An efficient combination method of Laplace transform and mixed multiscale finite-element method for coupling partial differential equations of flow in a dual-permeability system is present. First, the time terms of parabolic equation with unknown pressure term are removed by the Laplace transform. Then the transformed equations are solved by mixed FEMs which can provide the numerical approximation formulas for pressure and velocity at the same time. With some assumptions, the multiscale basis functions are constructed by utilizing the effects of fine-scale heterogeneities through basis functions formulation computed from local flow problems. Without time step in discrete process, the present method is efficient when solving spatial discrete problems. At last, the associated pressure transform is inverted by the method of numerical inversion of the Laplace transform.
1. Introduction
In recent years, there have been some important developments in numerical methods of flow in fractured porous media. And an important numerical method is mixed finite-element method. Due to the heterogeneous properties and the multiscale nature of media, a complete analysis of these problems is extremely difficult, and efficient numerical solvers usually require an extremely large amount of computer storage and CPU time. Developing multiscale methods allow us to overcome above difficulty while retaining a satisfactory accuracy. The popular multiscale methods include multiscale finite-element method, numerical upscaling method, and multiscale finite volume method [1, 2]. In general, multiscale finite-element methods are regarded as numerical methods and strategies in which basis functions are computed by solving local homogeneous PDEs subject to special boundary conditions [1, 3–11].
Multiscale FEMs are efficiency and convenience for elliptic equation of steady flow model. Despite their advantage, they still have some practical limitations in solving parabolic equations for nonsteady flow model [12]. First, their major drawback is that it is necessary to take small time steps. A severe limitation on the time step may require an excessive amount of computer time. Second, all the interior velocity or pressure must be computed at each time step. We can see that numerical methods for parabolic equation with general FEMs require the solution of some simultaneous algebraic equations at each time step. On the other hand, these approaches tend to increase cost when the solutions must be carried out over long time periods [9].
The present method, the combined use of the Laplace transform and the multiscale finite-element method, is used to solve some problems of flow in fractured porous media. It is efficient to overcome the above difficulty.
2. Mathematical Model and Laplace Transform
Incompressible flow in fractured porous media can be described by some coupled partial differential equations. We consider the following mathematical model consisting of four coupling equations for pressure p1, p2 and Darcy velocity u1, u2 in a spatial domain [13, 14]:
(2.1)(Φ1C1)∂p1(x,t)∂t-div(u1(x,t))-λα(p2(x,t)-p1(x,t))=0,x∈Ω,t>0,(2.2)(Φ2C2)∂p2(x,t)∂t-div(u2(x,t))+λα(p2(x,t)-p1(x,t))=0,x∈Ω,t>0,(2.3)u1(x,t)=-λK1(x)∇p1(x,t),(2.4)u2(x,t)=-λK2(x)∇p2(x,t),(2.5)p1(x,0)=p10,(2.6)p2(x,0)=p20,
where Ω is a domain in Rd(d=2), Ci(i=1,2) denote the compressibility, ϕi(i=1,2) are the porosity, and ki(x)(i=1,2) denote the heterogeneous field permeability tensor which have scale separation and periodicity. We assume that only ui,pi(i=1,2)are unknown expressions. And here we assume that the equations are equipped with Neumann boundary conditions
(2.7)u→1⋅n→=0on∂Ω,(2.8)u→2⋅n→=0on∂Ω,
where n→is the outward normal of ∂Ω.
In order to remove time dependence from the governing equation and boundary conditions, the scheme of the Laplace transform will be utilized. The Laplace transform of the function p and its inversion formulas are defined as
(2.9)P-(x,s)=L[p(x,t)]=∫0∞e-stp(x,t)dt,p(x,t)=L-1[p-(x,s)]=12πi∫c-i∞c+i∞estp-(x,s)ds,
where s=c+iω,c,ω∈R.
