Comput Mech (2014) 53:1251–1264

DOI 10.1007/s00466-013-0969-x

ORIGINAL PAPER

A meshfree unification: reproducing kernel peridynamics

Received: 5 December 2013 / Accepted: 18 December 2013 / Published online: 19 January 2014
© Springer-Verlag Berlin Heidelberg 2014

Abstract This paper is the first investigation establish- theory: (1) the theoretical aspects, since it is presented as
ing the link between the meshfree state-based peridynam- an alternative formulation to classical continuum mechanics
ics method and other meshfree methods, in particular with in the sense that the equations of motion are derived from
the moving least squares reproducing kernel particle method integral expressions instead of partial differential equations;
(RKPM). It is concluded that the discretization of state-based (2) the numerical aspects, since its natural discretization leads
peridynamics leads directly to an approximation of the deriv- to a meshfree strong form method that has been used to solve
atives that can be obtained from RKPM. However, state- different solid mechanics problems.
based peridynamics obtains the same result at a significantly In terms of the theoretical aspects, the work of Silling and
lower computational cost which motivates its use in large- Lehoucq [2] is of particular importance since it proved that
scale computations. In light of the findings of this study, an the state-based peridynamic theory converges to the classical
update to the method is proposed such that the limitations theory of elasticity. This implies that any constitutive model
regarding application of boundary conditions and the use of formulated in continuum mechanics can be used in state-
non-uniform grids are corrected by using the reproducing based peridynamics which was not the case for the previous
kernel approximation. theory, called bond-based peridynamics [3]. This opened the
way so that different authors could apply state-based peridy-
Keywords State-based peridynamics · Meshfree methods · namics to different solid mechanics problems. Warren et al.
Reproducing kernel · Moving least squares · Strong form [4] applied this method to predict fracture for both notched
discretization · Weak form discretization and pristine bars subjected to tension. Foster et al. [5] imple-
mented an elastic viscoplastic constitutive model in order to
reproduce the experimental results from Taylor impact tests
1 Introduction of aluminum. Tupek et al. [6] implemented a classical con-
tinuum damage model by modifying the peridynamic weight
In this article we will explore the recently developed state- function according to the state of accumulated damage.
based peridynamic theory that was introduced by Silling et However, the numerical aspects of state-based peridynam-
al. [1]. There are fundamentally two points of interest in this ics are the main focus of this work. In particular, the goal is
to find the missing link between this strong form method and
other meshfree methods [7,8]. In addition, the reproducing
M. A. Bessa · T. Belytschko · W. K. Liu (B)
Department of Mechanical Engineering, Northwestern University, kernel peridynamics method is introduced such that some
2145 Sheridan Rd., Evanston, IL 60208, USA of the limitations of the original formulation are mitigated.
e-mail: [email protected] More importantly, understanding the underlying numerics of
M. A. Bessa state-based peridynamics may open new avenues for future
e-mail: [email protected] research in the method and offer explanations for some of
the results obtained previously in the literature.
J. T. Foster
Mechanical Engineering Department, The University of Texas at San This article is organized as follows. Section 2 reviews how
Antonio, One UTSA Circle, San Antonio, TX 78249, USA meshfree methods using polynomial basis functions can be

