# DG-GMsFEM for Problems in Perforated Domains with Non-Homogeneous Boundary Conditions

^{1}

^{2}

^{3}

^{4}

^{*}

Previous Article in Journal

Multiscale Model Reduction Laboratory, North-Eastern Federal University, 677000 Yakutsk, Russia

Yakutsk Branch of the Regional Scientific and Educational Mathematical Center “Far Eastern Center of Mathematical Research”, North-Eastern Federal University, 677000 Yakutsk, Russia

Center of Innovation for Flow through Porous Media, University of Wyoming, Laramie, WY 82071, USA

Department of Mathematics, The Chinese University of Hong Kong, Shatin, New Territories, Hong Kong, China

Author to whom correspondence should be addressed.

Academic Editor: Ali Cemal Benim

Received: 27 May 2021
/
Revised: 22 June 2021
/
Accepted: 25 June 2021
/
Published: 29 June 2021

Problems in perforated media are complex and require high resolution grid construction to capture complex irregular perforation boundaries leading to the large discrete system of equations. In this paper, we develop a multiscale model reduction technique based on the Discontinuous Galerkin Generalized Multiscale Finite Element Method (DG-GMsFEM) for problems in perforated domains with non-homogeneous boundary conditions on perforations. This method implies division of the perforated domain into several non-overlapping subdomains constructing local multiscale basis functions for each. We use two types of multiscale basis functions, which are constructed by imposing suitable non-homogeneous boundary conditions on subdomain boundary and perforation boundary. The construction of these basis functions contains two steps: (1) snapshot space construction and (2) solution of local spectral problems for dimension reduction in the snapshot space. The presented method is used to solve different model problems: elliptic, parabolic, elastic, and thermoelastic equations with non-homogeneous boundary conditions on perforations. The concepts for coarse grid construction and definition of the local domains are presented and investigated numerically. Numerical results for two test cases with homogeneous and non-homogeneous boundary conditions are included, as well. For the case with homogeneous boundary conditions on perforations, results are shown using only local basis functions with non-homogeneous boundary condition on subdomain boundary and homogeneous boundary condition on perforation boundary. Both types of basis functions are needed in order to obtain accurate solutions, and they are shown for problems with non-homogeneous boundary conditions on perforations. The numerical results show that the proposed method provides good results with a significant reduction of the system size.

Problems in perforated domains are of great interest additionally proposing many real-world applications. Take, for example, diffusion in perforated domains, mechanical processes in granular media, pore-scale flows in porous media, and so on [1,2,3,4]. Problems in perforated domains with non-homogeneous boundary conditions on perforations have great importance for a lot of applications in physics, biology, geology, and chemistry [5,6,7,8]. The main characteristic of perforated domains is the multiscale nature of the underlying processes. The solution techniques for these problems require high resolution in grid construction to capture complex irregular boundaries of perforations. Using direct numerical methods to solve these problems is computationally challenging and expensive. In order to reduce the size of the system with accurate approximation, we use homogenization techniques and multiscale methods.

Reduction techniques are widely used today for problems in heterogeneous perforated media to reduce computational cost. In recent years, many methods have been developed focused on obtaining solutions on a coarse mesh. The homogenization method can be used for problems in perforated domains with scale separation [8,9,10]. For example, in Reference [11], the author considers the periodic homogenization problem in perforated domains that are formed by removing a periodic array of small holes from a fixed open bounded and connected domain with regular boundary. In the numerical homogenization method, the approximation of the solution on a coarse grid is constructed by calculating effective properties for coarse grid cells [12]. In Reference [13], the authors construct the machine learning methods for a fast calculation of effective characteristic for domains with random inclusions. In addition, for the perforated media without scale separation, the multiscale techniques are widely used [14,15]. Multiscale methods construct multiscale basis functions in each coarse-grid cell and couple these basis functions in a global formulation. Several studies have shown employment of multiscale methods as solution to problems in perforated domains [16,17,18,19]. There are also studies presenting the multiscale approach based on Crouzeix-Raviart coupling of multiscale finite element basis [16,17]. The generalization of the heterogeneous multiscale finite element method for elliptic problems in perforated domains is presented in work of Henning and Ohlberger (2009) [18]. In order to avoid a limited number of degrees of freedom per coarse element, this paper considers the Generalized Multiscale Finite Element Method (GMsFEM) for solution problems in perforated domains [19,20,21,22]. The GMsFEM is a general multiscale procedure, wherein the model reduction is based on some local multiscale basis functions that are constructed using local spectral decomposition. In our earlier study [19], we presented an accurate multiscale approximation using GMsFEM with several basis functions in each local domain. The problems with non-homogeneous boundary conditions on perforations are studied in References [23,24], where authors evaluate construction of the additional basis functions for perforation boundary. The multiscale model reduction technique, based on Discontinuous Galerkin Generalized Multiscale Finite Element Method (DG-GMsFEM), is presented in Reference [22], where we consider the solution of the problems with homogeneous boundary conditions. Additional points of multiscale methods can distinguish methods, such as upscaling method using non-local multicontinuum method (NLMC) [24,25], which can be effective for heterogeneous. The main part of the method is the construction of suitable local basis functions with the capability of capturing multiscale features and non-local effects. Another method that deserves attention is the constraint energy minimizing generalized multiscale finite element method (CEM-GMsFEM) [26,27]. The main purpose of this method is that the convergence of the method is independent of contrast and decreases linearly with respect to the cell size if the oversampling size is chosen appropriately. In addition, for punctured tasks, the oversampling method can be especially useful and improve accuracy and convergence [28,29,30]. Besides, the oversampling strategy is used to reduce the mismatching effects of boundary conditions imposed artificially in the construction of snapshot basis functions.

This paper presents an extension of the DG-GMsFEM for problems with non-homogeneous boundary conditions on perforations. The presented approach is based on the construction of a separate multiscale space for coarse cell interfaces and for local perforation boundary. We present a unified approach for different types of problems: elliptic, parabolic, elastic, and thermoelastic problems. Using the DG-GMsFEM for coarse grid coupling, we construct multiscale basis functions to generate an accurate lower-dimensional model on a coarse grid. The paper starts with the construction of the snapshot space that contains a set of local solutions in each coarse cell with various boundary conditions. In order to perform a dimension reduction of the snapshot space and define multiscale basis functions, we solve a local spectral problem in the snapshot space. The construction of two types of multiscale basis functions for subdomain boundary and perforation boundary are presented separately. The presented perforation boundary basis functions are used to approximate non-homogeneous boundary conditions on perforations. The numerical results for two-dimensional problems with homogeneous and non-homogeneous boundary conditions on perforations are presented. We investigate a different number of subdomain and perforation boundary multiscale basis functions and show that an accurate solution can be obtained with a large reduction of the system size with a sufficient number of multiscale basis functions. The coarse grid construction and definition of the local domains are presented and investigated numerically for each of the considered equations.

This paper is organized as follows. Section 2 contains description of the problem formulation. The fine-scale approximation using the interior penalty discontinuous Galerkin (IPDG) method is given in Section 3. The multiscale method based on DG-GMsFEM is presented in Section 4 with the description of the construction of the multiscale basis functions. Section 5 presents numerical results for the two-dimensional problem in a perforated domain with homogeneous and non-homogeneous boundary conditions on perforations. Results are shown for the different number of multiscale basis functions, as well as results for different types of problems (elliptic, parabolic, elastic, and thermoelastic). The results for quasi-structured and unstructured coarse grid are presented. Finally, the conclusion is given in Section 6.

Let $\mathsf{\Omega}$ be a perforated domain and ${\mathsf{\Gamma}}_{p}$ be the perforation boundary. In this work, we consider the elliptic equation in perforated domain
with
and non-homogeneous boundary conditions on perforation boundary
where f is a given source vector, $\mathcal{L}$ denotes a linear differential operator, $\mathcal{B}$ is a normal derivative operator, and $\partial \mathsf{\Omega}={\mathsf{\Gamma}}_{g}\cup {\mathsf{\Gamma}}_{p}$.

$$\mathcal{L}\left(u\right)=f,\phantom{\rule{1.em}{0ex}}x\in \mathsf{\Omega},$$

$$u={g}_{g},\phantom{\rule{1.em}{0ex}}x\in {\mathsf{\Gamma}}_{g},$$

$$\mathcal{B}\left(u\right)=\alpha u+{g}_{p},\phantom{\rule{1.em}{0ex}}x\in {\mathsf{\Gamma}}_{p},$$

- For the Laplace operator, we have$$\mathcal{L}\left(u\right)=\nabla \xb7q\left(u\right),\phantom{\rule{1.em}{0ex}}\mathcal{B}\left(u\right)=q\xb7n,$$$$q\left(u\right)=-k\nabla u,$$
- For the elasticity operator, we have$$\mathcal{L}\left(u\right)=\nabla \xb7\sigma \left(u\right),\phantom{\rule{1.em}{0ex}}\mathcal{B}\left(u\right)=\sigma \xb7n,$$$$\sigma \left(u\right)=2\mu \epsilon \left(u\right)+\lambda \nabla \xb7u\phantom{\rule{0.166667em}{0ex}}\mathcal{I},\phantom{\rule{1.em}{0ex}}\epsilon \left(u\right)=\frac{1}{2}(\nabla u+\nabla {u}^{Tr}),$$

This section shows an approximation using the interior penalty discontinuous Galerkin method (IPDG) on the fine grid that resolves all perforations on the grid level.

Let ${\mathcal{T}}^{h}$ be a fine-grid partition of the domain $\mathsf{\Omega}$ given by
where ${N}_{cell}^{h}$ is the number of fine grid cells. We use ${\mathcal{E}}^{h}$ to denote the set of facets in ${\mathcal{T}}^{h}$ with ${\mathcal{E}}^{h}={\mathcal{E}}_{o}^{h}\cup {\mathcal{E}}_{b}^{h}$, where ${\mathcal{E}}_{o}^{h}$ and ${\mathcal{E}}_{b}^{h}$ are the set of interior and boundary facets (see Figure 1).

$${\mathcal{T}}^{h}=\bigcup _{i=1}^{{N}_{cell}^{h}}{K}_{i},$$

Defining the jump $\left[u\right]$ and the average $\left\{u\right\}$ of a function u on the interior facet by
where ${u}_{+}{=u|}_{{K}^{+}}$, ${u}_{-}{=u|}_{{K}^{-}}$, and ${K}^{+}$ and ${K}^{-}$ are the two cells sharing the facet E. For boundary facets, we have ${\left[u\right]=u|}_{E}$ and ${\left\{u\right\}=u|}_{E}$, $E\in {\mathcal{E}}_{b}^{h}$.

$$\left[u\right]={u}_{+}-{u}_{-},\phantom{\rule{1.em}{0ex}}\left\{u\right\}=\frac{{u}_{+}+{u}_{-}}{2},$$

For IPDG approximation, the following variational formulation of the problem is used: find ${u}_{h}\in {V}_{h}$ such that
where:

$$\begin{array}{c}\hfill a({u}_{h},v)=l\left(v\right),\phantom{\rule{1.em}{0ex}}\forall v\in {V}_{h}\end{array},$$

- for the Laplace operator, we have ${V}_{h}=\{v\in {L}^{2}\left(\mathsf{\Omega}\right):\phantom{\rule{0.166667em}{0ex}}v{|}_{K}\in {\mathbb{P}}_{1}\left(K\right),\phantom{\rule{0.166667em}{0ex}}\forall K\in {\mathcal{T}}^{h}\}$ and$$\begin{array}{cc}\hfill a(u,v)& =\sum _{K\in {\mathcal{T}}^{h}}{\int}_{K}(k\nabla u,\nabla v)\phantom{\rule{0.166667em}{0ex}}dx+\sum _{E\in {\mathcal{E}}_{b,p}^{h}}{\int}_{E}\alpha \phantom{\rule{0.166667em}{0ex}}u\phantom{\rule{0.166667em}{0ex}}v\phantom{\rule{0.166667em}{0ex}}ds\hfill \\ & -\sum _{E\in {\mathcal{E}}_{o}^{h}\cup {\mathcal{E}}_{b,g}^{h}}{\int}_{E}\left(\{k\nabla u\xb7n\}\xb7\left[v\right]+\{k\nabla v\xb7n\}\xb7\left[u\right]-\frac{{\gamma}_{f}}{h}\left\{k\right\}\left[u\right]\xb7\left[v\right]\right)\phantom{\rule{0.166667em}{0ex}}ds,\hfill \\ \hfill l\left(v\right)& =\sum _{K\in {\mathcal{T}}^{h}}{\int}_{K}f\phantom{\rule{4pt}{0ex}}v\phantom{\rule{4pt}{0ex}}dx+\sum _{E\in {\mathcal{E}}_{b,p}^{h}}{\int}_{E}{g}_{p}\phantom{\rule{0.166667em}{0ex}}v\phantom{\rule{0.166667em}{0ex}}ds\hfill \\ & +\sum _{E\in {\mathcal{E}}_{b,g}^{h}}{\int}_{E}\left(\frac{{\gamma}_{f}}{h}\phantom{\rule{0.166667em}{0ex}}k\phantom{\rule{0.166667em}{0ex}}v-k\nabla v\xb7n\right)\phantom{\rule{0.166667em}{0ex}}{g}_{g}\phantom{\rule{0.166667em}{0ex}}ds.\hfill \end{array}$$
- for the elasticity operator, we have ${V}_{h}=\{v\in {\left[{L}^{2}\left(\mathsf{\Omega}\right)\right]}^{2}:\phantom{\rule{0.166667em}{0ex}}v{|}_{K}\in {\left[{\mathbb{P}}_{1}\left(K\right)\right]}^{2},\phantom{\rule{0.166667em}{0ex}}\forall K\in {\mathcal{T}}^{h}\},$ and$$\begin{array}{cc}\hfill a(u,v)& =\sum _{K\in {\mathcal{T}}^{h}}{\int}_{K}(\sigma \left(u\right),\epsilon \left(v\right))dx+\sum _{E\in {\mathcal{E}}_{b,p}^{h}}{\int}_{E}\alpha \phantom{\rule{0.166667em}{0ex}}u\phantom{\rule{0.166667em}{0ex}}v\phantom{\rule{0.166667em}{0ex}}ds\hfill \\ & -\sum _{E\in {\mathcal{E}}_{o}^{h}\cup {\mathcal{E}}_{b,g}^{h}}{\int}_{E}\left(\left\{\tau \left(u\right)\right\}\left[v\right]+\left\{\tau \left(v\right)\right\}\left[u\right]-\frac{{\gamma}_{f}}{h}\{\lambda +2\mu \}\left[u\right]\left[v\right]\right)\phantom{\rule{0.166667em}{0ex}}ds,\hfill \\ \hfill l\left(v\right)& =\sum _{K\in {\mathcal{T}}^{h}}{\int}_{K}f\xb7v\phantom{\rule{4pt}{0ex}}dx+\sum _{E\in {\mathcal{E}}_{b,p}^{h}}{\int}_{E}{g}_{p}\xb7v\phantom{\rule{0.166667em}{0ex}}ds\hfill \\ & +\sum _{E\in {\mathcal{E}}_{b,g}^{h}}{\int}_{E}\left(\frac{{\gamma}_{f}}{h}\phantom{\rule{0.166667em}{0ex}}(\lambda +2\mu )\phantom{\rule{0.166667em}{0ex}}v-\tau \left(v\right)\right)\xb7{g}_{g}\phantom{\rule{0.166667em}{0ex}}ds\hfill \end{array},$$

Here, ${\gamma}_{f}$ is the penalty parameter.

