Loading [MathJax]/extensions/TeX/boldsymbol.js
Research article

Empirical likelihood for varying coefficient partially nonlinear model with missing responses

  • Received: 20 January 2021 Accepted: 08 April 2021 Published: 28 April 2021
  • MSC : 62G05, 62G20

  • In this paper, we consider the statistical inferences for varying coefficient partially nonlinear model with missing responses. Firstly, we employ the profile nonlinear least squares estimation based on the weighted imputation method to estimate the unknown parameter and the nonparametric function, meanwhile the asymptotic normality of the resulting estimators is proved. Secondly, we consider empirical likelihood inferences based on the weighted imputation method for the unknown parameter and nonparametric function, and propose an empirical log-likelihood ratio function for the unknown parameter vector in the nonlinear function and a residual-adjusted empirical log-likelihood ratio function for the nonparametric component, meanwhile construct relevant confidence regions. Thirdly, the response mean estimation is also studied. In addition, simulation studies are conducted to examine the finite sample performance of our methods, and the empirical likelihood approach based on the weighted imputation method (IEL) is further applied to a real data example.

    Citation: Liqi Xia, Xiuli Wang, Peixin Zhao, Yunquan Song. Empirical likelihood for varying coefficient partially nonlinear model with missing responses[J]. AIMS Mathematics, 2021, 6(7): 7125-7152. doi: 10.3934/math.2021418

    Related Papers:

    [1] Leilei Wei, Xiaojing Wei, Bo Tang . Numerical analysis of variable-order fractional KdV-Burgers-Kuramoto equation. Electronic Research Archive, 2022, 30(4): 1263-1281. doi: 10.3934/era.2022066
    [2] E. A. Abdel-Rehim . The time evolution of the large exponential and power population growth and their relation to the discrete linear birth-death process. Electronic Research Archive, 2022, 30(7): 2487-2509. doi: 10.3934/era.2022127
    [3] Li Tian, Ziqiang Wang, Junying Cao . A high-order numerical scheme for right Caputo fractional differential equations with uniform accuracy. Electronic Research Archive, 2022, 30(10): 3825-3854. doi: 10.3934/era.2022195
    [4] Li-Bin Liu, Ying Liang, Jian Zhang, Xiaobing Bao . A robust adaptive grid method for singularly perturbed Burger-Huxley equations. Electronic Research Archive, 2020, 28(4): 1439-1457. doi: 10.3934/era.2020076
    [5] Bidi Younes, Abderrahmane Beniani, Khaled Zennir, Zayd Hajjej, Hongwei Zhang . Global solution for wave equation involving the fractional Laplacian with logarithmic nonlinearity. Electronic Research Archive, 2024, 32(9): 5268-5286. doi: 10.3934/era.2024243
    [6] Jun Pan, Yuelong Tang . Two-grid H1-Galerkin mixed finite elements combined with L1 scheme for nonlinear time fractional parabolic equations. Electronic Research Archive, 2023, 31(12): 7207-7223. doi: 10.3934/era.2023365
    [7] Jun Liu, Yue Liu, Xiaoge Yu, Xiao Ye . An efficient numerical method based on QSC for multi-term variable-order time fractional mobile-immobile diffusion equation with Neumann boundary condition. Electronic Research Archive, 2025, 33(2): 642-666. doi: 10.3934/era.2025030
    [8] Jingyun Lv, Xiaoyan Lu . Convergence of finite element solution of stochastic Burgers equation. Electronic Research Archive, 2024, 32(3): 1663-1691. doi: 10.3934/era.2024076
    [9] Qingming Hao, Wei Chen, Zhigang Pan, Chao Zhu, Yanhua Wang . Steady-state bifurcation and regularity of nonlinear Burgers equation with mean value constraint. Electronic Research Archive, 2025, 33(5): 2972-2988. doi: 10.3934/era.2025130
    [10] Nelson Vieira, M. Manuela Rodrigues, Milton Ferreira . Time-fractional telegraph equation of distributed order in higher dimensions with Hilfer fractional derivatives. Electronic Research Archive, 2022, 30(10): 3595-3631. doi: 10.3934/era.2022184
  • In this paper, we consider the statistical inferences for varying coefficient partially nonlinear model with missing responses. Firstly, we employ the profile nonlinear least squares estimation based on the weighted imputation method to estimate the unknown parameter and the nonparametric function, meanwhile the asymptotic normality of the resulting estimators is proved. Secondly, we consider empirical likelihood inferences based on the weighted imputation method for the unknown parameter and nonparametric function, and propose an empirical log-likelihood ratio function for the unknown parameter vector in the nonlinear function and a residual-adjusted empirical log-likelihood ratio function for the nonparametric component, meanwhile construct relevant confidence regions. Thirdly, the response mean estimation is also studied. In addition, simulation studies are conducted to examine the finite sample performance of our methods, and the empirical likelihood approach based on the weighted imputation method (IEL) is further applied to a real data example.



    Fractional calculus (FC) theory was proposed by N. H. Abel and J. Liouville, and a description of their work is presented in [1]. By using FC, integer derivatives, and integrals can be generalized to real or variable derivatives and integrals. FC is studied since fractional differential equations (FDEs) are better suited to modeling natural physics processes and dynamic systems than integer differential equations. Furthermore, FDEs that incorporate memory effects are better suited to describing natural processes that have memory and hereditary properties. In other words, because fractional derivatives have memory effects, FDEs are more accurate in describing physical phenomena with memory or hereditary characteristics. There was a trend to consider FC to be an esoteric theory with no application until the last few years. Now, more and more researchers are investigating how it can be applied to economics, control system and finance. As a result, many fractional order differential operators were developed, such as Hadamard, Riemann-Liouville, Caputo, Riesz, Grünwald-Letnikov, and variable order differential operators. The researchers have devoted considerable effort to solving FDEs numerically so that they can be applied to a variety of problems [2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20]. Several numerical approaches have been proposed in the literature, including eigenvector expansion, the fractional differential transform technique [21], the homotopy analysis technique [22], the homotopy perturbation transform technique [23], the generalized block pulse operational matrix technique [24] and the predictor-corrector technique [25]. In addition, the use of Legendre wavelets to integrate and differentiate fractional order matrices has been suggested as a numerical method [26,27].