The Laplace transform of (2.1)-(2.2) is:
(2.10)Φ1C1(sp-(x,s)-p1(x,0))-div(u-1(x,s))-λα(p-2-p-1)=0,(2.11)Φ2C2(sp-2(x,s)-p2(x,0))-div(u-2(x,s))+λα(p-2-p-1)=0.
The expression (2.10)-(2.11) can be written as
(2.12)(Φ1C1s+λα)p-1-λαp-2-Φ1C1p10=div(u-1),(Φ2C2s+λα)p-2-λαp-1-Φ2C2p20=div(u-2).
Now the unknown function u-i(x,s) and p-i(x,s),i=1,2, will be solved.
By the Laplace transform of (2.3)-(2.4), we can get the following:
(2.13)u-1(x,t)=-λK1(x)∇p-1(x,t),(2.14)u-2(x,t)=-λK2(x)∇p-2(x,t).
By the Laplace transform of Neumann boundary condition (2.7)-(2.8), we can know
(2.15)u-→i⋅n→=0,on∂Ω,i=1,2.
Next we consider mixed multiscale finite-element method to (2.10)–(2.15).
3. Mixed Finite-Element Method
The mixed finite-element method is based on a mixed formulation [6–8, 15]. For our model problem (2.10)–(2.15), the mixed formulation is as follows.
Find (u-i,p-i)(i=1,2)∈H01,div(Ω)×L2(Ω), such that
(3.1)∫Ωv⋅(kiλ)-1u-idx+∫Ωv⋅∇p-idx=0,(3.2)∫Ωdiv(u-i)wdx-∫((ΦiCis+λα)p-i-λαp-3-i-ΦiCipi0)wdx=0,
which hold for all v∈H01,div(Ω) and w∈L2(Ω),
(3.3)H01,div(Ω)={v∈(L2(Ω))d:div(v)∈L2(Ω),v⋅n=0on∂Ω},
where L2(Ω) is the space of square integral function in Ω. The above formulas (3.1)-(3.2) can provides the numerical results of p-i(x,s) and u-i(x,s)(i=1,2) at the same time.
In mixed finite-element method, L2(Ω) and H01,div(Ω) are approximated by finite-dimensional subspaces W and V, respectively [1, 6]. For instance, L2(Ω) is replaced by W: (3.4)W={w∈L2(Ω):w|Ωk=const∀Ωk∈Ω}.
And H01,div(Ω) is replaced by V:
(3.5)V={v∈H01,div(Ω):div(v)∈L2(Ω),∀Ωi∈Ω,(v⋅nij)|γij=const,∀γij∈Ω,andv⋅nijiscontinuousarossγijH01,div}.
Here nij is the unit normal to γij pointing from Ωi to Ωj.
Denoting the approximation of (p-i,u-i)(i=1,2) individually by (p-ih,u-ih)(i=1,2), the discrete formulation reads as follows.
Find (p-ih,u-ih)(i=1,2)∈W×V, such that (3.1)-(3.2) hold for all w∈W and v∈V.
By applying Green formulation for (3.1), we can conclude the following:
(3.6)∫Ωv⋅(k1λ)-1u-1hdx-∫Ωp-1h∇⋅vdx=0,(3.7)∫Ωv⋅(k2λ)-1u-2hdx-∫Ωp-2h∇⋅vdx=0.
Now letting V=span{ψj} and w=span{φj}, we let v be equal to velocity basis function ψk and w equal pressure basis function φk (k=1,2,…,N.). And the approximations of u-1h,u-2h and p-1h,p-2h can be written as follows:
(3.8)u-1h=∑j=1NU~1j(s)⋅ψj(x),u-2h=∑j=1NU~2j(s)⋅ψj(x),p-1h=∑j=1NP~j1(s)⋅φj(x),p-2h=∑j=1NP~j2(s)⋅φj(x).