The above systems can be written in the matrix form as follows:
where
with ${u}_{h}={\sum}_{i}{u}_{i}{\phi}_{i}$ and ${U}_{h}=\left[{u}_{j}\right]$.

$${A}_{h}{U}_{h}={F}_{h},$$

$${A}_{h}=\left[{a}_{i,j}\right],\phantom{\rule{1.em}{0ex}}{a}_{i,j}=a({\phi}_{i},{\phi}_{j}),\phantom{\rule{1.em}{0ex}}{F}_{h}=\left[{f}_{j}\right],\phantom{\rule{1.em}{0ex}}{f}_{j}=l\left({\phi}_{j}\right),$$

This section describes the construction of multiscale approximation on the coarse grid using the Discontinuous Galerkin Generalized Multiscale Finite Element method (DG-GMsFEM) [22,31,32].

Let ${\mathcal{T}}^{H}$ be a coarse-grid partition of the domain $\mathsf{\Omega}$ with coarse mesh size H (see Figure 2).
where ${N}_{cell}^{H}$ is the number of coarse grid cells, and ${K}_{i}$ is coarse cell (local domain). Use ${\mathcal{E}}^{H}$ to denote the set of facets in ${\mathcal{T}}^{H}$ with ${\mathcal{E}}^{H}={\mathcal{E}}_{o}^{H}\cup {\mathcal{E}}_{b}^{H}$. Note that, in DG-GMsFEM, the multiscale basis functions are supported in each coarse cell ${K}_{i}$.

$${\mathcal{T}}^{H}=\bigcup _{i=1}^{{N}_{cell}^{H}}{K}_{i},$$

Defining ${V}_{H}$ as the multiscale space
parabolic, elastic, and thermoelastic ${N}_{u}=\mathrm{dim}\left({V}_{H}\right)$ is the number of basis functions.

$${V}_{H}=\mathrm{span}{\left\{{\varphi}_{i}\right\}}_{i=1}^{{N}_{u}},$$

For the coarse grid approximation, we use a DG approach and have the following variational formulation: find ${u}_{H}\in {V}_{H}$ such that

$$a({u}_{H},v)=l\left(v\right),\phantom{\rule{1.em}{0ex}}\forall v\in {V}_{H}.$$

Note that the coarse-scale system can be formed by projecting the fine-scale system onto the coarse grid. The projection matrix can be assembled using the multiscale basis functions.

To construct the multiscale space, we start with the construction of the snapshot space that contains a set of basis functions formed by the solution of local problems with all possible boundary conditions up to the fine grid resolution in each coarse cell ${K}_{i}$ (local domain) for $i=1,\cdots ,N$, parabolic, elastic, and thermoelastic N is the number of coarse blocks in $\mathsf{\Omega}$. After that, we solve a spectral problem to select dominant modes of the snapshot space.

Next, we present the details of constructing the multiscale basis functions considering two types of basis functions related to the two boundaries: (1) subdomain boundaries ${\mathsf{\Gamma}}_{g}^{i}$ and (2) perforation boundary ${\mathsf{\Gamma}}_{p}^{i}$.

$$\mathcal{L}\left({u}_{l}^{i}\right)=0,\phantom{\rule{1.em}{0ex}}x\in {K}_{i},$$

$${u}_{l}^{i}={g}_{i}^{l},\phantom{\rule{1.em}{0ex}}x\in {\mathsf{\Gamma}}_{g}^{i},$$

$$\mathcal{B}\left({u}_{l}^{i}\right)=\alpha {u}_{l}^{i},\phantom{\rule{1.em}{0ex}}x\in {\mathsf{\Gamma}}_{p}^{i},$$

The collection of the solutions of the above local problems generates the snapshot space in the local domain ${K}_{i}$

$${V}_{g}^{i,\mathrm{snap}}=\{{u}_{l}^{i}:1\le l\le {L}_{i}^{g}\},\phantom{\rule{1.em}{0ex}}{R}_{g}^{i,\mathrm{snap}}={\left[{u}_{1}^{i},\dots ,{u}_{{L}_{i}^{g}}^{i}\right]}^{Tr}.$$

To reduce the size of the snapshot space, we solve the following local spectral problem in the snapshot space ${V}_{g}^{i,\mathrm{snap}}$
where

$${\tilde{A}}_{g}^{{K}_{i}}{\tilde{\psi}}_{g}^{i}={\lambda}_{o}{\tilde{S}}_{g}^{{K}_{i}}{\tilde{\psi}}_{g}^{i},$$

$${\tilde{A}}_{g}^{{K}_{i}}={R}_{g}^{i,\mathrm{snap}}{A}_{h}^{{K}_{i}}{\left({R}_{g}^{i,\mathrm{snap}}\right)}^{Tr},\phantom{\rule{1.em}{0ex}}{\tilde{S}}_{g}^{{K}_{i}}={R}_{g}^{i,\mathrm{snap}}{S}_{h}^{{K}_{i}}{\left({R}_{g}^{i,\mathrm{snap}}\right)}^{Tr}.$$

Here, ${A}_{h}^{{K}_{i}}$ and ${S}_{h}^{{K}_{i}}$ are the matrix representation of the bilinear forms ${a}^{{K}_{i}}(u,v)$ and ${s}^{{K}_{i}}(u,v)$

- For Laplace operator:$$\begin{array}{cc}\hfill {a}^{{K}_{i}}(u,v)\phantom{\rule{1.em}{0ex}}& =\sum _{K\in {\mathcal{T}}^{h}\left({K}_{i}\right)}{\int}_{K}k\nabla u\xb7\nabla v\phantom{\rule{0.166667em}{0ex}}dx\hfill \\ \hfill \phantom{\rule{1.em}{0ex}}& -\sum _{E\in {\mathcal{E}}_{o}^{h}\left({K}_{i}\right)}{\int}_{E}\left(\{k\nabla u\xb7n\}\xb7\left[v\right]+\{k\nabla v\xb7n\}\xb7\left[u\right]-\frac{{\gamma}_{f}}{h}\left\{k\right\}\left[u\right]\xb7\left[v\right]\right)\phantom{\rule{0.166667em}{0ex}}ds,\hfill \\ \hfill {s}^{{K}_{i}}(u,v)& =\sum _{E\in {\mathcal{E}}_{b}^{h}\left({K}_{i}\right)}{\int}_{E}k\phantom{\rule{4pt}{0ex}}u\phantom{\rule{0.166667em}{0ex}}v\phantom{\rule{0.166667em}{0ex}}ds.\hfill \end{array}$$
- For elasticity operator:$$\begin{array}{cc}\hfill {a}^{{K}_{i}}(u,v)\phantom{\rule{1.em}{0ex}}& =\sum _{K\in {\mathcal{T}}^{h}\left({K}_{i}\right)}{\int}_{K}(\sigma \left(u\right),\epsilon \left(v\right))dx\hfill \\ \hfill \phantom{\rule{1.em}{0ex}}& -\sum _{E\in {\mathcal{E}}_{o}^{h}\left({K}_{i}\right)}{\int}_{E}\left(\left\{\tau \left(u\right)\right\}\left[v\right]+\left\{\tau \left(v\right)\right\}\left[u\right]-\frac{{\gamma}_{f}}{h}\{\lambda +2\mu \}\left[u\right]\left[v\right]\right)\phantom{\rule{0.166667em}{0ex}}ds,\hfill \\ \hfill {s}^{u,{K}_{i}}(u,v)& =\sum _{E\in {\mathcal{E}}_{b}^{h}\left({K}_{i}\right)}{\int}_{E}(\lambda +2\mu )u\xb7v\phantom{\rule{0.166667em}{0ex}}ds.\hfill \end{array}$$

It should be noted that the integral in ${s}^{{K}_{i}}(u,v)$ is defined on the boundary of the coarse block ${K}_{i}$ and based on the definition of the snapshot space used to extract dominant modes related to the outer boundary of the local domain.

Next, we arrange the eigenvalues in increasing order and choose the first eigenvectors corresponding to the first smallest eigenvalues ${\psi}_{g,k}^{i}={\left({R}_{g}^{i,\mathrm{snap}}\right)}^{Tr}{\tilde{\psi}}_{g,k}^{i}$ as the basis functions ($k=1,\cdots ,{M}_{g}^{i}$)
The first five eigenvectors for some local domains ${K}_{i}$ are depicted in Figure 3 for Laplace operator and in Figure 4 for elasticity operator.

$${V}_{H}^{g}=\mathrm{span}\{{\psi}_{g,k}^{i}:\phantom{\rule{0.166667em}{0ex}}1\le i\le {N}_{cell}^{H},\phantom{\rule{0.166667em}{0ex}}1\le k\le {M}_{g}^{i}\}.$$

$$\mathcal{L}\left({u}_{l}^{i}\right)=0,\phantom{\rule{1.em}{0ex}}x\in {K}_{i},$$

$${u}_{l}^{i}=0,\phantom{\rule{1.em}{0ex}}x\in {\mathsf{\Gamma}}_{g}^{i},$$

$$\mathcal{B}\left({u}_{l}^{i}\right)=\alpha {u}_{l}^{i}+{g}_{i}^{l},\phantom{\rule{1.em}{0ex}}x\in {\mathsf{\Gamma}}_{p}^{i},$$

We form a snapshot space using local solutions

$${V}_{p}^{i,\mathrm{snap}}=\{{u}_{l}^{i}:1\le l\le {L}_{i}^{p}\},\phantom{\rule{1.em}{0ex}}{R}_{p}^{i,\mathrm{snap}}={\left[{u}_{1}^{i},\dots ,{u}_{{L}_{i}^{p}}^{i}\right]}^{Tr}.$$

We perform a dimension reduction in the snapshot space using the local spectralproblem
where

$${\tilde{A}}_{p}^{{K}_{i}}{\tilde{\psi}}_{g,k}^{i}=\eta {\tilde{S}}_{p}^{{K}_{i}}{\tilde{\psi}}_{g,k}^{i},$$

$${\tilde{A}}_{p}^{{K}_{i}}={R}_{p}^{i,\mathrm{snap}}{A}_{h}^{{K}_{i}}{\left({R}_{p}^{i,\mathrm{snap}}\right)}^{Tr},\phantom{\rule{1.em}{0ex}}{\tilde{S}}_{p}^{{K}_{i}}={R}_{p}^{i,\mathrm{snap}}{S}_{h}^{{K}_{i}}{\left({R}_{p}^{i,\mathrm{snap}}\right)}^{Tr}.$$

The eigenvalues are arranged in increasing order, and, by choosing the first eigenvectors corresponding to the first smallest eigenvalues, we define perforation boundary multiscale basis functions
where ${\psi}_{g,k}^{i}={\left({R}_{p}^{i,\mathrm{snap}}\right)}^{Tr}{\tilde{\psi}}_{g,k}^{i}$ for $k=1,\cdots ,{M}_{p}^{i}$, and ${N}_{cell,p}$ is the number of local domains with perforations. It should be noted that the first eigenvector was not taken because constant valued vector already exists in outer boundary multiscale space. The perforation boundary of multiscale basis functions are presented in Figure 5 for Laplace problem and in Figure 6 for elasticity problem.

$${V}_{H}^{p}=\mathrm{span}\{{\psi}_{g,k}^{i}:\phantom{\rule{0.166667em}{0ex}}1\le i\le {N}_{cell,p}^{H},\phantom{\rule{0.166667em}{0ex}}2\le k\le {M}_{p}^{i}\},$$

$$\mathcal{L}\left({\varphi}^{i}\right)=f,\phantom{\rule{1.em}{0ex}}x\in {K}_{i},$$

$${\varphi}^{i}=0,\phantom{\rule{1.em}{0ex}}x\in {\mathsf{\Gamma}}_{g}^{i},\phantom{\rule{1.em}{0ex}}\mathcal{B}\left(\varphi \right)=\alpha \varphi ,\phantom{\rule{1.em}{0ex}}x\in {\mathsf{\Gamma}}_{p}^{i}.$$

Finally, there is the following multiscale space:
where ${N}_{p}={N}_{cell,p}$, and ${N}_{g}={N}_{cell}$

$${V}_{H}=\mathrm{span}\{{\psi}_{g,k}^{i},{\psi}_{p,m}^{j},{\varphi}^{i}:\phantom{\rule{0.166667em}{0ex}}1\le i\le {N}_{g},\phantom{\rule{0.166667em}{0ex}}1\le j\le {N}_{p},\phantom{\rule{0.166667em}{0ex}}1\le k\le {M}_{g}^{i},\phantom{\rule{0.166667em}{0ex}}2\le m\le {M}_{p}^{i}\},$$

$$R={\left[{\psi}_{g,1}^{1},\dots ,{\psi}_{g,{M}_{g}^{{N}_{g}}}^{{N}_{g}},{\psi}_{p,1}^{2},\dots ,{\psi}_{p,{M}_{p}^{{N}_{p}}}^{{N}_{p}},{\varphi}_{1},\dots ,{\varphi}_{{N}_{cell}^{H}}\right]}^{Tr}.$$

Using projection matrix, we have the following coarse grid system in matrix form:
where

$${A}_{H}{U}_{H}={F}_{H},$$

$${A}_{H}=R{A}_{h}{R}^{Tr},\phantom{\rule{1.em}{0ex}}{F}_{H}=R{F}_{h}.$$

After the solution of the coarse-scale system, the fine-scale solution for displacement can be recovered ${U}_{ms}={R}^{Tr}{U}_{H}$.