    In this paper, we study the numerical solution of the time-fractional Burger's equation (TFBE) [28] as follows:

    γU(x,t)tγ+U(x,t)U(x,t)xv2U(x,t)x2=f(x,t), (1)

    which is subject to the following boundary conditions (BCs):

    U(a,t)=l1(t),U(b,t)=l2(t),axb,t[0,tf], (2)

    and the following initial condition (IC):

    U(x,0)=g(x)andaxb, (3)

    in which 0<γ1 is a parameter representing the order of the fractional time, v denotes a viscosity parameter and g(x),l1(t)andl2(t) are given functions of their argument. The TFBE is a kind of sub-diffusion convection, which is widely adopted to describe many physical problems such as unidirectional propagation of weakly nonlinear acoustic waves, shock waves in flow systems, viscous media, compressible turbulence, electromagnetic waves and weak shock propagation [29,30,31]. In recent years, there has been some technique development in the study of Burger's equation: an implicit difference scheme and algorithm implementation [32], pointwise error analysis of the third-order backward differentiation formula (BDF3) [33], pointwise error estimates of a compact difference scheme [34], efficient (BDF3) finite-difference scheme [35], semi-analytical methods [36], composite spectral methods [37], least-squares methods [38], geometric analysis methods [39], error and stability estimate techniques [40].

    Definition 1. Suppose that m is the smallest integer exceeding γ; the Caputo time fractional derivative operator of order γ>0 can be defined as follows [41]:

    CDγ0,tu(x,t)={mu(x,t)tmγ=mN1Γ(mγ)t0(tω)mγ1mu(x,ω)ωmdω,m1<γ<m,mN, (4)

    where u(x,t) is the unknown function that is (m1) times continuously differentiable and Γ(.) denotes the usual gamma function. The finite-element method has been an important method for solving both ordinary and partial differential equation, therefore, in recent research, it has been applied to solve the TFBE. In what follows, we describe the solution process by using the finite-element scheme for solving the TFBE.

    To discretize the TFBE (1), first let us define the cubic B-spline base function. We partition the interval [a,b], which represents the solution domain of (1) into M uniformly spaced points xm such that a=x0<x1<<xM1<xM=b and h=(xm+1xm). Then, the cubic B-spline Cm(x),(m=1(1)(M+1), at the knots xm which form basis on the solution interval [a,b], is defined as follows [42]:

    Cm(x)=1h3{(xxm2)3,ifx[xm2,xm1],h3+3h2(xxm1)+3h(xxm1)23(xm+1x)3,ifx[xm1,xm],h3+3h2(xm+1x)+3h(xm+1x)23(xm+1x)3,ifx[xm,xm+1],(5)(xm+2x)3,ifx[xm+1,xm+2],o,otherwise. (5)

    where the set of cubic B-splines (C1(x),C0(x),,CM(x),CM+1(x)) is a basis for the functions defined over interval [a,b]. Thus, the numerical solution UM(x,t) to the analytic solution U(x,t) can be illustrated as

    UM(x,t)=M+1m=1σm(t)Cm(x), (6)

    where σm(t) are unknown time-dependent parameters to be determined from the initial, boundary and weighted residual conditions. Since each cubic B-spline covers four consecutive elements, each element [xm,xm+1] is also covered by four cubic B-splines. So, the nodal values Um and its first and second derivatives U'm, U"m can be respectively computed in terms of the element parameter σm(t), at the knot xm as follows:

    Um=σm1+4σm+σm+1,U'm=3h(σm1σm+1),U"m=6h2(σm12σm+σm+1), (7)

    and by means of the local coordinate transformation [43] as follows:

    hη=xxm,0η1. (8)

    A cubic B-spline shape function in terms of η over the element [xm,xm+1] is formulated as:

    Cm1=(1η)3,Cm1=1+3(1η)+3(1η)23(1η)3,Cm+1=1+3η+3η23η3,Cm+2=η3 (9)

    and the variation of UM(η,t) over the typical element [xm,xm+1] is represented as

    UM(x,t)=m+2j=m1σj(t)Cj(η), (10)

    in which B-splines Cm1(η),Cm(η),Cm+1(η), Cm+2(η) and σm1(t), σm(t),σm+1 and σm+2(t) are element shape functions and element parameters, respectively.

    Based on the Galerkin's method with weight function W(x)>0, we get the following weak formula of (1):

    baW(γUtγ+UUxv2Ux2)dx=baWf(x,t); (11)

    using transformation (8) and by apply partial integration we obtain:

    10(WγUtγ+λWUη+ΦWηUη)dη=ΦWUη10+baWƑ(η,t)dη, (12)

    where λ=1hÛ, Φ=vh2 and Û=U(η,t) which is considered to be a constant on an element to simplify the integral [43]; replace the weight function W by quadratic B-spline {B}_{\boldsymbol{m}}\left(\boldsymbol{x}\right), \mathrm{m} = -1\left(1\right)M , at the knots {x}_{m} , which forms a basis on the solution interval [a, b] , introduced as follows [44]:

    {B_m}\left( x \right) = \frac{1}{{{h^2}}}\left\{ {\begin{array}{*{20}{l}} {{{\left( {{x_{m + 2}} - x} \right)}^2} - 3{{\left( {{x_{m + 1}} - x} \right)}^2} + 3{{\left( {{x_m} - x} \right)}^2}, \;\;{\rm{if}}\;\;x \in \left[ {{x_{m - 1}}, {x_m}} \right], }\\ {{{\left( {{x_{m + 2}} - x} \right)}^2} - 3{{\left( {{x_{m + 1}} - x} \right)}^2}, \;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;{\rm{if}}\;x \in \left[ {{x_m}, {x_{m + 1}}} \right], }\\ {{{({x_{m + 2}} - x)}^2}, \;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;{\rm{if}}\;x \in \left[ {{x_{m + 1}}, {x_{m + 2}}} \right], }\\ {\;\;\;\;0, \;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;{\rm{otherwise}}.} \end{array}} \right. (13)

    where {({B}_{-1}\left(x\right), B}_{0}\left(x\right), \dots , {B}_{M}\left(x\right)) is the set of splines for the basis of functions introduced on [a, b] . The numerical solution {U}_{M}(x, t) to the analytic solution U(x, t) is expanded by

    {U}_{M}(x, t) = \sum \limits_{m = -1}^{M}{{\vartheta }_{m}\left(t\right)B}_{m}\left(x\right), (14)

    where {\vartheta }_{m} are unknown time-dependent parameters, and by using local coordinate transformation (8), the quadratic B-spline shape functions for the typical element [ {x}_{m}, {x}_{m+1} ] are given as

    \begin{array}{c} {B}_{m-1} = {(1-\eta )}^{2} \\{B}_{m} = 1+2\eta -2{\eta }^{2} \\ {B}_{m+1} = {\eta }^{2} \end{array} (15)

