The interpolating moving least-squares (IMLS) method is discussed in detail, and a simpler formula of the shape function of the IMLS method is obtained. Then, based on the IMLS method and the Galerkin weak form, an interpolating element-free Galerkin (IEFG) method for two-point boundary value problems is presented. The IEFG method has high computing speed and precision. Then error analysis of the IEFG method for two-point boundary value problems is presented. The convergence rates of the numerical solution and its derivatives of the IEFG method are presented. The theories show that, if the original solution is sufficiently smooth and the order of the basis functions is big enough, the solution of the IEFG method and its derivatives are convergent to the exact solutions in terms of the maximum radius of the domains of influence of nodes. For the purpose of demonstration, two selected numerical examples are given to confirm the theories.
1. Introduction
Conventional computational methods, such as the finite element method (FEM) and the boundary element method (BEM), cannot be applied well to some engineering problems. For the extremely large deformation and crack growth problems, the remeshing technique must be used. Meshless methods have been developed in recent years. The most important common feature of meshless methods is that the trial function is constructed from a set of nodes with no meshing at all. Then some complex problems, such as the large deformation and crack growth problems, can be simulated with the method without the remeshing techniques [1, 2].
Many kinds of meshless methods have been proposed, such as element-free Galerkin (EFG) method [3–5], complex variable meshless method [6–11], mesh-free reproducing kernel particle Ritz method [12], radial basis function (RBF) method [13], finite point method (FPM) [14–16], meshless local Petrov-Galerkin (MLPG) method [17], reproducing kernel particle method (RKPM) [18–22], meshless manifold method [23–29], boundary element-free method (BEFM) [30–40], and local boundary integral equation (LBIE) method [41, 42].
The element-free Galerkin (EFG) method is one of the most powerful meshless methods [3]. The EFG method can obtain a solution with high precision. Various problems have been successfully analyzed by the EFG method. By using the orthogonal function system with a weight function as the basis function, Zhang et al. presented the improved element-free Galerkin method [43–48], which has high computational efficiency. By combining the complex variable moving least-squares (CVMLS) approximation and the EFG method, Peng et al. presented the complex variable element-free Galerkin (CVEFG) method [49–56]. Compared with the conventional EFG methods, the CVEFG method has greater computational precision and efficiency.
The EFG method is constructed based on the moving least-squares (MLS) approximation. The shape function that is formed with MLS approximation can obtain a solution with high precision. However, a disadvantage of the MLS approximation is that its shape function does not satisfy the property of Kronecker δ function. Then the EFG method based on the MLS approximation cannot apply the essential boundary conditions directly and easily. The essential boundary conditions need to be introduced by additional approaches, such as Lagrange multipliers [3] and penalty methods [57]. However, for Lagrange multipliers, the corresponding discrete system will introduce additional unknowns which are not directly associated with the solution themselves. Furthermore, the banded structure of the matrix equation system is seriously worsened, as well as the conditioning properties. And, for penalty methods, the optimal value of penalty factor is hard to be set, which always affects the accuracy of the final solution.
To overcome this problem, Most and Bucher designed a regularized weight function with a regularization parameter ε, by which the MLS approximation can almost fulfill the interpolation with high accuracy [58]. Most and Bucher enhanced the regularized weighting function to obtain a true interpolation MLS approximation [59]. Another possible solution for this problem is the interpolating moving least-squares (IMLS) method presented by Lancaster and Salkauskas [60]. The IMLS method is established based on the MLS approximation by using singular weight functions. The shape function of the IMLS method satisfies the property of Kronecker δ function. Thus the meshless methods based on the IMLS method can apply the essential boundary condition directly without any additional numerical effort. Based on the IMLS method, Kaljević and Saigal [61] presented an improved EFG method, in which the boundary condition is applied directly. Ren improved the expression of the shape function of the IMLS method and then presented the interpolating element-free Galerkin (IEFG) method and the interpolating boundary element-free (IBEFG) method for two-dimensional potential and elasticity problems [62–65]. To overcome the singularity of the weight function in the IMLS method, Wang et al. presented the improved interpolating moving least-squares (IIMLS) method, in which nonsingular weight function is used [66–68]. In the IEFG method, the essential boundary conditions are applied directly and easily, and the number of unknown coefficients in the trial function of the IMLS method is less than that in the trial function of the MLS approximation. Therefore, the IEFG method based on the IMLS method has high computational efficiency and precision.
Error estimation for meshless method is certainly important to increase the reliability and reduce the cost of numerical computations in many engineering problems. Some error analyses have been done for the MLS approximation and the meshless method based on it [69–74]. Krysl and Belytschko studied the convergence of the continuous and discontinuous shape functions of the second-order elliptic partial differential equations [75]. Chung and Belytschko proposed the local and global error estimates using the difference between the values of the projected stress and these given directly by the EFG solution [76]. Dolbow and Belytschko studied the integration error [77]. Gavete et al. presented a procedure to estimate the error in elliptic equations and then proposed a posteriori error approximation [78, 79]. R. J. Cheng and Y. M. Cheng studied the error estimate of the finite point method based on the MLS approximation [80] and the error estimates of element-free Galerkin method for potential and elasticity problems [81, 82]. For the IEFG method, since the essential boundary condition is applied directly, then the error estimate of the IEFG method is no doubt different from that of the EFG method. However, until now the error analysis of the IEFG method has not been seen in the recent literature.