We present numerical results for the model problems in the perforated domain $\mathsf{\Omega}=[0,{L}_{x}]\times [0,{L}_{y}]$ with ${L}_{x}={L}_{y}=1$. The computational domain with coarse and fine grids is presented in Figure 7. The fine grid contains 14,648 vertices and 28,410 cells. The coarse grid contains 121 vertices and 100 cells (local domains). To construct the computational domain with computational mesh, we use the mesh generator Gmsh [33].

To investigate the presented multiscale method for problems solving in perforated domains, the following tests are considered: (1) elliptic equation and (2) elasticity equation. The extension of the method for solution of the parabolic and thermoelasticity problems are presented in Appendix A and Appendix B. The numerical implementation of the model problems is based on the FEniCS computing platform [34]. There are two test cases for each problem:

- Case 1. homogeneous boundary conditions on perforations;
- Case 2. non-homogeneous boundary conditions on perforations.

We calculate the following relative errors in ${L}^{2}$ and energy norm (semi-norm ${H}^{1}$) between multiscale and fine-grid solution:
where u and ${u}_{ms}$ are the fine-scale and multiscale solutions, respectively.

$${e}_{{L}^{2}}=\sqrt{\frac{{\int}_{\mathsf{\Omega}}{({u}_{ms}-u)}^{2}\phantom{\rule{0.166667em}{0ex}}dx}{{\int}_{\mathsf{\Omega}}{u}^{2}\phantom{\rule{0.166667em}{0ex}}dx}}\xb7100\%,\phantom{\rule{1.em}{0ex}}{e}_{{H}^{1}}=\sqrt{\frac{{\int}_{\mathsf{\Omega}}a({u}_{ms}-u,{u}_{ms}-u)\phantom{\rule{0.166667em}{0ex}}dx}{{\int}_{\mathsf{\Omega}}a(u,u)\phantom{\rule{0.166667em}{0ex}}dx}}\xb7100\%,\phantom{\rule{1.em}{0ex}}$$

To start, there are results for elliptic and elastic equations. Whereat there is evaluation of the unstructured coarse grids and heterogeneous perforated domains. The results for parabolic and thermoelasticity problems are given in Appendix A and Appendix B.

We consider the elliptic equation in the perforated domain $\mathsf{\Omega}$ with $k=1$ and the following boundary conditions:

- Case 1. Homogeneous boundary conditions on perforations:$$u=1,\phantom{\rule{1.em}{0ex}}x\in {\mathsf{\Gamma}}_{g},\phantom{\rule{1.em}{0ex}}-k\nabla u\xb7n=\alpha u,\phantom{\rule{1.em}{0ex}}x\in {\mathsf{\Gamma}}_{p}.$$
- Case 2. Non-homogeneous boundary conditions on perforations:$$u=0,\phantom{\rule{1.em}{0ex}}x\in {\mathsf{\Gamma}}_{g},\phantom{\rule{1.em}{0ex}}-k\nabla u\xb7n=\alpha (u-1),\phantom{\rule{1.em}{0ex}}x\in {\mathsf{\Gamma}}_{p}.$$

We consider three different $\alpha =1$, 25, and 100.

Table 1 and Table 2 show relative errors for elliptic problem with different perforation boundary coefficient $\alpha =1,25$, and 100. Here, M is the number of multiscale basis functions, and $DO{F}_{h}$ and $DO{F}_{H}$ are the numbers of degrees of freedom for fine-grid (reference) solution and multiscale solution. Here, $DO{F}_{h}=3\xb7{N}_{cell}^{h}$ for fine grid solution, where ${N}_{cell}^{h}$ is the number of fine grid cells. For multiscale solution, $DO{F}_{H}=({M}_{g}+1)\xb7{N}_{g}+{M}_{p}\xb7{N}_{p}$, where ${N}_{g}={N}_{cell}^{H}$ is the total number of coarse grid cells, and ${N}_{p}={N}_{cell,p}^{H}$ is the number of coarse grid cells with perforations. For the problem with homogeneous boundary conditions (Case 1), we use multiscale space without perforation basis functions (${M}_{p}=0$). Results for multiscale spaces with and without perforation boundary basis functions (${M}_{p}=M$) are presented for Case 2 with non-homogeneous boundary conditions. For Case 1, we observe that the relative error in ${L}^{2}$ norm is reduced to $1\%$ for $\alpha =1,25$, and 100, when we take the sufficient number of multiscale basis functions (${M}_{g}=12$ and ${M}_{p}=0$). For Case 2 of the boundary conditions, the multiscale method gives a solution with a large error when perforation boundary multiscale basis functions is not taken (${M}_{p}=0$). However, we obtain good results after adding perforation boundary bases (${M}_{p}=M$). For example, the relative ${L}^{2}$ error for $M=12$ multiscale basis functions is $12\%$ for ${M}_{p}=0$ and reduced to $1.2\%$ for ${M}_{p}=M$ ($\alpha =25$). For the test problem with larger $\alpha =100$, ${L}^{2}$ error from $21\%$ to $0.67\%$ is reduced. The graphical representation of the Table 1 and Table 2 is given in Figure 8.

Figure 9 and Figure 10 show the fine-grid and multiscale solutions for the elliptic problem with $\alpha =100$ for Case 1 and Case 2, respectively. $DO{F}_{h}=\mathrm{85,230}$ is a reference (fine-grid) solution. For multiscale solution, we used 12 multiscale basis functions ($M=12$) for both Cases. We have $DO{F}_{H}=1300$ for multiscale solution without perforation boundary basis functions in Case 1 and Case 2 (${M}_{g}=M$, ${M}_{p}=0$), and $DO{F}_{H}=1776$ for multiscale solution with perforation boundary basis functions in Case 2 (${M}_{g}={M}_{p}=M$). Figure test1-elliptic shows a reference (fine grid) solution on the left and a multiscale solution on the right for Case 1. For Case 2, there are three pictures in Figure 10: reference solution on the left, multiscale solution with ${M}_{g}=M$, ${M}_{p}=0$ in the center, and multiscale solution with ${M}_{g}={M}_{p}=M$ on the right. For Case 1, the perforation boundary multiscale basis functions are not needed because of set homogeneous boundary conditions on perforations (${M}_{p}=0$). For Case 2, the multiscale solution without perforation boundary bases (${M}_{p}=0$) gives large errors, but very good results are obtained for multiscale solver after adding perforation bases to handle non-homogeneous boundary conditions on perforations (${M}_{p}=M$).

For Case 1 with homogeneous boundary conditions on perforations and Case 2 with non-homogeneous boundary conditions on perforations, based on results, it could be said that the presented multiscale method shows good results with small errors and demonstrates high accuracy for the both cases. Using non-homogeneous boundary conditions on perforations, the multiscale method using perforation boundary basis functions greatly improves results.

Let $\partial \mathsf{\Omega}={\mathsf{\Gamma}}_{t}\cup {\mathsf{\Gamma}}_{b}\cup {\mathsf{\Gamma}}_{l}\cup {\mathsf{\Gamma}}_{r}\cup {\mathsf{\Gamma}}_{p}$, where ${\mathsf{\Gamma}}_{t},{\mathsf{\Gamma}}_{b},{\mathsf{\Gamma}}_{l},{\mathsf{\Gamma}}_{r}$ be the top, bottom, left, and right boundaries.

We consider the elasticity equation with $\lambda =\mu =1$ with the following boundary conditions:
and

$${u}_{x}=0,\phantom{\rule{1.em}{0ex}}{\sigma}_{y}=0,\phantom{\rule{1.em}{0ex}}x\in {\mathsf{\Gamma}}_{l},\phantom{\rule{1.em}{0ex}}{u}_{y}=0,\phantom{\rule{1.em}{0ex}}{\sigma}_{x}=0,\phantom{\rule{1.em}{0ex}}x\in {\mathsf{\Gamma}}_{b},$$

- Case 1. Homogeneous boundary conditions on perforations:$$\sigma \xb7n=1,\phantom{\rule{1.em}{0ex}}x\in {\mathsf{\Gamma}}_{t}\cup {\mathsf{\Gamma}}_{r},\phantom{\rule{1.em}{0ex}}\sigma \xb7n=0,\phantom{\rule{1.em}{0ex}}x\in {\mathsf{\Gamma}}_{p}.$$
- Case 2. Non-homogeneous boundary conditions on perforations:$$\sigma \xb7n=0,\phantom{\rule{1.em}{0ex}}x\in {\mathsf{\Gamma}}_{t}\cup {\mathsf{\Gamma}}_{r},\phantom{\rule{1.em}{0ex}}\sigma \xb7n=-0.01,\phantom{\rule{1.em}{0ex}}x\in {\mathsf{\Gamma}}_{p}.$$

Figure 11 and Figure 12 show the results of the reference (fine-grid) and multiscale solutions for elasticity problem for Case 1 and Case 2. For illustration 24, multiscale basis functions ($M=24$) are used. For fine grid solution, we have $DO{F}_{h}=\mathrm{170,460}$. $DO{F}_{H}=4900$ is for the multiscale solution without perforation basis functions (Case 1 and Case 2) and $DO{F}_{H}=5988$ for the multiscale solution with perforation boundary basis functions in Case 2. The displacements X-component is presented on the first row and Y-component on the second row. The first column shows the solutions on the fine grid. The second column shows the multiscale solution using outer boundary multiscale basis functions with ${M}_{p}=0$ and ${M}_{g}=24$. The third column for Case 2 shows a multiscale solution using outer and perforation boundary multiscale basis functions (${M}_{g}={M}_{p}=24$). Here, $DO{F}_{h}=6\xb7{N}_{cell}^{h}$, $DO{F}_{H}=(2\xb7{M}_{g}+1)\xb7{N}_{cell}^{H}+2\xb7{M}_{p}\xb7{N}_{cell,p}^{H}$.

Table 3 and Table 4 and Figure 13 present relative errors for elasticity problem for Case 1 and Case 2, respectively. Evidently the ${L}^{2}$ error reduce from $117\%$ for ${M}_{p}=0$ to $2.3\%$ for ${M}_{p}=M$ in Case 2 with $M=24$. For Case 1, we have $2\%$ of ${L}^{2}$ error for ${M}_{g}=M=24$ with ${M}_{p}=0$. For both cases, we obtain good results with a sufficient number of multiscale basis functions.

In Case 1 with homogeneous boundary conditions for elasticity on perforations, good results with outer basis functions are obtained. In Case 2 with non-homogeneous boundary conditions for elasticity on perforations, it is seen that, for displacement, we should use perforation boundary basis functions in order to approximate the non-homogeneous boundary condition.

We consider a solution of the elliptic and elasticity problems using the presented multiscale method on quasi-structured and unstructured coarse grids. Figure 14 presents quasi-structured and unstructured coarse grids with an illustration of the local domains. Coarse grids contain 100 local domains. Problems with Case 2 boundary conditions are considered for numerical investigation.

Table 5 presents results for elliptic equation with $\alpha =100$ for non-homogeneous boundary conditions (Case 2). The results for elasticity equation are presented in Table 6. Numerical results are shown for multiscale space with outer and boundary basis functions (${M}_{g}={M}_{p}=M$) on quasi-structured and unstructured coarse grids with 100 local domains. Using 32 multiscale basis functions (${M}_{p}={M}_{g}=32$), similar results are obtained with less than $1\%$ of ${L}^{2}$ error for structured, quasi-structured, and unstructured coarse grids with 100 local domains with $DO{F}_{H}=3844$ ($4.5\%$ of $DO{F}_{f}=\mathrm{85,230}$). By comparing results with less number of multiscale basis functions (${M}_{p}={M}_{g}=8$), it is observed that the error for structured coarse grid is smaller ($1\%$ and $19\%$ for ${L}^{2}$ and ${H}^{1}$ errors) than for unstructured coarse grid ($4.9\%$ and $33\%$ for ${L}^{2}$ and ${H}^{1}$ errors). Furthermore, the results show that the quasi-structured grid give a smaller errors than unstructured coarse grid ($2\%$ and $23\%$ for ${L}^{2}$ and ${H}^{1}$ errors). For the elasticity problem with $M=32$, we have $2\%$ and $13\%$ for ${L}^{2}$ and ${H}^{1}$ errors for structured grid, $3\%$ and $15\%$ for ${L}^{2}$ and ${H}^{1}$ errors for quasi-structured grid, and $5\%$ and $22\%$ for ${L}^{2}$ and ${H}^{1}$ errors for unstructured grid.

