Breast cancer mitotic cell detection using cascade convolutional neural network with U-Net

Xi Lu; Zejun You; Miaomiao Sun; Jing Wu; Zhihong Zhang; Xi Lu; Zejun You; Miaomiao Sun; Jing Wu; Zhihong Zhang

doi:10.3934/mbe.2021036

Mathematical Biosciences and Engineering

2021, Volume 18, Issue 1: 673-695. doi: 10.3934/mbe.2021036

Previous Article Next Article

Research article

Breast cancer mitotic cell detection using cascade convolutional neural network with U-Net

1.
School of Mechanical Engineering, Southeast University, Nanjing 211189, China
2.
Department of Pathology, The First Affiliated Hospital of Nanjing Medical University, Nanjing 210029, China

Received: 23 September 2020 Accepted: 15 December 2020 Published: 18 December 2020

The number of mitotic tumor cells detected in each slide is one of the key indicators of breast cancer prognosis. However, accurate mitotic cell counts are still a difficult problem for pathologists and related experts. Traditional methods use manual design algorithms to extract features of mitotic cells, and most methods rely on sliding windows to achieve pixel-level classification through deep learning. However, the complex background and high resolution of pathological images make the above methods time-consuming and ineffective. In order to solve the above problems, we propose a new cascaded convolutional neural network UBCNN (cascaded CNN based on UNet), which consists of three parts: semantic segmentation and classification to detect mitosis. First, we use an improved UNet ++ segmentation network to locate the candidate set of mitotic targets. Secondly, an adequately labeled cell nucleus data set is sent to an improved two-dimensional VNet network, and the cell nucleus is located by means of semantic segmentation to obtain accurate image blocks of mitotic and non-mitotic cells. Finally, the obtained cell image block is used to train a convolutional neural network to achieve binary classification, and the candidate set area is screened to retain the final result of mitosis cells. This paper verifies the detection effect of the above-mentioned cascade detection algorithm on the ICPR 2012 and 2014 mitosis automatic detection competition data sets. The evaluation indicators include accuracy, recall and F-score. Our cascade detection algorithm based on segmentation and classification reached 0.831 on the ICPR 2012 data set and 0.576 on the ICPR 2014 data set. Compared with other existing algorithms, the detection effect was improved, which was very competitive.

Keywords:

Citation: Xi Lu, Zejun You, Miaomiao Sun, Jing Wu, Zhihong Zhang. Breast cancer mitotic cell detection using cascade convolutional neural network with U-Net[J]. Mathematical Biosciences and Engineering, 2021, 18(1): 673-695. doi: 10.3934/mbe.2021036

Related Papers:

[1]	Siqin Tang, Hong Li . A Legendre-tau-Galerkin method in time for two-dimensional Sobolev equations. AIMS Mathematics, 2023, 8(7): 16073-16093. doi: 10.3934/math.2023820
[2]	Jae-Myoung Kim . Time decay rates for the coupled modified Navier-Stokes and Maxwell equations on a half space. AIMS Mathematics, 2021, 6(12): 13423-13431. doi: 10.3934/math.2021777
[3]	Xiaojun Zhou, Yue Dai . A spectral collocation method for the coupled system of nonlinear fractional differential equations. AIMS Mathematics, 2022, 7(4): 5670-5689. doi: 10.3934/math.2022314
[4]	Yuanqiang Chen, Jihui Zheng, Jing An . A Legendre spectral method based on a hybrid format and its error estimation for fourth-order eigenvalue problems. AIMS Mathematics, 2024, 9(3): 7570-7588. doi: 10.3934/math.2024367
[5]	A.S. Hendy, R.H. De Staelen, A.A. Aldraiweesh, M.A. Zaky . High order approximation scheme for a fractional order coupled system describing the dynamics of rotating two-component Bose-Einstein condensates. AIMS Mathematics, 2023, 8(10): 22766-22788. doi: 10.3934/math.20231160
[6]	Xingyang Ye, Chuanju Xu . A posteriori error estimates of spectral method for the fractional optimal control problems with non-homogeneous initial conditions. AIMS Mathematics, 2021, 6(11): 12028-12050. doi: 10.3934/math.2021697
[7]	Chuanhua Wu, Ziqiang Wang . The spectral collocation method for solving a fractional integro-differential equation. AIMS Mathematics, 2022, 7(6): 9577-9587. doi: 10.3934/math.2022532
[8]	Ishtiaq Ali . Long time behavior of higher-order delay differential equation with vanishing proportional delay and its convergence analysis using spectral method. AIMS Mathematics, 2022, 7(4): 4946-4959. doi: 10.3934/math.2022275
[9]	Sharifah E. Alhazmi, M. A. Abdou, M. Basseem . Physical phenomena of spectral relationships via quadratic third kind mixed integral equation with discontinuous kernel. AIMS Mathematics, 2023, 8(10): 24379-24400. doi: 10.3934/math.20231243
[10]	Manal Alqhtani, Khaled M. Saad . Numerical solutions of space-fractional diffusion equations via the exponential decay kernel. AIMS Mathematics, 2022, 7(4): 6535-6549. doi: 10.3934/math.2022364

Abstract

1. Introduction

It is well known that the spectral method has high-order accuracy for smooth problems. The spectral method together with the difference method and the finite element method has become an important method for the numerical solution of partial differential equations (PDEs), and has been successfully applied to solve many practical problems. In recent years, with regard to the differential equations of time evolution, the high-order discrete scheme in time has received widespread attention and has become one of the hot spots in the field of numerical computing. The discontinuous Galerkin method in time is constantly developing, and a better higher-order discrete scheme in time is established ^[1,2,3]. The explicit, implicit and implicit-explicit Runge-Kutta methods have also made great progress: a local discontinuous Galerkin method with implicit-explicit time-marching is used to solve the multi-dimensional convection-diffusion problems and time-dependent incompressible fluid flow in ^[4,5,6]. In ^[7,8,9], the spectral method in time and the time multi-interval spectral method are also proposed. The single interval and multi-interval Legendre spectral methods in time are established for the parabolic equations, in which the $L^2$ -optimal error estimate in space is obtained in ^[10].

The Maxwell equation is a set of important PDEs that describes electromagnetic field phenomena, and some effective numerical methods have been established for the Maxwell equation by scholars ^[11,12,13]. The finite-difference time-domain method (also called Yee's scheme) for the Maxwell equation is proposed in ^[14]. In ^[15,16], an energy-conserved splitting spectral method for solving the Maxwell equation is given. For the 2-D Maxwell equation, a Legendre-Galerkin method in space and the energy-conserved splitting spectral method in time is constructed ^[17]. In previous work, the different method is used in the time direction. For the 1-D Maxwell equation of inhomogeneous media with discontinuous solutions, the multidomain Legendre-Galerkin and the multidomain Legendre-tau method are established in ^[18,19], and the optimal error estimates of the semi-discrete schemes are given.

Consider the following 1-D Maxwell equation ^[20]

$\begin{equation} \left\{\begin{array}{ll} \epsilon \partial_t E_z = \partial_x H_y, &(x, t)\in \Omega, \\ \mu \partial_t H_y = \partial_x E_z, &(x, t)\in \Omega, \\ E_z(-1, t) = E_z(1, t) = 0, &t\in I_t, \\ E_z(x, 0) = E_{z0}(x), \; H_y(x, 0) = H_{y0}(x), \; \; &x\in I_x, \end{array}\right. \end{equation}$

(1.1)

where $I_x = (-1, 1)$ , $I_t = (0, T]$ , and $\Omega = I_x\times I_t$ . $E_z$ and $H_y$ stand for the electric field and the magnetic field, respectively. The positive constants $\epsilon$ and $\mu$ stand for the electric permeability and the magnetic permeability, respectively.

In ^[21,22], an $h$ - $p$ version of the Petrov-Galerkin time stepping method is used to solve the nonlinear initial value problems by transforming the second-order problem into a first-order system. For the linear second-order wave equation, it is often transformed into the first-order system similar to equation (1.1) by using the substitution $v = \frac{ \partial u}{ \partial t}, \; w = \frac{ \partial u}{ \partial x}$ ^[23]. It is interesting to note that some methods use the derivative as the main unknown function, and $u$ is expressed as the integral of $w$ .

In this paper, a Legendre-tau space-time (LT-ST) spectral method is developed to solve the 1-D Maxwell equation $(1.1)$ and a time multi-interval Legendre-tau spectral method is considered. The scheme is based on the Legendre-tau method, which uses polynomials of different degrees are used to approximate the electric field $E_z$ and magnetic field $H_y$ , respectively, so that they can be decoupled in computation. After decoupling, it is an equation only about $E_z$ , which can be solved by the method in ^[10]. The method is also applied to the numerical solutions of the 1-D nonlinear Maxwell equation.

The paper is organized as follows. In Section 2, a Legendre-tau space-time spectral method for $(1.1)$ is presented, and stability analysis and error estimate are given. In Section 3, a time multi-interval Legendre-tau spectral method is developed, and its error estimate is also obtained. Some numerical results are given in Section 4. Finally, the method is applied to the numerical solution of the 1-D nonlinear Maxwell equation in Section 5.

2. Legendre-tau space-time spectral method

In this section, a Legendre-tau space-time spectral method is presented for the problem $(1.1)$ . Moreover, the stability and the error estimate of this method are given.

2.1. Preliminaries

Let $(\cdot, \cdot)_Q$ and $\|\cdot\|_Q$ be the inner product and the norm of $L^2(Q)$ , where Q stands for $\Omega$ , $I_x$ and $I_t$ , respectively. For a nonnegative integer $m$ , let $\|\cdot\|_{m, I}$ and $|\cdot|_{m, I}$ be the norm and the semi-norm of the classical Sobolev space $H^m(I)$ , where $I$ stands for $I_x$ or $I_t$ , respectively. Define

${H^1_{0}(I) = \{ v \in H^1 (I): v(-1) = v(1) = 0 \}.}$

For a pair of positive integers $N$ and $M$ , define $L = (N, M)$ . Let $\mathbb P_{N}(I_x)$ be the space of polynomials of degree at most $N$ on $I_x$ . Define the polynomial space

${V_{N} = \{v \in \mathbb P_{N}(I_x)\}, }$

and the approximation space in space

$\begin{equation} V_N^{0} = H^1_{0}(I_x)\cap V_{N} , \; \; \; \; V_{N-1} = \{v \in \mathbb P_{N-1}(I_x)\}. \end{equation}$

(2.1)

Let $\mathbb P_{M}(I_t)$ be the space of polynomials of degree at most $M$ on $I_t$ , we define the approximation space in time

$\begin{equation} V_{M} = \{v\in \mathbb P_{M}(I_t)\}, \; \; \; \; V_{M-1} = \{v\in \mathbb P_{M-1}(I_t)\}. \end{equation}$

(2.2)

Let $x^C_j$ and $\omega_{j}^{C}(0\leq j\leq N)$ be the Chebyshev-Gauss-Lobatto (CGL) points and the corresponding weights on $I_x$ . We define the CGL interpolation operator $I_N^C v \in V_N$ :

$I_N^C v (x_j^C ) = v(x_j^C ), \; \; \; \; 0\leq j\leq N.$

Similarly, let $x^L_j$ and $\omega_{j}^{L}(0\leq j\leq N)$ be the Legendre-Gauss-Lobatto (LGL) points and the corresponding weights on $I_x$ . $I_N^L v \in V_N$ denotes the LGL interpolation operator, and

$I_N^L v (x_j^L ) = v(x_j^L ), \; \; \; \; 0\leq j\leq N.$

We denote by $P_{N}: L^{2}(I_x) \rightarrow V_N$ the $L^2(I_x)$ -Legendre projection operator and define $P_{N}^1:H^{1}(I_{x}) \rightarrow V_{N}$ by

$\begin{equation} P_{ N}^1 u (x) = u(-1) + \int_{-1}^{ x} P_{N -1} \partial_x u(y) {\rm dy}, \; \; x\in I_x. \end{equation}$

(2.3)

It is easy to see that

$\begin{equation} {P}^1_{N}{u}(-1) = {u}(-1), \; \; \; \; \; \; {P}^1_{N}{u}(1) = {u}(1), \end{equation}$

(2.4)

$\begin{equation} ( \partial_x P^1_{N}u - \partial_x u, v) = (P_{N-1} \partial_x u- \partial_x u, v) = 0, \; \; \forall v\in V_{N-1}. \end{equation}$

(2.5)

Let $C$ be a generic positive constant independent of $N$ , and the following approximation results can be found in ^[10,24].