    The variation of the function U(\eta , t) is approximated by

    {U}_{M}\left(\eta , t\right) = \sum \limits_{i = m-1}^{m+1}{{\vartheta }_{i}\left(t\right)B}_{i}\left(\eta \right), (16)

    where {\vartheta }_{m-1}\left(t\right) , {\vartheta }_{m}\left(t\right) and {\vartheta }_{m+1}\left(t\right) act as element parameters and B-splines {B}_{m-1}\left(\eta \right), {B}_{m}\left(\eta \right) and {B}_{m+1}\left(\eta \right) as element shape functions based on the above; (12) will be in the following form:

    \begin{array}{l} \;\;\;\;\;\;\;\;\;\;\;\;\;\;\sum \limits_{j = m-1}^{m+2}[{\int }_{0}^{1}{B}_{i}{C}_{j}d\eta ]\dot{\sigma }+\sum \limits_{j = m-1}^{m+2}[{\int }_{0}^{1}(\mathrm{\lambda }{B}_{i}{C}_{j}^{\text{'}}+\mathit{\Phi} {B}_{i}^{\text{'}}{C}_{j}^{\text{'}})d\eta -\mathit{\Phi} {B}_{i}{C}_{j}^{\text{'}}{|}_{0}^{1}]\sigma \\ = {\int }_{0}^{1}{B}_{i}\mathrm{Ƒ}\left(\eta , t\right)d\eta , i = m-1, m, m+1, \end{array} (17)

    in which "Dot" represents the \sigma \mathrm{t}\mathrm{h} fractional derivative with respect to time. We can write (17) in matrix notation as follows:

    {X}_{ij}^{e}{\dot{\sigma }}^{e}+({\mathrm{\lambda }Y}_{ij}^{e}+\mathit{\Phi} ({Z}_{ij}^{e}-{Q}_{ij}^{e}\left)\right){\sigma }^{e} = {E}_{i}^{e}, (18)

    in which {\sigma }^{e} = {({\sigma }_{m-1}, {\sigma }_{m}, {\sigma }_{m+1}, {\sigma }_{m+2})}^{T} are the element parameters. The element matrices {X}_{ij}^{e}, {Y}_{ij}^{e}, {Z}_{ij}^{e}, {Q}_{ij}^{e} and {E}_{i}^{e} are rectangular 3\times 4 matrices introduced through the following integrals:

    {X}_{ij}^{e} = \underset{0}{\overset{1}{\int }}{B}_{i}{C}_{j}d\eta = \frac{1}{60}\left[\begin{array}{cccc}10& 71& 38& 1\\ 19& 221& 221& 19\\ 1& 38& 71& 10\end{array}\right],
    {Y}_{ij}^{e} = \underset{0}{\overset{1}{\int }}{B}_{i}{C}_{j}^{\text{'}}d\eta = \frac{1}{10}\left[\begin{array}{cccc}-6& -7& 12& 1\\ -13& -41& 41& 13\\ -1& -12& 7& 6\end{array}\right],
    {Z}_{ij}^{e} = \underset{0}{\overset{1}{\int }}{B}_{i}^{\text{'}}{C}_{j}^{\text{'}}d\eta = \frac{1}{2}\left[\begin{array}{cccc}3& 5& -7& -1\\ -2& 2& 2& -2\\ -1& -7& 5& 3\end{array}\right],
    {Q}_{ij}^{e} = {B}_{i}{C}_{j}^{\text{'}}{\mid }_{0}^{1} = 3\left[\begin{array}{cccc}1& 0& -1& 0\\ 1& -1& -1& 1\\ 0& -1& 0& 1\end{array}\right] \;\;{\rm{and}}
    {E}_{i}^{e} = {\int }_{0}^{1}{B}_{i}\mathrm{Ƒ}\left(\eta , t\right)d\eta ,

    where i and j take only the values (m-1, m, m+1) and (m-1, m, m+1, m+2) respectively, and a lumped value for λ is defined by \mathrm{\lambda } = \frac{1}{2h}({\sigma }_{m-1}+{5\sigma }_{m}+5{\sigma }_{m+1}+{\sigma }_{m+2}) .

    By assembling all contributions from all elements, we get the following matrix equation:

    X\dot{\sigma }+\left(\mathrm{\lambda }Y+\mathit{\Phi} \left(Z-Q\right)\right)\sigma = E, (19)

    where \sigma = {({\sigma }_{-1}, {\sigma }_{0}, {\sigma }_{1}, \dots , {\sigma }_{M}, {\sigma }_{\mathrm{M}+1})}^{T} denotes a global element parameter. The matrices \mathrm{X}, \;\mathrm{Z} and Y represent rectangular, septa-diagonal and every sub-diagonal matrices, which include the following forms:

    X = \frac{1}{60}(1, 57, 302, 302, 57, 1, 0) ,
    Z = \frac{1}{2}(-1, -9, 10, 10, -9, -1, 0) ,
    \begin{array}{l} \lambda Y = \frac{1}{10}({-\lambda }_{1}, -12{\lambda }_{1}-13{\lambda }_{2}, 7{\lambda }_{1}-41{\lambda }_{2}-6{\lambda }_{3}, 6{\lambda }_{1}+41{\lambda }_{2}-7{\lambda }_{3}, 13{\lambda }_{2}+\\ 12{\lambda }_{3}, {\lambda }_{3}, 0), \end{array}

    in which,

    {\lambda }_{1} = \frac{1}{2h}({\sigma }_{m-2}+5{\sigma }_{m-1}+5{\sigma }_{m}+{\sigma }_{m+1}) ,
    {\lambda }_{2} = \frac{1}{2h}({\sigma }_{m-1}+5{\sigma }_{m}+5{\sigma }_{m+1}+{\sigma }_{m+2}) ,
    {\lambda }_{3} = \frac{1}{2h}({\sigma }_{m}+5{\sigma }_{m+1}+5{\sigma }_{m+2}+{\sigma }_{m+3}) .

    Following [45], we can approximate the temporal Caputo derivative with the help of the L1 formula:

    \frac{{d}^{\gamma }f\left(t\right)}{d{t}^{\gamma }}{\mid }_{{t}_{f}} = \frac{{\left(\Delta t\right)}^{-\gamma }}{\mathit{\Gamma} \left(2-\gamma \right)}\sum \limits_{k = 0}^{m-1}{b}_{k}^{\gamma }\left[f\right({t}_{m-k}-f\left({t}_{m-1k}\right)]+O({\Delta t)}^{2-\gamma },

    where {b}_{k}^{\gamma } = ({k+1)}^{1-\gamma }-{k}^{1-\gamma }\;\mathrm{a}\mathrm{n}\mathrm{d}\;\Delta t = \frac{{t}_{f}-0}{N} , and {t}_{f} = n\left(\Delta t\right), n = \mathrm{0, 1}, \dots N, \mathrm{w}\mathrm{h}\mathrm{e}\mathrm{r}\mathrm{e}N represents a positive integer. Now, we recall the following lemma.

    Lemma 1: Suppose that 0 < \gamma < 1and{b}_{k}^{\gamma } = ({k+1)}^{1-\gamma }-{k}^{1-\gamma }, k = \mathrm{0, 1}, \dots ;then, 1 = {b}_{0}^{\gamma } > {b}_{1}^{\gamma } > \dots > {b}_{k}^{\gamma }\to 0, ask\to \mathrm{\infty } [46].