Two-point boundary value problems occur in applied mathematics, theoretical physics, engineering, and optimization theory. Since it is usually impossible to obtain analytical solutions to two-point boundary value problems met in practice, these problems must be attacked by numerical methods. Many numerical methods have been proposed for the solutions of these problems, such as the Galerkin and collocation methods, boundary value method, variational iteration method, and meshless method based radial basis functions.
In this paper, the IMLS method is discussed in detail. The computation of the shape function of this paper is simpler than the corresponding expression presented by Lancaster and Salkauskas. Then based on the IMLS method of this paper and the Galerkin weak form, an IEFG method for two-point boundary value problems is presented. Since the shape function of the IMLS method satisfies the property of Kronecker δ function, then the IEFG method can apply the essential boundary condition directly. And as the number of the coefficients in the trial function of the IMLS method is less than that in the MLS approximation, then fewer nodes are selected in the entire domain in the IEFG method than in the conventional EFG method. Hence, the IEFG method has high computing speed and precision.
Then the error analysis of the IEFG method for two-point boundary value problems is presented. The convergence rates of the numerical solution and its derivatives of the IEFG method are presented. The theoretical results show that if the exact solution is smooth enough and the order of the polynomial basis functions is big enough, then the solution of the IEFG method and its derivatives are convergent to the exact solutions in terms of the maximum radius of the domains of influence of nodes. For the purpose of demonstration, some selected numerical examples are given to confirm the theory.
2. Interpolating Moving Least-Squares Method
Let X={x1,x2,…,xM} be a set of all nodes in the bounded domain Ω⊂Rn, where M is the number of nodes. The parameter ρI denotes the radius of the domain of influence of node xI, and ∥·∥ denotes the Euclidean norm. The domain of influence of xI is defined by
(1)ΩI={x∣∥x-xI∥≤ρI,x∈Ω}.
Let u(x) be the function of the field variable defined in Ω. The approximation function of u(x) is denoted by uh(x). In this paper, the following weight function is used:
(2)w(x,xI)=mI(x)∥x-xIρI∥-α,
where mI(x)=m(x-xI)∈Cl(Ω) satisfying mI(x)>0 for ∥x-xI∥<ρI and mI(x)=0 for ∥x-xI∥≥ρI. In general, mI(x) can be chosen to be any weight function of the MLS approximation.
For a given point x, the inner product is defined as
(3)(f,g)x=∑I=1nw(x,xI)f(xI)g(xI),∀f,g∈C0(Ω),
where n is the number of nodes whose compact support domains cover x.
Let p0(x)≡1 and let p1(x),…,pm(x) be given basis functions. Then a new set of basis functions will be generated from these given basis. Let
(4)p~0(x;x-)=1[∑I=1nw(x,xI)]1/2,(5)p~i(x;x-)=pi(x-)-Spi(x),i=1,2,…,m,
where S is a linear operator defined as
(6)Spi(x)=∑I=1nv(x,xI)pi(xI),(7)v(x,xI)=w(x,xI)∑J=1nw(x,xJ).
Define a local approximation as
(8)uh(x,x-)=p~0(x;x-)a0(x)+∑i=1mp~i(x;x-)ai(x),
where x- is the point in the domain of influence of x, and ai(x) (i=0,1,…,m) are the unknown coefficients of basis functions.
Then define a functional as
(9)κ=∑I=1nw(x,xI)[uh(x,xI)-uI]2,
where w(x,xI) shown in (2) is a weight function with compact support, xI are the nodes with domains of influence that cover the point x, and uI=u(xI).
By minimizing the functional κ, we have
(10)(u(·)-uh(x,·),p~0)x=0,(11)(u(·)-uh(x,·),p~i)x=0,i=1,2,…,m.
From (4) and (10), we have
(12)p~0(x;x-)a0(x)=∑I=1nv(x,xI)u=Su.
Then (11) reduces to
(13)∑i=1mai(x)(p~i,p~j)x=(u-Su,p~j)x,j=1,2,…,m.
In [60], the unknown parameters ai(x)(i=1,2,…,m) are solved from (13). In fact, (13) can be simplified.
If the weight function of (2) is used, ∀x∈Ω, it can be proved that there exists
(14)(Su,p~i)x=0,i=1,2,…,m.
Then (13) can be simplified as
(15)∑i=1mai(x)(p~i,p~j)x=(u,p~j)x,j=1,2,…,m.
Equation (15) is simpler than the expression (13) presented in [60] and can be rewritten as
(16)A(x)a(x)=FW(x)u,
where
(17)aT(x)=(a1(x),a2(x),…,am(x)),(18)uT=(u1,u2,…,un),(19)A(x)=FW(x)FT(x),(20)F(x)=[p~1(x;x1)p~1(x;x2)⋯p~1(x;xn)p~2(x;x1)p~2(x;x2)⋯p~2(x;xn)⋮⋮⋱⋮p~m(x;x1)p~m(x;x2)⋯p~m(x;xn)],
and FW(x)=(ϖkJ(x))m×n is a m×n matrix, and
(21)ϖkJ(x)={w(x,xJ)p~k(x;xJ),x≠xJ;∑I=1,I≠Jnw(xJ,xI)[pk(xJ)-pk(xI)],x=xJ.
From (16), we have
(22)a(x)=A-1(x)FW(x)u.
Then the local approximation function is obtained as
(23)uh(x,x-)=Su+∑i=1mai(x)p~i(x;x-).