Lemma 2.1. If $u\in H^r(I_x)$ , then

$\begin{aligned} & \|P_Nu-u\|_{I_x}\le CN^{-r}|u|_{r, I_x}, \; \; \; & r\ge 0, \; \; \; &\\ & \|I^L_Nu-u\|_{I_x}\le CN^{l-r}|u|_{r, I_x}, \; \; \; & r\ge 1, \; \; \; &l = 0, 1.\\ &|P^1_Nu-u|_{l, I_x}\le CN^{l-r}|u|_{r, I_x}, \; \; \; & r\ge 1, \; \; \; &l = 0, 1. \end{aligned}$

Let $t^C_j$ and $\omega_{j}^{C}(0\leq j\leq M)$ be the CGL points and the corresponding weights on $I_t$ , and let $t^L_j$ and $\omega_{j}^{L}(0\leq j\leq M)$ be the LGL points and the corresponding weights on $I_t$ . We denote by $P_M:L^{2}(I_t)\rightarrow V_{M}$ the $L^{2}(I_t)$ -Legendre projection operator and define $P^1_{M}:H^1(I_t)\rightarrow V_{M}$ as

$\begin{equation} {P}^1_{M}{v}({t}) = {v}(0) + \int_{0}^{{t}}{P}_{M -1}\partial_{{t}}{v}(s) {\rm ds}, \; \; {t}\in {I_t}. \end{equation}$

(2.6)

It is easy to find that

$\begin{equation} {P}^1_{M}{u}(-1) = {u}(-1), \; \; \; \; \; \; {P}^1_{M}{u}(1) = {u}(1), \end{equation}$

(2.7)

$\begin{equation} (\partial_{t}{P}^1_{M}u - \partial_{t}{u}, {v}) = ({P}_{M-1}\partial_{{t}}{u}-\partial_{{t}}{u}, {v}) = 0, \; \; \forall v\in V_{M-1}. \end{equation}$

(2.8)

The following approximation result can be found in ^[10].

Lemma 2.2. If $u\in H^{\sigma}(I_t)$ and $\sigma \geq 1$ , then

$|P_M^{1}v-v|_{l, I_t}\leq CM^{l-\sigma}|v|_{\sigma, I_t}, \; \; \; \; l = 0, 1,$

where $C$ is a positive constant independent of $M$ .

The problem $(1.1)$ is expressed in a weak form: Find $E_{z} \in H_0^{1}(I_x)\otimes H^{1}(I_t)$ and $H_{y} \in L^2(I_x)\otimes H^{1}(I_t)$ such that

$\begin{equation} \left\{\begin{array}{ll} ({ \epsilon} \partial_t E_{ z }, v)_{\Omega} + (H_{ y }, \partial_x v)_{\Omega} = 0, \; \; &\forall\; v \in H_0^{1}(I_x)\otimes L^2(I_t), \; \\ ({\mu} \partial_t H_{y}, w)_{\Omega} - ( \partial_x E_{ z }, w)_{\Omega} = 0, \; \; &\forall\; w\in L^2(I_x)\otimes L^2(I_t), \\ E_z(x, 0) = E_{z0}(x), \; H_y(x, 0) = H_{y0}(x), \; \; &\forall x\in I_x. \end{array}\right. \end{equation}$

(2.9)

The LT-ST scheme to the problem $(1.1)$ is: Find $E_{zL} \in V_N^{0}\otimes V_{M}$ and $H_{yL} \in V_{N-1}\otimes V_{M}$ such that

$\begin{equation} \left\{\begin{array}{ll} ({ \epsilon} \partial_t E_{zL}, v)_{\Omega}+(H_{yL}, \partial_x v)_{\Omega} = 0, \; \; &\forall\; v \in V_N^{0}\otimes V_{M-1}, \; \\ ({\mu} \partial_t H_{yL}, w)_{\Omega}-( \partial_x E_{zL}, w)_{\Omega} = 0, \; \; &\forall\; w\in V_{N-1}\otimes V_{M-1}, \\ E_{zL}(x, 0) = I_N^L E_{z0}(x), \; \; H_{yL}(x, 0) = P_{N-1}I_N^L H_{y0}(x), \; \; &\forall x\in I_x. \end{array}\right. \end{equation}$

(2.10)

2.2. Stability analysis

In the following section, the stability analysis of (2.10) is considered. Suppose that there are perturbations $\tilde f_i(i = 1, 2)$ on the right-hand side. For simplicity, the original notations $E_{zL}$ and $H_{yL}$ are used to represent the solutions to the perturbation problem, which satisfies the following perturbation equation:

$\begin{equation} \left\{\begin{array}{ll} ( \epsilon \partial_t E_{zL}, v)_{\Omega} + ( H_{yL}, \partial_x v)_{\Omega} = ( \tilde f_1, v)_{\Omega}, \; \; &\forall\; v \in V_N^{0}\otimes V_{M-1}, \; \\ (\mu \partial_t H_{yL}, w)_{\Omega} - ( \partial_x E_{zL}, w)_{\Omega} = ( \tilde f_2, w)_{\Omega}, \; \; &\forall\; w\in V_{N-1}\otimes V_{M-1}, \\ E_{zL}(x, 0) = 0, \; \; H_{yL}(x, 0) = 0, \; \; &\forall x\in I_x. \end{array}\right. \end{equation}$

(2.11)

Theorem 2.1. Let $E_{zL}$ and $H_{yL}$ are the solutions to (2.11). Suppose that $\tilde f_i(i = 1, 2)$ are perturbations on the right-hand side, such that

$\begin{equation} \|\sqrt{\epsilon} E_{zL}\|_{\Omega}^2 +\|\sqrt{\mu}H_{yL}\|_{\Omega}^2 +T({\|\sqrt{ \epsilon}E_{zL}(T)\|_{I_x}^2+\|\sqrt{\mu}H_{yL}(T)\|_{I_x}^2}) \leq CT^2({\|\tilde{f_{ 1 }}\|_{\Omega}^2+\|\tilde{f_{ 2 }} \|_{\Omega}^2}). \end{equation}$

(2.12)

Proof. Taking $v = \tilde{E}_{zL}: = t^{-1}E_{zL}\in V_{N}^{0}\otimes V_{M-1}$ and $w = \tilde{H}_{yL}: = t^{-1}H_{yL}\in V_{N-1}\otimes V_{M-1}$ in (2.11), we get

$\begin{equation} \left\{\begin{array}{ll} ( \epsilon \partial_t (t\tilde{E}_{zL}), \tilde{E}_{zL})_{\Omega} +( t\tilde{H}_{yL}, \partial_x \tilde{E}_{zL})_{\Omega} = ( \tilde f_1, \tilde{E}_{zL})_{\Omega}, \\ (\mu \partial_t (t\tilde{H}_{yL}), \tilde{H}_{yL})_{\Omega} -(t \partial_x\tilde{E}_{zL}, \tilde{H}_{yL})_{\Omega} = (\tilde f_2, \tilde{H}_{yL})_{\Omega}, \end{array}\right. \end{equation}$

(2.13)

which leads to

$\begin{equation} ( \epsilon \partial_t (t\tilde{E}_{zL}), \tilde{E}_{zL})_{\Omega} + (\mu \partial_t (t\tilde{H}_{yL}), \tilde{H}_{yL})_{\Omega} = ( \tilde f_1, \tilde{E}_{zL})_{\Omega}+ ( \tilde f_2, \tilde{H}_{yL})_{\Omega}. \end{equation}$

(2.14)

By integration by parts,

$\begin{equation} \begin{aligned} &( \epsilon \partial_t(t\tilde{E}_{zL}), \tilde{E}_{zL})_{\Omega} = ( \epsilon\tilde{E}_{zL}, \tilde{E}_{zL})_{\Omega} +( \epsilon t \partial_t \tilde{E}_{zL}, \tilde{E}_{zL})_{\Omega}\\ & = \|\sqrt {\epsilon}\tilde{E}_{zL}\|^2_{\Omega}+\frac{1}{2}T\|\sqrt {\epsilon}\tilde{E}_{zL}(T)\|_{I_x}^2 -\frac{1}{2}\|\sqrt {\epsilon}\tilde{E}_{zL}\|_{\Omega}^2 = \frac{1}{2}({\|\sqrt {\epsilon}\tilde{E}_{zL}\|^2_{\Omega}+T\|\sqrt {\epsilon}\tilde{E}_{zL}(T)\|_{I_x}^2}), \\ &(\mu \partial_t(t\tilde{H}_{zL}), \tilde{H}_{zL})_{\Omega} = (\mu\tilde{H}_{zL}, \tilde{H}_{zL})_{\Omega} +(\mu t \partial_t \tilde{H}_{zL}, \tilde{H}_{zL})_{\Omega}\\ & = \frac{1}{2}({\|\sqrt \mu \tilde{H}_{zL}\|^2_{\Omega}+T\|\sqrt \mu \tilde{H}_{zL}(T)\|_{I_x}^2}), \\ \end{aligned} \end{equation}$

(2.15)

and using the Cauchy-Schwarz inequality

$\begin{equation} \begin{aligned} |( \tilde{f_1}, \tilde E_{zL})_{\Omega}+( \tilde{f_2}, \tilde H_{yL})_{\Omega}| &\leq\|\tilde{f_1}\|_{\Omega}\|\tilde E_{zL}\|_{\Omega} +\|\tilde{f_2}\|_{\Omega}\|\tilde H_{yL}\|_{\Omega}\\ &\leq\frac14({\|\sqrt {\epsilon}\tilde E_{zL }\|_{\Omega}^2+\|\sqrt \mu \tilde{H}_{zL}\|_{\Omega}^2}) +\frac{1}{ \epsilon}\|\tilde{f_1}\|_{\Omega}^2 +\frac{1}{\mu}\|\tilde{f_2}\|_{\Omega}^2, \end{aligned} \end{equation}$

(2.16)

where $\tilde{E}_{zL}(T)$ = $\tilde{E}_{zL}(x, T)$ and $\tilde{H}_{yL}(T)$ = $\tilde{H}_{yL}(x, T)$ . Substituting (2.15)-(2.16) into (2.14),

$\begin{equation} \begin{aligned} \frac{1}{4}({\|\sqrt {\epsilon}\tilde{E}_{zL}\|^2_{\Omega}+\|\sqrt \mu \tilde{H}_{zL}\|^2_{\Omega}}) &+\frac{T}{2}({\|\sqrt {\epsilon}\tilde{E}_{zL}(T)\|_{I_x}^2+\|\sqrt \mu \tilde{H}_{zL}(T)\|_{I_x}^2})\\ &\leq\frac{1}{ \epsilon}\|\tilde{f_1}\|_{\Omega}^2+\frac{1}{\mu}\|\tilde{f_2}\|_{\Omega}^2. \end{aligned} \end{equation}$

(2.17)

and noting that $\|E_{zL}\|_\Omega\leq T\|\tilde E_{zL}\|_\Omega, \; \; \|H_{yL}\|_\Omega\leq T\|\tilde H_{yL}\|_\Omega$ , we get the result of (2.12).