Then formulas (3.2)–(3.7) can show that
(3.9)∫Ωψk⋅(k1λ)-1(∑j=1NU~1j(s)⋅ψj(x))dx-∫Ω(∑j=1NP~1j(s)⋅φj(x))∇⋅ψkdx=0,∫Ωψk⋅(k2λ)-1(∑j=1NU~2j(s)⋅ψj(x))dx-∫Ω(∑j=1NP~2j(s)⋅φj(x))∇⋅ψkdx=0,∫Ωdiv(∑j=1NU~1j(s)⋅ψj(x))φkdx=∫Ω((Φ1C1s+λα)∑j=1NP~1j(s)⋅φj(x)-λα∑j=1NP~2j(s)⋅φj(x)-Φ1C1p10)φkdx,∫Ωdiv(∑j=1NU~2j(s)⋅ψj(x))φkdx=∫Ω((Φ2C2s+λα)∑j=1NP~2j(s)⋅φj(x)-λα∑j=1NP~1j(s)⋅φj(x)-Φ2C2p20)φkdx,k=1,2,…,N.
The above four formulas can be rearranged as follows:
∫Ωψk⋅(k1λ)-1(∑j=1NU~1j(s)⋅ψj(x))dx-∫Ω(∑j=1NP~1j(s)⋅φj(x))∇⋅ψkdx=0,∫Ωdiv(∑j=1NU~1j(s)⋅ψj(x))φkdx-∫Ω((Φ1C1s+λα)∑j=1NP~1j(s)⋅φj(x))φkdx+∫Ωλα∑j=1NP~2j(s)⋅φj(x)dx=-∫ΩΦ1C1p10φkdx,∫Ωλα∑j=1NP~1j(s)⋅φj(x)dx-∫Ω((Φ2C2s+λα)∑j=1NP~2j(s)⋅φj(x))φkdx+∫Ωdiv(∑j=1NU~2j(s)⋅ψj(x))φkdx=-∫ΩΦ2C2p20φkdx-∫Ω(∑j=1NP~2j(s)⋅φj(x))∇⋅ψkdx+∫Ωψk⋅(k2λ)-1(∑j=1NU~2j(s)⋅ψj(x))dx=0,k=1,2,…,N.
Then, we obtain the approximations:
(3.11)u-1h=∑j=1NU~1j(s)⋅ψj(x),u-2h=∑j=1NU~2j(s)⋅ψj(x),p-1h=∑j=1NP~1j(s)⋅φj(x),p-2h=∑j=1NP~2j(s)⋅φj(x),
where the coefficients U~1j, P~1j, P~2j and U~2j are obtained by solving the following linear system:
(3.12)[A1B100B2D1D200D3D4B300B4A2][U~1jP~1jP~2jU~2j]=[00F1F2],
where
(3.13)A1=[∫Ωψk⋅(λK1)-1ψjdx],A2=[∫Ωψk⋅(λK2)-1ψjdx],B1=B4=[-∫Ωφjdiv(ψk)dx],B2=B3=[∫Ωφkdiv(ψj)dx],D1=[-∫Ω(Φ1C1s+λα)φjφkdx],D2=D3=[∫Ωλαφjdx],D4=[-∫Ω(Φ2C2s+λα)φjφkdx],F1=[-∫ΩΦ1C1p10φkdx],F2=[-∫ΩΦ2C2p20φkdx](k,j=1,2,…,N).
Then the linear system (3.12) can be written as
(3.14)[A1B100B2D1D200D2D4B200B1A2][U~1jP~1jP~2jU~2j]=[00F1F2].
4. Multiscale Basis Functions
The main idea of the MMsFEM (Mixed Multiscale Finite-Element Method) is to construct special local basis functions that are adaptive to the local properties of the elliptic differential operator [6]. In this paper, two sets of grids are considered: a fine grid and a coarsened grid in which each coarse block ΩH consists of connected cells from the underlying fine grid. Local basis functions φiH for pressure and ψiH for velocity in each coarse block ΩH generally are constructed by solving
(4.1)∇⋅ψiH={TH(x),forx∈ΩH,0,else,ψiH=-λK∇φiH,ψiH⋅n=0,on∂ΩH,
where subscript i denotes the order number of the basis function.