Thus the global interpolating approximation function of u(x) can be obtained as
(24)uh(x)=Su+∑i=1mai(x)gi(x)≡𝚽T(x)u,
where 𝚽T(x) is the shape function vector as
(25)𝚽T(x)=vT+pT(x)A-1(x)FW(x),(26)vT=(v(x,x1),v(x,x2),…,v(x,xn)),(27)pT(x)=(g1(x),g2(x),…,gm(x)),(28)gi(x)=pi(x)-Spi(x).
Equation (25) is the shape function of the IMLS method, and then the IMLS method is presented.
Equations (13) and (15) show that the computation of the shape function of this paper is simpler than the corresponding expression in [60]. The reduction of computational amount is related to the order of the operations. And, in any case, the calculation amount of the shape functions of this paper is at least n2 multiplication operations less than that of the IMLS method in [60].
3. Interpolating Element-Free Galerkin Method for Two-Point Boundary Value Problems
Consider the following two-point boundary value problem:
(29)-ddx(pdudx)+qdudx+gu=f,x∈Γ=(a,b),u(a)=u(b)=0,
where p,q,g, and f are known sufficiently smooth functions and p(x)≥pmin>0. ∀f∈C(Γ), suppose that problem (29) has a unique solution.
The Galerkin weak form of (29) is
(30)∫Γδu,xpu,xdx+∫Γδuqu,xdx+∫Γδugudx=∫Γδufdx,
where u,x=du/dx.
From the IMLS method, the unknown solution u(x) at arbitrary field point x in the interval Γ can be expressed as
(31)u(x)≈𝚽T(x)u=∑I=1nΦI(x)uI,
where n is the number of nodes whose compact support domains cover the point x.
Substituting (31) into (30) yields
(32)∫ΓδuT𝚽,x𝚽,xTupdx+∫ΓδuT𝚽𝚽,xTuqdx+∫ΓδuT𝚽𝚽Tugdx=∫ΓδuT𝚽fdx.
Because the nodal test function δu is arbitrary, the final discretized equation of (29) is obtained as
(33)Ku=F,
where
(34)K=∫Γp𝚽,x𝚽,xTdx+∫Γq𝚽𝚽,xTdx+∫Γg𝚽𝚽Tdx,F=∫Γ𝚽fdx.
Since the shape function of the IMLS method satisfies the property of Kronecker δ function, then the essential boundary conditions can be applied directly. Substituting the boundary conditions into (33) directly, we can obtain the unknowns at nodes by solving the linear equations (33).
To evaluate the integrals in (34), it is necessary to generate integration cells over the whole domain of the problem. These cells can be defined arbitrarily, but a sufficient number of quadrature points must be used to obtain a well-conditioned and nonsingular system of (33). In one dimension, one example of a cell structure is to set the quadrature cells equal to the intervals between the nodes. Once the cells and corresponding quadrature points are defined, the discrete equations are assembled by looping over each quadrature point.
The numerical procedure of IEFG method for two-point boundary value problems is listed as follows.
Looping over background cells to determine all Gauss points to find out its location and weight.
Looping over Gauss points for integration of (34) to
determine the support domain for specified Gauss point and select neighboring nodes based on a defined criterion;
compute shape function and its derivatives for each Gauss point;
assemble the contribution of each Gauss point to form system equation.
Solving the system equation to obtain nodal displacements.
Thus the IEFG method is presented for two-point boundary value problems.
4. Error Estimates
In this section, the error analysis of the IEFG method for two-point boundary value problems is presented. The convergence rates of the numerical solution and its derivatives of the IEFG method are presented.
Let z={x1,x2,…,xM} be a set of all nodes in the interval [a,b], where M is the number of nodes. Let ε=minxI,xJ∈z,I≠J{∥xI-xJ∥} and ρ=maxxI∈z{ρI}, where ρI is the radius of the domain of influence of node xI. For a given x, ρx denotes the maximum radius of the influence domains of nodes whose compact support domains cover x. And suppose that there exist constants cε and cI such that ρ≤cεε and ρ≤cIρI, respectively.
Define the Lp Lebesgue space as
(35)Lp(Γ):={f(x):∥f∥Lp(Γ)<∞},1≤p<∞,
where
(36)∥f∥Lp(Γ):=(∫Γfp(x)dx)1/p.
The Sobolev space Hk(Γ) is defined as
(37)Hk(Γ):={f∈Lloc1(Γ):∥f∥Hk(Γ)<∞},
where
(38)∥f∥Hk(Γ):=(∑s≤k∥dsf(x)dxs∥L2(Γ))1/2.
Then the variational problem in accordance with (29) is to find u∈H01(Γ) such that
(40)a(u,v)=(f,v),∀v∈H01(Γ),
where
(41)a(u,v)=∫Γpu,xv,xdx+∫Γqu,xvdx+∫Γguvdx,(f,v)=∫Γfvdx.
Suppose that the bilinear form a(·,·) on the Sobolev space H01(Γ) is bounded and coercive; that is, there exist constants α->0 and M-<∞ such that
(42)|a(u,v)|≤M-∥u∥H1(Γ)∥v∥H1(Γ),∀u,v∈H01(Γ),a(v,v)≥α-∥v∥H1(Γ)2,∀v∈H01(Γ).
Since the shape function of the IMLS method satisfies the property of Kronecker δ function, then the finite-dimensional solution space of the IEFG method for two-point boundary value problems is defined as
(43)Vρ(Γ)={v∣v∈span{ΦI(x),1≤I≤M},v(a)=v(b)=0}.