2.3. Error estimate

In the following section, the error estimate of (2.10) is given. In order to deal with the error of the initial value, the following auxiliary problem is considered ^[25]

$\begin{equation} \left\{\begin{array}{ll} \epsilon \partial_t E = \partial_x H, &(x, t)\in \Omega, \\ \mu \partial_t H = \partial_x E, &(x, t)\in \Omega, \\ E(-1, t) = E(1, t) = 0, &t\in I_t, \\ E(x, 0) = I_N^L E_{z0}(x), \; H(x, 0) = P_{N-1}I_N^L H_{y0}(x), \; \; &x\in I_x . \end{array}\right. \end{equation}$

(2.18)

Firstly, the estimate between the two solutions to (2.10) and (2.18) is considered. We define

$\begin{equation} \label{{bjhs}}{E_{a} = P_{N}^{1}P_{M}^{1}E, \; \; \; H_{a} = P_{N-1}P_{M}^{1}H.} \end{equation}$

(2.19)

By (2.5) and (2.8), we have

$\begin{aligned} &( \partial_t P_{M}^1 E, v)_{I_t} = ( \partial_t E, v)_{I_t}, \; &\forall\; v \in V_{M-1}, \\ &( \partial_x P_{N}^1 E, w)_{I_x} = ( \partial_x E, w)_{I_x}, \; &\forall\; w\in V_{N-1}, \end{aligned}$

and

$\begin{equation} \left\{\begin{array}{ll} ({ \epsilon} \partial_t E_{a}, v)_{\Omega}+(H_{a}, \partial_x v)_{\Omega} = ({ \epsilon}P_{N}^{1} \partial_t E, v)_{\Omega}-(P_{M}^{1} \partial_x H, v)_{\Omega}, \; \; &\forall\; v \in V_N^{0}\otimes V_{M-1}, \; \\ ({\mu} \partial_t H_{a}, w)_{\Omega}-( \partial_x E_{a}, w)_{\Omega} = ({\mu} \partial_t H, w)_{\Omega}-(P_{M}^{1} \partial_x E, w)_{\Omega}, \; \; &\forall\; w\in V_{N-1}\otimes V_{M-1}. \end{array}\right. \end{equation}$

(2.20)

Let $e_z = E_{zL}-E_{a}$ and $e_y = H_{yL}-H_{a}$ . By (2.10) and (2.20), the following error equation is obtained

$\begin{equation} \left\{\begin{array}{ll} ( \epsilon \partial_t e_z , v)_{\Omega} + (e_y , \partial_x v)_{\Omega} = ( f_1, v)_{\Omega}, \; \; &\forall\; v \in V_N^{0}\otimes V_{M-1}, \; \\ (\mu \partial_t e_y , w)_{\Omega} - ( \partial_x e_z , w)_{\Omega} = ( f_2, w)_{\Omega}, \; \; &\forall\; w\in V_{N-1}\otimes V_{M-1}, \\ e_{z}(x, 0) = 0, \; \; e_{y}(x, 0) = 0, \; \; &\forall x\in I_x. \end{array}\right. \end{equation}$

(2.21)

Due to (2.18), we have $\epsilon \partial_t E = \partial_x H$ , $\mu \partial_t H = \partial_x E$ , and

$f_1 = \epsilon(P_{M}^{1}-I) \partial_t E+(I-P_{N}^{1}) \partial_x H , \; \; \; \; f_2 = \mu(P_{M}^{1}-I) \partial_t H.$

Similar to the proof of Theorem 2.1, we obtain the following error estimate.

Theorem 2.2. Let $E_a$ and $H_a$ be the projections (2.19) of $E$ and $H$ (2.18), respectively. Let $E_{zL}$ and $H_{yL}$ be the solutions to (2.10), respectively. Assuming that $\sigma\geq1$ , $r\geq2$ , $E, H\in C([0, T]; H^r(I_x))\cap L^2(I_x; H^\sigma(I_t))$ , and then there exists a positive constant $C$ such that

$\begin{array}{l} \left\|\sqrt{\epsilon}\left(E_{z L}-E_{a}\right)\right\|_{\Omega}^{2}+\left\|\sqrt{\mu}\left(H_{y L}-H_{a}\right)\right\|_{\Omega}^{2}+T\left(\left\|\sqrt{\epsilon}\left(E_{z L}-E_{a}\right)(T)\right\|_{I_{x}}^{2}+\left\|\sqrt{\mu}\left(H_{y L}-H_{a}\right)(T)\right\|_{I_{x}}^{2}\right) \\ \leq C T^{2}\left[M^{2(1-\sigma)}\left(\left\|\partial_{t}^{\sigma} E\right\|_{\Omega}^{2}+\left\|\partial_{t}^{\sigma} H\right\|_{\Omega}^{2}\right)+N^{2(1-r)}\left\|\partial_{x}^{r} H\right\|_{\Omega}^{2}\right] \end{array}$

(2.22)

Proof. By (2.12) and (2.21), we have

$\begin{equation} \|\sqrt{ \epsilon}e_{z}\|_{\Omega}^2+\|\sqrt{\mu}e_{ y }\|_{\Omega}^2 +T({\|\sqrt{ \epsilon}e_{z}(T)\|_{I_x}^2+ \|\sqrt{\mu}e_{y}(T)\|_{I_x}^2}) \leq CT^2({\| f_1 \|_{\Omega}^2+\|\ f_2 \|_{\Omega}^2}). \end{equation}$

(2.23)

According to Lemma 2.1 and 2.2, it follows that

$\begin{equation} \begin{aligned} \|f_1\|^2_{\Omega}\leq C({M^{2(1-\sigma)}\| \partial_t^{\sigma}E\|^2_{\Omega}+N^{2(1-r)}\| \partial_x^{r}H\|^2_{\Omega}}), \end{aligned} \end{equation}$

(2.24)

$\begin{equation} \begin{aligned} \|f_2\|^2_{\Omega}\leq C M^{2(1-\sigma)}\| \partial_t^{\sigma}H\|^2_{\Omega}. \end{aligned} \end{equation}$

(2.25)

Substituting (2.24)-(2.25) into (2.23), the error estimate (2.22) is obtained.

Next, the error estimate between the solutions to (2.10) and (1.1) is considered.

Theorem 2.3. Let $E_z$ , $H_y$ , $E_{zL}$ , and $H_{yL}$ be the solutions to (1.1) and (2.10), respectively. Assume that $\sigma\geq1$ , $r\geq2$ , $E, H\in C([0, T]; H^r(I_x))\cap L^2(I_x; H^\sigma(I_t))$ , and then there exists a positive constant $C$ such that

$\begin{array}{l} \left\|\sqrt{\epsilon}\left(E_{z L}-E_{z}\right)\right\|_{\Omega}^{2}+\left\|\sqrt{\mu}\left(H_{y L}-H_{y}\right)\right\|_{\Omega}^{2}+T\left(\left\|\sqrt{\epsilon}\left(E_{z L}-E_{z}\right)(T)\right\|_{I_{x}}^{2}+\left\|\sqrt{\mu}\left(H_{y L}-H_{y}\right)(T)\right\|_{I_{x}}^{2}\right) \\ \leq C T^{2}\left[M^{2(1-\sigma)}\left(\left\|\partial_{t}^{\sigma} E\right\|_{\Omega}^{2}+\left\|\partial_{t}^{\sigma} H\right\|_{\Omega}^{2}\right)+N^{2(1-r)}\left\|\partial_{x}^{r} H\right\|_{\Omega}^{2}\right] \\ \quad+C T N^{-2 r}\left(\left\|E_{z}\right\|_{L^{\infty}\left(0, T ; H^{r}\left(I_{x}\right)\right)}^{2}+\left\|H_{y}\right\|_{L^{\infty}\left(0, T ; H^{r}\left(I_{x}\right)\right)}^{2}\right) \end{array}$

(2.26)

Proof. Firstly, the error between the solutions to (2.18) and (1.1) is estimated. Let $e_z = E-E_{z}$ and $e_y = H-H_{y}$ . By (1.1) and (2.18), we get the following error equation

$\begin{equation} \left\{\begin{array}{ll} \epsilon \partial_t e_z = \partial_x e_y, &(x, t)\in \Omega, \\ \mu \partial_t e_y = \partial_x e_z, &(x, t)\in \Omega. \end{array}\right. \end{equation}$

(2.27)

Then, we consider the inner product on $I_x$

$\begin{equation} \left\{\begin{array}{ll} ( \epsilon \partial_t e_z, e_z)_{I_x} = -(e_y , \partial_x e_z)_{I_x}, \\ (\mu \partial_t e_y, e_y)_{I_x} = ( \partial_x e_z, e_y)_{I_x}, \end{array}\right. \end{equation}$

(2.28)

which leads to

$\begin{equation} \| \sqrt{ \epsilon} e_{ z } (t)\|_{I_x}^2 + \| \sqrt{\mu} e_{ y } (t)\|_{I_x}^2 = \| \sqrt{ \epsilon} e_{ z } (0)\|_{I_x}^2 + \| \sqrt{\mu} e_{ y } (0)\|_{I_x}^2, \; \; \; \forall t > 0. \end{equation}$

(2.29)

Next, integrating over $t$

$\begin{equation} \begin{aligned} \| \sqrt{ \epsilon }e_z \|^2_{\Omega} + \|\sqrt{\mu}e_y \|^2_{\Omega} = T(\| \sqrt{ \epsilon} e_{ z } (0)\|^2_{I_x} + \| \sqrt{\mu} e_{ y } (0)\|^2_{I_x}), \end{aligned} \end{equation}$

(2.30)

and taking $t = T$ , we have

$\begin{equation} \begin{aligned} \|\sqrt{ \epsilon }e_z(T)\|^2_{I_x}+\|\sqrt{\mu }e_y(T)\|^2_{I_x} = \| \sqrt{ \epsilon} e_{ z } (0)\|^2_{I_x} + \| \sqrt{\mu} e_{ y } (0)\|^2_{I_x}. \end{aligned} \end{equation}$

(2.31)

According to Lemma 2.1, it follows that

$\|e_z(0)\|_{I_x} = \|E_{z0}-I_N^L E_{z0}\|_{I_x}\leq CN^{-r}|E_{z0}|_{r, I_x}, \\ \|e_y(0)\|_{I_x} = \|H_{y0}-P_{N-1}I_N^L H_{y0}\|_{I_x}\\ \leq\|H_{y0}-P_{N-1} H_{y0} \|_{I_x}+\|P_{N-1}(H_{y0}-I_N^L H_{y0})\|_{I_x}\leq CN^{-r}|H_{y0}|_{r, I_x}.$

(2.32)

Substituting (2.32) into (2.30)-(2.31), we obtain

$\begin{array}{l} \left\|\sqrt{\epsilon}\left(E-E_{z}\right)\right\|_{\Omega}^{2}+\left\|\sqrt{\mu}\left(H-H_{y}\right)\right\|_{\Omega}^{2}+T\left(\left\|\sqrt{\epsilon}\left(E-E_{z}\right)(T)\right\|_{I_{x}}^{2}+\left\|\sqrt{\mu}\left(H-H_{y}\right)(T)\right\|_{I_{x}}^{2}\right) \\ \leq C T N^{-2 r}\left(\left|E_{z}\right|_{r, I_{x}}^{2}+\left|H_{y 0}\right|_{r, I_{x}}^{2}\right) . \end{array}$

(2.33)

On the other hand, by Lemmas 2.1-2.2, we have

$\begin{array}{l} \left\|\sqrt{\epsilon}\left(E_{a}-E\right)\right\|_{\Omega}^{2}+\left\|\sqrt{\mu}\left(H_{a}-H\right)\right\|_{\Omega}^{2}+T\left(\left\|\sqrt{\epsilon}\left(E_{a}-E\right)(T)\right\|_{I_{x}}^{2}+\left\|\sqrt{\mu}\left(H_{a}-H\right)(T)\right\|_{I_{x}}^{2}\right) \\ \leq C T N^{-2 r}\left(\left.\left|E_{z} 0_{r, I_{x}}^{2}+\right| H_{y 0}\right|_{r, I_{x}} ^{2}\right) . \end{array}$

(2.34)

From (2.22) and (2.33)-(2.34), the error estimate (2.26) is obtained.