For elasticity problem with a smaller number of multiscale basis functions, there less errors for a structured grid than an unstructured grid. By comparison of the quasi-structured and unstructured grids, we see less errors for the quasi-structured coarse grid. Good results with less errors for any coarse grids can be obtained with a large number of multiscale basis functions. It should be noted that the main advantage of using unstructured coarse grids is the similar number of cells in each local domain (load balancing). Conforming triangulation of the domain with coarse edges is used in construction of the structured grid, which can be difficult for complex geometries with a large number of perforations. Though the conforming construction is not needed in the quasi-structured grid. Numerical results show that the presented multiscale method works very well with any coarse grid construction concepts.

Finally, we consider the efficiency of the presented method for the solution of the elliptic and elasticity problems with heterogeneous coefficients ($k=k\left(x\right)$ and $E=E\left(x\right)$). The heterogeneous elasticity parameter and heterogeneous diffusion coefficient are presented in Figure 15. The parameters $\lambda $ and $\mu $ are given as follows:
where $E=E\left(x\right)$ is the elasticity parameter, and $\nu =0.3$ is the constant Poisson’s ratio.

$$\mu \left(x\right)=\frac{E\left(x\right)}{2(1+\nu )},\phantom{\rule{1.em}{0ex}}\lambda \left(x\right)=\frac{E\left(x\right)\nu}{(1+\nu )(1-2\nu )},$$

Table 7 presents the relative ${L}_{2}$ and ${H}_{1}$ errors for elliptic (left) and elasticity (right) problems with heterogeneous coefficients. Case 2 with non-homogeneous boundary conditions for perforation is considered (${M}_{g}={M}_{p}=M$). For elliptic problem, the results are presented with $\alpha =100$. Numerical results for elliptic and elasticity problems show that the presented multiscale method works well with heterogeneous coefficients.

This paper presents the multiscale method for solution problems in the perforated domain with non-homogeneous boundary conditions on perforations. For approximation on the fine grid that resolved perforations on the grid level, we apply the Discontinuous Galerkin finite element method and use the solution as a reference solution. To reduce the size of the fine grid system, we present a Discontinuous Galerkin Generalized Multiscale Finite Element Method (DG-GMsFEM) with the construction of two types of multiscale basis functions for subdomain boundary and perforation boundary, separately. Presented perforation boundary basis functions are used to approximate non-homogeneous boundary conditions on perforations. Numerical investigation of the presented method was performed for four model problems: (1) elliptic, (2) parabolic, (3) elasticity, and (4) thermoelasticity problems in the perforated domain. The results for two cases with homogeneous and non-homogeneous boundary conditions are given. For the case with homogeneous boundary conditions on perforations, the given results are obtained using only subdomain boundary basis functions. However, for a non-homogeneous boundary condition, both subdomain and perforation boundary basis functions should be used. Numerical results are presented for different concepts of the coarse grid construction (structured, quasi-structured, and unstructured coarse grids). Numerical results show that the proposed method can provide good results and give a significant reduction of the system size with appropriate choosing construction of the multiscale basis functions for problems with homogeneous and non-homogeneous perforation boundary conditions. In future works, we plan to consider oversampling techniques for multiscale basis function construction, construct a multiscale solver for three-dimensional problems, and consider construction of the accurate multiscale techniques for nonlinear problems in heterogeneous perforated media.

Conceptualization, M.V. and E.T.C.; methodology, M.V., E.C. and V.A.; software, V.A. and U.K.; validation, V.A., M.V. and U.K.; formal analysis, M.V.; investigation, V.A.; resources, V.A., U.K.; data curation, M.V., V.A.; writing—original draft preparation, V.A.; writing—review and editing, M.V., U.K., E.T.C. and V.A.; visualization, V.A.; supervision, M.V., E.T.C.; project administration, M.V.; funding acquisition, V.A. All authors have read and agreed to the published version of the manuscript.

V.A.’s, U.K.’s, and V.M.’s works are supported by the mega-grant of the Russian Federation Government N14.Y26.31.0013. V.A.’s work is supported by the Ministry of Science and Higher Education of the Russian Federation, supplementary agreement No. 075-02-2020-1543/1, 29 April 2020. U.K.’s work is supported by the RFBR (project number 19-31-90117).

Not applicable.

Not applicable.

Not applicable.

The authors declare no conflict of interest.

We consider the parabolic problem in perforated domain $\mathsf{\Omega}$:
with the following boundary conditions
and initial condition $T={T}_{0}$ in $\mathsf{\Omega}$ for $t=0$.

$$\frac{\partial u}{\partial t}-\nabla \xb7(k\nabla u)=0,\phantom{\rule{1.em}{0ex}}x\in \mathsf{\Omega}$$

$$u={g}_{g},\phantom{\rule{1.em}{0ex}}x\in {\mathsf{\Gamma}}_{g},\phantom{\rule{1.em}{0ex}}-k\nabla u\xb7n=\alpha (u-{g}_{p}),\phantom{\rule{1.em}{0ex}}x\in {\mathsf{\Gamma}}_{p},$$

For approximation by time, we use an implicit approximation with a time step $\tau $. We have the following fine grid approximation on ${\mathcal{T}}^{h}$ using the IPDG method: find ${u}_{h}\in {V}_{h}$ such that
where
and ${\stackrel{\u02c7}{u}}_{h}$ is the solution from the previous time step.

$$\frac{1}{\tau}m({u}_{h}-{\stackrel{\u02c7}{u}}_{h},v)+a({u}_{h},v)=l\left(v\right),\phantom{\rule{1.em}{0ex}}\forall v\in {V}_{h},$$

$$\begin{array}{cc}\hfill a(u,v)& =\sum _{K\in {\mathcal{T}}^{h}}{\int}_{K}(k\nabla u,\nabla v)\phantom{\rule{0.166667em}{0ex}}dx+\sum _{E\in {\mathcal{E}}_{b,p}^{h}}{\int}_{E}\alpha \phantom{\rule{0.166667em}{0ex}}u\phantom{\rule{0.166667em}{0ex}}v\phantom{\rule{0.166667em}{0ex}}ds\hfill \\ & -\sum _{E\in {\mathcal{E}}_{o}^{h}\cup {\mathcal{E}}_{b,g}^{h}}{\int}_{E}\left(\{k\nabla u\xb7n\}\xb7\left[v\right]+\{k\nabla v\xb7n\}\xb7\left[u\right]-\frac{{\gamma}_{f}}{h}\left\{k\right\}\left[u\right]\xb7\left[v\right]\right)\phantom{\rule{0.166667em}{0ex}}ds,\hfill \\ \hfill l\left(v\right)& =\sum _{E\in {\mathcal{E}}_{b,p}^{h}}{\int}_{E}\alpha \phantom{\rule{4pt}{0ex}}{g}_{p}\phantom{\rule{0.166667em}{0ex}}v\phantom{\rule{0.166667em}{0ex}}ds+\sum _{E\in {\mathcal{E}}_{b,g}^{h}}{\int}_{E}\left(\frac{{\gamma}_{f}}{h}\phantom{\rule{0.166667em}{0ex}}k\phantom{\rule{0.166667em}{0ex}}v-k\nabla v\xb7n\right)\phantom{\rule{0.166667em}{0ex}}{g}_{g}\phantom{\rule{0.166667em}{0ex}}dx,\phantom{\rule{1.em}{0ex}}m(u,v)=\sum _{K\in {\mathcal{T}}^{h}}{\int}_{K}\phantom{\rule{0.166667em}{0ex}}u\phantom{\rule{0.166667em}{0ex}}v\phantom{\rule{0.166667em}{0ex}}ds,\hfill \end{array}$$

The matrix form is following
for ${u}_{h}={\sum}_{i}{U}_{i}{\phi}_{i}$ and with

$$\frac{1}{\tau}{M}_{h}({U}_{h}-{\stackrel{\u02c7}{U}}_{h})+{A}_{h}\left({U}_{h}\right)={F}_{h},$$

$${M}_{h}=[{m}_{i,j}:=m({\phi}_{i},{\phi}_{j})],\phantom{\rule{1.em}{0ex}}{A}_{h}=[{a}_{i,j}:=a({\phi}_{i},{\phi}_{j})],\phantom{\rule{1.em}{0ex}}{F}_{h}=[{f}_{j}:=l\left({\phi}_{j}\right)].$$

In the multiscale method, we construct a multiscale space similar to the Laplace problem and generate a projection matrix

$$R={\left[{\phi}_{g,1}^{1},\dots ,{\phi}_{g,{M}_{g}}^{{N}_{g}},{\phi}_{p,1}^{2},\dots ,{\phi}_{p,{M}_{p}}^{{N}_{p}},{\phi}_{1},\dots ,{\phi}_{{N}_{cell}^{H}}\right]}^{Tr}.$$

Finally, we obtain the following coarse grid system for a parabolic problem
where
and calculation of the coarse-scale solution, we reconstruct the fine-scale solution ${U}_{ms}={R}^{Tr}{U}_{H}$.

$$\frac{1}{\tau}{M}_{H}({U}_{H}-{\stackrel{\u02c7}{U}}_{H})+{A}_{H}\left({U}_{H}\right)={F}_{H},$$

$${M}_{H}=R{M}_{h}{R}^{Tr},\phantom{\rule{1.em}{0ex}}{A}_{H}=R{A}_{h}{R}^{Tr},\phantom{\rule{1.em}{0ex}}{F}_{H}=R{F}_{h},$$

- Case 1. Homogeneous boundary conditions on perforations:$$u=1,\phantom{\rule{1.em}{0ex}}x\in {\mathsf{\Gamma}}_{g},\phantom{\rule{1.em}{0ex}}-k\nabla u\xb7n=\alpha u,\phantom{\rule{1.em}{0ex}}x\in {\mathsf{\Gamma}}_{p},$$
- Case 2. Non-homogeneous boundary conditions on perforations:$$u=0,\phantom{\rule{1.em}{0ex}}x\in {\mathsf{\Gamma}}_{g},\phantom{\rule{1.em}{0ex}}-k\nabla u\xb7n=\alpha (u-1),\phantom{\rule{1.em}{0ex}}x\in {\mathsf{\Gamma}}_{p}.$$

We consider three different $\alpha =1$, 25, and 100. We set zero initial condition and perform a simulation with 40 time steps.

M | ${\mathit{D}\mathit{O}\mathit{F}}_{\mathit{H}}$ | $\mathit{\alpha}=1$ | $\mathit{\alpha}=25$ | $\mathit{\alpha}=100$ | |||
---|---|---|---|---|---|---|---|

${\mathit{e}}_{{\mathit{L}}^{2}}$ | ${\mathit{e}}_{{\mathit{H}}^{1}}$ | ${\mathit{e}}_{{\mathit{L}}^{2}}$ | ${\mathit{e}}_{{\mathit{H}}^{1}}$ | ${\mathit{e}}_{{\mathit{L}}^{2}}$ | ${\mathit{e}}_{{\mathit{H}}^{1}}$ | ||

${\mathit{M}}_{\mathit{g}}=\mathit{M}$, ${\mathit{M}}_{\mathit{p}}=\mathbf{0}$ | |||||||

1 | 200 | 86.14 | 99.75 | 101.6 | 134.7 | 83.13 | 157.8 |

2 | 300 | 85.57 | 99.22 | 92.55 | 124.4 | 67.41 | 130.3 |

4 | 500 | 18.58 | 34.51 | 9.460 | 32.42 | 5.393 | 31.97 |

6 | 700 | 8.323 | 28.86 | 5.001 | 27.98 | 3.436 | 27.71 |

8 | 900 | 4.445 | 20.53 | 2.774 | 20.93 | 1.959 | 20.08 |

12 | 1300 | 2.870 | 14.32 | 1.734 | 13.33 | 1.164 | 13.10 |

16 | 1700 | 2.784 | 13.96 | 1.675 | 12.79 | 1.109 | 12.52 |

20 | 2100 | 2.693 | 13.28 | 1.635 | 12.06 | 1.076 | 11.84 |

24 | 2500 | 2.666 | 13.23 | 1.622 | 11.99 | 1.066 | 11.76 |

32 | 3300 | 2.625 | 13.09 | 1.604 | 11.79 | 1.058 | 11.56 |

The reference and multiscale solutions at the different times (${t}_{m}$ with $m=5,15$, and 40) are presented in Figure A1 for Case 1 and in Figure A2 for Case 2. The multiscale solution is obtained using 12 multiscale basis functions ($M=12$). In the first row, there is a fine-scale solution with $DO{F}_{h}=\mathrm{85,230}$. In the second row, there is a multiscale solution with $DO{F}_{H}=1300$ for ${M}_{p}=0$. In the third row, we give a multiscale solution with $DO{F}_{H}=1776$ for ${M}_{p}=M$ in Case 2. We observe a bad multiscale approximation near the perforation boundary for Case 2 with ${e}_{{L}^{2}}=22\%$ at final time, if a perforation boundary basis functions is not used to handle non-homogeneous boundary conditions. For ${M}_{p}=M$ in Case 2, we observe a good multiscale solution with ${e}_{L62}=0.7\%$ at final time.