    Then, we can we write the parameter {\sigma }_{m}^{ \cdot } as follows:

    \begin{array}{l} {\sigma }_{m}^{ \cdot } = \frac{{d}^{\gamma }\sigma }{d{t}^{\gamma }} = \frac{{(\Delta t)}^{-\gamma }}{\mathit{\Gamma} (2-\gamma )}\sum \limits_{k = 0}^{m-1}{b}_{k}^{\gamma }\left[\right({\sigma }_{m-1}^{n-k+1}-{\sigma }_{m-1}^{n-k})+4\left({\sigma }_{m}^{n-k+1}-{\sigma }_{m}^{n-k}\right)\\ \;\;\;\;\;\;\;\;\;\;\;\;+({\sigma }_{m+1}^{n-k+1}-{\sigma }_{m+1}^{n-k})]+O({\Delta t)}^{2-\gamma }, \;\;\;\;\;\;\;\;{b}_{k}^{\gamma }\\ \;\;\;\;\;\;\;\;\;\;\;\; = ({k+1)}^{1-\gamma }-{k}^{1-\gamma }, \end{array}

    while the parameter \sigma by the Crank-Nicolson scheme, is as follows:

    {\sigma }_{m} = \frac{1}{2}({\sigma }_{m}^{n}+{\sigma }_{m}^{n+1}) .

    Substitution both parameters above into \left(18\right) , we obtain the (M+2)\times (M+3) matrix system:

    \begin{array}{l} \;\;\; \;\;\; \;\;\;\left[X+\frac{\left[{\left(\Delta t\right)}^{-\gamma }\mathit{\Gamma} \left(2-\gamma \right)\left(\lambda Y+\mathit{\Phi} (Z-Q\right)\right]}{2}\right]{\sigma }^{n+1}\\ \;\;\; \;\;\; \;\;\; = \left[X-\frac{\left[{\left(\Delta t\right)}^{-\gamma }\mathit{\Gamma} \left(2-\gamma \right)\left(\lambda Y+\mathit{\Phi} (Z-Q\right)\right]}{2}\right]{\sigma }^{n} \\ -X\sum \limits_{k = 1}^{n}{b}_{k}^{\gamma }\left[\left({\sigma }_{m-1}^{n-k+1}-{\sigma }_{m-1}^{n-k}\right)+4\left({\sigma }_{m}^{n-k+1}-{\sigma }_{m}^{n-k}\right)+\left({\sigma }_{m+1}^{n-k+1}-{\sigma }_{m+1}^{n-k}\right)\right]\\ \;\;\; \;\;\; \;\;\;+{\left(\Delta t\right)}^{-\gamma }\mathit{\Gamma} \left(2-\gamma \right)E, \end{array} (20)

    where \sigma = {({\sigma }_{m-2}+{\sigma }_{m-1}+{\sigma }_{m}+{\sigma }_{m+1}+{\sigma }_{m+1}+{\sigma }_{m+2}+{\sigma }_{m+3})}^{T} ; to make the matrix equation be square, we need to find an additional constraint of BC (2) and their second derivatives and we obtain discard {\sigma }_{-1} from system (20) as follows:

    {\sigma }_{-1}\left(t\right) = -4{\sigma }_{0}\left(t\right)-{\sigma }_{1}\left(t\right){+U(x}_{0}, t) ;

    the variables {\sigma }_{-1}^{n} and {\sigma }_{M+1}^{n} can be ignored from system (20) and then the system can be converted to an (M+1)\times (M+1) matrix system. The initial vector of parameter {\sigma }^{0} = ({\sigma }_{0}^{0}, {\sigma }_{1}^{0}, \dots , {\sigma }_{M}^{0}) should be obtained to iterate system (20); the approximation of (6) has been reformulated on the interval [a, b] when time t = 0 as follows:

    {U}_{N}(x, 0) = \sum \limits_{m = 0}^{M}{C}_{m}{\sigma }_{m}^{0},

    where U\left(x, 0\right) fulfills the following equation at node {x}_{m}:

    {U}_{M}\left({x}_{m}, 0\right) = U\left({x}_{m}, 0\right), \;\;m = \mathrm{0, 1}, \dots , M+1
    {U}_{M}^{\text{'}}\left({x}_{0}, 0\right) = {U}^{\text{'}}\left({x}_{M}, 0\right) = 0,
    {U}_{M}^{\text{'}\text{'}}\left({x}_{0}, 0\right) = {U}^{\text{'}\text{'}}\left({x}_{M}, 0\right) = 0.

    Therefore, we can obtain the following system:

    \left[\begin{array}{c}{\sigma }_{0}^{0}\\ {\sigma }_{1}^{0}\\ ⋮\\ {\sigma }_{M-1}^{0}\\ {\sigma }_{M}^{0}\end{array}\begin{array}{cccccccc}6& 0& 0& & & & & 0\\ 1& 4& 1& & & & & 0\\ & 1& 4& 4& & & & \\ & & & \ddots & \ddots & & & \\ & & & & \ddots & 1& 4& 1\\ 0& & & & & & 0& 6\end{array}\right]\left[\begin{array}{c}{\sigma }_{0}^{0}\\ {\sigma }_{1}^{0}\\ ⋮\\ {\sigma }_{M-1}^{0}\\ {\sigma }_{M}^{0}\end{array}\right] = \left[\begin{array}{c}U\left({x}_{0}, 0\right)-\frac{{h}^{2}}{6}{g}^{"}\left(a\right)\\ U\left({x}_{1}, 0\right)\\ ⋮\\ U\left({x}_{M-1}, 0\right)\\ U\left({x}_{M}, 0\right)-\frac{{h}^{2}}{6}{g}^{"}\left(b\right)\end{array}\right]

    and we solve this identity matrix by applying the Jain algorithm [47].