The weighting function TH(x) plays an important role to distribute div(u-i)(i=1,2) onto the fine grid appropriately. It can be chosen on the form TH(x)=l(x)/(∫ΩHl(x)dx) with various choices of l(x) [6].
Now denote Vms=span{ψiH} the approximation space of V and W=span{φjH}, we let vH equal velocity basis function ψiH and wH equal pressure basis function φiH (i,j=1,2,…N′.). And the approximations of u-1H,u-2H and p-1H,p-2H can be written as follows:
(4.2)u-1H=∑j=1N′U~1H,j(s)⋅ψj(x),u-2H=∑j=1N′U~2H,j(s)⋅ψj(x),p-1H=∑j=1N′P~1H,j(s)⋅φj(x),p-2H=∑j=1N′P~2H,j(s)⋅φj(x).
Then the MMsFEM seeks (p-iH,u-iH)(i=1,2)∈W×Vms,such that (3.1)-(3.2) hold for all vH∈Vms and wH∈W. We can get the following formulas on a similar conclusion from (3.6)-(3.7) to expression (3.14):
(4.3)[A1′B1′00B2′D1′D2′00D4′D4′B2′00B1′A2′][U~1H,jP~H,j1P~H,j2U~2H,j]=[00F1′F2′],(4.4)A1′=[∫ΩψkH⋅(λK1)-1ψjHdx],A2′=[∫ΩψkH⋅(λK2)-1ψjHdx],B1′=B4′=[-∫ΩφjHdiv(ψkH)dx],B2′=B3′=[∫ΩφkHdiv(ψjH)dx],D1′=[-∫Ω(Φ1C1s+λα)φjHφkHdx],D2′=D3′=[∫ΩλαφjHdx],D4′=[-∫Ω(Φ2C2s+λα)φjHφkHdx]F1′=[-∫ΩΦ1C1p10φkHdx],F2′=[-∫ΩΦ2C2p20φkHdx],(k,j=1,2,…,N).
5. Numerical Inversion of the Pressure and Velocity Term
In this section we compute the numerical value of UiH,j(t)(i=1,2;j=1,2,…,N′) and PiH,j(t)(i=1,2;j=1,2…,N′) by U~iH,j(s)(i=1,2;j=1,2…N′) and P~iH,j(s)(i=1,2;j=1,2…N′) using the numerical inversion formula of Laplace transform. In accordance with the method of Durbin [16], the Fourier series expansion of UiH,j(t) can be derived as follows [9, 17]:
(5.1)UiH,j(t)=1Tevt{-12Re[U~iH,j(v)]+∑k=0∞{Re[U~iH,j(Sk)]coskπtT-Im[U~iH,j(Sk)]sinkπtT}}-U~0(v,t,T),Sk=v+ikπT,k=0,1,…,N.
And here the discrete error is given by
(5.2)U~0(v,t,T)=∑k=1∞e-2vkTU~iH,j(2kT+t).
So the numerical approximate value of UiH,j(t) is
(5.3)U~i,NH,j(t)=1Tevt{-12Re[U~iH,j(v)]}+1Tevt∑k=0N{Re[U~iH,j(Sk)]coskπtT-Im[U~iH,j(Sk)]sinkπtT},
where 2T is the period of Fourier series approximate inversion function on the interval [0,2T] and N and v*T are free parameters. All U~iH,j(s) are computed by formulation (4.3).
Obviously, there occurs a truncation error given by
(5.4)U~i,AH,j(t)=1Tevt{∑k=N+1∞{Re[U~iH,j(Sk)]coskπtT-Im[U~iH,j(Sk)]sinkπtT}}.