M | ${\mathit{D}\mathit{O}\mathit{F}}_{\mathit{H}}$ | $\mathit{\alpha}=1$ | $\mathit{\alpha}=25$ | $\mathit{\alpha}=100$ | |||
---|---|---|---|---|---|---|---|

${\mathit{e}}_{{\mathit{L}}^{2}}$ | ${\mathit{e}}_{{\mathit{H}}^{1}}$ | ${\mathit{e}}_{{\mathit{L}}^{2}}$ | ${\mathit{e}}_{{\mathit{H}}^{1}}$ | ${\mathit{e}}_{{\mathit{L}}^{2}}$ | ${\mathit{e}}_{{\mathit{H}}^{1}}$ | ||

${\mathit{M}}_{\mathit{g}}=\mathit{M}$, ${\mathit{M}}_{\mathit{p}}=\mathbf{0}$ | |||||||

1 | 200 | 91.97 | 100.0 | 88.25 | 106.2 | 87.32 | 113.5 |

2 | 300 | 84.84 | 98.29 | 79.71 | 107.8 | 80.87 | 119.0 |

4 | 500 | 13.88 | 79.31 | 17.52 | 146.2 | 31.49 | 209.5 |

6 | 700 | 9.709 | 74.94 | 15.97 | 148.1 | 27.40 | 228.8 |

8 | 900 | 8.124 | 72.15 | 14.66 | 147.7 | 22.90 | 240.5 |

12 | 1300 | 7.552 | 71.39 | 14.83 | 147.6 | 22.84 | 238.9 |

16 | 1700 | 7.330 | 68.91 | 14.40 | 146.1 | 22.90 | 238.7 |

20 | 2100 | 7.289 | 68.60 | 14.41 | 145.9 | 22.96 | 238.4 |

24 | 2500 | 7.280 | 68.49 | 14.42 | 145.8 | 22.99 | 238.2 |

32 | 3300 | 7.264 | 68.39 | 14.42 | 145.8 | 23.01 | 238.1 |

${\mathit{M}}_{\mathit{g}}={\mathit{M}}_{\mathit{p}}=\mathit{M}$ | |||||||

1 | 251 | 85.93 | 79.07 | 77.39 | 103.4 | 63.56 | 116.3 |

2 | 402 | 75.40 | 75.78 | 57.40 | 99.07 | 38.96 | 101.3 |

4 | 704 | 7.938 | 38.96 | 5.154 | 35.27 | 6.326 | 36.65 |

6 | 1006 | 4.084 | 29.37 | 2.857 | 27.33 | 2.897 | 26.93 |

8 | 1308 | 2.339 | 21.96 | 1.566 | 20.95 | 1.394 | 19.50 |

12 | 1776 | 1.289 | 14.29 | 0.819 | 13.35 | 0.742 | 12.92 |

16 | 2244 | 1.192 | 13.19 | 0.735 | 12.54 | 0.654 | 12.24 |

20 | 2644 | 1.129 | 12.19 | 0.701 | 11.78 | 0.619 | 11.59 |

24 | 3044 | 1.117 | 11.90 | 0.691 | 11.61 | 0.603 | 11.49 |

32 | 3844 | 1.093 | 11.59 | 0.677 | 11.36 | 0.588 | 11.28 |

M | ${\mathit{D}\mathit{O}\mathit{F}}_{\mathit{H}}$ | Quasi-Structured | Unstructured | ||
---|---|---|---|---|---|

${\mathit{e}}_{{\mathit{L}}^{2}}$ | ${\mathit{e}}_{{\mathit{H}}^{1}}$ | ${\mathit{e}}_{{\mathit{L}}^{2}}$ | ${\mathit{e}}_{{\mathit{H}}^{1}}$ | ||

1 | 251 | 54.82 | 96.18 | 53.32 | 93.48 |

2 | 402 | 30.56 | 79.76 | 41.23 | 94.00 |

4 | 704 | 8.184 | 45.51 | 11.60 | 56.13 |

6 | 1006 | 4.889 | 34.72 | 6.561 | 41.29 |

8 | 1308 | 3.434 | 25.08 | 4.976 | 35.25 |

12 | 1776 | 3.944 | 23.14 | 3.725 | 27.85 |

16 | 2244 | 1.526 | 17.72 | 2.761 | 23.90 |

20 | 2644 | 1.209 | 16.25 | 2.063 | 21.00 |

24 | 3044 | 0.999 | 15.26 | 1.775 | 18.64 |

32 | 3844 | 0.803 | 14.43 | 1.429 | 16.74 |

The relative errors for parabolic problem (Case 1) between reference solution and multiscale solution for $\alpha =1,25$, and 100 are shown in Table A1 and in Figure A3 (left picture). We have a good multiscale solution with nearly 1–2% of ${L}^{2}$ error when we take 12 outer boundary multiscale basis functions. Table A2 and Figure A3 (right picture) present relative errors for Case 2. Similar to the previous results for the elliptic equation, we see large errors for ${M}_{p}=0$ and can obtain a good multiscale solution when we take the sufficient number of multiscale basis functions with outer and perforation boundary multiscale basis functions. For example, if 12 multiscale basis functions ($M=12$) are taken, we have 7–22% of ${L}^{2}$ errors for ${M}_{p}=0$ and reduce errors to $1\%$ when we add perforation boundary basis functions for $\alpha =1,25$, and 100.

Table A3 shows results for non-homogeneous boundary conditions (Case 2) on quasi-structured and unstructured coarse grids with 100 local domains. We present results for $\alpha =100$. Using 32 multiscale basis functions (${M}_{p}={M}_{g}=32$), we obtain similar results with less than $1\%$ of ${L}^{2}$ error for structured, quasi-structured, and unstructured coarse grids with 100 local domains with $DO{F}_{H}=3844$ ($4.5\%$ of $DO{F}_{f}$ = 85,230).

We consider the thermoelasticity problem in $\mathsf{\Omega}$ that is described by a system of equations for temperature and displacement:

$$\begin{array}{c}\hfill \nabla \xb7\sigma \left(u\right)-\beta \nabla (T-{T}_{\ast})=0,\phantom{\rule{1.em}{0ex}}x\in \mathsf{\Omega},\\ \hfill c\frac{\partial T}{\partial t}+\beta \frac{\partial \nabla \xb7u}{\partial t}-\nabla \xb7(k\nabla T)=0,\phantom{\rule{1.em}{0ex}}x\in \mathsf{\Omega},\end{array}$$

We consider Equation (A2) with the following boundary conditions:
and initial condition $T={T}_{0}$, $u=0$ in $\mathsf{\Omega}$ for $t=0$.

$$u={g}_{g}^{u},\phantom{\rule{1.em}{0ex}}T={g}_{g}^{T},\phantom{\rule{1.em}{0ex}}x\in {\mathsf{\Gamma}}_{g},\phantom{\rule{1.em}{0ex}}\sigma \xb7n={g}_{p}^{u},\phantom{\rule{1.em}{0ex}}-k\nabla T\xb7n=\alpha (T-{g}_{p}^{T}),\phantom{\rule{1.em}{0ex}}x\in {\mathsf{\Gamma}}_{p},$$

For the IPDG method on the fine grid ${\mathcal{T}}^{h}$, we have the following variational formulation: find $({u}_{h},{T}_{h})\in {W}_{h}\times {Q}_{h}$ such that
where
with

$$\begin{array}{cc}& {a}^{u}({u}_{h},v)+b({T}_{h},v)={l}^{u}\left(v\right),\phantom{\rule{1.em}{0ex}}\forall v\in {W}_{h}\hfill \\ & \frac{1}{\tau}m({T}_{h}-{\stackrel{\u02c7}{T}}_{h},q)+\frac{1}{\tau}d({u}_{h}-{\stackrel{\u02c7}{u}}_{h},q)+{a}^{T}({T}_{h},q)={l}^{T}\left(q\right),\phantom{\rule{1.em}{0ex}}\forall q\in {Q}_{h}\hfill \end{array},$$

$${Q}_{h}=\{T\in {L}^{2}\left(\mathsf{\Omega}\right):\phantom{\rule{0.166667em}{0ex}}T{|}_{K}\in {\mathbb{P}}_{1}\left(K\right),\phantom{\rule{0.166667em}{0ex}}\forall K\in {\mathcal{T}}^{h}\}$$

$${W}_{h}=\{v\in {\left[{L}^{2}\left(\mathsf{\Omega}\right)\right]}^{2}:\phantom{\rule{0.166667em}{0ex}}v{|}_{K}\in {\left[{\mathbb{P}}_{1}\left(K\right)\right]}^{2},\phantom{\rule{0.166667em}{0ex}}\forall K\in {\mathcal{T}}^{h}\},$$

$$\begin{array}{cc}\hfill {a}^{T}(T,q)& =\sum _{K\in {\mathcal{T}}^{h}}{\int}_{K}(k\nabla T,\nabla q)\phantom{\rule{0.166667em}{0ex}}dx+\sum _{E\in {\mathcal{E}}_{b,p}^{h}}{\int}_{E}\alpha \phantom{\rule{0.166667em}{0ex}}T\phantom{\rule{0.166667em}{0ex}}q\phantom{\rule{0.166667em}{0ex}}ds\hfill \\ & -\sum _{E\in {\mathcal{E}}_{o}^{h}\cup {\mathcal{E}}_{b,g}^{h}}{\int}_{E}\left(\{k\nabla T\xb7n\}\xb7\left[q\right]+\{k\nabla q\xb7n\}\xb7\left[T\right]-\frac{{\gamma}_{f}}{h}\left\{k\right\}\left[T\right]\xb7\left[q\right]\right)\phantom{\rule{0.166667em}{0ex}}ds,\hfill \\ \hfill l\left(q\right)& =\sum _{E\in {\mathcal{E}}_{b,p}^{h}}{\int}_{E}\alpha \phantom{\rule{4pt}{0ex}}{g}_{p}^{T}\phantom{\rule{0.166667em}{0ex}}q\phantom{\rule{0.166667em}{0ex}}ds+\sum _{E\in {\mathcal{E}}_{b,g}^{h}}{\int}_{E}\left(\frac{{\gamma}_{f}}{h}\phantom{\rule{0.166667em}{0ex}}k\phantom{\rule{0.166667em}{0ex}}q-k\nabla q\xb7n\right)\phantom{\rule{0.166667em}{0ex}}{g}_{g}^{T}\phantom{\rule{0.166667em}{0ex}}dx,\hfill \\ \hfill m(T,q)& =\sum _{K\in {\mathcal{T}}^{h}}{\int}_{K}c\phantom{\rule{0.166667em}{0ex}}T\phantom{\rule{0.166667em}{0ex}}q\phantom{\rule{0.166667em}{0ex}}dx,\hfill \\ \hfill {a}^{u}(u,v)& =\sum _{K\in {\mathcal{T}}^{h}}{\int}_{K}(\sigma \left(u\right),\epsilon \left(v\right))dx\hfill \\ & -\sum _{E\in {\mathcal{E}}_{o}^{h}\cup {\mathcal{E}}_{b,g}^{h}}{\int}_{E}\left(\left\{\tau \left(u\right)\right\}\left[v\right]+\left\{\tau \left(v\right)\right\}\left[u\right]-\frac{{\gamma}_{f}}{h}\{\lambda +2\mu \}\left[u\right]\left[v\right]\right)\phantom{\rule{0.166667em}{0ex}}ds,\hfill \\ \hfill {l}^{u}\left(v\right)& =\sum _{E\in {\mathcal{E}}_{b,p}^{h}}{\int}_{E}{g}_{p}^{u}\xb7v\phantom{\rule{0.166667em}{0ex}}ds+\sum _{E\in {\mathcal{E}}_{b,g}^{h}}{\int}_{E}\left(\frac{{\gamma}_{f}}{h}\phantom{\rule{0.166667em}{0ex}}(\lambda +2\mu )\phantom{\rule{0.166667em}{0ex}}v-\tau \left(v\right)\right)\xb7{g}_{g}^{u}\phantom{\rule{0.166667em}{0ex}}ds,\hfill \\ \hfill b(T,v)& =\sum _{K\in {\mathcal{T}}^{h}}{\int}_{K}\beta \phantom{\rule{4pt}{0ex}}\nabla T\xb7v\phantom{\rule{4pt}{0ex}}dx,\phantom{\rule{1.em}{0ex}}d(u,q)=\sum _{K\in {\mathcal{T}}^{h}}{\int}_{K}\phantom{\rule{0.166667em}{0ex}}\beta \phantom{\rule{0.166667em}{0ex}}\nabla \xb7u\phantom{\rule{0.166667em}{0ex}}q\phantom{\rule{0.166667em}{0ex}}dx.\hfill \end{array}$$

The matrix form is the following:
for
with

$$\begin{array}{cc}& {A}_{h}^{u}\left({U}_{h}\right)+{B}_{h}\left({T}_{h}\right)={F}_{h}^{u}\hfill \\ & \frac{1}{\tau}{M}_{h}({T}_{h}-{\stackrel{\u02c7}{T}}_{h})+\frac{1}{\tau}{D}_{h}({U}_{h}-{\stackrel{\u02c7}{U}}_{h})+{A}_{h}^{T}\left({T}_{h}\right)={F}_{h}^{T},\hfill \end{array}$$

$${u}_{h}=\sum _{i}{U}_{i}{\psi}_{i};\phantom{\rule{1.em}{0ex}}{T}_{h}=\sum _{i}{T}_{i}{\phi}_{i},$$

$${M}_{h}=[{m}_{i,j}:=m({\phi}_{i},{\phi}_{j})],\phantom{\rule{1.em}{0ex}}{A}_{h}^{T}=[{a}_{i,j}^{T}:={a}^{T}({\phi}_{i},{\phi}_{j})],\phantom{\rule{1.em}{0ex}}{A}_{h}^{u}=[{a}_{i,j}^{u}:={a}^{u}({\psi}_{i},{\psi}_{j})],$$

$${B}_{h}=[{b}_{i,j}:=b({\psi}_{i},{\phi}_{j})],\phantom{\rule{1.em}{0ex}}{D}_{h}=[{d}_{i,j}:=d({\phi}_{i},{\psi}_{j})],$$

$${F}_{h}^{u}=[{f}_{j}^{u}:={l}^{u}\left({\psi}_{j}\right)],\phantom{\rule{1.em}{0ex}}{F}_{h}^{T}=[{f}_{j}^{T}:={l}^{T}\left({\phi}_{j}\right)].$$