123
1254 Comput Mech (2014) 53:1251–1264

The above expression, or particular cases of it, has been methods is given by what independent variable is allowed to
rediscovered several times in different fields, as will be clari- vary: x or x . If x is the one varying around a fixed point x,
fied in the next subsection. Note that Eq. (19) can be derived then the approximation method is usually renamed to WLS
from a particular case of MLS approximation: the WLS. This instead of MLS because the coefficients a(x) become con-
follows from the fact that the coefficients a(x I ) are constant stant. In addition, if a unity weight function ω(x − x J ) is
because they are calculated at fixed points. used then the approximation method is referred to as LS.
Appendix provides a short review of the meshfree meth-
2.3 Unification of different meshfree methods using ods based on the LS approximation due to their importance
polynomial basis for our comparison with state-based peridynamics. Providing
additional details about the remaining methods mentioned
The most remarkable aspect of Eq. (11) is that it is sufficiently is considered unnecessary since they are well documented
general so that particular cases of this equation lead to several in the given references. The interested reader is referred to
different numerical methods, often times used in completely [8,11,30] for a review of these and other meshfree methods.
different scientific disciplines. The aim of this section is to highlight that the general
equation (11) is the origin for different meshfree methods,
– Fixed x and varying x :
and that the differences between those methods arise from:
1. Least Squares (LS) approximation methods, where (1) the use of constant or non-constant coefficients; (2) the
ω(x − x J ) = 1: number of monomials m of the local polynomial approxima-
(a) L = m (overfitting): tion when compared to the number of local points L in the
i. Classic and generalized Finite Difference supporting size; (3) the presence or absence of the weight
Method (FDM). [26] function.
(b) L < m (smooth approximations or fitting): The first point listed above is related to the use of weak
i. Discretized Lanczos derivatives. [27] or strong forms, since constant coefficients are calculated
ii. Savitzky-Golay filters. [28] at fixed points that can be used for collocation, while non-
2. Weighted Least Squares (WLS) approximation meth- constant coefficients are well suited for variational meth-
ods, where ω(x − x J ) = 1: ods because they form shape functions defined in the entire
(a) L = m (overfitting): domain. The second point is related to having an interpola-
i. Studied in [19]. tory method (L = m) where there is overfitting of the data,
(b) L < m (fitting): or related to having a smoothing method (L > m) where
i. Finite Point Method (FPM). [19] the data is fit according to LS, WLS or MLS. Interpolatory
ii. RKPM using synchronized derivatives. [22] methods lead to shape functions that have the Kronecker delta
iii. Meshfree Point Collocation Method. [23– property, since by definition the polynomial approximation
25] passes through the data points, at the expense that these meth-
ods tend to produce noisy responses in the presence of noisy
– Varying x and fixed x :
data [26]. Smoothing methods like MLS/RKPM do not have
1. Moving Least Squares (MLS) approximation meth- the Kronecker delta property [9], which causes difficulties
ods: in the imposition of essential boundary conditions, as will
(a) L = m (overfitting) and ω(x − x J ) = 1: be reviewed towards the end of this article. Finally, the third
i. Polynomial Point Interpolation Method point has been proven to be related to a smaller (or larger)
(PPIM). [29] sensitivity of the approximation provided by the methods
(b) L < m (fitting) and ω(x − x J ) = 1: when the weight function is present (or not present) [19].
i. MLS/RKPM original formulations. [7,13, We want to highlight that only meshfree methods with
15] polynomial basis functions are being considered here because
(c) m = 1, i.e. p(x) = p(x) = 1, and ω(x−x J ) = 1: it was found that state-based peridynamics can be included
i. Smooth Particle Hydrodynamics (SPH). in these types of meshfree methods (as will be shown later in
[18] the paper). Consequently, the increasingly attractive mesh-
free formulations using radial basis functions that are locally
The above diagram summarizes different meshfree meth- supported [31,32] or globally supported [33,34] are out of the
ods that are based on locally supported polynomial basis scope of this work. For the same reason, the recent advances
functions. All of the above mentioned methods are special on methods using spline basis functions such as the isogeo-
cases of Eq. (11). The most fundamental difference between metric collocation method [35,36] are not included here.