Then the IEFG method according to (29) is to find uh∈Vρ(Γ) such that
(44)a(uh,v)=(f,v),∀v∈Vρ(Γ).
Obviously, there exists Vρ(Γ)⊂H01(Γ). Hence, the IEFG method for two-point boundary value problems has a unique solution. And the following theorem can be obtained.
Theorem 1.
Suppose that u is the solution of the variational problem (40) and uh is the solution of the IEFG method (44). Then there exist
a(u-uh,v)=0, ∀v∈Vρ(Γ);
a(u-uh,u-uh)=infv∈Vρ(Γ)a(u-v,u-v);
∥u-uh∥H1(Γ)≤Cinfv∈Vρ(Γ)∥u-v∥H1(Γ).
In fact, the approximation function of the IMLS method provides a linear operator 𝒜 defined as
(45)𝒜u≡Su+∑i=1mai(x)gi(x)=𝚽T(x)u.
If u∈Hm+1(Γ), then we have proved that there exist bounded function Ck′(x) and constant Ck such that
(46)dkdxkΦI(x)=Ck′(x)ρx-k,(47)∥𝒜u-u∥Hk(Γ)≤Ckρm+1-k∥u∥Hm+1(Γ),hhhhhhhhh0≤k≤m.
Let ∥u-uh∥L2=a(u-uh,u-uh). Then the following error estimates of the energy norm and the H1 norm can be obtained.
Theorem 2.
Suppose that u∈Hm+1(Γ). Let u and uh be, respectively, the solutions of the problems (40) and (44). Then there exist C1 and C2, which are independent with the parameter ρ, such that
(48)∥u-uh∥L≤C1ρm∥u∥Hm+1(Γ),∥u-uh∥H1(Γ)≤C2ρm∥u∥Hm+1(Γ).
Proof.
From Theorem 1 and (47) we have
(49)∥u-uh∥L2=a(u-uh,u-uh)=infv∈Vρ(Γ)a(u-v,u-v)≤a(u-𝒜u,u-𝒜u)≤M∥u-𝒜u∥H1(Γ)2≤C1ρ2m∥u∥Hm+1(Γ)2,∥u-uh∥H1(Γ)≤Cinfv∈Vρ(Γ)∥u-v∥H1(Γ)≤C∥𝒜u-u∥H1(Γ)≤C2ρm∥u∥Hm+1(Γ).
Then this theorem holds.
By using the Aubin-Nitsche method, the following error estimates in the L2 norm can be obtained.
Theorem 3.
Suppose that u∈Hm+1(Γ). Let u and uh be, respectively, the solutions of the problems (40) and (44). Then there exists a constant C, which is independent with the parameter ρ, such that
(50)∥u-uh∥L2(Γ)≤Cρm+1∥u∥Hm+1(Γ).
Proof.
For ∀g∈L2(Γ), let φ∈H01∩H2 be the solution to the adjoint problems
(51)a(φ,v)=(g,v),∀v∈H01(Γ).
If the coefficients of the bilinear form a(·,·) are sufficiently smooth, then there exists the following estimate:
(52)∥φ∥H2(Γ)≤∥g∥L2(Γ).
Let v=u-uh. Then we have
(53)a(φ,u-uh)=(g,u-uh).
For arbitrary vh∈Vρ, from Theorem 1, we have
(54)a(vh,u-uh)=0.
It follows from (53) and (54) that
(55)a(φ-vh,u-uh)=(g,u-uh).
In (55), if we let 𝒱𝓀=𝒜φ and g=u-uh, then there exists
(56)∥u-uh∥L2(Γ)2=(u-uh,u-uh)=a(φ-𝒜φ,u-uh)≤M∥φ-𝒜φ∥H1(Γ)∥u-uh∥H1(Γ)≤MC1ρ∥φ∥H2(Γ)C2ρm∥u∥Hm+1(Γ).
From (52) and (56), we have
(57)∥u-uh∥L2(Γ)≤Cρm+1∥u∥Hm+1(Γ).
Then this theorem is proved.
To study the error estimates of the high derivatives of the numerical solution of the IEFG method, we need to firstly prove the following inverse estimates of the function in the shape function space.
Theorem 4.
Suppose that ΦI(x) is defined by (25). Then ∀vh(x)∈span{ΦI(x),1≤I≤M}; there exists a constant C, which is independent with the parameter ρ, such that
(58)∥vh(x)∥Hk≤Cρn-k∥vh(x)∥Hn,0≤k≤m,-m≤n≤m.
Proof.
From (46), there exists a bounded function Ck(x) independent with ρ such that
(59)∂kvh(x)=Ck(x)ρx-k,
where ∂kvh(x)=(dk/dxk)vh(x).
It is obvious that there exist bounded functions C1(x) and C2(x) independent with ρ such that
(60)C1(x)ρ≤ρx≤C2(x)ρ.
Then we have
(61)|vh(x)|Hk2=∫Γ[∂kvh(x)]2dx≤∫Γ[Ck(x)C2(x)ρ-k]2dx,hhhhhhhhhhhh0≤k≤m,|vh(x)|Hn2=∫Γ[∂nvh(x)]2dx≥∫Γ[Ck(x)C2(x)ρ-n]2dx.hhhhhhhhhhhh0≤n≤m.
Then it follows from (61) that
(62)∥vh(x)∥Hk≤Cρn-k∥vh(x)∥Hn,0≤k,n≤m,
where C is independent with ρ.
From (62), we have
(63)∥vh(x)∥Hk≤Cρ-k∥vh(x)∥H0,0≤k≤m.