Because numerical inversion formula requires the value of transformed function at each Sk(k=0,1,2,3,…,N), it is needed to solve the equations for any S=Sk.
The approximate value of PiH,j(t)(i=1,2;j=1,2,…,N′) can be computed in the same way. (5.5)P~i,NH,j(t)=1Tevt{-12Re[P~iH,j(v)]}+1Tevt∑k=0N{Re[P~iH,j(Sk)]coskπtT-Im[P~iH,j(Sk)]sinkπtT}.
The solution at a specific node in the given domain can be obtained from (4.3) and (5.3)–(5.5). This strategy is different from that of classical FEM, which must compute all nodal values for each time step until the specific time is reached. The present method can compute the specific nodal value at a specific time. In addition, it is obvious that the present method takes less computer time if lengthy solutions are required.
6. Conclusion
This paper focuses on efficient numerical method which combines the numerical inversion of Laplace transform with mixed FEM for coupling partial differential equations of flow in a dual-permeability system. The present combined method can save computing cost by multiscale method in spatial domain, and overcome some disadvantages coming from numerical processing of different time step size by the numerical inversion of Laplace transform. However, the present method can be extended to other two- or three-dimensional linear time-dependent problems.
Acknowledgments
This research was supported by Natural Science-Technology Major Special Task of China (2011ZX05008-004-44), Natural Science Foundation of China (11161002), and Natural Science Foundation of Jiangxi province (20114BAB201016).
KippeV.AarnesJ. E.LieK.-A.A comparison of multiscale methods for elliptic problems in porous media flow200812337739810.1007/s10596-007-9074-62434951DingL.HanB.LiuJ.-Q.A wavelet multiscale method for inversion of Maxwell equations2009308103510442-s2.0-6894921712910.1007/s10483-009-0810-1AarnesJ. E.EfendievY.Mixed multiscale finite element methods for stochastic porous media flows20083052319233910.1137/07070108X2429468ChuC.-C.GrahamI. G.HouT.-Y.A new multiscale finite element method for high-contrast elliptic interface problems2010792721915195510.1090/S0025-5718-2010-02372-52684351ZBL1202.65154KrogstadS.LieK.-A.NilsenH. M. M.NatvigJ. R.SkaflestadB.AarnesJ. E.SINTEF. A multiscale mixed dinite-element solver for three-phase black-oil flowProceedings of the Society of Petroleum Engineers (SPE 118993)2009113EfendievY.HouT. Y.2009New York, NY, USASpringer2477579MaX.ZabarasN.A stochastic mixed finite element heterogeneous multiscale method for flow in porous media2011230124696472210.1016/j.jcp.2011.03.0012795977ArbogastT.201283Lecture Notes in Computational Science and Engineering10.1007/978-3-642-22061-6RenL.A hybrid Laplace transform finite element method for solute radial dispersion problem in subsurface flow1994913743HeX. G.RenL.Adaptive multiscale finite element method for unsaturated flow in heterogeneous porous media I. Numerical scheme20094013845XueY. Q.YeS. J.XieC. H.ZhangY.Application of multi-scale finite element method to simulation of groundwater flow20047713ChenH. T.ChenC. K.Hybrid laplace transform/finite element method for two dimensional heat conduction problem198821313610.2514/3.58GeJ. L.2003Beijing, ChinaPetroleum Industry PressVogelT.GerkeH. H.ZhangR.Van GenuchtenM. T.Modeling flow and transport in a two-dimensional dual-permeability system with spatially variable hydraulic properties20002381-278892-s2.0-003473615610.1016/S0022-1694(00)00327-9HuangA.ZhouT. X.2009Beijing, ChinaScience PressDurbinF.Numerical inversion of Laplace transforms: an efficient improvement to Dubner and Abate's method1974173713760356449ZBL0288.65072HonigG.HirdesU.A method for the numerical inversion of Laplace transforms198410111313210.1016/0377-0427(84)90075-X736594ZBL0535.65090