3. Time multi-interval Legendre-tau spectral method

In this section, a time multi-interval Legendre-tau spectral scheme is developed and its error estimate is obtained.

3.1. Preliminaries

Let $K$ be a positive integer and a partition of the computational interval $I_t$ is given as

$\begin{equation} I_t = \bigcup\limits^K_{k = 1}I_k, \; \; I_k = (a_{k-1}, a_{k}), \; \; \tau_k = a_k-a_{k-1}, \; \; 1\leq k\leq K, \end{equation}$

(3.1)

where

$0 = a_0 < a_1 < \cdot\cdot\cdot < a_k < \cdot\cdot\cdot < a_{K} = T.$

Let ${\cal M} = (M_1, \cdots, M_K)$ and $L = (N, {\cal M})$ . We define the space of approximate functions in time as

$\begin{equation} X_{\cal M} = W_{\cal M}\cap H^{1}(I_t), \; \; \; \; W_{\cal M} = \{v:v|_{I_k}\in {\mathbb P}_{M_k}(I_k), \; 1\le k\le K\}, \end{equation}$

(3.2)

where ${\mathbb P}_{M_k}(I_k)$ denotes the space of polynomials of degree at most $M_k$ on $I_k$ . We define the space of the test functions in time as

$\begin{equation} W_{{\cal M}-1} = \{v:v|_{I_k}\in {\mathbb P}_{M_k-1}(I_k), \; 1\le k\le K\}, \end{equation}$

(3.3)

where ${\cal M}-1 = (M_1-1, \cdots, M_K-1)$ .

Let $\hat{I} = (-1, 1)$ be a reference interval, $\hat{t}_j^{k}$ and $\hat{\omega}_{j}^{k}(0\leq j\leq M_k)$ be the LGL points and the corresponding weights on $\hat{I}$ . We denote by $\{\hat{t}^{k}_j\}$ and $\{\hat{\omega}_{j}^{k}\}$ be the LGL points and the corresponding weights on $I_k$ . Next, we define

$I^{k}_{\cal M} = \{t_{j}^{k}: t_{j}^{k} = \frac{\tau_k\hat{t}_{j}^{k}+a_{k-1}+a_k}{2}, \; 0\leq j\leq M_{k}, 1\leq k\leq K\},$

where $\tau_k = a_k-a_{k-1}$ .

Letting $v^k\equiv v|_{I_k}$ , for any $u, v\in C(\bar{I})$ and $\omega_j^k = \frac{1}{2}\tau_k\hat{\omega}_{j}^{k}$ , we define

$(u, v)_{{\cal M}, I_k} = \sum\limits^{M_k}_{j = 0}u^k(t_j^k)v^k(t_j^k)\omega_j^k, \; \; \; \; (u, v)_{\cal M} = \sum\limits^{K}_{k = 1}(u, v)_{{\cal M}, I_k}.$

Similarly, we denote $\hat{t}_j^{k, C}$ and $\hat{\omega}_j^{k, C}$ be the CGL points and the corresponding weights on $\hat{I}$ . Let $\{\hat{t}^{k, C}_j\}$ and $\{\hat{\omega}_{j}^{k, C}\}$ be the CGL points and the corresponding weights on $I_k$ .

We define LGL interpolation operator $I_{\cal M}^L:C(\bar{I})\rightarrow W_{\cal M}$ by

$I_{\cal M}^L u(t_j^{k}) = u(t_j^{k}), \; \; 0\leq M_k, \; \; 1\leq k\leq K.$

Similarly, for the CGL interpolation operator $I_{\cal M}^L:C(\bar{I})\rightarrow W_{\cal M}$ , which satisfies

$I_{\cal M}^C u(t_j^{k, C}) = u(t_j^{k, C}), \; \; 0\leq M_k, \; \; 1\leq k\leq K.$

Define the following relation

$v(t) = \hat v(\hat t), \; \; \; \; t = \frac{1}{2}(\tau_k\hat t+a_{k-1}+a_{k}), \; \; \; \; a_{k-1}\leq t\leq a_{k}.$

Let $\hat P_{M_k -1}: L^{2}(\hat I) \rightarrow \mathbb P_{M_k -1}$ the $L^2$ -Legendre projection operator by $P_{{\cal M}-1}: L^{2}(I_t) \rightarrow W_{{\cal M} -1}$ such that

$(P_{{\cal M}-1} v)|_{I_k}(t) = \hat P_{M_k -1} \widehat{(v|_{I_k})} (\hat t).$

Let $\hat P_{1, M_k}: H^{1}(\hat I) \rightarrow \mathbb P_{M_k}$ be the Legendre projection operator, which satisfies

$\hat P_{1, M_k}\hat v (\hat t) = \hat v(-1) + \int_{-1}^{\hat t} \hat {P}_{M_k -1} \partial_{\hat t} \hat v(s){\rm d}s,$

and $P_{1, {\cal M}}$ be generated by $P_{1, M_k}:H^1(I_k)\to {\mathbb P}_{M_k}(I_k)$ such that

$\begin{equation} (P_{1, {\cal M}} v)|_{I_k}(t)\equiv P_{1, M_k}v|_{I_k}(t) = \hat P_{1, M_k} \widehat{(v|_{I_k})}(\hat t). \end{equation}$

(3.4)

The following approximation results can be found in ^[10].

Lemma 3.1. If $v\in H^{\sigma}(I_t)$ and $\sigma\geq1$ , then

$|P^1_{\cal M} v-v|_{l, I_t}\leq C ({\sum\limits^K_{k = 1}(\tau_k^{-1} M_k)^{2(l-\sigma)}|v|^2_{\sigma, I_k}})^\frac{1}{2}, \; \; \; l = 0, 1,$

where $C$ is a generic positive constant independent of $\tau_k$ , $M_k$ .

The time multi-interval Legendre-tau spectral method for the problem (1.1) is : Find $E_{zN}^{k} \in V_N^{0}\otimes W_M$ and $H_{yN}^{k}\in {V}_{N-1}\otimes W_M$ such that

$\begin{equation} \left\{\begin{array}{ll} ( \epsilon \partial_t E_{zL}^{K}, v)_{\Omega} + (H_{yL}^{K}, \partial_x v)_{\Omega} = 0, \; \; &\forall\; v \in V_N^{0}\otimes W_{{\cal M}-1}, \; \\ (\mu \partial_t H_{yL}^{K}, w)_{\Omega} - ( \partial_x E_{zL}^{K}, w)_{\Omega} = 0 , \; \; &\forall\; w\in V_{N-1}\otimes W_{{\cal M}-1}, \\ E_{zL}^{K}(x, 0) = I_N^L E_{z0}(x), \; \; \; \; H_{yL}^{K}(x, 0) = P_{N-1} I_N^L H_{y0}(x), \; \; &\forall x\in I_x. \end{array}\right. \end{equation}$

(3.5)

We set

${v^{k}(x, t) = v(x, t+a_{k-1}), \; \; \; t\in \hat{I}_k = (0, \tau_k), \; \; \; 1\leq k\leq K.}$

Let $\Omega_k = I_x\times \hat{I}_k$ , and (3.5) can be written as: For $1\leq k \leq K$ , find $E_{zL}^{k} \in V_N^{0}\otimes {\mathbb P}_{M_k}(\hat{I}_k)$ and $H_{yL}^{k}\in {V}_{N-1}\otimes {\mathbb P}_{M_k}(\hat{I}_k)$ such that

$\begin{equation} \left\{\begin{array}{ll} ( \epsilon \partial_t E_{zL}^{k}, v^{k})_{\Omega_k} + (H_{yL}^{k}, \partial_x v^{k})_{\Omega_k} = 0, \; \; &\forall\; v^{k} \in V_N^{0}\otimes{\mathbb P}_{M_k-1}(\hat{I}_k), \\ (\mu \partial_t H_{yL}^{k}, w^{k})_{\Omega_k} - ( \partial_x E_{zL}^{k}, w^{k})_{\Omega_k} = 0, \; \; &\forall\; w^{k}\in V_{N-1}\otimes {\mathbb P}_{M_k-1}(\hat{I}_k), \\ E_{zL}^{k}(x, 0) = E_{zL}^{k-1}(x, \tau_{k-1}), \; \; \; H_{yL}^{k}(x, 0) = H_{yN}^{k-1}(x, \tau_{k-1}), \; \; &x\in I_x , \end{array}\right. \end{equation}$

(3.6)

where $E_{zL}^{0}(x, \tau_0) = I_N^L E_{z0}(x), \; \; H_{yL}^{0}(x, \tau_0) = P_{N-1} I_N^L H_{y0}(x)$ when $k = 1$ .

3.2. Error estimate

In the following, we present the error estimate. In order to deal with the error of the initial value, we consider the following auxiliary problems on $\Omega_k, \; \; 1\le k\le K$ ,

$\begin{equation} \left\{\begin{array}{ll} \epsilon \partial_t E^k = \partial_x H^k, &(x, t)\in \Omega_k, \\ \mu \partial_t H^k = \partial_x E^k, &(x, t)\in \Omega_k, \\ E^{k}(x, 0) = E_{zL}^{k}(x, 0), \; \; H^{k}(x, 0) = H_{yL}^{k}(x, 0), \; \; &x\in I_x.\\ \end{array}\right. \end{equation}$

(3.7)

Similar to the process of the single-interval, we define $E_{a}^{k} = P_{N}^{1}P^1_{M_k}E^{k}$ , $H_{a}^{k} = P_{N-1}P^1_{M_k}H^{k}$ , and denote

$f_1^k = \epsilon (P^1_{M_k}-I) \partial_t E^k + (I-P_{N}^{1}) \partial_x H^k, \; \; \; \; f_2^k = \mu(P^1_{M_k}-I) \partial_t H^k.$

Let $e_z^k = E_{ zL }^k-E^{k}_{a}$ and $e_y^k = H_{ yL }^k-H^{k}_{a}$ , the following error equation is obtained

$\begin{equation} \left\{\begin{array}{ll} ( \epsilon \partial_t e_z^{k} , v^{k})_{\Omega_k} + (e_y^{k} , \partial_x v^{k})_{\Omega_k} = ( f_1^{k}, v^{k})_{\Omega_k}, \; \; &\forall\; v^{k} \in V_N^{0}\otimes {\mathbb P}_{M_k-1}, \; \\ (\mu \partial_t e_y^{k} , w)_{\Omega_k} - ( \partial_x e_z^{k} , w^{k})_{\Omega_k} = ( f_2^{k}, w^{k})_{\Omega_k}, \; \; &\forall\; w^{k}\in V_{N-1}\otimes {\mathbb P}_{M_k-1}, \\ e_{z}^{k}(x, 0) = 0, \; \; e_{y}^{k}(x, 0) = 0, \; \; &\forall x\in I_x. \end{array}\right. \end{equation}$

(3.8)

For each subinterval in the multi-interval, using Theorem 2.2 and Lemma 3.1, the error estimate between the solution to (3.6) and the projection of the solution to (3.7) is obtained

$\begin{array}{l} \left\|\sqrt{\epsilon}\left(E_{z L}^{k}-E_{a}^{k}\right)\right\|_{\Omega_{k}}^{2}+\left\|\sqrt{\mu}\left(H_{y L}^{k}-H_{a}^{k}\right)\right\|_{\Omega_{k}}^{2}+\tau_{k}\left(\left\|\sqrt{\epsilon}\left(E_{z L}^{k}-E_{a}^{k}\right)\left(\tau_{k}\right)\right\|_{I_{x}}^{2}+\left\|\sqrt{\mu}\left(H_{y L}^{k}-H_{a}^{k}\right)\left(\tau_{k}\right)\right\|_{I_{x}}^{2}\right) \\ \leq C \tau_{k}^{2}\left[\left(\tau_{k}^{-1} M_{k}\right)^{2(1-\sigma)}\left(\left\|\partial_{t}^{\sigma} E^{k}\right\|_{\Omega_{k}}^{2}+\left\|\partial_{t}^{\sigma} H^{k}\right\|_{\Omega_{k}}^{2}\right)+N^{2(1-r)}\left\|\partial_{x}^{r} H^{k}\right\|_{\Omega_{k}}^{2}\right] . \end{array}$