And there exists
(64)∥vh(x)∥H02≤∥vh(x)∥Hn∥vh(x)∥H-n≤∥vh(x)∥Hn·Cρn∥vh(x)∥H0,-m≤n≤0.
From (63) and (64), we have
(65)∥vh(x)∥Hk≤Cρn-k∥vh(x)∥Hn,0≤k≤m,-m≤n≤0.
Then from (62) and (65), this theorem holds.
By using the inverse estimates, the following error estimates of the high derivatives of the numerical solution can be obtained.
Theorem 5.
Suppose that u∈Hm+1(Γ). Let u and uh be, respectively, the solutions of the problems (40) and (44). Then there exists a constant C independent with ρ, such that
(66)∥u-uh∥Hs(Γ)≤Cρm+1-s∥u∥Hm+1(Γ),0≤s≤m.
Proof.
From Theorem 4 we have
(67)∥uh-𝒜u∥Hs(Γ)≤Ch1-s∥uh-𝒜u∥H1(Γ),0≤s≤m.
Then it follows from Theorem 2 and (47) that
(68)∥uh-𝒜u∥Hs(Γ)≤Ch1-s[∥uh-u∥H1(Γ)+∥u-𝒜u∥H1(Γ)]≤Chm+1-s∥u∥Hm+1(Γ).
There certainly exits
(69)∥u-uh∥Hs(Γ)≤∥u-𝒜u∥Hs(Γ)+∥𝒜u-uh∥Hs(Γ).
From (68), (69), and (47), we have
(70)∥u-uh∥Hs(Γ)≤Cρm+1-s∥u∥Hm+1(Γ).
Then this theorem holds.
5. Numerical Examples
In this section, two numerical examples are presented to show the applicability and the theoretical error estimates of the IEFG method of this paper. In our numerical computation, the nodes are arranged regularly, and the radius of the domain of influence of node xI is determined by ρI=dmax·|xI-xI-1|, where dmax is a positive scalar. The value of dmax must be chosen so that the solution of (22) exists. In our following examples, dmax=2.5 and α=4. Define the error norms as
(71)|ek|≡∥dkdxkuh(x)-dkdxku(x)∥L2(Γ)=[∫Γ(dkdxkuh-dkdxku)2dΓ]1/2,
where u and uh are, respectively, the analytical and numerical solutions. The integration in (71) is obtained numerically by fourth-order Gaussian quadrature. The mI(x) in (2) is chosen to be the cubic spline weight function.
The first example considered is a linear elastostatics problem. A one-dimensional bar of unit length is subjected to a body force of magnitude x. The displacement of the bar is fixed at the left end, and the right end is traction free. The bar has a constant cross sectional area of the unit value, and the elastic modulus is E. Then the equilibrium equation and boundary conditions of this problem are
(72)Eu,xx+x=0,0<x<1,u(0)=0,u,x(1)=0.
The analytical solution to the above problem is
(73)u(x)=1E(x2-x36).
Let E=1. Under the quadratic basis functions and 21 regular distributed nodes, the analytical and numerical solutions of the displacement and strains along the bar are shown, respectively, in Figures 1 and 2, where the numerical values of the IEFG method are in good agreement with the exact ones.
The analytical and numerical displacement.
The analytical and numerical strain.
The absolute errors of the displacements obtained by the IEFG and EFG methods are shown in Figure 3. Here, the essential boundary conditions are enforced by the penalty method in the EFG method, and the penalty factor is chosen to be 108. The CPU times to obtain these results by using the IEFG and EFG methods are, respectively, 0.0267 s and 0.0253 s. It can be seen that the IEFG method in this paper has higher precision than the EFG method under the similar CPU time.
The absolute error of the EFG and IEFG methods.
The error norms of |e0|, |e1|, and |e2| under the quadratic and linear basis functions are, respectively, shown in Figures 4 and 5. The convergence rates of |e0|, |e1|, and |e2| with quadratic basis are, respectively, about 3, 2, and 1, and the convergence rates with linear basis are 2, 1, and -0.1. It is also shown that the second derivatives of the numerical solution of the IEFG method are not convergent to the exact values on the radius ρ when the linear basis is used. It can be seen that these numerical results are in excellent agreement with the ones of the theories of the paper.
Error norms of |e0|, |e1|, and |e2| with quadratic basis functions.
Error norms of |e0|, |e1|, and |e2| with linear basis functions.
The second example considers the following equilibrium equation:
(74)u,xx+π2u=2π2sin(πx),0<x<1,
with the boundary conditions
(75)u(0)=u(1)=0.
The analytical solution of this example is
(76)u(x)=sin(πx).
When the quadratic basis functions and 21 regular distributed nodes are used, the analytical and numerical solutions of u and u,x are shown, respectively, in Figures 6 and 7, where the numerical values of the IEFG method are also in accordance well with the exact ones.
Analytical and numerical solutions of u.
Analytical and numerical solutions of u,x.
The absolute errors of u obtained by the IEFG and EFG methods are shown in Figure 8. In the EFG method, the cubic spline weight function is used, and the essential boundary conditions are enforced by the penalty method. The penalty factor is chosen to be 1010. The CPU times to obtain these results by the IEFG and EFG methods are, respectively, 0.0361 s and 0.0325 s. Again, the IEFG method has higher precision than the EFG method under the similar CPU time.
The absolute error of the EFG and IEFG methods.