123
1264 Comput Mech (2014) 53:1251–1264

ments. Comput Methods Appl Mech Eng 139(1–4):3. doi:10.1016/ 29. Liu GR, Gu Y (2001) A point interpolation method for two-
S0045-7825(96)01078-X dimensional solids. Int J Numer Methods Eng 50(4):937
10. Li S, Liu WK (2004) Meshfree particle methods. Springer, Berlin 30. Liu GR (2010) Meshfree methods: moving beyond the finite ele-
11. Nguyen VP, Rabczuk T, Bordas S, Duflot M (2008) Meshless meth- ment method. CRC Press, Boca Raton
ods: a review and computer implementation aspects. Math Comput 31. Wang JG, Liu GR (2002) A point interpolation meshless method
Simul 79(3):763. doi:10.1016/j.matcom.2008.01.003 based on radial basis functions. Int J Numer Methods Eng
12. Nayroles B, Touzot G, Villon P (1992) Generalizing the finite ele- 54(11):1623. doi:10.1002/nme.489
ment method: diffuse approximation and diffuse elements. Comput 32. Liu G, Kee BB, Chun L (2006) A stabilized least-squares radial
Mech 10(5):307. doi:10.1007/BF00364252 point collocation method (LS-RPCM) for adaptive analysis. Com-
13. Belytschko T, Lu YY, Gu L (1994) Element-free Galerkin meth- put Methods Appl Mech Eng 195(37):4843
ods. Int J Numer Methods Eng 37(2):229. doi:10.1002/nme. 33. Hu HY, Chen JS, Hu W (2007) Weighted radial basis collocation
1620370205 method for boundary value problems. Int J Numer Methods Eng
14. Liu WK, Jun S, Li S, Adee J, Belytschko T (1995) Reproducing 69(13):2736. doi:10.1002/nme.1877
kernel particle methods for structural dynamics. Int J Numer Meth- 34. Wang L, Chen JS, Hu HY (2010) Subdomain radial basis collo-
ods Eng 38(10):1655. doi:10.1002/nme.1620381005 cation method for fracture mechanics. Int J Numer Methods Eng
15. Liu WK, Li S, Belytschko T (1997) Moving least-square 83(7):851. doi:10.1002/nme.2860
reproducing kernel methods (I) Methodology and convergence. 35. Auricchio F, Veiga LBD, Hughes TJR, Reali A, Sangalli G (2010)
Comput Methods Appl Mech Eng 143(1–2):113. doi:10.1016/ Isogeometric collocation methods. Math Models Methods Appl Sci
S0045-7825(96)01132-2 20(11):2075. doi:10.1142/S0218202510004878
16. Liu WK, Chen Y, Uras R, Chang CT (1996) Generalized multiple 36. Auricchio F, da Veiga LB, Hughes T, Reali A, Sangalli G (2012)
scale reproducing kernel particle methods. Comput Methods Appl Isogeometric collocation for elastostatics and explicit dynamics.
Mech Eng 139(1–4):91. doi:10.1016/S0045-7825(96)01081-X Comput Methods Appl Mech Eng 2:249–252. doi:10.1016/j.cma.
17. Liu W, Chen Y, Chang C, Belytschko T (1996) Advances in multi- 2012.03.026
ple scale kernel particle methods. Comput Mech 18(2):73. doi:10. 37. Lehoucq R, Silling S (2008) Force flux and the peridynamic stress
1007/BF00350529 tensor. J Mech Phys Solids 56(4):1566. doi:10.1016/j.jmps.2007.
18. Gingold RA, Monaghan JJ (1977) Smoothed particle 08.004
hydrodynamics-theory and application to non-spherical stars. 38. Zhou K, Du Q (2010) Mathematical and numerical analysis of lin-
Mon Notices R Astron Soc 181:375 ear peridynamic models with nonlocal boundary conditions. SIAM
19. Oñ ate E, Idelsohn S, Zienkiewicz OC, Taylor RL (1996) Finite J Numer Anal 48(5):1759. doi:10.1137/090781267
point method in computational mechanics. Applications to convec- 39. Silling S, Lehoucq R (2010) Peridynamic theory of solid mechan-
tive transport and fluid flow. Int J Numer Methods Eng 39(22):3839 ics. Adv Appl Mech 44:73. doi:10.1016/S0065-2156(10)44002-8
20. Kim DW, Liu WK (2006) Maximum principle and convergence 40. Seleson P, Parks M (2011) On the role of the influence function in
analysis for the meshfree point collocation method. SIAM J Numer the peridynamic theory. Int J Multiscale Comput Eng 9(6):689
Anal 44(2):515 41. Diekema E, Koornwinder TH (2012) Differentiation by integra-
21. Hu HY, Chen JS, Hu W (2011) Error analysis of collocation method tion using orthogonal polynomials, a survey. J Approx Theory
based on reproducing kernel approximation. Numer Methods Par- 164(5):637. doi:10.1016/j.jat.2012.01.003
tial Differ Equ 27(3):554. doi:10.1002/num.20539 42. Foster J, Silling SA, Chen W (2011) An energy based failure crite-
22. Li S, Liu WK (1998) Synchronized reproducing kernel interpolant rion for use with peridynamic states. Int J Multiscale Comput Eng
via multiple wavelet expansion. Comput Mech 21(1):28. doi:10. 9(6):675
1007/s004660050281 43. Zhu T, Atluri S (1998) A modified collocation method and a penalty
23. Kim DW, Yoon YC, Liu WK, Belytschko T (2007) Extrinsic mesh- formulation for enforcing the essential boundary conditions in the
free approximation using asymptotic expansion for interfacial dis- element free Galerkin method. Comput Mech 21(3):211
continuity of derivative. J Comput Phys 221(1):370. doi:10.1016/ 44. Ferná ndez-Mé ndez S, Huerta A (2004) Imposing essential bound-
j.jcp.2006.06.023 ary conditions in mesh-free methods. Comput Methods Appl Mech
24. Yoon YC, Song JH (2013) Extended particle difference method for Eng 193(12):1257
weak and strong discontinuity problems: part II. Formulations and 45. Tian R (2013) Extra-dof-free and linearly independent enrichments
applications for various interfacial singularity problems. Comput in GFEM. Comput Meth Appl Mech Eng 266:1. doi:10.1016/j.cma.
Mech 13:1–24. doi:10.1007/s00466-013-0951-7 2013.07.005
25. Yoon YC, Song YH (2013) Extended particle difference method for 46. Chen JS, Hu W, Hu HY (2008) Reproducing kernel enhanced
weak and strong discontinuity problems: part I. Derivation of the local radial basis collocation method. Int J Numer Methods Eng
extended particle derivative approximation for the representation 75(5):600. doi:10.1002/nme.2269
of weak and strong discontinuities. Comput Mech 3:1–17. doi:10. 47. Chi SW, Chen JS, Luo H, Hu HY, Wang L (2012) Dispersion and
1007/s00466-013-0950-8 stability properties of radial basis collocation method for elasto-
26. Fornberg B (1988) Generation of finite difference formulas on arbi- dynamics. Numer Methods Partial Diff Equ 29:818–842. doi:10.
trarily spaced grids. Math Comput 51(184):699 1002/num.21732
27. Lanczos C (1956) Applied analysis. Prentice-Hall, Old Tappan 48. Hamming RW (1989) Digital filters. Courier Dover Publications,
28. Savitzky A, Golay MJE (1964) Smoothing and differentiation of Cambridge
data by simplified least squares procedures. Anal Chem 36(8):1627.
doi:10.1021/ac60214a047

123