(3.9)

Let $\mathbf{e}_z^k = E^k-E^k_{z}$ and $\mathbf{e}_y^k = H^k-H^k_{y}$ , the results are similar to (2.30)-(2.31) for the multi-interval case,

$\left\|\sqrt{\epsilon} \mathbf{e}_{z}^{k}\right\|_{\Omega_{k}}^{2}+\left\|\sqrt{\mu} \mathbf{e}_{y}^{k}\right\|_{\Omega_{k}}^{2} = \tau_{k}\left(\left\|\sqrt{\epsilon} \mathbf{e}_{z}^{k}(0)\right\|_{I_{x}}^{2}+\left\|\sqrt{\mu} \mathbf{e}_{y}^{k}(0)\right\|_{I_{x}}^{2}\right),$

(3.10)

$\left\|\sqrt{\epsilon} \mathbf{e}_{z}^{k}\left(\tau_{k}\right)\right\|_{I_{x}}^{2}+\left\|\sqrt{\mu} \mathbf{e}_{y}^{k}\left(\tau_{k}\right)\right\|_{I_{x}}^{2} = \left\|\sqrt{\epsilon} \mathbf{e}_{z}^{k}(0)\right\|_{I_{x}}^{2}+\left\|\sqrt{\mu} \mathbf{e}_{y}^{k}(0)\right\|_{I_{x}}^{2} .$

(3.11)

Using the triangle inequality, we get

$\begin{array}{l} \sqrt{\left\|\sqrt{\epsilon} \mathbf{e}_{z}^{k}(0)\right\|_{x_{x}}^{2}+\left\|\sqrt{\mu} \mathbf{e}_{y}^{k}(0)\right\|_{I_{x}}^{2}} = \sqrt{\left\|\sqrt{\epsilon}\left(E_{z L}^{k-1}-E_{z}^{k-1}\right)\left(\tau_{k-1}\right)\right\|_{I_{x}}^{2}+\left\|\sqrt{\mu}\left(H_{y L}^{k-1}-H_{y}^{k-1}\right)\left(\tau_{k-1}\right)\right\|_{I_{x}}^{2}} \\ \leq \sqrt{\left\|\sqrt{\epsilon}\left(E_{z L}^{k-1}-E^{k-1}\right)\left(\tau_{k-1}\right)\right\|_{L_{x}}^{2}+\left\|\sqrt{\mu}\left(H_{y L}^{k-1}-H^{k-1}\right)\left(\tau_{k-1}\right)\right\|_{l_{x}}^{2}} \\ \quad+\sqrt{\left\|\sqrt{\epsilon} \mathbf{e}_{z}^{k-1}\left(\tau_{k-1}\right)\right\|_{I_{x}}^{2}+\left\|\sqrt{\mu} \mathbf{e}_{y}^{k-1}\left(\tau_{k-1}\right)\right\|_{I_{x}}^{2}} \\ = \sqrt{\left\|\sqrt{\epsilon}\left(E_{z L}^{k-1}-E^{k-1}\right)\left(\tau_{k-1}\right)\right\|_{I_{x}}^{2}+\left\|\sqrt{\mu}\left(H_{y L}^{k-1}-H^{k-1}\right)\left(\tau_{k-1}\right)\right\|_{I_{x}}^{2}} \\ +\sqrt{\left\|\sqrt{\epsilon} \mathbf{e}_{z}^{k-1}(0)\right\|_{I_{x}}^{2}+\left\|\sqrt{\mu} \mathbf{e}_{y}^{k-1}(0)\right\|_{I_{x}}^{2}} \end{array}$

which leads to

$\begin{array}{l} \sqrt{\left\|\sqrt{\epsilon} \mathbf{e}_{z}^{k}(0)\right\|_{I_{x}}^{2}+\left\|\sqrt{\mu} \mathbf{e}_{y}^{k}(0)\right\|_{I_{x}}^{2}} \leq \sum\limits_{m = 1}^{k-1} \sqrt{\left\|\sqrt{\epsilon}\left(E_{z L}^{m}-E^{m}\right)\left(\tau_{m}\right)\right\|_{I_{x}}^{2}+\left\|\sqrt{\mu}\left(H_{y L}^{m}-H^{m}\right)\left(\tau_{m}\right)\right\|_{I_{x}}^{2}} \\ +\sqrt{\left\|\sqrt{\epsilon} \mathbf{e}_{z}^{1}(0)\right\|_{I_{x}}^{2}+\left\|\sqrt{\mu} \mathbf{e}_{y}^{1}(0)\right\|_{I_{x}}^{2}}, \quad \forall k \geq 2 \end{array}$

(3.12)

By the Cauchy-Schwarz inequality, $\sum_{m = 1}^{k-1}\tau_m = a_{k-1}$ , and (3.9), we derive

(3.13)

According to (2.7) and Lemma 2.1, it follows that

$\begin{array}{l} \left(\sum\limits_{m = 1}^{k-1} \sqrt{\left\|\sqrt{\epsilon}\left(E_{a}^{m}-E^{m}\right)\left(\tau_{m}\right)\right\|_{I_{x}}^{2}+\left\|\sqrt{\mu}\left(H_{a}^{m}-H^{m}\right)\left(\tau_{m}\right)\right\|_{I_{x}}^{2}}\right)^{2} \\ = \left(\sum\limits_{m = 1}^{k-1} \sqrt{\left\|\sqrt{\epsilon}\left(P_{N}^{1}-I\right) E^{m}\left(\tau_{m}\right)\right\|_{I_{x}}^{2}+\left\|\sqrt{\mu}\left(P_{N-1}-I\right) H^{m}\left(\tau_{m}\right)\right\|_{I_{x}}^{2}}\right)^{2} \\ \leq C a_{k-1} \sum\limits_{m = 1}^{k-1} \tau_{m}^{-1} N^{-2 r}\left(\left|E^{m}\left(\tau_{m}\right)\right|_{r, I_{x}}^{2}+\left|H^{m}\left(\tau_{m}\right)\right|_{r, I_{x}}^{2}\right) \end{array}$

(3.14)

As (2.32), we have

$\left\|\sqrt{\epsilon} \mathbf{e}_{z}^{1}(0)\right\|_{I_{x}}^{2}+\left\|\sqrt{\mu} \mathbf{e}_{y}^{1}(0)\right\|_{I_{x}}^{2} \leq C N^{-2 r}\left(\left|E_{z 0}\right|_{r, I_{x}}^{2}+\left|H_{y 0}\right|_{r, I_{x}}^{2}\right) .$

Substituting the above estimation results into (3.10)-(3.11), we obtain

$\begin{aligned} \| & \sqrt{\epsilon}\left(E^{k}-E_{z}^{k}\right)\left\|_{\Omega_{k}}^{2}+\right\| \sqrt{\mu}\left(H^{k}-H_{y}^{k}\right) \|_{\Omega_{k}}^{2}+\tau_{k}\left(\left\|\sqrt{\epsilon}\left(E^{k}-E_{z}^{k}\right)\left(\tau_{k}\right)\right\|_{I_{x}}^{2}+\left\|\sqrt{\mu}\left(H^{k}-H_{y}^{k}\right)\left(\tau_{k}\right)\right\|_{I_{x}}^{2}\right) \\ \leq & C a_{k-1} \tau_{k} \sum\limits_{m = 1}^{k-1}\left[\left(\tau_{m}^{-1} M_{m}\right)^{2(1-\sigma)}\left(\left\|\partial_{t}^{\sigma} E^{m}\right\|_{\Omega_{m}}^{2}+\left\|\partial_{t}^{\sigma} H^{m}\right\|_{\Omega_{m}}^{2}\right)+N^{2(1-r)}\left\|\partial_{x}^{r} H^{m}\right\|_{\Omega_{m}}^{2}\right] \\ &+C a_{k-1} \tau_{k} \sum\limits_{m = 1}^{k-1} \tau_{m}^{-1} N^{-2 r}\left(\left|E^{m}\left(\tau_{m}\right)\right|_{r, I_{x}}^{2}+\left|H^{m}\left(\tau_{m}\right)\right|_{r, I_{x}}^{2}\right)+C \tau_{k} N^{-2 r}\left(\left|E_{z 0}\right|_{r, I_{x}}^{2}+\left|H_{y 0}\right|_{r, I_{x}}^{2}\right) \end{aligned}$

(3.15)

By (2.7), Lemma 2.1 and 2, we get

$\begin{aligned} \| & \sqrt{\epsilon}\left(E_{a}^{k}-E^{k}\right)\left\|_{\Omega_{k}}^{2}+\right\| \sqrt{\mu}\left(H_{a}^{k}-H^{k}\right) \|_{\Omega_{k}}^{2}+\tau_{k}\left(\left\|\sqrt{\epsilon}\left(E_{a}^{k}-E^{k}\right)\left(\tau_{k}\right)\right\|_{I_{x}}^{2}+\left\|\sqrt{\mu}\left(H_{a}^{k}-H^{k}\right)\left(\tau_{k}\right)\right\|_{I_{x}}^{2}\right) \\ \leq & C\left[\left(\tau_{k}^{-1} M_{k}\right)^{-2 \sigma}\left(\left\|\partial_{t}^{\sigma} E^{k}\right\|_{\Omega_{k}}^{2}+\left\|\partial_{t}^{\sigma} H^{k}\right\|_{\Omega_{k}}^{2}\right)+N^{-2 r}\left(\left\|\partial_{x}^{r} E^{k}\right\|_{\Omega_{k}}^{2}+\partial_{x}^{r} H^{k} \|_{\Omega_{k}}^{2}\right)\right] \\ &+C \tau_{k} N^{-2 r}\left(\left|E^{k}\left(\tau_{k}\right)\right|_{r, I_{x}}^{2}+\left|H^{k}\left(\tau_{k}\right)\right|_{r, I_{x}}^{2}\right) . \end{aligned}$

(3.16)

If $\tau_k\equiv \tau, M_k\equiv M$ for simplicity, and combining (3.9) and (3.15)-(3.16), we get the following error estimate.

Theorem 3.1. Let $E_z$ and $H_y$ be solutions to (1.1), respectively. Let $E_{zL}^K$ and $H_{yL}^K$ be solutions to (3.5), respectively. Let $E^k$ and $H^k$ be solutions to (3.7), respectively. Assuming that $\sigma\geq1$ , $r\geq2$ , $E_z, H_y\in C([0, T]; H^r(I_x))\cap L^2(I_x; H^\sigma(I_t))$ , $E^k, H^k\in C([0, \tau_k]; H^r(I_x))\cap L^2(I_x; H^\sigma(\hat{I}^k))$ , and then there exists a positive constant $C$ such that

$\begin{array}{l} \left\|\sqrt{\epsilon}\left(E_{z L}^{K}-E_{z}\right)\right\|_{\Omega}^{2}+\left\|\sqrt{\mu}\left(H_{y L}^{K}-H_{y}\right)\right\|_{\Omega}^{2}+\sum\limits_{k = 1}^{K} \tau_{k}\left(\left\|\sqrt{\epsilon}\left(E_{z L}^{k}-E_{z}^{k}\right)\left(\tau_{k}\right)\right\|_{I_{x}}^{2}+\left\|\sqrt{\mu}\left(H_{y L}^{k}-H_{y}^{k}\right)\left(\tau_{k}\right)\right\|_{I_{x}}^{2}\right) \\ \leq C\left[\left(\tau^{-1} M\right)^{2(1-\sigma)}+N^{2(1-r)}+\tau^{-2} N^{-2 r}\right] \end{array}$

(3.17)

4. Numerical examples

In this section, some numerical results are presented. We define

$\begin{aligned} &\mathbb{E}_{\infty}(E_z) = \max\limits_{0\le j\le N}|E_{zL}(x_j^C, t)-E_{z}(x_j^C, t)|, \\ &\mathbb{E}_{\infty}(H_y) = \max\limits_{0\le j\le N}|H_{yL}(x_j^C, t)-H_{y}(x_j^C, t)|. \end{aligned}$

Example 4.1. The LT-ST spectral method for the 1-D Maxwell equation

Consider the problem (1.1) with ${I}_x = (0, 1)$ , ${I}_t = (0, 1)$ , ${\Omega} = {I}_x\times {I}_t$ , $\epsilon = 1$ , and $\mu = 1$ . The solution is as

$\begin{equation} \rm \left\{\begin{array}{ll}\rm E_{z}(x, t) = cos(3\pi t)sin(3\pi x), &(x, t)\in {\Omega}, \\\rm H_{y}(x, t) = sin(3\pi t)cos(3\pi x), &(x, t)\in {\Omega}. \end{array}\right. \end{equation}$

(4.1)

In , the values of $\rm log_{10}$ $\mathbb{E}_{\infty}(E_z)$ and $\rm log_{10}$ $\mathbb{E}_{\infty}(H_y)$ is obtained when $t = 1$ . It can be seen from Figure 1 that the LT-ST method has spectral accuracy both in the time and space, which is consistent with the results of theoretical analysis.