    This section adopts the von Neumann stability analysis to investigate the stability of approximation obtained by scheme (20). First, we introduce the recurrence relationship between successive time levels relating unknown element parameters {\sigma }_{m}^{n+1}\left(t\right) , as follows:

    \begin{array}{l} {q}_{1}{\sigma }_{m-2}^{n+1}+{q}_{2}{\sigma }_{m-1}^{n+1}+{q}_{3}{\sigma }_{m}^{n+1}+{q}_{4}{\sigma }_{m+1}^{n+1}+{q}_{5}{\sigma }_{m+2}^{n+1}+{q}_{6}{\sigma }_{m+3}^{n+1}\\ = {q}_{6}{\sigma }_{m-2}^{n}+{q}_{5}{\sigma }_{m-1}^{n}+{q}_{4}{\sigma }_{m}^{n}+{q}_{3}{\sigma }_{m+1}^{n}+{q}_{2}{\sigma }_{m+2}^{n}+{q}_{1}{\sigma }_{m+3}^{n}\\ -20\sum \limits_{k = 1}^{n}{b}_{k}^{\gamma }[(({\sigma }_{m-2}^{n-k+1}-{\sigma }_{m-2}^{n-k})+4({\sigma }_{m-2}^{n-k+1}-{\sigma }_{m-2}^{n-k})+({\sigma }_{m-2}^{n-k+1}-{\sigma }_{m-2}^{n-k}))\\ +57(({\sigma }_{m-1}^{n-k+1}-{\sigma }_{m-1}^{n-k})+4({\sigma }_{m-1}^{n-k+1}-{\sigma }_{m-1}^{n-k})++({\sigma }_{m-1}^{n-k+1}-{\sigma }_{m-1}^{n-k}))\\ +302(({\sigma }_{m}^{n-k+1}-{\sigma }_{m}^{n-k})+4({\sigma }_{m}^{n-k+1}-{\sigma }_{m}^{n-k})+({\sigma }_{m}^{n-k+1}-{\sigma }_{m}^{n-k}))\\ +302(({\sigma }_{m+1}^{n-k+1}-{\sigma }_{m+1}^{n-k})+4({\sigma }_{m+1}^{n-k+1}-{\sigma }_{m+1}^{n-k})+({\sigma }_{m+1}^{n-k+1}-{\sigma }_{m+1}^{n-k}))\\ +57(({\sigma }_{m+2}^{n-k+1}-{\sigma }_{m+2}^{n-k})+4({\sigma }_{m+2}^{n-k+1}-{\sigma }_{m+2}^{n-k})+({\sigma }_{m+2}^{n-k+1}-{\sigma }_{m+2}^{n-k}))\\ +(({\sigma }_{m+3}^{n-k+1}-{\sigma }_{m+3}^{n-k})+4({\sigma }_{m+3}^{n-k+1}-{\sigma }_{m+3}^{n-k})\\ +({\sigma }_{m+3}^{n-k+1}-{\sigma }_{m+3}^{n-k}))] \end{array} (21)

    where

    {q}_{1} = 20-300\mathit{\Phi} \alpha -60\lambda \alpha , \;\;{q}_{2} = 1140-2700\mathit{\Phi} \alpha -1500\lambda \alpha , \;\;{q}_{3} = 6040+3000\mathit{\Phi} \alpha -2400\lambda \alpha
    {q}_{4} = 6040+3000\mathit{\Phi} \alpha +2400\lambda \alpha , \;\;{q}_{5} = 1140-2700\mathit{\Phi} \alpha +1500\lambda \alpha , \;\;{q}_{6} = 20-300\mathit{\Phi} \alpha +60\lambda \alpha

    and \alpha = {\left(\Delta t\right)}^{-\gamma }\mathit{\Gamma} \left(2-\gamma \right) .

    The growth factor of the typical Fourier mode is defined as

    {\sigma }_{m}^{n} = {\xi }^{n}{e}^{i\beta mh} (22)

    where, i = \sqrt{-1, }\beta is a mode number and h is the element size. Substitution of (22) into (21) yields

    \begin{array}{l} \;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\; {\xi }^{n+1}\left({q}_{1}{e}^{-2i\beta h}+{q}_{2}{e}^{-i\beta h}+{q}_{3}+{q}_{4}{e}^{i\beta h}+{q}_{5}{e}^{2i\beta h}+{q}_{6}{e}^{3i\beta h}\right) = \\ \;\;\;\;\;\;\;{\xi }^{n}\left({q}_{6}{e}^{-2i\beta h}+{q}_{5}{e}^{-i\beta h}+{q}_{4}+{q}_{3}{e}^{i\beta h}+{q}_{2}{e}^{2i\beta h}+{q}_{1}{e}^{3i\beta h}\right) \\ -20\sum \limits_{k = 1}^{n}{b}_{k}^{\gamma }\left[\left(\left({\sigma }_{m-2}^{n-k+1}-{\sigma }_{m-2}^{n-k}\right)+4\left({\sigma }_{m-2}^{n-k+1}-{\sigma }_{m-2}^{n-k}\right)+\left({\sigma }_{m-2}^{n-k+1}-{\sigma }_{m-2}^{n-k}\right)\right)({e}^{-2i\beta h}302\\ \;\;\;\;\;\;\;\;\;+302{e}^{i\beta h}+57{e}^{2i\beta h}+{e}^{2i\beta h})\right]; \end{array} (23)

    let {\xi }^{n+1} = Ϋ{\xi }^{n} and assume that Ϋ\equiv Ϋ\left(\theta \right) is independent of time, therefore, we can write Ϋ as follows:

    Ϋ = \frac{A-iB}{A+iB},

    where

    A = \left(6040+3000\mathit{\Phi} \alpha \right)\mathrm{cos}\left(\frac{\theta }{2}\right)h+\left(1140-2700\mathit{\Phi} \alpha \right)\mathrm{cos}\left(\frac{3\theta }{2}\right)h
    +\left(20-300\mathit{\Phi} \alpha \right)\mathrm{cos}\left(\frac{5\theta }{2}\right)h,
    B = \left(2400\lambda \alpha \right)\mathrm{sin}\left(\frac{\theta }{2}\right)h++\left(1500\lambda \alpha \right)\mathrm{sin}\left(\frac{3\theta }{2}\right)h+\left(60\lambda \alpha \right)\mathrm{sin}\left(\frac{\theta }{2}\right)h,

    Obviously note that \left|\mathrm{Ϋ}\right|\le 1 . Therefore, according to the Fourier condition, the scheme (20) is unconditionally stable.

    This section introduces two numerical examples, which highlight numerical results for the TFBE with different IC and BCs given by the CBSGM with quadratic weight function. In this section, we use the {L}_{2} and {L}_{\infty } to calculate the accuracy of the CBSGM with a quadratic weight function, which has been employed in this study; we will also show how the analytical results and the numerical results are close to each other. To do this, first we will find the exact solutions to the problem (1) by applying the following problems; then, we compare the results with the numerical solution obtained from the given method. To this aim, the {\boldsymbol{L}}_{{\infty }} and {\boldsymbol{L}}_{2} error norms are respectively defined as [48]

    {L}_{\infty } = {||U-{U}_{M}||}_{\infty }\cong \underset{j}{\mathit{max}}\left|{U}_{j}-{\left({U}_{M}\right)}_{j}\right|,
    {L}_{2} = {||U-{U}_{M}||}_{2}\cong \sqrt{h\sum \limits_{j = 0}^{M}{\left|{U}_{j}-{\left({U}_{M}\right)}_{j}\right|}^{2}}

    where U and {U}_{M} represent the exact solution and numerical solution, respectively.