The convergence rates of |e0|, |e1|, and |e2| are shown in Figures 9 and 10, respectively, with the quadratic and linear basis functions. The convergence rates of |e0|, |e1|, and |e2| with the quadratic basis are, respectively, 3, 2, and 1. And the corresponding rates with linear basis are, respectively, 2, 1, and -0.004. Figure 10 also shows that the second derivatives of the numerical solution of the IEFG method are not convergent to the exact values on the radius ρ under the linear basis. It is also evident that these numerical results agree well with the ones of the theories of the paper.
Error norms of |e0|, |e1|, and |e2| with quadratic basis functions.
Error norms of |e0|, |e1|, and |e2| with linear basis functions.
6. Conclusions
In this paper, the IMLS method is discussed in detail. The computation of the shape function of this paper is simpler than the corresponding expression presented by Lancaster and Salkauskas. Then, based on the IMLS method of this paper and the Galerkin weak form, an IEFG method for two-point boundary value problems is presented. Since the shape function of the IMLS method satisfies the property of Kronecker δ function, then the IEFG method can apply the essential boundary condition directly. And as the number of the coefficients in the trial function of the IMLS method is less than that in the MLS approximation, then fewer nodes are selected in the entire domain in the IEFG method than in the conventional EFG method. Hence, the IEFG method has high computing speed and precision.
The error analysis of the IEFG method for two-point boundary value problems is presented. The convergence rates of the numerical solution and its derivatives of the IEFG method are presented. The theories of this paper show that if the analytical solution is sufficiently smooth and the order of the polynomial basis functions is big enough, then the solution of the IEFG method and its derivatives are convergent to the analytical solutions in terms of the maximum radius of the domains of influence of nodes. For the purpose of demonstration, some selected numerical examples are given to confirm the theories.
Conflict of Interests
The authors declare that there is no conflict of interests regarding the publication of this paper.
Acknowledgments
This work was supported by the National Natural Science Foundation of China (no. 11171208), Shanghai Leading Academic Discipline Project (no. S30106), and the Natural Science Foundation of Ningbo (no. 2013A610103).
LiewK. M.RenJ.ReddyJ. N.Numerical simulation of thermomechanical behaviours of shape memory alloys via a non-linear mesh-free Galerkin formulation2005637101410402-s2.0-2084443415610.1002/nme.1320LiuG. R.GuY. T.A point interpolation method for two-dimensional solids2001504937951ZBL1050.74057BelytschkoT.LuY. Y.GuL.Element-free Galerkin methods199437222925610.1002/nme.1620370205MR1256818ZBL0796.73077ZBL0796.73077Ju-FengW.Feng-XinS.Rong-JunC.Element-free Galerkin method for a kind of KdV equation20101962-s2.0-7795359328610.1088/1674-1056/19/6/060201060201ChengR. J.ChengY. M.A meshless method for the compound KdV-Burgers equation20112072-s2.0-7996079182510.1088/1674-1056/20/7/070206070206ChengY. M.PengM. J.LiJ. H.The complex variable moving least-square approximation and its application2005376719723MR2223274ChengY. M.LiJ. H.A meshless method with complex variables for elasticity2005541044634471MR2201966ZBL1202.74163ZBL1202.74163ChengY. M.LiJ.A complex variable meshless method for fracture problems200649146592-s2.0-3364465300910.1007/s11433-004-0027-yZBL1147.74410LiewK. M.FengC.ChengY. M.KitipornchaiS.Complex variable moving least-squares method: a meshless approximation technique2007701467010.1002/nme.1870MR2307709ZBL1194.74554ZBL1194.74554WangJ. F.ChengY. M.A new complex variable meshless method for transient heat conduction problems20122112120206WangJ. F.ChengY. M.New complex variable meshless method for advection-diffusion problems2013223030208ChengR. J.LiewK. M.Analyzing two-dimensional sine-Gordon equation with the mesh-free reproducing kernel particle Ritz method2012245-24613214310.1016/j.cma.2012.07.010MR2969191DaiB. D.ChengY. M.Local boundary integral equation method based on radial basis functions for potential problems2007562597603MR2309608ZBL1150.31300ZBL1150.31300ChengR. J.ChengY. M.A meshless method for solving the inverse heat conduction problem with a source parameter2007561055695574MR2493027ZBL1150.80304ZBL1150.80304ChengR. J.ChengY. M.The meshless method for a two-dimensional inverse heat conduction problem with a source parameter20073968438472-s2.0-37349102322WangJ. F.BaiF.