Figure 1.

$L^{\infty}$ -error at

$t = 1$ of the LT-ST method (2.10).

DownLoad: Full-Size Img PowerPoint

To check the high accuracy, we compare the numerical errors of our scheme (2.10) with the Legendre-tau spectral method in space and the leapfrog-Crank-Nicolson method in time (LT-LFCN) ^[19]. For convenience of notation, let $(N, \tau)$ be the degree of the polynomial in the space approximation and the time step for the LT-LFCN method.

The $L^{\infty}$ -error of the LT-LFCN scheme and our method (2.10) at $t = 1$ are listed in Table 1. It can be seen from Table 1 that on the same PC machine, the proposed method takes shorter time than the LT-LFCN method.

Table 1.

$L^{\infty}$ -error of the LT-LFCN method and the LT-ST method (2.10).

LT-LFCN				LT-ST
$(N, \tau)$	$\mathbb{E}_{\infty}(E_z)$	$\mathbb{E}_{\infty}(H_y)$	time	$(N, M)$	$\mathbb{E}_{\infty}(E_z)$	$\mathbb{E}_{\infty}(H_y)$	time
(8, 1e-02)	5.14e-04	6.85e-03	0.19s	(8, 8)	4.04e-03	1.99e-02	0.08s
(12, 1e-03)	4.35e-06	6.97e-06	0.49s	(12, 12)	9.38e-06	3.99e-05	0.10s
(16, 1e-04)	1.29e-09	6.96e-07	4.94s	(16, 16)	6.57e-09	3.34e-08	0.10s
(20, 1e-05)	1.03e-13	6.98e-09	52.81s	(20, 20)	1.38e-12	7.86e-12	0.10s
(24, 1e-06)	2.83e-14	6.97e-11	598.48s	(24, 24)	1.69e-15	2.99e-15	0.11s

| Show Table

DownLoad: CSV

Example 4.2. The time multi-interval Legendre-tau spectral method for the 1-D Maxwell equation

Further, the method (3.6) is used to solve Example 1 of $N = M_k = 24$ and $0\leq t\leq 5$ , and the numerical results are shown in Table 2.

Table 2.

$L^{\infty}$ -error of the time five-interval Legendre-tau spectral method (3.6) (

$N = M_k = 24$ ).

$t$	$\mathbb{E}_{\infty}(E_z)$	$\mathbb{E}_{\infty}(H_y)$	time
$1.00$	1.69e-15	2.99e-15	0.11s
$2.00$	3.10e-15	3.44e-15	0.12s
$3.00$	3.38e-15	3.44e-15	0.13s
$4.00$	5.82e-15	7.10e-15	0.16s
$5.00$	9.49e-15	7.71e-15	0.19s

| Show Table

DownLoad: CSV

5. Application to the 1-D nonlinear Maxwell equation

In this section, the proposed method is applied to the numerical solution of the 1-D nonlinear Maxwell equation. The approximating of the nonlinear term is calculated by interpolation at the CGL point, and implemented with the help of Fast Legendre transformation.

5.1. Scheme

Now, we apply the LT-ST method to solve the 1-D nonlinear Maxwell equation as ^[26]

$\begin{equation} \left\{\begin{array}{ll} \epsilon \partial_t E_z+J(E_z)- \partial_x H_y = 0, &(x, t)\in \Omega, \\ \mu \partial_t H_y- \partial_x E_z = 0, &(x, t)\in \Omega, \\ E_z(-1, t) = E_z(1, t) = 0, &t\in I_t, \\ E_z(x, 0) = E_{z0}(x), \; H_y(x, 0) = H_{y0}(x), \; \; &x\in I_x, \end{array}\right. \end{equation}$

(5.1)

where the nonlinear function $J(E_z) = \sigma(|E_z|)E_z$ with $\sigma(s)$ is a real valued function representing the electric conductivity.

The problem (5.1) can be written in a weak form: Find $E_{z} \in H_0^{1}(I_x)\otimes H^{1}(I_t)$ and $H_{y} \in L^2(I_x)\otimes H^{1}(I_t)$ such that

$\begin{equation} \left\{\begin{array}{ll} ({ \epsilon} \partial_t E_{ z }, v)_{\Omega} +(J(E_z), v)_{\Omega} +(H_{ y }, \partial_x v)_{\Omega} = 0, \; \; &\forall\; v \in H_0^{1}(I_x)\otimes L^2(I_t), \; \\ ({\mu} \partial_t H_{y}, w)_{\Omega} - ( \partial_x E_{ z }, w)_{\Omega} = 0, \; \; &\forall\; w\in L^2(I_x)\otimes L^2(I_t), \\ E_z(x, 0) = E_{z0}(x), \; H_y(x, 0) = H_{y0}(x), \; \; &\forall x\in I_x. \end{array}\right. \end{equation}$

(5.2)

Combining the interpolation operator both in space and time, a 2-D interpolation is defined as $I_{(N, M)}^L$ . The LT-ST method to the problem $(5.1)$ is: Find $E_{zL} \in V_N^{0}\otimes V_{M}$ and $H_{yL} \in V_{N-1}\otimes V_{M}$ such that

$\begin{equation} \left\{\begin{array}{ll} ({ \epsilon} \partial_t E_{ zL }, v)_{\Omega}+(I_{(N, M)}^C J(E_{zL}), v)_{\Omega} +(H_{ yL }, \partial_x v)_{\Omega} = 0, \; \; &\forall\; v \in V_N^{0}\otimes V_{M-1}, \; \\ ({\mu} \partial_t H_{yL}, w)_{\Omega}-( \partial_x E_{ zL }, w)_{\Omega} = 0, \; \; &\forall\; w\in V_{N-1}\otimes V_{M-1}, \\ E_{zL}(x, 0) = I_N^L E_{z0}(x), \; \; H_{yL}(x, 0) = P_{N-1}I_N^L H_{y0}(x), \; \; &\forall x\in I_x, \end{array}\right. \end{equation}$

(5.3)

We briefly describe the implementation of scheme $(5.3)$ . For simplicity, taking $\Omega = [-1, 1]\times [-1, 1]$ . Let $L_k$ be the Legendre polynomial of degree $k$ , and the basis functions in space are

${\Phi(x) = ({\frac{1-x}{2}, \frac{1+x}{2}, \phi_2(x), ..., \phi_{N}(x)}), }$

${\Phi_{0}(x) = ({\phi_2(x), ..., \phi_{N}(x)}), \; \; \; \; L(x) = ({L_0(x), L_1(x), ..., L_{N-1}(x)}), }$

where $\phi_k(x) = L_k(x)-L_{k-2}(x)$ .

The basis functions in time are

${\Psi(t) = (1, 1+t, \phi_{2}(t), ..., \phi_{M}( t)), \; \; \; \; L(t) = ({L_0(t), L_1(t), ..., L_{M-1}(t)}), }$

where $\phi_k(t) = L_k(t)-L_{k-2}(t)$ .

The approximate solutions and the test functions are expressed as

$\begin{array}{ll} E_{z L}(x, t) = \Psi(t) \hat{E} \Phi^{T}(x), & H_{y L}(x, t) = \Psi(t) \hat{H} L^{T}(x) \\ v(x, t) = L(t) \hat{v} \Phi_{0}^{T}(x), & w(x, t) = L(t) \hat{w} L^{T}(x) \end{array}$

The interpolation polynomial of the nonlinear term can be expressed as $I_{(N, M)}^CJ(E_{zL}) = \Psi(t)\hat{J}\Phi^T(x)$ . The following algebraic equation is obtained from $(5.3)$

$\begin{equation} \left\{\begin{array}{ll} \epsilon ( \partial_t \Psi, L)_{I_t}\hat{E}(\Phi_{0}, \Phi)_{I_x}+( \Psi, L)_{I_t}\hat{J}(\Phi_{0}, \Phi)_{I_x} +(\Psi, L)_{I_t}\hat{H}( \partial_x \Phi_{0}, L)_{I_x} = 0, \\ \mu ( \partial_t \Psi, L)_{I_t}\hat{H}(L, L)_{I_x}-(\Psi, L)_{I_t}\hat{E}(L, \partial_x\Phi)_{I_x} = 0, \end{array}\right. \end{equation}$

(5.4)

where $\hat{E}$ and $\hat{H}$ are matrices composed of coefficients of approximate solutions $E_{zL}$ and $H_{yL}$ , respectively. For simplicity, (5.4) can be rewritten in matrix form as

$\begin{equation} \left\{\begin{array}{ll} \epsilon K^t\hat{E}M^x+M^t\hat{J}M^x+M^t\hat{H}K^x_0 = 0, \\ \mu K^t\hat{H}D-M^t\hat{E}K^{xT} = 0. \end{array}\right. \end{equation}$

(5.5)

A simple implicit-explicit iteration method is used to solve (5.5). In order to separate the initial conditions from the coefficient matrix, $\hat{E}$ , $\hat{H}$ , $M_t$ is divided into the following forms as

$\begin{equation} \label{{eq321}}{ \hat{E} = \begin{bmatrix} \hat{E}_i\\ \hat{E}_0 \end{bmatrix}, \qquad \hat{H} = \begin{bmatrix} \hat{H}_i\\ \hat{H}_0 \end{bmatrix}, \qquad M^t = \begin{bmatrix} M^t_i& M^t_0 \end{bmatrix}, } \end{equation}$

(5.6)

where $\hat{E_i}$ and $\hat{H_i}$ are the first rows of the coefficient matrix $\hat{E}$ and $\hat{H}$ respectively, corresponding to the initial value, $M^t_i$ is the first column of $M^t$ . By the properties of the basis function and the orthogonality of Legendre polynomials show that both $K^t$ and $D$ are diagonal matrices, and the elements on the diagonal of $K^t$ are 2 except that the first element is zero. Thus, (5.5) can be expressed as

$\begin{equation} 4 \epsilon\mu\hat{E}_0M^x+(M^t_0)^2\hat{E}_0 K^{xx} = -2\mu M^t\hat{J}M^x-2\mu M^t_i\hat{H_i}K^x_0-M^t_0M^t_i\hat{E_i}K^{xx}, \\ \end{equation}$

(5.7)

$\begin{equation} 2\mu\hat{H}_0 = M^t_0\hat{E}_0K^{xT}D^{-1}+M^t_i\hat{E_i}K^{xT}D^{-1}. \end{equation}$

(5.8)

Let

${G = -2\mu M^t_i\hat{H_i}K^x_0-M^t_0M^t_i\hat{E_i}K^{xx}, }$

In computations. We use the following simple explicit-implicit iteration scheme for (5.7),

$\begin{equation} 4 \epsilon\mu\hat{E}_0^{[k+1]}M^x+(M^t_0)^2\hat{E}_0^{[k+1]}K^{xx} = -2\mu M^t\hat{J}^{[k]}M^x+G, \; \; \; k = 0, 1, \cdots, \end{equation}$

(5.9)

when $k = 0$ , using the initial information of $E_{zL}$ in (5.3), and taking $E^{[0]}_{zL}(t)\equiv E_{zL}(0)$ as the initial guess of the iteration. The iterative scheme (5.9) is a linear equation of $\hat{E}_0^{[k+1]}$ , which can be solved by the method in ^[10].