For the construction of the multiscale method for the thermoelasticity problem, we construct multiscale spaces for displacement and temperature, separately. Then, we generate projection matrices using multiscale basis functions

$${R}_{u}={\left[{\psi}_{g,1}^{1},\dots ,{\psi}_{g,{M}_{g}^{u}}^{{N}_{g}},{\psi}_{p,1}^{2},\dots ,{\psi}_{p,{M}_{p}^{u}}^{{N}_{p}},{\varphi}_{1},\dots ,{\varphi}_{{N}_{cell}^{H}}\right]}^{Tr},$$

$${R}_{T}={\left[{\phi}_{g,1}^{1},\dots ,{\phi}_{g,{M}_{g}^{T}}^{{N}_{g}},{\phi}_{p,1}^{2},\dots ,{\phi}_{p,{M}_{p}^{T}}^{{N}_{p}},{\phi}_{1},\dots ,{\phi}_{{N}_{cell}^{H}}\right]}^{Tr}.$$

Finally, we construct the following coarse grid system for the thermoelasticity problem
where

$$\begin{array}{cc}& {A}_{H}^{u}\left({U}_{H}\right)+{B}_{H}\left({T}_{H}\right)={F}_{H}^{u}\hfill \\ & \frac{1}{\tau}{M}_{H}({T}_{H}-{\stackrel{\u02c7}{T}}_{H})+\frac{1}{\tau}{D}_{H}({U}_{H}-{\stackrel{\u02c7}{U}}_{H})+{A}_{H}^{T}\left({T}_{H}\right)={F}_{H}^{T},\hfill \end{array}$$

$${M}_{H}={R}_{T}{M}_{h}{R}_{T}^{Tr},\phantom{\rule{1.em}{0ex}}{A}_{H}^{T}={R}_{T}{A}_{h}^{T}{R}_{T}^{Tr},\phantom{\rule{1.em}{0ex}}{A}_{H}^{u}={R}_{u}{A}_{h}^{u}{R}_{u}^{Tr},$$

$${B}_{H}={R}_{u}{B}_{h}{R}_{T}^{Tr},\phantom{\rule{1.em}{0ex}}{D}_{H}={R}_{T}{D}_{h}{R}_{u}^{Tr},\phantom{\rule{1.em}{0ex}}{F}_{H}^{u}={R}_{u}{F}_{h}^{u},\phantom{\rule{1.em}{0ex}}{F}_{H}^{T}={R}_{T}{F}_{h}^{T}.$$

After calculation of the coarse-scale solution ${U}_{h},{T}_{H}$, we reconstruct a fine-scale solution, ${U}_{ms}={R}_{u}^{Tr}{U}_{H}$ and ${T}_{ms}={R}_{T}^{Tr}{T}_{H}$.

$$\begin{array}{cc}& {u}_{x}=0,\phantom{\rule{1.em}{0ex}}{\sigma}_{y}=0,\phantom{\rule{1.em}{0ex}}x\in {\mathsf{\Gamma}}_{l}\cup {\mathsf{\Gamma}}_{r},\phantom{\rule{1.em}{0ex}}{u}_{y}=0,\phantom{\rule{1.em}{0ex}}{\sigma}_{x}=0,\phantom{\rule{1.em}{0ex}}x\in {\mathsf{\Gamma}}_{b}\cup {\mathsf{\Gamma}}_{t},\hfill \\ & -k\frac{\partial T}{\partial n}=0,\phantom{\rule{1.em}{0ex}}x\in {\mathsf{\Gamma}}_{l}\cup {\mathsf{\Gamma}}_{t}\cup {\mathsf{\Gamma}}_{r}\cup {\mathsf{\Gamma}}_{b},\hfill \end{array}$$

- Case 1. Non-homogeneous boundary condition for elasticity and homogeneous boundary condition for temperature on perforations:$$\sigma \xb7n=-0.01,\phantom{\rule{1.em}{0ex}}-k\frac{\partial T}{\partial n}=\alpha T,\phantom{\rule{1.em}{0ex}}x\in {\mathsf{\Gamma}}_{p},$$
- Case 2. Homogeneous boundary condition for elasticity and non-homogeneous boundary condition for temperature on perforations:$$\sigma \xb7n=0,\phantom{\rule{1.em}{0ex}}-k\frac{\partial T}{\partial n}=\alpha (T-1),\phantom{\rule{1.em}{0ex}}x\in {\mathsf{\Gamma}}_{p},$$

We set $\lambda =\mu =k=c=1$, ${T}_{\ast}=0$, $\beta =0.0001$, and $\alpha =25$. We set zero initial conditions for displacement and temperature and perform simulations with 40 time steps.

Table A4 and Figure A4 show relative errors for the thermoelasticity problem between a fine-grid and multiscale solution with different multiscale basis functions at the final time. In presented results, we tested different approaches for choosing ${M}_{g}^{u},{M}_{p}^{u}$ and ${M}_{g}^{T},{M}_{p}^{T}$ for the coupled process. Here, ${M}^{v}$ are the number of multiscale basis functions for displacement ($v=u$) and temperature ($v=T$), $DO{F}_{h}^{v}$ and $DO{F}_{H}^{v}$ are the numbers of degrees of freedom for reference (fine-grid) and multiscale solution, and $e{\left(v\right)}_{{L}^{2}}$ and $e{\left(v\right)}_{{H}^{1}}$ are the ${L}^{2}$ and ${H}^{1}$ relative errors for $v=u,T$. Here, $DO{F}_{h}=DO{F}_{h}^{u}+DO{F}_{h}^{T}$ with $DO{F}_{h}^{u}=6\xb7{N}_{cell}^{h}$ and $DO{F}_{h}^{T}=3\xb7{N}_{cell}^{h}$ for fine grid solution. For multiscale solution, we have $DO{F}_{H}=DO{F}_{H}^{u}+DO{F}_{H}^{T}$ with $DO{F}_{H}^{u}=(2\xb7{M}_{g}^{u}+1)\xb7{N}_{cell}^{H}+2\xb7{M}_{p}^{u}\xb7{N}_{cell,p}^{H}$ and $DO{F}_{H}^{T}=({M}_{g}^{T}+1)\xb7{N}_{cell}^{H}+{M}_{p}^{T}\xb7{N}_{cell,p}^{H}$.

The results for quasi-structured and unstructured coarse grids with 100 local domains are presented in Table A5. Numerical results are shown for multiscale space with outer and boundary basis functions (${M}_{g}={M}_{p}=M$) for Case 2. We obtain similar results with $4\%$ and $0.3\%$ for displacement and temperature ${L}^{2}$ errors for all grids for $M=32$.

Figure A1 and Figure A2 show the results of the fine-grid and multiscale solvers for the thermoelasticity problem at the final time on structured coarse grid. The temperature distribution is presented on the left figure, the displacement X and Y components are shown in the center and right figures. In presented results, we used 24 multiscale basis functions for displacement and temperature (${M}^{u}={M}^{T}=24$, ${M}_{g}^{u}={M}_{p}^{u}={M}^{u}$, and ${M}_{g}^{T}={M}_{p}^{T}={M}^{T}$). Reference solution with $DO{F}_{h}^{u}=\mathrm{170,460}$ for displacement, $DO{F}_{h}^{T}=\mathrm{85,230}$ for temperature is shown in the first row. Solution on a coarse grid using outer and perforation boundary multiscale basis functions with $DO{F}_{H}^{u}=5988$ for displacement and $DO{F}_{H}^{T}=3044$ for temperature is shown in the second row.

M | ${\mathit{DOF}}_{\mathit{H}}^{\mathit{u}}$ | ${\mathit{DOF}}_{\mathit{H}}^{\mathit{T}}$ | Case 1 | Case 2 | ||||||
---|---|---|---|---|---|---|---|---|---|---|

$\mathit{e}{\left(\mathit{u}\right)}_{{\mathit{L}}^{\mathbf{2}}}$ | $\mathit{e}{\left(\mathit{u}\right)}_{{\mathit{H}}^{\mathbf{1}}}$ | $\mathit{e}{\left(\mathit{T}\right)}_{{\mathit{L}}^{\mathbf{2}}}$ | $\mathit{e}{\left(\mathit{T}\right)}_{{\mathit{H}}^{\mathbf{1}}}$ | $\mathit{e}{\left(\mathit{u}\right)}_{{\mathit{L}}^{\mathbf{2}}}$ | $\mathit{e}{\left(\mathit{u}\right)}_{{\mathit{H}}^{\mathbf{1}}}$ | $\mathit{e}{\left(\mathit{T}\right)}_{{\mathit{L}}^{\mathbf{2}}}$ | $\mathit{e}{\left(\mathit{T}\right)}_{{\mathit{H}}^{\mathbf{1}}}$ | |||

${\mathit{M}}_{\mathit{g}}^{\mathit{u}}={\mathit{M}}^{\mathit{u}}$, ${\mathit{M}}_{\mathit{p}}^{\mathit{u}}=\mathbf{0}$ and ${\mathit{M}}_{\mathit{g}}^{\mathit{T}}={\mathit{M}}^{\mathit{T}}$, ${\mathit{M}}_{\mathit{p}}^{\mathit{T}}=\mathbf{0}$ | ||||||||||

1 | 300 | 200 | 98.00 | 99.98 | 100.0 | 100.0 | 99.32 | 98.98 | 86.48 | 105.9 |

2 | 500 | 300 | 92.01 | 98.38 | 100.6 | 100.4 | 98.04 | 97.07 | 74.55 | 106.3 |

4 | 900 | 500 | 72.44 | 93.63 | 100.9 | 98.37 | 90.47 | 73.54 | 14.16 | 142.4 |

6 | 1300 | 700 | 36.91 | 87.77 | 100.3 | 93.44 | 90.01 | 55.30 | 13.32 | 143.3 |

8 | 1700 | 900 | 32.73 | 85.34 | 95.06 | 87.08 | 91.23 | 50.41 | 12.36 | 143.4 |

12 | 2500 | 1300 | 31.88 | 84.32 | 93.31 | 85.43 | 91.41 | 48.40 | 12.25 | 142.2 |

16 | 3300 | 1700 | 31.03 | 83.99 | 93.04 | 85.05 | 91.85 | 47.36 | 12.26 | 142.1 |

20 | 4100 | 2100 | 30.86 | 83.82 | 92.43 | 84.44 | 91.79 | 47.07 | 12.28 | 142.0 |

24 | 4900 | 2500 | 30.37 | 84.07 | 92.89 | 84.64 | 91.69 | 47.56 | 12.30 | 141.9 |

32 | 6500 | 3300 | 29.31 | 82.66 | 89.80 | 81.68 | 91.67 | 45.92 | 12.31 | 141.8 |

${\mathit{M}}_{\mathit{g}}^{\mathit{u}}={\mathit{M}}^{\mathit{u}}$, ${\mathit{M}}_{\mathit{p}}^{\mathit{u}}=\mathbf{0}$ and ${\mathit{M}}_{\mathit{g}}^{\mathit{T}}={\mathit{M}}_{\mathit{p}}^{\mathit{T}}={\mathit{M}}^{\mathit{T}}$ | ||||||||||

1 | 300 | 251 | 98.00 | 99.98 | 99.98 | 100.0 | 99.28 | 97.66 | 77.45 | 123.9 |

2 | 500 | 402 | 92.01 | 98.38 | 100.9 | 100.3 | 96.78 | 94.60 | 56.93 | 110.3 |

4 | 900 | 704 | 72.44 | 93.63 | 101.1 | 96.89 | 93.48 | 70.98 | 3.306 | 31.18 |

6 | 1300 | 1006 | 36.91 | 87.77 | 101.0 | 89.01 | 99.56 | 52.87 | 1.984 | 25.50 |

8 | 1700 | 1308 | 32.73 | 85.34 | 95.78 | 82.69 | 100.7 | 48.14 | 1.057 | 19.77 |

12 | 2500 | 1776 | 31.88 | 84.32 | 93.97 | 80.77 | 101.3 | 46.12 | 0.441 | 12.77 |

16 | 3300 | 2244 | 31.03 | 83.99 | 93.70 | 80.21 | 102.1 | 45.10 | 0.406 | 12.17 |

20 | 4100 | 2644 | 30.86 | 83.82 | 93.10 | 79.59 | 102.1 | 44.78 | 0.373 | 11.40 |

24 | 4900 | 3044 | 30.37 | 84.07 | 93.56 | 79.69 | 102.0 | 45.30 | 0.374 | 11.34 |

32 | 6500 | 3844 | 29.31 | 82.66 | 90.46 | 76.90 | 102.0 | 43.54 | 0.369 | 11.10 |

${\mathit{M}}_{\mathit{g}}^{\mathit{u}}={\mathit{M}}_{\mathit{p}}^{\mathit{u}}={\mathit{M}}^{\mathit{u}}$ and ${\mathit{M}}_{\mathit{g}}^{\mathit{T}}={\mathit{M}}^{\mathit{T}}$, ${\mathit{M}}_{\mathit{p}}^{\mathit{T}}=\mathbf{0}$ | ||||||||||