-N.ChengY. M.A meshless method for the nonlinear generalized regularized long wave equation20112032-s2.0-7995566605410.1088/1674-1056/20/3/030206030206AtluriS. N.ZhuT.A new Meshless Local Petrov-Galerkin (MLPG) approach in computational mechanics199822211712710.1007/s004660050346MR1649420ZBL0932.76067ZBL0932.76067ChenL.ChengY. M.Reproducing kernel particle method with complex variables for elasticity2008571110MR2415775ChenL.ChengY. M.Complex variable reproducing kernel particle method for transient heat conduction problems2008571060476055MR2492810ZBL1187.80046ZBL1187.80046ChenL.ChengY. M.The complex variable reproducing kernel particle method for elasto-plasticity problems20105359549652-s2.0-7795314956910.1007/s11433-010-0186-yChenL.ChengY. M.The complex variable reproducing kernel particle method for two-dimensional elastodynamics20101992-s2.0-7864933877110.1088/1674-1056/19/9/090204090204WengY. J.ChengY. M.Analyzing variable coefficient advection-diffusion problems via complex variable reproducing kernel particle method2013229090204LiS. C.ChengY. M.Meshless numerical manifold method based on unit partition2004364496500LiS. C.ChengY. M.Numerical manifold method and its applications in rock mechanics2004344446454LiS.ChengY. M.WuY.-F.Numerical manifold method based on the method of weighted residuals20053564704802-s2.0-1744436976510.1007/s00466-004-0636-3ZBL1109.74373LiS. C.LiS. C.ChengY. M.Enriched meshless manifold method for two-dimensional crack modeling20054432342482-s2.0-2954444019910.1016/j.tafmec.2005.09.002LiS.-C.ChengY. M.LiS.-C.Meshless manifold method for dynamic fracture mechanics2006559476047662-s2.0-33750033641GaoH. F.ChengY. M.Complex variable numerical manifold method for elasticity2009414480488MR2589292GaoH.ChengY. M.A complex variable meshless manifold method for fracture problems201071558110.1142/S0219876210002064MR2646928ZBL1267.74124ZBL1267.74124ChengY. M.ChenM. J.A boundary element-free method for linear elasticity2003352181186ChengY. M.PengM.Boundary element-free method for elastodynamics20054866416572-s2.0-3364466025110.1360/142004-25LiewK. M.ChengY. M.KitipornchaiS.Boundary Element-Free Method (BEFM) for two-dimensional elastodynamic analysis using Laplace transform20056412161016272-s2.0-2834444246410.1002/nme.1417ZBL1122.74533KitipornchaiS.LiewK. M.ChengY. M.A Boundary Element-Free Method (BEFM) for three-dimensional elasticity problems200536113202-s2.0-1774436546210.1007/s00466-004-0638-1ZBL1109.74372LiewK. M.ChengY. M.KitipornchaiS.Boundary Element-Free Method (BEFM) and its application to two-dimensional elasticity problems2006658131013322-s2.0-3244443997510.1002/nme.1489ZBL1147.74047QinY.-X.ChengY. M.Reproducing kernel particle boundary element-free method2006557321532222-s2.0-33747469895LiewK. M.ChengY. M.KitipornchaiS.Analyzing the 2D fracture problems via the enriched boundary element-free method20074411-12422042332-s2.0-3394757791310.1016/j.ijsolstr.2006.11.018ZBL05198864QinY. X.ChengY. M.Reproducing kernel particle boundary element-free method for potential problems2009416898905MR2667866LiewK. M.ChengY. M.Complex variable boundary element-free method for two-dimensional elastodynamic problems200919849–523925393310.1016/j.cma.2009.08.020MR2557480ZBL1231.74502ZBL1231.74502PengM.ChengY. M.A Boundary Element-Free Method (BEFM) for two-dimensional potential problems2009331778210.1016/j.enganabound.2008.03.005MR2466280ZBL1160.65348ZBL1160.65348ChengY. M.LiewK. M.KitipornchairS.Reply to “Comments on Boundary Element-Free Method (BEFM) and its application to two-dimensional elasticity problems’20097810125812602-s2.0-6544913274310.1002/nme.2544AtluriS. N.SladekJ.SladekV.ZhuT.The Local Boundary Integral Equation (LBIE) and it's meshless implementation for linear elasticity2000252-31801982-s2.0-003388523810.1007/s004660050468DaiB.ChengY. M.An improved local boundary integral equation method for two-dimensional potential problems2010224214362-s2.0-8345520797010.1142/S1758825110000561ZhangZ.LiewK. M.ChengY. M.LeeY. Y.Analyzing 2D fracture problems with the improved element-free Galerkin method20083232412502-s2.0-3834910498610.1016/j.enganabound.2007.08.012ZBL1244.74240ZhangZ.LiewK. M.ChengY. M.Coupling of the improved element-free Galerkin and boundary element methods for two-dimensional elasticity problems20083221001072-s2.0-3664901750910.1016/j.enganabound.2007.06.006ZBL1244.74204ZhangZ.LiD.-M.ChengY. M.LiewK. M.The improved element-free Galerkin method for three-dimensional wave equation201228380881810.1007/s10409-012-0083-xMR2950238ZhangZ.WangJ. F.ChengY. M.LiewK. M.The improved element-free Galerkin method for three-dimensional transient heat conduction problems20135681568158010.1007/s11433-013-5135-0ZhangZ.HaoS. Y.LiewK. M.ChengY. M.The improved element-free Galerkin method for two-dimensional elastodynamics problems201337121576158410.1016/j.enganabound.2013.08.017MR3124864PengM. J.LiR. X.ChengY. M.Analyzing three-dimensional viscoelasticity problems via the Improved Element-Free Galerkin (IEFG) method20144010411310.1016/j.enganabound.2013.11.018MR3161271PengM.LiuP.ChengY. M.The Complex Variable Element-Free Galerkin (CVEFG) method for two-dimensional elasticity problems2009123673852-s2.0-7795316267810.1142/S1758825109000162PengM.LiD.ChengY. M.The Complex Variable Element-Free Galerkin (CVEFG) method for elasto-plasticity problems20113311271352-s2.0-7864944557410.1016/j.engstruct.2010.09.025ChengY. M.WangJ. F.BaiF. N.A new complex variable element-free Galerkin method for two-dimensional potential problems201221909020310.1088/1674-1056/21/9/090203ChengY. M.LiR. X.PengM. J.Complex Variable Element-Free Galerkin (CVEFG) method for viscoelasticity problems201221909020510.1088/1674-1056/21/9/090205ChengY. M.WangJ. F.LiR. X.The Complex Variable Element-Free Galerkin (CVEFG) method for two-dimensional elastodynamics problems2012442312500410.1142/S1758825112500421LiD.BaiF.ChengY. M.LiewK. M.A novel complex variable element-free Galerkin method for two-dimensional large deformation problems2012233–23611010.1016/j.cma.2012.03.015MR2924016ZBL1253.74106ZBL1253.74106BaiF.