Combining the interpolation operator in space and the multi-interval interpolation operator in time in Section 3, a 2-D interpolation is defined as $I_{(N, \cal{M})}^L$ . The time multi-interval Legendre-tau spectral method for (5.1) is: Find $E_{zN}^{k} \in V_N^{0}\otimes W_{\cal M}$ and $H_{yN}^{k}\in {V}_{N-1}\otimes W_{\cal M}$ such that

$\begin{equation} \left\{\begin{array}{ll} ( \epsilon \partial_t E_{zL}^{K}, v)_{\Omega}+(I_{(N, \cal{M})}^L J(E_{zL}^{K}), v)_{\Omega} + (H_{yL}^{K}, \partial_x v)_{\Omega} = 0, \; \; &\forall\; v \in V_N^{0}\otimes W_{{\cal M}-1}, \; \\ (\mu \partial_t H_{yL}^{K}, w)_{\Omega}-( \partial_x E_{zL}^{K}, w)_{\Omega} = 0, \; \; &\forall\; w\in V_{N-1}\otimes W_{{\cal M}-1}, \\ E_{zL}^{K}(x, 0) = I_N^L E_{z0}(x), \; \; \; \; H_{yL}^{K}(x, 0) = P_{N-1} I_N^L H_{y0}(x), \; \; &\forall x\in I_x, \end{array}\right. \end{equation}$

(5.10)

In computation, the interval is shifted to $\hat{I}_k = (0, \tau_k)$ . Let $\Omega_k = I_x\times \hat{I}_k$ , and then (5.10) can be written as: Find $E_{zL}^{k} \in V_N^{0}\otimes {\mathbb P}_{M_k}(\hat{I}_k)$ and $H_{yL}^{k}\in {V}_{N-1}\otimes {\mathbb P}_{M_k}(\hat{I}_k), \; 1\leq k \leq K$ , such that

$\begin{equation} \left\{\begin{array}{ll} ( \epsilon \partial_t E_{zL}^{k}, v^{k})_{\Omega_k} +(I_{(N, M_k)}^L J(E_{zL}^{k}), v^k)_{\Omega_k} +(H_{yL}^{k}, \partial_x v^{k})_{\Omega_k} = 0, \; \; &\forall\; v^{k} \in V_N^{0}\otimes{\mathbb P}_{M_k-1}(\hat{I}_k), \\ (\mu \partial_t H_{yL}^{k}, w^{k})_{\Omega_k}-( \partial_x E_{zL}^{k}, w^{k})_{\Omega_k} = 0, \; \; &\forall\; w^{k}\in V_{N-1}\otimes {\mathbb P}_{M_k-1}(\hat{I}_k), \\ E_{zL}^{k}(x, 0) = E_{zL}^{k-1}(x, \tau_{k-1}), \; \; \; H_{yL}^{k}(x, 0) = H_{yN}^{k-1}(x, \tau_{k-1}), \; \; &x\in I_x , \end{array}\right. \end{equation}$

(5.11)

where $E_{zL}^{0}(x, \tau_0) = I_N^L E_{z0}(x)$ and $H_{yL}^{0}(x, \tau_0) = P_{N-1} I_N^L H_{y0}(x)$ when $k = 1$ .

5.2. Numerical examples

Example 5.1. The LT-ST method for the 1-D nonlinear Maxwell equation

Consider the problem (5.1), and set the right-hand function of the first equation to $f(x, t)$ . According to ^[26], the nonlinear term is given as

${J(E_z) = (|E_z|^{2}-|E_z|^{4})E_z, }$

where ${I}_x = (0, 1)$ , ${I}_t = (0, 1)$ , ${\Omega} = {I}_x\times {I}_t$ , and $\epsilon = \mu = 1$ . The solution is

(5.12)

and the right-hand side of the first equation is

$\begin{equation} f(x, t) = \cos(3\pi t)^{3}\sin(3\pi x)^{3}-\cos(3\pi t)^{5}\sin(3\pi x)^{5}, \; \; \; \; (x, t)\in \Omega. \end{equation}$

(5.13)

The scheme (5.3) is used to solve Example 5.1, and the values of $\rm log_{10}$ $\mathbb{E}_{\infty}(E_z)$ and $\rm log_{10}$ $\mathbb{E}_{\infty}(H_y)$ are obtained when $t = 1$ . It can be seen from Figure 2 that the method has high accuracy both in time and space.

Figure 2.

$L^{\infty}$ -error at

$t = 1$ of the LT-ST method (5.3).

DownLoad: Full-Size Img PowerPoint

ItNum represents the number of iterations. Further, the method (5.11) is used to solve Example 5.1 in the case of $N = M_k = 24$ and $0\leq t\leq 5$ , the numerical results are shown in Table 3.

Table 3.

$L^{\infty}$ -error of the time five-interval Legendre-tau spectral method (5.11) (

$N = M_k = 24$ ).

$t$	$\mathbb{E}_{\infty}(E_z)$	$\mathbb{E}_{\infty}(H_y)$	time	ItNum
$1.00$	1.72e-15	2.83e-15	0.17s	10
$2.00$	3.72e-15	3.44e-15	0.32s	10
$3.00$	4.11e-15	4.10e-15	0.49s	10
$4.00$	6.30e-15	6.55e-15	0.65s	10
$5.00$	1.04e-14	8.93e-15	0.81s	10

| Show Table

DownLoad: CSV

Example 5.2. Comparison of the LT-ST method of 1-D nonlinear Maxwell equation and related computation results

Consider the same problem as in Example 5.1, but the nonlinear is given as ^[26]

${J(E_z) = |E_z|^{\frac{1}{2}}E_z.}$

Taking the same solution (5.12), the right-hand function of the first equation is

$\begin{equation} f(x, t) = \cos(3\pi t)\sin(3\pi x)\sqrt{|\cos(3\pi t)\sin(3\pi x)|}, \; \; \; \; (x, t)\in \Omega. \end{equation}$

(5.14)

The Scheme (5.3) is applied to Example 5.1, and the values of $\rm log_{10}$ $\mathbb{E}_{\infty}(E_z)$ and $\rm log_{10}$ $\mathbb{E}_{\infty}(H_y)$ is obtained when $t = 1$ . Computational results are given in Figure 3 to show that the LT-ST method has high accuracy both in time and space.

Figure 3.

$L^{\infty}$ -error at

$t = 1$ of the LT-ST method (5.3).

DownLoad: Full-Size Img PowerPoint

In order to compare the accuracy with the LT-LFCN method, we use it and the LT-ST method to computate Example 5.2, respectively. The LT-LFCN method is as follows:

Let $\tau$ be the time step, $t_k = k\tau\, \, (k = 0, 1, \cdot\cdot\cdot, n_T; T = n_T \tau).$ Denote $u^k(x): = u(x, k\tau)$ , and we define

${u^k_{\hat{t}} = \frac{u^{k+1}-u^{k-1}}{2\tau}, \; \; \; \; u^{\bar k} = \frac{u^{k+1}+u^{k-1}}{2}.}$

The LT-LFCN scheme to the problem $(5.1)$ is: For $1\leq k\leq n_T-1$ , find $E^k_{zN}\in V^0_N$ and $H^k_{yN}\in V_{N-1}$ such that

$\begin{equation} \left\{\begin{array}{ll} ( \epsilon E_{zN\hat{t}}^k, v )+( H^{\bar k}_{yN}, \partial_x v )+(I_N J({E}^k_{zN}), v) = 0, \; \; \; \; &\forall\; v \in V_N^0, \\ ( \mu H_{yN\hat{t}}^k, w)-( \partial_x E_{zN}^{\bar k}, w ) = 0, \; \; \; \; &\forall\; w \in V_{N-1}\\ E^0_{zN} = I_N^L E_{z0}, \; E^1_{zN} = I_N^L[E_{z0}+\tau \partial_t E_{z}(0)], \\ H^0_{yN} = P_{N-1}^LI_N H_{y0}, \; H^1_{yN} = P_{N-1}^LI_N [H_{y0}+\tau \partial_t H_{y}(0)]. \end{array}\right. \end{equation}$

(5.15)

The $L^{\infty}$ -error of the LT-LFCN method (5.15) and the proposed method (5.3) at $t = 1$ are shown in Table 4. The results in Table 4 demonstrate that on the same PC machine, the proposed method provides more accurate results using less time than the LT-LFCN method.

Table 4.

$L^{\infty}$ -error of the LT-LFCN method (5.15) and the LT-ST method (5.3).

LT-LFCN				LT-ST
$(N, \tau)$	$\mathbb{E}_{\infty}(E_z)$	$\mathbb{E}_{\infty}(H_y)$	time	$(N, M)$	$\mathbb{E}_{\infty}(E_z)$	$\mathbb{E}_{\infty}(H_y)$	time
(8, 1e-02)	1.86e-03	1.98e-02	0.21s	(8, 8)	4.13e-03	1.99e-02	0.16s
(12, 1e-03)	2.22e-05	1.46e-04	0.71s	(12, 12)	9.41e-06	3.98e-05	0.17s
(16, 1e-04)	1.65e-07	1.56e-06	7.86s	(16, 16)	6.55e-09	3.33e-08	0.18s
(20, 1e-05)	1.50e-09	1.56e-08	82.99s	(20, 20)	1.37e-12	7.86e-12	0.18s
(24, 1e-06)	1.59e-11	1.58e-10	863.42s	(24, 24)	1.77e-15	2.77e-15	0.19s

| Show Table

DownLoad: CSV

The scheme (5.11) is also used to solve Example 5.2 for long-time computation. Numerical results are given in with $N = M_k = 24$ and $0\leq t\leq 5$ to show the effectiveness of the LT-ST method.

Table 5.

$L^{\infty}$ -error of the time five-interval Legendre-tau spectral method (5.11) (

$N = M_k = 24$ ).

$t$	$L^{\infty}(E_z)$	$L^{\infty}(H_y)$	time	ItNum
$1.00$	1.77e-15	2.77e-15	0.19s	12
$2.00$	3.33e-15	4.88e-15	0.36s	11
$3.00$	3.77e-15	4.21e-15	0.55s	12
$4.00$	6.77e-15	7.21e-15	0.73s	11
$5.00$	9.85e-15	9.35e-15	0.91s	11

| Show Table

DownLoad: CSV

6. Conclusions

In this paper, the LT-ST method is investigated for the 1-D Maxwell equation and the time multi-interval Legendre-tau spectral method is considered. Error estimates for the method of single and multidomain are given, respectively. Numerical results are consistent with the theoretical analysis. Compared with the LT-LFCN method, the proposed method has advantages in accuracy and computation time. Moreover, the space-time spectral method is developed for the numerical solutions of the 1-D nonlinear Maxwell equation. In the future, the multidomain spectral method in space will be developed to solve the case of inhomogeneous media.

Acknowledgments

The research was supported by the National Natural Science Foundation of China (Grants No. 11971016).

Conflict of interest

The authors declare no conflict of interest.