1 | 402 | 200 | 90.81 | 68.11 | 91.36 | 96.15 | 99.42 | 95.38 | 86.48 | 105.9 |

2 | 704 | 300 | 82.37 | 57.77 | 82.90 | 91.90 | 97.80 | 90.97 | 74.55 | 106.3 |

4 | 1308 | 500 | 45.43 | 40.88 | 32.78 | 53.70 | 80.07 | 71.81 | 14.16 | 142.4 |

6 | 1912 | 700 | 12.48 | 29.04 | 20.93 | 35.71 | 69.01 | 63.08 | 13.32 | 143.3 |

8 | 2516 | 900 | 7.990 | 21.90 | 12.90 | 25.77 | 69.64 | 61.46 | 12.36 | 143.4 |

12 | 3452 | 1300 | 6.163 | 17.55 | 9.528 | 20.61 | 65.62 | 56.87 | 12.25 | 142.2 |

16 | 4388 | 1700 | 3.989 | 16.40 | 8.771 | 18.66 | 66.02 | 54.52 | 12.26 | 142.1 |

20 | 5188 | 2100 | 3.692 | 15.82 | 8.873 | 17.36 | 68.14 | 48.12 | 12.28 | 142.0 |

24 | 5988 | 2500 | 3.255 | 15.99 | 8.274 | 16.95 | 64.84 | 46.52 | 12.30 | 141.9 |

32 | 7588 | 3300 | 2.283 | 12.90 | 7.604 | 16.74 | 59.34 | 43.95 | 12.31 | 141.8 |

${\mathit{M}}_{\mathit{g}}^{\mathit{u}}={\mathit{M}}_{\mathit{p}}^{\mathit{u}}={\mathit{M}}^{\mathit{u}}$and ${\mathit{M}}_{\mathit{g}}^{\mathit{T}}={\mathit{M}}_{\mathit{p}}^{\mathit{T}}={\mathit{M}}^{\mathit{T}}$ | ||||||||||

1 | 402 | 251 | 90.81 | 68.11 | 90.19 | 102.7 | 98.54 | 88.17 | 77.45 | 123.9 |

2 | 704 | 402 | 82.37 | 57.77 | 77.77 | 99.16 | 95.03 | 80.46 | 56.93 | 110.3 |

4 | 1308 | 704 | 45.43 | 40.88 | 31.77 | 54.53 | 62.22 | 56.58 | 3.306 | 31.18 |

6 | 1912 | 1006 | 12.48 | 29.04 | 19.71 | 35.62 | 12.12 | 34.48 | 1.984 | 25.50 |

8 | 2516 | 1308 | 7.990 | 21.90 | 11.79 | 24.15 | 9.885 | 29.21 | 1.057 | 19.77 |

12 | 3452 | 1776 | 6.163 | 17.55 | 8.775 | 19.12 | 7.687 | 25.99 | 0.441 | 12.77 |

16 | 4388 | 2244 | 3.989 | 16.40 | 7.970 | 17.15 | 5.304 | 24.32 | 0.406 | 12.17 |

20 | 5188 | 2644 | 3.692 | 15.82 | 7.535 | 16.35 | 5.153 | 23.90 | 0.373 | 11.40 |

24 | 5988 | 3044 | 3.255 | 15.99 | 8.087 | 16.37 | 5.023 | 24.03 | 0.374 | 11.34 |

32 | 7588 | 3844 | 2.283 | 12.90 | 7.530 | 15.66 | 4.698 | 22.09 | 0.369 | 11.10 |

M | ${\mathit{DOF}}_{\mathit{H}}^{\mathit{u}}$ | ${\mathit{DOF}}_{\mathit{H}}^{\mathit{T}}$ | Quasi-Structured | Unstructured | ||||||
---|---|---|---|---|---|---|---|---|---|---|

$\mathit{e}{\left(\mathit{u}\right)}_{{\mathit{L}}^{\mathbf{2}}}$ | $\mathit{e}{\left(\mathit{u}\right)}_{{\mathit{H}}^{\mathbf{1}}}$ | $\mathit{e}{\left(\mathit{T}\right)}_{{\mathit{L}}^{\mathbf{2}}}$ | $\mathit{e}{\left(\mathit{T}\right)}_{{\mathit{H}}^{\mathbf{1}}}$ | $\mathit{e}{\left(\mathit{u}\right)}_{{\mathit{L}}^{\mathbf{2}}}$ | $\mathit{e}{\left(\mathit{u}\right)}_{{\mathit{H}}^{\mathbf{1}}}$ | $\mathit{e}{\left(\mathit{T}\right)}_{{\mathit{L}}^{\mathbf{2}}}$ | $\mathit{e}{\left(\mathit{T}\right)}_{{\mathit{H}}^{\mathbf{1}}}$ | |||

1 | 402 | 251 | 98.47 | 91.35 | 76.32 | 103.8 | 98.44 | 93.52 | 72.88 | 95.42 |

2 | 704 | 402 | 94.78 | 82.68 | 52.28 | 92.76 | 94.43 | 86.17 | 54.50 | 93.85 |

4 | 1308 | 704 | 72.68 | 64.89 | 6.404 | 37.24 | 79.52 | 73.01 | 15.27 | 53.38 |

6 | 1912 | 1006 | 51.18 | 54.39 | 3.044 | 32.04 | 62.35 | 63.75 | 7.285 | 37.87 |

8 | 2516 | 1308 | 21.07 | 36.85 | 1.496 | 22.45 | 46.90 | 55.00 | 3.408 | 30.38 |

12 | 3452 | 1776 | 13.06 | 31.33 | 0.877 | 18.19 | 21.52 | 39.67 | 1.495 | 23.10 |

16 | 4388 | 2244 | 8.606 | 27.89 | 0.667 | 16.36 | 12.78 | 32.80 | 0.941 | 20.06 |

20 | 5188 | 2644 | 6.745 | 26.25 | 0.543 | 15.19 | 8.371 | 29.68 | 0.667 | 18.02 |

24 | 5988 | 3044 | 5.701 | 25.20 | 0.462 | 14.47 | 6.383 | 28.10 | 0.500 | 16.17 |

32 | 7588 | 3844 | 4.834 | 24.01 | 0.395 | 13.68 | 4.941 | 26.57 | 0.384 | 14.81 |

Case 1 with non-homogeneous boundary conditions for elasticity and homogeneous boundary conditions for temperature on perforations shows that, for displacement, we should use perforation boundary basis functions in order to approximate the non- homogeneous boundary condition for the elasticity part of the coupled system. For temperature, only outer boundary basis functions can be used because we have homogeneous boundary conditions for temperature. For Case 1 with ${M}_{g}^{u}={M}_{p}^{u}=24$, ${M}_{g}^{T}=24$, and ${M}_{p}^{T}=0$, we obtain a good results with $3\%$ and $8\%$ of ${L}^{2}$ errors for displacement and temperature, respectively. When we use only outer boundary basis functions for displacement (${M}_{p}^{u}=0$), we obtain a large error for both displacement and temperature. In Case 2 with homogeneous boundary conditions for elasticity and non-homogeneous boundary conditions for temperature on perforations, we can obtain good results with outer and perforation boundary basis functions for both displacement and temperature (${M}_{g}^{u}={M}_{p}^{u}={M}^{u}$ and ${M}_{g}^{T}={M}_{p}^{T}={M}^{T}$). It should be noted that perforation boundary basis functions should be added for displacement due to the coupled process for accurate multiscale approximation even for non-homogeneous boundary conditions for elasticity on perforations.

- Heinze, S. Wave solutions to reaction-diffusion systems in perforated domains. Z. Anal. Und Ihre Anwendungen
**2001**, 20, 661–676. [Google Scholar] [CrossRef] - Amaziane, B.; Goncharenko, M.; Pankratov, L. Homogenization of a convection–diffusion equation in perforated domains with a weak adsorption. Z. Angew. Math. Und Phys.
**2007**, 58, 592–611. [Google Scholar] [CrossRef] - Höhner, D.; Wirtz, S.; Scherer, V. A study on the influence of particle shape on the mechanical interactions of granular media in a hopper using the Discrete Element Method. Powder Technol.
**2015**, 278, 286–305. [Google Scholar] [CrossRef] - Sadeghi, R.; Shadloo, M.S.; Hopp-Hirschler, M.; Hadjadj, A.; Nieken, U. Three-dimensional lattice Boltzmann simulations of high density ratio two-phase flows in porous media. Comput. Math. Appl.
**2018**, 75, 2445–2465. [Google Scholar] [CrossRef] - Korneev, S.; Battiato, I. Sequential homogenization of reactive transport in polydisperse porous media. Multiscale Model. Simul.
**2016**, 14, 1301–1318. [Google Scholar] [CrossRef] - Iliev, O.; Lakdawala, Z.; Neßler, K.H.; Prill, T.; Vutov, Y.; Yang, Y.; Yao, J. On the pore-scale modeling and simulation of reactive transport in 3D geometries. Math. Model. Anal.
**2017**, 22, 671–694. [Google Scholar] [CrossRef] - Allaire, G.; Mikelić, A.; Piatnitski, A. Homogenization approach to the dispersion theory for reactive transport through porous media. Siam J. Math. Anal.
**2010**, 42, 125–144. [Google Scholar] [CrossRef] - Sánchez-Palencia, E. Non-Homogeneous Media and Vibration Theory; Lecture Notes in Physics; Springer: Berlin, Germany, 1980; Volume 127. [Google Scholar]
- Bakhvalov, N.S.; Panasenko, G. Homogenisation: Averaging Processes in Periodic Media: Mathematical Problems in the Mechanics of Composite Materials; Springer Science & Business Media: Dordrecht, The Netherlands, 2012; Volume 36. [Google Scholar]
- Allaire, G. Homogenization of the Navier-Stokes equations in open sets perforated with tiny holes II: Non-critical sizes of the holes for a volume distribution and a surface distribution of holes. Arch. Ration. Mech. Anal.
**1991**, 113, 261–298. [Google Scholar] [CrossRef] - Jing, W. A unified homogenization approach for the Dirichlet problem in perforated domains. SIAM J. Math. Anal.
**2020**, 52, 1192–1220. [Google Scholar] [CrossRef] - Vasilyeva, M.; Tyrylgin, A. Convolutional neural network for fast prediction of the effective properties of domains with random inclusions. J. Phys.
**2019**, 1158, 042034. [Google Scholar] [CrossRef] - Vasilyeva, M.; Tyrylgin, A. Machine learning for accelerating macroscopic parameters prediction for poroelasticity problem in stochastic media. Comput. Math. Appl.
**2021**, 84, 185–202. [Google Scholar] [CrossRef] - Efendiev, Y.; Hou, T.Y. Multiscale Finite Element Methods: Theory and Applications; Springer Science & Business Media: New York, NY, USA, 2009; Volume 4. [Google Scholar]
- Hou, T.Y.; Wu, X.H. A multiscale finite element method for elliptic problems in composite materials and porous media. J. Comput. Phys.
**1997**, 134, 169–189. [Google Scholar] [CrossRef] - Muljadi, B.P.; Narski, J.; Lozinski, A.; Degond, P. Nonconforming multiscale finite element method for stokes flows in heterogeneous media. Part I: Methodologies and numerical experiments. Multiscale Model. Simul.
**2015**, 13, 1146–1172. [Google Scholar] [CrossRef] - Le Bris, C.; Legoll, F.; Lozinski, A. An MsFEM type approach for perforated domains. Multiscale Model. Simul.
**2014**, 12, 1046–1077. [Google Scholar] [CrossRef] - Henning, P.; Ohlberger, M. The heterogeneous multiscale finite element method for elliptic homogenization problems in perforated domains. Numer. Math.
**2009**, 113, 601–629. [Google Scholar] [CrossRef] - Chung, E.T.; Efendiev, Y.; Li, G.; Vasilyeva, M. Generalized multiscale finite element methods for problems in perforated heterogeneous domains. Appl. Anal.
**2016**, 95, 2254–2279. [Google Scholar] [CrossRef] - Chung, E.T.; Efendiev, Y.; Leung, W.T.; Vasilyeva, M.; Wang, Y. Online adaptive local multiscale model reduction for heterogeneous problems in perforated domains. Appl. Anal.
**2017**, 96, 2002–2031. [Google Scholar] [CrossRef] - Chung, E.T.; Leung, W.T.; Vasilyeva, M.; Wang, Y. Multiscale model reduction for transport and flow problems in perforated domains. J. Comput. Appl. Math.
**2018**, 330, 519–535. [Google Scholar] [CrossRef] - Chung, E.T.; Efendiev, Y.; Vasilyeva, M.; Wang, Y. A multiscale discontinuous Galerkin method in perforated domains. In Proceedings of the Institute of Mathematics and Mechanics, Maringá, Brazil, 22–23 September 2016; Volume 42, pp. 212–229. [Google Scholar]
- Spiridonov, D.; Vasilyeva, M.; Leung, W.T. A Generalized Multiscale Finite Element Method (GMsFEM) for perforated domain flows with Robin boundary conditions. J. Comput. Appl. Math.
**2019**, 357, 319–328. [Google Scholar] [CrossRef] - Vasilyeva, M.; Chung, E.T.; Leung, W.T.; Wang, Y.; Spiridonov, D. Upscaling method for problems in perforated domains with non-homogeneous boundary conditions on perforations using Non-Local Multi-Continuum method (NLMC). J. Comput. Appl. Math.
**2019**, 357, 215–227. [Google Scholar] [CrossRef] - Zhao, L.; Chung, E.T. An analysis of the NLMC upscaling method for high contrast problems. J. Comput. Appl. Math.
**2020**, 367, 112480. [Google Scholar] [CrossRef] - Chung, E.T.; Efendiev, Y.; Leung, W.T. Constraint energy minimizing generalized multiscale finite element method. Comput. Methods Appl. Mech. Eng.
**2018**, 339, 298–319. [Google Scholar] [CrossRef] - Fu, S.; Chung, E.; Mai, T. Constraint energy minimizing generalized multiscale finite element method for nonlinear poroelasticity and elasticity. J. Comput. Phys.
**2020**, 417, 109569. [Google Scholar] [CrossRef] - Efendiev, Y.; Galvis, J.; Li, G.; Presho, M. Generalized multiscale finite element methods: Oversampling strategies. Int. J. Multiscale Comput. Eng.
**2014**, 12, 465–484. [Google Scholar] [CrossRef] - Henning, P.; Peterseim, D. Oversampling for the multiscale finite element method. Multiscale Model. Simul.
**2013**, 11, 1149–1175. [Google Scholar] [CrossRef] - Calo, V.M.; Efendiev, Y.; Galvis, J.; Li, G. Randomized oversampling for generalized multiscale finite element methods. Multiscale Model. Simul.
**2016**, 14, 482–501. [Google Scholar] [CrossRef] - Chung, E.T.; Efendiev, Y.; Leung, W.T. An online generalized multiscale discontinuous Galerkin method (GMsDGM) for flows in heterogeneous media. arXiv
**2015**, arXiv:1504.04417. [Google Scholar] [CrossRef] - Chung, E.T.; Efendiev, Y.; Leung, W.T. An adaptive generalized multiscale discontinuous Galerkin method for high-contrast flow problems. Multiscale Model. Simul.
**2018**, 16, 1227–1257. [Google Scholar] [CrossRef] - Geuzaine, C.; Remacle, J.F. Gmsh: A 3-D finite element mesh generator with built-in pre-and post-processing facilities. Int. J. Numer. Methods Eng.
**2009**, 79, 1309–1331. [Google Scholar] [CrossRef] - Logg, A.; Mardal, K.A.; Wells, G. Automated Solution of Differential Equations by the Finite Element Method: The FEniCS Book; Springer Science & Business Media: Berlin, Germany, 2012; Volume 84. [Google Scholar]