-N.LiD.-M.WangJ. F.ChengY. M.An improved complex variable element-free Galerkin method for two-dimensional elasticity problems20122122-s2.0-8485734499010.1088/1674-1056/21/2/020204020204LiD. M.LiewK. M.ChengY. M.An improved complex variable element-free Galerkin method for two-dimensional large deformation elastoplasticity problems2014269728610.1016/j.cma.2013.10.018MR3144633ZhuT.AtluriS. N.A modified collocation method and a penalty formulation for enforcing the essential boundary conditions in the element free Galerkin method199821321122210.1007/s004660050296MR1633725ZBL0947.74080ZBL0947.74080MostT.BucherC.A moving least squares weighting function for the element-free Galerkin method which almost fulfills essential boundary conditions20052133153322-s2.0-2804445060710.12989/sem.2005.21.3.315MostT.BucherC.New concepts for moving least squares: an interpolating non-singular weighting function and weighted nodal least squares20083264614702-s2.0-4364910493810.1016/j.enganabound.2007.10.013ZBL1244.74228LancasterP.SalkauskasK.Surfaces generated by moving least squares methods19813715514115810.2307/2007507MR616367ZBL0469.41005ZBL0469.41005KaljevićI.SaigalS.An improved element free Galerkin formulation1997401629532974MR1461997ZBL0895.73079ZBL0895.73079RenH.-P.ChengY. M.ZhangW.An Improved Boundary Element-Free Method (IBEFM) for two-dimensional potential problems20091810406540732-s2.0-7035051016210.1088/1674-1056/18/10/002HongpingR.YuminC.WuZ.An Interpolating Boundary Element-Free Method (IBEFM) for elasticity problems20105347587662-s2.0-7795316758210.1007/s11433-010-0159-1RenH. P.ChengY. M.The Interpolating Element-Free Galerkin (IEFG) method for two-dimensional elasticity problems20113473575810.1142/S1758825111001214RenH.ChengY. M.The Interpolating Element-Free Galerkin (IEFG) method for two-dimensional potential problems201236587388010.1016/j.enganabound.2011.09.014MR2880734WangJ. F.SunF. X.ChengY. M.An improved interpolating element-free Galerkin method with nonsingular weight function for two-dimensional potential problems2012219090204WangJ.WangJ.SunF.ChengY. M.An interpolating boundary element-free method with nonsingular weight function for two-dimensional potential problems201310623135004310.1142/S0219876213500436MR3056497SunF. X.WangJ. F.ChengY. M.An improved interpolating element-free Galerkin method for elasticity20132212120203LevinD.The approximation power of moving least-squares1998672241517153110.1090/S0025-5718-98-00974-0MR1474653ZBL0911.41016ZBL0911.41016ArmentanoM. G.DuránR. G.Error estimates for moving least square approximations200137339741610.1016/S0168-9274(00)00054-4MR1822748ZBL0984.65096ZBL0984.65096ZuppaC.Error estimates for moving least square approximations200334223124910.1007/s00574-003-0010-7MR1992639ZBL1040.65034ZBL1040.65034LiX.ZhuJ.A Galerkin boundary node method and its convergence analysis2009230131432810.1016/j.cam.2008.12.003MR2532313ZBL1189.65291ZBL1189.65291LiX.Meshless Galerkin algorithms for boundary integral equations with moving least square approximations201161121237125610.1016/j.apnum.2011.08.003MR2851120ZBL1232.65160ZBL1232.65160LiX.The meshless Galerkin boundary node method for Stokes problems in three dimensions201188544247210.1002/nme.3181MR2842410ZBL1242.76244ZBL1242.76244KryslP.BelytschkoT.Element-free Galerkin method: convergence of the continuous and discontinuous shape functions19971483-425727710.1016/S0045-7825(96)00007-2MR1465821ZBL0918.73125ZBL0918.73125ChungH.-J.BelytschkoT.An error estimate in the EFG method19982129110010.1007/s004660050286MR1616071ZBL0910.73060ZBL0910.73060DolbowJ.BelytschkoT.Numerical integration of the Galerkin weak form in meshfree methods199923321923010.1007/s004660050403MR1691704ZBL0963.74076ZBL0963.74076GaveteL.CuestaJ. L.RuizA.A procedure for approximation of the error in the EFG method200253367769010.1002/nme.307MR1876499ZBL1112.74564ZBL1112.74564GaveteL.GaveteM. L.AlonsoB.MartínA. J.A posteriori error approximation in EFG method200358152239226310.1002/nme.850MR2020718ZBL1047.74081ZBL1047.74081ChengR. J.ChengY. M.Error estimates for the finite point method200858688489810.1016/j.apnum.2007.04.003MR2420624ZBL1145.65086ZBL1145.65086ChengR. J.ChengY. M.Error estimates of the element-free Galerkin method for potential problems2008571060376046MR2492809ChengR. J.ChengY. M.Error estimate of element-free Galerkin method for elasticity20116072-s2.0-79961107471070206