    Example 1: Let us consider the TFBE (1) with the BCs

    U\left(O, t\right) = {l}_{1}\left(t\right) = {t}^{2} , \;\;\;\; U\left(1, t\right) = {l}_{2}\left(t\right) = -{t}^{2}, \;\;\;\;t\ge 0 ,

    and IC

    U\left(x, 0\right) = g\left(x\right) = 0, \;\;\;\;\;\; 0\le x\le 1 ,

    such that the forcing term f(x, t) is achieved as [45]

    f\left(x, t\right) = \frac{2{t}^{2-\gamma }{e}^{x}}{\mathit{\Gamma} (3-\gamma )}+{t}^{4}{e}^{2x}-v{t}^{2}{e}^{x},

    where the analytic solution is obtained as

    U\left(x, t\right) = {t}^{2}{e}^{x} .

    Numerical results are reported in Tables 13 and Figure 1. Table 1 lists the numerical solutions and the {L}_{2} and {L}_{\infty } error norms with \gamma = 0.5, \Delta t = 0.0025, {t}_{f} = 0.05\;\mathrm{a}\mathrm{n}\mathrm{d}\;v = 1 for various numbers of partitions M. As seen in Table 1, we notice that when the number of partitions M are increased, the {L}_{\infty } and {L}_{2} error norms will decrease considerably. Table 2 displays the numerical solutions with \gamma = 0.5, M = 40, t = 1, {t}_{f} = 0.05\;\mathrm{a}\mathrm{n}\mathrm{d}\;v = 1 for various values of \Delta t. In view of Table 2, we can see that when \Delta t decreases, the {L}_{\infty } and {L}_{2} error norms decrease, as was expected. Table 3 shows the numerical solutions with \Delta t = 0.00025, M = 40, t = 1, {t}_{f} = 0.05, v = 1 for various values of \gamma . As observed in Table 3, the {L}_{\infty } and {L}_{2} error norms decrease when γ increases. A comparison between the results of our proposed strategy and two other methods is demonstrated in detail, the researchers of which relied on their work on a weight function corresponding to the spline function in terms of degree; see [44,45]. Figure 1 represents the surfaces of the exact and numerical solutions of the TFBE in Example (1).

    Table 1.  Numerical solutions with \gamma = 0.5, \Delta t = 0.0025, {t}_{f} = 0.05, v = 1 for various numbers of partitions M .
    x M = 10 M = 20 M = 40 M = 80 Exact
    0.0 0.000000 0.000000 0.000000 0.000000 0.000000
    0.1 1.104360 1.105211 1.105166 1.105122 1.105101
    0.2 1.222151 1.222040 1.221593 1.221555 1.221511
    0.3 1.351010 1.350426 1.350012 1.349831 1.349789
    0.4 1.493377 1.492288 1.491990 1.491910 1.491844
    0.5 1.650589 1.650001 1.649822 1.648889 1.648731
    0.6 1.824211 1.823336 1.822449 1.822214 1.822110
    0.7 2.015587 2.014111 2.013822 2.013776 2.013692
    0.8 2.227577 2.226110 2.225699 2.225611 2.225562
    0.9 2.461410 2.461101 2.460893 2.459550 2.459491
    1.0 2.718202 2.718202 2.718202 2.718202 2.718202
    {\boldsymbol{L}}_{\bf{2}}{\bf{\times}} {\bf{10}}^{\bf{3}} 1.631895 0.440555 0.160761 0.062504
    {\boldsymbol{L}}_{\bf{2}}{\bf{\times}} {\bf{10}}^{\bf{3}} [44] 1.764966 0.465690 0.167743 0.095754
    {\boldsymbol{L}}_{\bf{2}}{\bf{\times}} {\bf{10}}^{\bf{3}} [45] 1.632995 0.447720 0.161833 0.082624
    {\boldsymbol{L}}_{{\infty }}{\bf{\times}} {\bf{10}}^{\bf{3}} 2.291578 0.64933 0.206677 0.032882
    {\boldsymbol{L}}_{{\infty }}{\bf{\times}} {\bf{10}}^{\bf{3}} [49] 3.101238 0.812842 0.209495 0.069208
    {\boldsymbol{L}}_{{\infty }}{\bf{\times}} {\bf{10}}^{\bf{3}} [50] 2.296683 0.625018 0.207352 0.033125

     | Show Table
    DownLoad: CSV
    Table 2.  Numerical solutions with \gamma = 0.5, M = 40, t = 1, {t}_{f} = 0.05, v = 1 for various values of \Delta t.
    x \Delta \mathrm{t} = 0.005 \Delta \mathrm{t} = 0.001 \Delta \mathrm{t} = 0.0005 \Delta \mathrm{t} = 0.00025 Exact
    0.0 0.000000 0.000000 0.000000 0.000000 0.000000
    0.1 1.105216 1.105211 1.105199 1.105186 1.105150
    0.2 1.221701 1.221601 1.221511 1.221445 1.221389
    0.3 1.350321 1.350188 1.350141 1.350110 1.349998
    0.4 1.492461 1.492211 1.492101 1.491879 1.491804
    0.5 1.649485 1.649112 1.648961 1.648822 1.648690
    0.6 1.822941 1.822675 1.822431 1.822310 1.822144
    0.7 2.014601 2.014201 2.014055 2.013979 2.013788
    0.8 2.226288 2.226001 2.225812 2.225699 2.225528
    0.9 2.260100 2.459980 2.459862 2.459785 2.459655
    1.0 2.718202 2.718202 2.718202 2.718202 2.718202
    {\boldsymbol{L}}_{\bf{2}}{\bf{\times}} {\bf{10}}^{\bf{3}} 0.659999 0.374901 0.232591 0.092489
    {\boldsymbol{L}}_{\bf{2}}{\bf{\times}} {\bf{10}}^{\bf{3}} [44] 0.176195 0.068869
    {\boldsymbol{L}}_{\bf{2}}{\bf{\times}} {\bf{10}}^{\bf{3}} [45] 0.375012 0.232768 0.092624
    {\boldsymbol{L}}_{\mathrm{\infty }}{\bf{\times}} {\bf{10}}^{\bf{3}} 0.936512 0.529997 0.326112 0.132945
    {\boldsymbol{L}}_{\mathrm{\infty }}{\bf{\times}} {\bf{10}}^{\bf{3}} [44] 0.665419 0.411883
    {\boldsymbol{L}}_{\mathrm{\infty }}{\bf{\times}} {\bf{10}}^{\bf{3}} [45] 0.530231 0.328303 0.133125

