augusto C. Ponce (2004). “An estimate in the spirit of Poincaré’s inequality”. J. Eur. Math.
Soc. (JEMS) 6.1, pp. 1–15. MR: 2041005 (cit. on p. 3566).
Bo Ren, CT Wu, and E Askari (2017). “A 3D discontinuous Galerkin finite element
method with the bond-based peridynamics model for dynamic brittle failure analysis”.
International Journal of Impact Engineering 99, pp. 14–25 (cit. on p. 3568).
Michael J Saxton (2012). “Wanted: a positive control for anomalous subdiffusion”. Bio-
physical journal 103.12, pp. 2411–2422 (cit. on p. 3577).
Pablo Seleson, Qiang Du, and Michael L. Parks (2016). “On the consistency between
nearest-neighbor peridynamic discretizations and discretized classical elasticity mod-
els”. Comput. Methods Appl. Mech. Engrg. 311, pp. 698–722. MR: 3564707 (cit. on
p. 3568).
Roman Shvydkoy and Eitan Tadmor (2017). “Eulerian dynamics with a commutator forc-
ing”. Transactions of Mathematics and its Applications 1.1 (cit. on p. 3580).
Stewart A. Silling (2000). “Reformulation of elasticity theory for discontinuities and long-
range forces”. J. Mech. Phys. Solids 48.1, pp. 175–209. MR: 1727557 (cit. on p. 3562).
– (2010). “Linearized theory of peridynamic states”. J. Elasticity 99.1, pp. 85–111. MR:
2592410 (cit. on p. 3567).
Stewart A. Silling, M. Epton, O. Weckner, J. Xu, and E. Askari (2007). “Peridynamic
states and constitutive modeling”. J. Elasticity 88.2, pp. 151–184. MR: 2348150 (cit.
on p. 3567).
Stewart A. Silling and Richard B. Lehoucq (2010). “Peridynamic theory of solid mechan-
ics”. In: Advances in applied mechanics. Vol. 44. Elsevier, pp. 73–168 (cit. on p. 3567).
Luis Silvestre (2014). “Regularity estimates for parabolic integro-differential equations
and applications”. In: Proceedings of the International Congress of Mathematicians—
Seoul 2014. Vol. III. Kyung Moon Sa, Seoul, pp. 873–894. MR: 3729056 (cit. on
p. 3579).
 3588 QIANG DU (杜强)

Amit Singer and Hau-Tieng Wu (2017). “Spectral convergence of the connection Lapla-
cian from random samples”. Inf. Inference 6.1, pp. 58–123. MR: 3636868 (cit. on
p. 3579).
Igor M Sokolov (2012). “Models of anomalous diffusion in crowded environments”. Soft
Matter 8.35, pp. 9043–9052 (cit. on pp. 3576, 3577).
Daniel A. Spielman (2010). “Algorithms, graph theory, and linear equations in Laplacian
matrices”. In: Proceedings of the International Congress of Mathematicians. Volume
IV . Hindustan Book Agency, New Delhi, pp. 2698–2722. MR: 2827990 (cit. on p. 3579).
Yunzhe Tao, Xiaochuan Tian, and Qiang Du (2017). “Nonlocal diffusion and peridynamic
models with Neumann type constraints and their numerical approximations”. Appl.
Math. Comput. 305, pp. 282–298. MR: 3621707 (cit. on p. 3568).
Hao Tian, Lili Ju, and Qiang Du (2017). “A conservative nonlocal convection-diffusion
model and asymptotically compatible finite difference discretization”. Comput. Meth-
ods Appl. Mech. Engrg. 320, pp. 46–67. MR: 3646345 (cit. on pp. 3568, 3576).
Hao Tian, Hong Wang, and Wenqia Wang (2013). “An efficient collocation method for
a non-local diffusion model”. Int. J. Numer. Anal. Model. 10.4, pp. 815–825. MR:
3125859 (cit. on p. 3568).
Xiaochuan Tian (2017). Nonlocal models with a finite range of nonlocal interactions.
Thesis (Ph.D.)–Columbia University. ProQuest LLC, Ann Arbor, MI, p. 221. MR:
3641030 (cit. on p. 3575).
Xiaochuan Tian and Qiang Du (2013). “Analysis and comparison of different approxima-
tions to nonlocal diffusion and linear peridynamic equations”. SIAM J. Numer. Anal.
51.6, pp. 3458–3482. MR: 3143839 (cit. on pp. 3568–3571).
– (2014). “Asymptotically compatible schemes and applications to robust discretization
of nonlocal models”. SIAM J. Numer. Anal. 52.4, pp. 1641–1665. MR: 3231986 (cit. on
pp. 3568, 3570, 3571, 3575).
– (2015). “Nonconforming discontinuous Galerkin methods for nonlocal variational prob-
lems”. SIAM J. Numer. Anal. 53.2, pp. 762–781. MR: 3323545 (cit. on pp. 3568, 3572).
– (2017). “Trace theorems for some nonlocal function spaces with heterogeneous local-
ization”. SIAM J. Math. Anal. 49.2, pp. 1621–1644. MR: 3640623 (cit. on pp. 3572–
3574).
Xiaochuan Tian, Qiang Du, and Max Gunzburger (2016). “Asymptotically compatible
schemes for the approximation of fractional Laplacian and related nonlocal diffusion
problems on bounded domains”. Adv. Comput. Math. 42.6, pp. 1363–1380. MR: 3571209
(cit. on p. 3571).
Nicolás Garcı́a Trillos and Dejan Slepčev (2016). “A variational approach to the consis-
tency of spectral clustering”. Applied and Computational Harmonic Analysis (cit. on
p. 3579).
 NONLOCAL MODELS 3589

Juan Luis Vázquez (2017). The mathematical theories of diffusion: Nonlinear and frac-
tional diffusion. Springer Lecture Notes in Mathematics, CIME Subseries (cit. on pp. 3561,
3578).
Hong Wang and Hao Tian (2012). “A fast Galerkin method with efficient matrix assembly
and storage for a peridynamic model”. J. Comput. Phys. 231.23, pp. 7730–7738. MR:
2972865 (cit. on p. 3580).
Bruce J. West (2016). Fractional calculus view of complexity. Tomorrow’s science, With
a foreword by Chris Arney. CRC Press, Boca Raton, FL, pp. xviii+285. MR: 3444156
(cit. on pp. 3561, 3578, 3579).
V. Zaburdaev, S. Denisov, and J. Klafter (2015). “Lévy walks”. Rev. Modern Phys. 87.2,
pp. 483–530. MR: 3403266 (cit. on p. 3580).
Xiaoping Zhang, Max Gunzburger, and Lili Ju (2016a). “Nodal-type collocation methods
for hypersingular integral equations and nonlocal diffusion problems”. Comput. Meth-
ods Appl. Mech. Engrg. 299, pp. 401–420. MR: 3434921 (cit. on p. 3568).
– (2016b). “Quadrature rules for finite element approximations of 1D nonlocal prob-
lems”. J. Comput. Phys. 310, pp. 213–236. MR: 3457967 (cit. on p. 3568).
Kun Zhou and Qiang Du (2010). “Mathematical and numerical analysis of linear peridy-
namic models with nonlocal boundary conditions”. SIAM J. Numer. Anal. 48.5, pp. 1759–
1780. MR: 2733097 (cit. on p. 3568).

Received 2017-12-03.

Q D (杜强)
[email protected]
D A P A M
C U
N Y , NY 10027
USA