References

[1]	C. W. Elston, I. O. Ellis, Pathological prognostic factors in breast cancer. I. The value of histological grade in breast cancer: Experience from a large study with long-term follow-up, Histopathology, 19 (1991), 403–410. doi: 10.1111/j.1365-2559.1991.tb00229.x
[2]	G. Jiménez, D. Racoceanu, Deep learning for semantic segmentation vs. classification in computational pathology: Application to mitosis analysis in breast cancer grading, Front. Bioeng. Biotechnol., (2019), 7–145.
[3]	Mitosis Detection Contest[EB/OL], http://ipal.cnrs.fr/ICPR2012/.
[4]	MITOS-ATYPIA-14 Contest[EB/OL], https://mitos-atypia-14.grand-challenge.org/home/.
[5]	C. Sommer, L. Fiaschi, F. A. Hamprecht, D. W. Gerlich, Learning-based mitotic cell detection in histopathological images, Proc. 21st Int. Conf. Pattern Recognit., (2012), 2306–2309.
[6]	S. Berg, D. Kutra, T. Kroeger, C. N. Straehle, B. X. Kausler, C. Haubold, et al., Ilastik: Interactive machine learning for (bio)image analysis, Nat. Methods, (2019), 1226–1232.
[7]	C. H. Huang, H. K. Lee, Automated mitosis detection based on exclusive independent component analysis, Proc. 21st Int. Conf. Pattern Recognit., (2012), 1856–1859.
[8]	A. M. Khan, H. El-Daly, N. M. Rajpoot, A gamma-gaussian mixture model for detection of mitotic cells in breast cancer histopathology images, Proc. 21st Int. Conf. Pattern Recognit., (2012), 149–152.
[9]	H. Irshad, Automated mitosis detection in histopathology using morphological and multi-channel statistics features, J. Pathol. Inf., 4 (2013), 10.
[10]	A. Tashk, M. S. Helfroush, H. Danyali, M. Akbarzadeh-jahromi, Automatic detection of breast cancer mitotic cells based on the combination of textural, statistical and innovative mathematical features, Appl. Math. Modell., 39 (2015), 6165–6182.
[11]	F. Pourakpour, H. Ghassemian, Automated mitosis detection based on combination of effective textural and morphological features from breast cancer histology slide images, Biomed. Eng. IEEE, 39 (2016), 214–223.
[12]	H. Wang, A. Cruzroa, A. Basavanhally, H. Gilmore, N. Shih, M. Feldman, et al., Mitosis detection in breast cancer pathology images by combining handcrafted and convolutional neural network features, J. Med. Imag., 1 (2014), 034003.
[13]	T. Kausar, M. Wang, B. Wu, M. Idrees, B. Kanwal, Multi-scale deep neural network for mitosis detection in histological images, Proc. Int. Conf. Intell. Informat. Biomed. Sci., (2018), 47–51.
[14]	D. C. Cireşan, A. Giusti, L. M. Gambardella, J. Schmidhuber, Mitosis detection in breast cancer histology images with deep neural networks, Int. Conf Med. Image Comput. Comput. Assisted Intervention, (2013), 411–418.
[15]	H. Chen, Q. Dou, X. Wang, J. Qin, P. A. Heng, Mitosis detection in breast cancer histology images via deep cascaded networks, Proc. 30th AAAI Conf. Artif. Intell., (2016), 1160–1166.
[16]	M. Ma, Y. Shi, W. Li, Y. Gao, J. Xu, A novel two-stage deep method for mitosis detection in breast cancer histology images, Proc. 24th Int. Conf. Pattern Recognit., (2018), 3892–3897.
[17]	N. Maroof, A. Khan, S. A. Qureshi, A. Rehman, R. K. Khalil, S. Shim, Mitosis detection in breast cancer histopathology images using hybrid feature space, Photodiagn. Photodyn. Ther., (2020), 101885.
[18]	Y. Zhang, J. Chen, X. Pan, Multi-feature fusion of deep networks for mitosis segmentation in histological images, Int. J. Imag. Syst. Technol., (2020), 1–13.
[19]	M. Hwang, D. Wang, C. Wu, W. C. Jiang, X. X. Kong, K. Hwang, et al., A fuzzy segmentation method to learn classification of mitosis, Int. J. Fuzzy Syst., (2020), 1653–1664.
[20]	C. Li, X. Wang, W. Liu, L. J. Latecki, DeepMitosis: Mitosis detection via deep detection, verification and segmentation networks, Med. Image Anal., (2018), 121–133.
[21]	S. Ren, K. He, R. Girshick, J. Sun, Faster R-CNN: Towards real-time object detection with region proposal networks, IEEE Trans. Pattern Anal. Mach. Intell. (2017), 1137–1149.
[22]	K. He, X. Zhang, S. Ren, J. Sun, Deep residual learning for image recognition, Proceed. IEEE Conf. Comput. Vision Pattern Recognit., (2016), 770–778.
[23]	H. Lei, S. Liu, H. Xie, J. Y. Kuo, B. Lei, An improved object detection method for mitosis detection, Proc. 41st Annu. Int. Conf. IEEE Eng. Med. Biol. Soc., (2019), 130–133.
[24]	A. Sengupta, Y. Ye, R. Wang, C. Liu, K. Roy, Going deeper in spiking neural networks: VGG and residual architectures, Front. Neurosci., (2019), 1–10
[25]	C. Li, X. Wang, W. Liu, L. J. Latecki, B. Wang, J. Huang, Weakly supervised mitosis detection in breast histopathology images using concentric loss, Med. Image Anal. (2019), 165–178.
[26]	K. He, G. Gkioxari, P. Dollár, R. Girshick, Mask R-CNN, Proceed. IEEE Int. Conf. Comput. Vision, (2017), 2980–2988.
[27]	V. Dodballapur, Y. Song, H. Huang, M. Chen, W. Chrzanowski, W. Cai, Mask-driven mitosis detection in histopathology images, Proceed. IEEE 16th Int. Symp. Biomed. Imaging, (2019), 1855–1859.
[28]	D. Cai, X. Sun, N. Zhou, X. Han, J. Yao, Efficient mitosis detection in breast cancer histology images by RCNN, Proceed. IEEE 16th Int. Symp. Biomed. Imaging, (2019), 919–922.
[29]	Y. Li, E. Mercan, S. Knezevitch, J. G. Elmore, L. G. Shapiro, Efficient and accurate mitosis detection, a lightweight RCNN approach, Proceed. 7th Int. Conf Pattern Recognit. Appl. Methods, (2018), 69–77.
[30]	H. J. Lei, S. M. Liu, A. Elazab, X. H. Gong, B. Y. Lei, Attention-guided multi-branch convolutional neural network for mitosis detection from histopathological images, J. Biomed. Health Inf., (2020), 2168–2194.
[31]	T. Mahmood, M. Arsalan, M. Owais, M. B. Lee, K. R. Park, Artificial intelligence-based mitosis detection in breast cancer histopathology images using faster R-CNN and deep CNNs, J. Clin. Med., (2020), 749
[32]	S. Meriem, T. Wang, S. A. Al-Fadhli, PartMitosis: A partially supervised deep learning framework for mitosis detection in breast cancer histopathology images, IEEE Access, (2020), 1–1.
[33]	S. Meriem, X. Wang, T. Wang, MaskMitosis: A deep learning framework for fully supervised, weakly supervised, and unsupervised mitosis detection in histopathology images, Med. Biol. Eng. Comput., (2020) 1603–1623.
[34]	N. Kumar, R. Verma, S. Sharma, S. Bhargava, A. Vahadane, A. Sethi, A dataset and a technique for generalized nuclear segmentation for computational pathology, IEEE Trans. Med. Imaging, 36 (2017), 1550–1560. doi: 10.1109/TMI.2017.2677499
[35]	O. Ronneberger, P. Fischer, T. Brox, U-Net: Convolutional networks for biomedical image segmentation, Int. Conf Med. Image Comput. Comput. Assisted Intervention, (2015), 234–241.
[36]	Z. W. Zhou, M. Mahfuzur, R. Siddiquee, N Tajbakhsh, J. M. Liang, UNet ++: A nested U-Net architecture for medical image segmentation, Dlmia, (2018), 1–7.
[37]	C. Y. Lee, S. Xie, P. Gallagher, Z. Zhang, Z. W. Tu, Deeply-supervised nets, Proceed. 2015 18th Int. Conf. Artif. Int. Stat., (2015), 562–570.
[38]	C. H. Sudre, W. Li, T. Vercauteren, S. Ourselin, M. J. Cardoso, Generalised dice overlap as a deep learning loss function for highly unbalanced segmentations, Comput. Vision Pattern Recogn., 2017.
[39]	C. Yamauchi, T. Hasebe, M. Iwasaki, S. Imoto, N. Wada, M. Fukayama, et al., Accurate assessment of lymph vessel tumor emboli in invasive ductal carcinoma of the breast according to tumor areas, and their prognostic significance, Hum. Pathol., 38 (2007), 247–259.
[40]	H. Wang, A. Cruz-Roa, A. Basavanhally, H. Gilmore, N. Shih, M. Feldman, et al., Cascaded ensemble of convolutional neural networks and handcrafted features for mitosis detection, Proc. SPIE, 9041 (2014).
[41]	F. Milletari, N. Navab, S. Ahmadi, V-Net: Fully convolutional neural networks for volumetric medical image segmentation, proc. Int. Conf. 3D Vision, (2016), 565–571.
[42]	K. He, X. Zhang, S. Ren, J. Sun, Deep residual learning for image recognition, Proc. IEEE Conf. Comput. Vis. Pattern Recognit., (2016), 770–778.
[43]	P. Vamplew, R. Dazele, C. Foale, Softmax exploration strategies for multi-objective reinforcement learning, Neuro comput., (2017), 74–86.
[44]	A. Vahadane, T. Peng, S. Albarqouni, M. Baust, K. Steiger, Structure-preserved color normalization for histological images, IEEE Int. Symp. Biomed. Imaging, (2015), 1012–1015.
[45]	A. Krizhevsky, I. Sutskever, G. E. Hinton, ImageNet classification with deep convolutional neural networks, Int. Conf. Neural Inf. Process. Syst., (2012), 1097–1105.
[46]	N. Otsu, A threshold selection method from gray-level histograms, IEEE Trans. Syst. Man. Cybern., 9 (1979), 62–66. doi: 10.1109/TSMC.1979.4310076
[47]	M. Veta, J. Pluim, Detecting mitotic figures in breast cancer histopathology images, SPIE Med. Imaging, 8676 (2013), 867607.
[48]	C. D. Malon, C. Eric, Classification of mitotic figures with convolutional neural networks and seeded blob features, J. Pathol. Inf., 4 (2013), 9.
[49]	M Veta, P. J. Diest, S. M. Willems, H. Wang, A. Madabhushi, A. Cruz-Roa, et al., Assessment of algorithms for mitosis detection in breast cancer histopathology images, Med. Image Anal., 20 (2015), 237–248.
[50]	H. Chen, X. Wang, P. A. Heng, Automated mitosis detection with deep regression networks, 2016 IEEE 13th Int. Symp. Biomed. Imaging, 2016, 1945–8452.
[51]	C. Li, X. Wang, W. Liu, L. J. Latecki, DeepMitosis: Mitosis detection via deep detection, verification and segmentation networks, Med. Image Anal., 45 (2018), 121–133.

This article has been cited by:

1.	Haizhou Liu, Yixin Huang, Yang Zhao, A step-by-step Chebyshev space-time spectral method for force vibration of functionally graded structures, 2025, 41, 0567-7718, 10.1007/s10409-024-24193-x
2.	Wenting Shao, Cheng Chen, A fourth order Runge-Kutta type of exponential time differencing and triangular spectral element method for two dimensional nonlinear Maxwell's equations, 2025, 207, 01689274, 348, 10.1016/j.apnum.2024.09.008
3.	Olga Podvigina, An efficient Galerkin method for problems with physically realistic boundary conditions, 2025, 309, 00104655, 109482, 10.1016/j.cpc.2024.109482

Reader Comments

Your name:*

Email:*
© 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)