     | Show Table
    DownLoad: CSV
    Table 3.  Numerical solutions with \Delta t = 0.00025, \;M = 40, \;t = 1, \;{t}_{f} = 0.05, \;v = 1 for various values of γ.
    x \gamma =0.10 \gamma =0.25 \gamma =0.75 \gamma =0.90 Exact
    0.0 0.000000 0.000000 0.000000 0.000000 0.000000
    0.1 1.105068 1.104981 1.104890 1.104899 1.104882
    0.2 1.221701 1.221601 1.221511 1.221445 1.221389
    0.3 1.350321 1.350188 1.350141 1.350110 1.349998
    0.4 1.492461 1.492211 1.492101 1.491879 1.491804
    0.5 1.649485 1.649112 1.648961 1.648822 1.648690
    0.6 1.822941 1.822675 1.822431 1.822310 1.822144
    0.7 2.014601 2.014201 2.014055 2.013979 2.013788
    0.8 2.226288 2.226001 2.225812 2.225699 2.225528
    0.9 2.260100 2.459980 2.459862 2.459785 2.459655
    1.0 2.718202 2.718202 2.718202 2.718202 2.718202
    {\boldsymbol{L}}_{\bf{2}}{\bf{\times}} {\bf{10}}^{\bf{3}} 0.659999 0.374901 0.232591 0.092489
    {\boldsymbol{L}}_{\bf{2}}{\bf{\times}} {\bf{10}}^{\bf{3}} [44] 0.096733 0.090053 0.035448 0.044398
    {\boldsymbol{L}}_{\bf{2}}{\bf{\times}} {\bf{10}}^{\bf{3}} [45] 0.167077 0.165443 0.159924 0.166085
    {\boldsymbol{L}}_{{\infty }}{\bf{\times}} {\bf{10}}^{\bf{3}} 0.936512 0.529997 0.328112 0.132945
    {\boldsymbol{L}}_{{\infty }}{\bf{\times}} {\bf{10}}^{\bf{3}} [44] 0.272943 0.258623 0.124569 0.066682
    {\boldsymbol{L}}_{{\infty }}{\bf{\times}} {\bf{10}}^{\bf{3}} [45] 0.235837 0.232645 0.224532 0.232565

     | Show Table
    DownLoad: CSV
    Figure 1.  The surfaces of the exact and numerical solutions of the TFBE in Example (1).

    Example 2: Finally, we consider the TFBE (1) with the BCs

    U\left(0, t\right) = 0 , \;\;\;\; U\left(1, t\right) = 0, \;\;\;\;t\ge 0 ,

    and IC

    U\left(x, 0\right) = 0, \;\;\;\; 0\le x\le 1 ,

    where the source term f\left(x, t\right) can be obtained as [44]

    f\left(x, t\right) = \frac{2{t}^{2-\gamma }\mathrm{sin}\left(2\pi x\right)}{\mathit{\Gamma} (3-\gamma )}+{2\pi t}^{4}\mathrm{sin}\left(2\pi x\right)\mathrm{cos}\left(2\pi x\right)+4v{t}^{2}{\pi }^{2}\mathrm{sin}\left(2\pi x\right).

    The exact solution is

    U\left(x, t\right) = {t}^{2}\mathrm{sin}\left(2\pi x\right) .

    Numerical results are represented in Tables 4 and 5 and Figure 2. Tables 4 and 5 report the numerical solutions for various numbers of partitions M and values of \Delta t . As seen in Tables 4 and 5, when the number of partitions M increased, the error norms {L}_{\infty } and {L}_{\bf{2}} will decrease considerably, while, in Table 5, we can see that when ∆t decrease, the error norms {L}_{\infty } and {L}_{\bf{2}} decrease. Figure 2 demonstrates the surfaces of the exact and numerical solutions of the TFBE in Example (2).

    Table 4.  Numerical solutions with \gamma = 0.5, \Delta t = 0.0025, {t}_{f} = 0.05, v = 1 for various numbers of partitions M .
    x M = 10 M = 20 M = 40 M = 80 Exact
    0.0 1,000,000 1,000,000 1,000,000 1,000,000 1,000,000
    0.1 0.951196 0.950876 0.951005 0.951077 0.951070
    0.2 0.808211 0.808681 0.808911 0.808988 0.808978
    0.3 0.587211 0.587513 0.587699 0.587761 0.587754
    0.4 0.308662 0.308901 0.308987 0.309011 0.309006
    0.5 0.000000 0.000000 0.000000 0.000000 0.000000
    0.6 -0.308662 -0.308843 -.308931 -0.309011 -0.309006
    0.7 -0.587194 -0.587501 -0.587694 -0.587737 -0.587732
    0.8 -0.808205 -0.808644 -0.808823 -0.808972 -0.808970
    0.9 -0.951211 -0.951661 -0.951811 -0.951965 -0.951960
    1.0 -1.000000 -1.000000 -1.000000 -1.000000 -1.000000
    {\boldsymbol{L}}_{\bf{2}}{\bf{\times}} {\bf{10}}^{\bf{3}} 0.435298 0.183971 0.041943 0.001960
    {\boldsymbol{L}}_{\bf{2}}{\bf{\times}} {\bf{10}}^{\bf{3}} [44] 1.224329 0.177703
    {\boldsymbol{L}}_{\bf{2}}{\bf{\times}} {\bf{10}}^{\bf{3}} [45] 2.899412 0.577143
    {\boldsymbol{L}}_{{\infty }}{\bf{\times}} {\bf{10}}^{\bf{3}} 0.731071 0.273289 0.063201 0.004168
    {\boldsymbol{L}}_{{\infty }}{\bf{\times}} {\bf{10}}^{\bf{3}} [44] 1.730469 0.253053
    {\boldsymbol{L}}_{{\infty }}{\bf{\times}} {\bf{10}}^{\bf{3}} [45] 4.063808 0.813220

     | Show Table
    DownLoad: CSV
    Table 5.  Numerical solutions with \gamma = 0.5, M = 80, t = 1, {t}_{f} = 0.05, v = 1 for various values of \Delta t. .
    x \Delta t = 0.005 \Delta t = 0.001 \Delta t = 0.0005 \Delta t = 0.00025 Exact
    0.0 1000000 1000000 1000000 1000000 1000000
    0.1 0.951196 0.950876 0.951005 0.951077 0.951070
    0.2 0.808211 0.808681 0.808911 0.808988 0.808978
    0.3 0.587211 0.587513 0.587699 0.587761 0.587754
    0.4 0.308662 0.308901 0.308987 0.309011 0.309006
    0.5 0.000000 0.000000 0.000000 0.000000 0.000000
    0.6 -0.308662 -0.308843 -.308931 -0.309011 -0.309006
    0.7 -0.587194 -0.587501 -0.587694 -0.587737 -0.587732
    0.8 -0.808205 -0.808644 -0.808823 -0.808972 -0.808970
    0.9 -0.951211 -0.951661 -0.951811 -0.951965 -0.951960
    1.0 -1.000000 -1.000000 -1.000000 -1.000000 -1.000000
    {\boldsymbol{L}}_{\bf{2}}{\bf{\times}} {\bf{10}}^{\bf{3}} 0.124034 0.054081 0.014255 0.001960
    {\boldsymbol{L}}_{\bf{2}}{\bf{\times}} {\bf{10}}^{\bf{3}} [44] 0.532436 0.188710
    {\boldsymbol{L}}_{\bf{2}}{\bf{\times}} {\bf{10}}^{\bf{3}} [45] 0.359489 0.017828
    {\boldsymbol{L}}_{{\infty }}{\bf{\times}} {\bf{10}}^{\bf{3}} 0.175611 0.077465 0.028523 0.004168
    {\boldsymbol{L}}_{{\infty }}{\bf{\times}} {\bf{10}}^{\bf{3}} [44] 0.753171 0.267546
    {\boldsymbol{L}}_{{\infty }}{\bf{\times}} {\bf{10}}^{\bf{3}} [45] 0.512105 0.0321162