M | ${\mathit{D}\mathit{O}\mathit{F}}_{\mathit{H}}$ | $\mathit{\alpha}=1$ | $\mathit{\alpha}=25$ | $\mathit{\alpha}=100$ | |||
---|---|---|---|---|---|---|---|

${\mathit{e}}_{{\mathit{L}}^{2}}$ | ${\mathit{e}}_{{\mathit{H}}^{1}}$ | ${\mathit{e}}_{{\mathit{L}}^{2}}$ | ${\mathit{e}}_{{\mathit{H}}^{1}}$ | ${\mathit{e}}_{{\mathit{L}}^{2}}$ | ${\mathit{e}}_{{\mathit{H}}^{1}}$ | ||

${\mathit{M}}_{\mathit{g}}=\mathit{M}$, ${\mathit{M}}_{\mathit{p}}=\mathbf{0}$ | |||||||

1 | 200 | 21.92 | 97.44 | 96.75 | 100.0 | 82.12 | 100.0 |

2 | 300 | 21.50 | 95.89 | 87.83 | 100.0 | 66.43 | 100.0 |

4 | 500 | 7.770 | 42.57 | 8.408 | 32.53 | 5.417 | 32.01 |

6 | 700 | 3.382 | 29.49 | 4.307 | 27.96 | 3.260 | 27.66 |

8 | 900 | 2.096 | 21.91 | 2.509 | 21.20 | 1.905 | 20.11 |

12 | 1300 | 1.201 | 14.33 | 1.417 | 13.33 | 1.053 | 13.08 |

16 | 1700 | 1.169 | 13.89 | 1.379 | 12.77 | 1.004 | 12.49 |

20 | 2100 | 1.127 | 13.10 | 1.349 | 12.03 | 0.972 | 11.81 |

24 | 2500 | 1.116 | 13.01 | 1.341 | 11.95 | 0.964 | 11.74 |

32 | 3300 | 1.098 | 12.80 | 1.329 | 11.74 | 0.957 | 11.53 |

M | ${\mathit{D}\mathit{O}\mathit{F}}_{\mathit{H}}$ | $\mathit{\alpha}=1$ | $\mathit{\alpha}=25$ | $\mathit{\alpha}=100$ | |||
---|---|---|---|---|---|---|---|

${\mathit{e}}_{{\mathit{L}}^{2}}$ | ${\mathit{e}}_{{\mathit{H}}^{1}}$ | ${\mathit{e}}_{{\mathit{L}}^{2}}$ | ${\mathit{e}}_{{\mathit{H}}^{1}}$ | ${\mathit{e}}_{{\mathit{L}}^{2}}$ | ${\mathit{e}}_{{\mathit{H}}^{1}}$ | ||

${\mathit{M}}_{\mathit{g}}=\mathit{M}$, ${\mathit{M}}_{\mathit{p}}=\mathbf{0}$ | |||||||

1 | 200 | 98.00 | 100.0 | 90.85 | 103.2 | 88.31 | 110.1 |

2 | 300 | 96.16 | 98.57 | 84.23 | 101.8 | 82.39 | 113.5 |

4 | 500 | 35.55 | 50.93 | 17.47 | 131.8 | 32.63 | 200.6 |

6 | 700 | 16.33 | 41.68 | 14.37 | 136.5 | 26.66 | 223.9 |

8 | 900 | 10.71 | 37.02 | 12.64 | 137.5 | 21.48 | 238.5 |

12 | 1300 | 6.896 | 33.31 | 12.29 | 136.9 | 21.52 | 237.5 |

16 | 1700 | 6.757 | 33.12 | 12.29 | 136.8 | 21.58 | 237.4 |

20 | 2100 | 6.580 | 32.80 | 12.29 | 136.7 | 21.63 | 237.1 |

24 | 2500 | 6.532 | 32.76 | 12.29 | 136.7 | 21.67 | 236.9 |

32 | 3300 | 6.457 | 32.68 | 12.29 | 136.7 | 21.68 | 236.8 |

${\mathit{M}}_{\mathit{g}}={\mathit{M}}_{\mathit{p}}=\mathit{M}$ | |||||||

1 | 251 | 96.63 | 97.45 | 82.96 | 123.0 | 66.74 | 129.7 |

2 | 402 | 93.85 | 95.23 | 67.28 | 116.0 | 43.93 | 112.6 |

4 | 704 | 27.84 | 39.03 | 6.909 | 32.68 | 3.523 | 33.20 |

6 | 1006 | 12.24 | 27.74 | 3.399 | 27.01 | 2.316 | 26.94 |

8 | 1308 | 7.909 | 20.65 | 1.747 | 20.01 | 1.201 | 19.30 |

12 | 1776 | 5.238 | 14.20 | 1.234 | 13.17 | 0.772 | 12.94 |

16 | 2244 | 5.108 | 13.80 | 1.197 | 12.64 | 0.735 | 12.37 |

20 | 2644 | 4.934 | 13.04 | 1.165 | 11.92 | 0.708 | 11.71 |

24 | 3044 | 4.887 | 12.97 | 1.155 | 11.86 | 0.699 | 11.66 |

32 | 3844 | 4.812 | 12.78 | 1.142 | 11.69 | 0.692 | 11.48 |

M | ${\mathit{D}\mathit{O}\mathit{F}}_{\mathit{H}}$ | ${\mathit{e}}_{{\mathit{L}}^{2}}$ | ${\mathit{e}}_{{\mathit{H}}^{1}}$ |
---|---|---|---|

${\mathit{M}}_{\mathit{g}}=\mathit{M}$, ${\mathit{M}}_{\mathit{p}}=\mathbf{0}$ | |||

1 | 300 | 95.47 | 100.0 |

2 | 500 | 94.41 | 99.24 |

4 | 900 | 73.76 | 82.19 |

6 | 1300 | 14.50 | 32.53 |

8 | 1700 | 8.312 | 26.82 |

12 | 2500 | 6.031 | 22.41 |

16 | 3300 | 2.566 | 18.72 |

20 | 4100 | 2.354 | 18.08 |

24 | 4900 | 2.117 | 17.59 |

32 | 6500 | 1.893 | 17.49 |

M | ${\mathit{D}\mathit{O}\mathit{F}}_{\mathit{H}}$ | ${\mathit{e}}_{{\mathit{L}}^{2}}$ | ${\mathit{e}}_{{\mathit{H}}^{1}}$ | M | ${\mathit{D}\mathit{O}\mathit{F}}_{\mathit{H}}$ | ${\mathit{e}}_{{\mathit{L}}^{2}}$ | ${\mathit{e}}_{{\mathit{H}}^{1}}$ |
---|---|---|---|---|---|---|---|

${\mathit{M}}_{\mathit{g}}=\mathit{M}$, ${\mathit{M}}_{\mathit{p}}=\mathbf{0}$ | ${\mathit{M}}_{\mathit{g}}=\mathit{M}$, ${\mathit{M}}_{\mathit{p}}=\mathbf{0}$ | ||||||

1 | 300 | 113.2 | 84.90 | 1 | 402 | 98.24 | 70.33 |

2 | 500 | 111.9 | 83.33 | 2 | 704 | 92.46 | 65.32 |

4 | 900 | 119.7 | 72.36 | 4 | 1308 | 52.99 | 45.22 |

6 | 1300 | 118.0 | 64.02 | 6 | 1912 | 8.847 | 26.20 |

8 | 1700 | 117.5 | 60.98 | 8 | 2516 | 6.835 | 20.38 |

12 | 2500 | 117.6 | 59.82 | 12 | 3452 | 5.762 | 17.84 |

16 | 3300 | 117.4 | 59.50 | 16 | 4388 | 2.745 | 16.50 |

20 | 4100 | 117.4 | 59.37 | 20 | 5188 | 2.568 | 16.20 |

24 | 4900 | 117.3 | 59.29 | 24 | 5988 | 2.324 | 14.24 |

32 | 6500 | 117.2 | 58.35 | 32 | 7588 | 2.019 | 13.20 |

M | ${\mathit{D}\mathit{O}\mathit{F}}_{\mathit{H}}$ | Quasi-Structured | Unstructured | ||
---|---|---|---|---|---|

${\mathit{e}}_{{\mathit{L}}^{2}}$ | ${\mathit{e}}_{{\mathit{H}}^{1}}$ | ${\mathit{e}}_{{\mathit{L}}^{2}}$ | ${\mathit{e}}_{{\mathit{H}}^{1}}$ | ||

1 | 251 | 58.12 | 101.1 | 56.66 | 98.80 |

2 | 402 | 35.39 | 87.82 | 45.87 | 104.1 |

4 | 704 | 6.539 | 41.82 | 16.53 | 60.94 |

6 | 1006 | 3.523 | 33.69 | 9.504 | 43.29 |

8 | 1308 | 2.034 | 23.72 | 4.909 | 33.93 |

12 | 1776 | 1.369 | 19.11 | 2.187 | 25.77 |

16 | 2244 | 1.052 | 17.04 | 1.390 | 22.29 |

20 | 2644 | 0.896 | 15.76 | 0.994 | 19.81 |

24 | 3044 | 0.783 | 15.06 | 0.734 | 17.48 |

32 | 3844 | 0.683 | 14.33 | 0.552 | 15.83 |

M | ${\mathit{D}\mathit{O}\mathit{F}}_{\mathit{H}}$ | Quasi-Structured | Unstructured | ||
---|---|---|---|---|---|

${\mathit{e}}_{{\mathit{L}}^{2}}$ | ${\mathit{e}}_{{\mathit{H}}^{1}}$ | ${\mathit{e}}_{{\mathit{L}}^{2}}$ | ${\mathit{e}}_{{\mathit{H}}^{1}}$ | ||

1 | 402 | 89.14 | 63.53 | 97.97 | 83.36 |

2 | 704 | 82.55 | 59.52 | 94.60 | 77.88 |

4 | 1308 | 62.82 | 49.96 | 76.75 | 66.07 |

6 | 1912 | 44.10 | 41.86 | 50.62 | 53.07 |

8 | 2516 | 20.71 | 30.19 | 29.94 | 43.22 |

12 | 3452 | 13.68 | 24.60 | 15.75 | 32.33 |

16 | 4388 | 8.648 | 20.83 | 10.57 | 27.54 |

20 | 5188 | 6.682 | 18.96 | 8.107 | 25.11 |

24 | 5988 | 5.309 | 17.49 | 6.705 | 23.75 |

32 | 7588 | 3.629 | 15.64 | 5.407 | 22.26 |

M | Elliptic Problem | Elasticity Problem | ||||
---|---|---|---|---|---|---|

${\mathit{D}\mathit{O}\mathit{F}}_{\mathit{H}}$ | ${\mathit{e}}_{{\mathit{L}}^{2}}$ | ${\mathit{e}}_{{\mathit{H}}^{1}}$ | ${\mathit{D}\mathit{O}\mathit{F}}_{\mathit{H}}$ | ${\mathit{e}}_{{\mathit{L}}^{2}}$ | ${\mathit{e}}_{{\mathit{H}}^{1}}$ | |

1 | 251 | 57.72 | 96.62 | 402 | 108.1 | 76.69 |

2 | 402 | 52.28 | 91.43 | 704 | 105.4 | 69.97 |

4 | 704 | 16.86 | 40.61 | 1308 | 44.45 | 44.88 |

6 | 1006 | 11.68 | 29.68 | 1912 | 9.455 | 29.43 |

8 | 1308 | 4.498 | 23.30 | 2516 | 6.824 | 23.97 |

12 | 1776 | 2.979 | 16.56 | 3452 | 4.753 | 21.43 |

16 | 2244 | 3.170 | 14.98 | 4388 | 3.511 | 20.25 |

20 | 2644 | 3.078 | 14.35 | 5188 | 3.327 | 19.97 |

24 | 3044 | 3.258 | 13.75 | 5988 | 3.235 | 19.90 |

32 | 3844 | 3.199 | 13.65 | 7588 | 3.024 | 17.95 |

Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations. |

© 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).