     | Show Table
    DownLoad: CSV
    Figure 2.  The surfaces of the exact and numerical solutions of the TFBE in Example (2).

    This paper presented a numerical approach based on the CBSGM with a quadratic weight function for the TFBE including the time Caputo derivative. Numerical results have shown that the proposed method is an appropriate and efficient scheme for solving such problems.

    The authors declare no conflict of interest.



    [1] T. Z. Li, C. L. Mei, Estimation and inference for varyig coefficient partially nonlinear models, J. Stat. Plann Inference, 143 (2013), 2023–2037. doi: 10.1016/j.jspi.2013.05.011
    [2] J. Yang, H. Yang, Smooth-threshold estimating equations for varying coefficient partially nonlinear models based on orthogonality-projection method, J. Comput. Appl. Math., 302 (2016), 24–37. doi: 10.1016/j.cam.2016.01.038
    [3] Y. Y. Qian, Z. S. Huang, Statistical inference for a varying-coefficient partially nonlinear model with measurement errors, Stat. Methodol., 32 (2016), 122–130. doi: 10.1016/j.stamet.2016.05.004
    [4] X. S. Zhou, P. X. Zhao, X. L. Wang, Empirical likelihood inferences for varying coefficient partially nonlinear models, J. Appl. Stat., 44 (2017), 474–492. doi: 10.1080/02664763.2016.1177496
    [5] Y. L. Jiang, Q. H. Ji, B. J. Xie, Robust estimation for the varying coefficient partially nonlinear models, J. Comput. Appl. Math., 326 (2017), 31–43. doi: 10.1016/j.cam.2017.04.028
    [6] Y. T. Xiao, Z. S. Chen, Bias-corrected estimations in varying-coefficient partially nonlinear models with measurement error in the nonparametric part, J. Appl. Stat., 45 (2018), 586–603. doi: 10.1080/02664763.2017.1288201
    [7] X. L. Wang, P. X. Zhao, H. Y. Du, Statistical inferences for varying coefficient partially nonlinear model with missing covariates, Commun. Stat. Theory Methods, (2019).
    [8] R. J. A. Little, D. B. Rubin, Statistical analysis with missing data, 2 Eds., New York: Wiley, 2002.
    [9] P. E. Cheng, Nonparametric estimation of mean functional with data missing at random, J. Am. Stat. Assoc., 89 (1994), 81–87. doi: 10.1080/01621459.1994.10476448
    [10] Q. H. Wang, J. N. K. Rao, Empirical likelihood for linear regression models under imputation for missing responses, Can. J. Stat., 29 (2001), 596–608.
    [11] Q. H. Wang, J. N. K. Rao, Empirical likelihood-based inference under imputation for missing response data, Ann. Stat., 30 (2002), 896–924.
    [12] Q. H. Wang, O. Linton, W. Härdle, Semiparametric regression analysis with missing response at random, J. Am. Stat. Assoc., 99 (2004), 334–345. doi: 10.1198/016214504000000449
    [13] L. G. Xue, Empirical likelihood for linear models with missing responses, J. Multivar. Anal., 100 (2009), 1353–1366. doi: 10.1016/j.jmva.2008.12.009
    [14] L. G. Xue, Empirical likelihood confidence intervals for response mean with data missing at Random, Scand. J. Stat., 36 (2009), 671–685. doi: 10.1111/j.1467-9469.2009.00651.x
    [15] X. L. Wang, F. Chen, L. Lin, Empirical likelihood inference for estimating equation with missing data, Sci. China Math., 56 (2013), 1233–1245. doi: 10.1007/s11425-012-4504-x
    [16] A. Owen, Empirical likelihood ratio confidence intervals for a single functional, Biometrika, 75 (1988), 237–249. doi: 10.1093/biomet/75.2.237
    [17] A. Owen, Empirical likelihood ratio confidence regions, Ann. Stat., 18 (1990), 90–120.
    [18] P. P. Chen, S. Y. Feng, L. G. Xue, Statistical inference for semiparametric varying coefficient partially linear model with missing data (in Chinese), Acta Math. Sci., 35 (2015), 345–358.
    [19] Y. P. Mack, B. W. Silverman, Weak and strong uniform consistency of kernel regression estimates, Z. Wahrsch. Verw. Gebiete., 61 (1982), 405–415. doi: 10.1007/BF00539840
    [20] J. Shi, T. S. Lau, Empirical Likelihood for partially linear models, Multivar. Anal., 72 (2000), 132–148. doi: 10.1006/jmva.1999.1866
    [21] X. L. Wang, Y. Q. Song, L. Lin, Handling estimating equation with nonignorably missing data based on SIR algorithm, J. Comput. Appl. Math., 326 (2017), 62–70. doi: 10.1016/j.cam.2017.05.016
    [22] X. L. Wang, Y. Q. Song, S. X. Zhang, An efficient estimation for the parameter in additive partially linear models with missing covariates, J. Korean Stat. Soc., 49 (2020), 779–801. doi: 10.1007/s42952-019-00036-6
    [23] H. M. Sun, Y. H. Luan, J. M. Jiang, A new classified mixed model predictor, J. Stat. Plann. Inference, 207 (2020), 45–54. doi: 10.1016/j.jspi.2019.11.001
    [24] Y. Fang, G. Cheng, Z. F. Qu, Optimal reinsurance for both an insurer and a reinsurer under general premium principles, AIMS Math., 5 (2020), 3231–3255. doi: 10.3934/math.2020208
    [25] Y. Fang, X. Wang, H. L. Liu, T. Li, Pareto-optimal reinsurance for both the insurer and the reinsurer with general premium principles, Commun. Stat. Theory Methods, 48 (2019), 6134–6154. doi: 10.1080/03610926.2018.1528364
  • Reader Comments
  • © 2021 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(3061) PDF downloads(114) Cited by(4)

Figures and Tables

Figures(4)  /  Tables(1)

Other Articles By Authors

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog