Download as pdf or txt
Download as pdf or txt
You are on page 1of 32

SPE 142206

Quasi-M-matrix Anisotropic Darcy-Flux Finite-volume Approximation for General


Grids and MPFA Decoupling Analysis
Michael G. Edwards Swansea University and Hongwen Zheng Computer Modelling Group

Copyright 2011, Society of Petroleum Engineers

This paper was prepared for presentation at the 2011 SPE Reservoir Simulation Symposium held in The Woodlands, Texas, USA, 21-23 February 2011.

This paper was selected for presentation by an SPE program committee following review of information contained in an abstract submitted by the author(s).
Contents of the paper have not been reviewed by the Society of Petroleum Engineers and are subject to correction by the author(s). The material does not
necessarily reflect any position of the Society of Petroleum Engineers, its officers, or members. Electronic reproduction, distribution, or storage of any part of
this paper without the written consent of the Society of Petroleum Engineers is prohibited. Permission to reproduce in print is restricted to an abstract of not
more than 300 words; illustrations may not be copied. The abstract must contain conspicuous acknowledgment of SPE copyright.

Abstract
This paper focuses on flux-continuous general tensor pressure equation approximation for strongly
anisotropic media. Earlier methods have focused on single-parameter pointwise schemes which are ef-
fective for smaller anisotropy ratios. However they have conditional M-matrices depending on the strength
of the tensor cross-terms. For strongly anisotropic full tensor cases, the methods can yield pressure solutions
with severe spurious oscillations. The spurious oscillations are shown to be caused by decoupling on both
structured and unstructured grids. Flow results and convergence tests confirm the decoupling theory.
Anisotropic flux-continuous full pressure continuity formulations are presented. The formulations lead
to improved discretization schemes for the complete range of elliptic tensors. The new schemes are shown
to be positive definite, have quasi-positive M-matrices for full-tensor fields with large anisotropy ratios, and
compute well resolved solutions that are essentially free of spurious oscillations.
Anisotropic Darcy-flux schemes are presented on structured, unstructured and anisotropy favoring
grids. The methods are applied to a range of geological test cases and grid-types including arbitrarily
high anisotropy ratio in two and three dimensions.

1. Introduction
This paper focuses on the development of flux-continuous finite-volume schemes for solving the
pressure equation resulting from Darcy’s law and mass conservation on grids comprised of general
cell type [14]. Key physical constraints of continuity in normal flux and full pressure continuity
are imposed at control-volume interfaces. Earlier families of efficient locally conservative point-
wise flux-continuous control volume distributed (CVD) finite-volume schemes for determining the
discrete pressure and velocity fields are presented in e.g. [9, 7, 8, 25, 24], these schemes are classified
2 SPE-142206

by the quadrature parameterization 0 < q ≤ 1. Schemes of this type are also called multi-point flux
approximation schemes or MPFA [17] where focus has been on a scheme that belongs to the above
mentioned family with (q = 1). Further schemes of this type are presented in [33, 34] and [35] via a
novel mixed method.
Other schemes that preserve flux continuity have been developed from variational frameworks,
using the mixed finite element method (MFE) e.g. [28, 31, 12, 5, 3, 32] and related work [16] and
discontinuous galerkin methods [29, 30], however these schemes require additional degrees of freedom.

When applied to strongly anisotropic full-tensor field problems the earlier point-wise continuous
subcell tetrahedral-pressure-support (TPS) methods can fail to satisfy the maximum principle (as
with other FEM and finite-volume methods) and result in spurious oscillations in the numerical
pressure solution. Stability of the TPS schemes in 2-D has been presented previously; M -matrix
conditions for families of schemes in two dimensions were first presented in [6, 9], and in [10] where
the 2-D TPS methods are shown to lead to decoupled (strongly oscillatory) solutions for strongly
anisotropic full-tensor fields. Monotone matrix conditions are presented in [2]. Grid optimization
techniques have been proposed for improving stability of the discrete system for variable anisotropy
[20]. Discretization schemes that help to improve stability for high anisotropy are presented in
[24, 10, 11] and [2]. Further methods have also been proposed, [23, 22, 27] (flux-splitting) and
[18, 19] (positivity preserving), which have been shown to yield numerical pressure solutions that
are free of spurious oscillations in two dimensions.
Anisotropic quadrature schemes were first presented in [11] in 2-D, their formulation result-
ing from a double-family framework. Three-dimensional multi-family flux-continuous finite volume
methods are presented in [14, 37]. In this work multi-family schemes and in particular anisotropic
quadrature schemes and their properties are explored together with performance. Formulation of
local algebraic flux and pressure continuity conditions for full-tensor discretization in three dimen-
sions on general structured and unstructured grids is given for completeness. The methods retain
a single degree of freedom per control-volume per phase continuity equation in the approximation
while maintaining flux continuity and full pressure continuity across control-volume subfaces. For
single phase flow each control-volume is assigned a single discrete pressure value. This contrasts with
the mixed finite element method, which requires that an additional degree of freedom corresponding
to every continuous interface condition in the grid be added to the global system matrix. The mixed
method thus requires a factor of four times as many degrees of freedom on a structured grid in 3-D
while having an indefinite matrix.
The quasi-positive formulation has full pressure support (FPS) over each control-volume sub-
cell. Full pressure support with full pressure continuity across subcell faces leads to an increased
quadrature range and is a significant advantage compared to the earlier point-wise continuous TPS
(triangle or tetrahedral pressure support) schemes. The increased quadrature range enables the FPS
schemes to compute discrete pressure solutions for strongly anisotropic full-tensor fields that are
essentially free of spurious oscillations. In contrast the earlier pointwise continuous TPS methods
are shown to permit a fundamental decoupled oscillating solution mode and consequently induce
spurious oscillations. The results presented demonstrate the benefits of the FPS schemes and in
particular anisotropic schemes on structured and unstructured grids in two and three dimensions.
M -matrix conditions for double families of schemes on structured and unstructured grids in
two dimensions are presented in [11]. M-matrix analysis of families of 3-D schemes is presented
3 SPE-142206

in [14, 37], which identifies bounding limits for the schemes to posses a local discrete maximum
principle in three dimensions. Further M-matrix stability analysis is given here together with new
observations on decoupling in 2-D and 3-D. Conditions for the schemes to be positive definite in 3-D
are also given.

2. Problem Description and General Tensor Equation

The problem is to find the pressure φ satisfying


Z
− ∂ξi (Ti,j ∂ξj φ)dΓ = M (1)

over an arbitrary volume Ω, where the general tensor T =| J | J−1 KJ−T has elements Ti,j
and is defined via the Piola transform. Here the summation convention over repeated indices i =
1, ...3, j = 1, ...3 applies. Eq.1 is solved subject to suitable (Neumann/Dirichlet) boundary conditions
on boundary ∂Ω. The right hand side term M represents a specified flow rate. Matrix K is a
diagonal or full symmetric elliptic cartesian permeability tensor with elements Ki,j . The tensor can
be discontinuous across internal boundaries of Ω. For incompressible flow pressure is specified at
at least one point in the domain. For reservoir simulation, Neumann boundary conditions on ∂Ω
requires zero normal flux on solid walls such that (K∇φ) · n̂ = 0, where n̂ is the outward normal
vector to ∂Ω.

3. Grid Definition

The primal grid considered here can be a hybrid composed of tetrahedra, prisms, pyramids
and hexahedra elements in 3-D [8, 24]. In principle the only restriction on grid structure is that
tetrahedra can only be joined to hexahedra through a pyramid interface [8]. The approximation is
vertex centred. A polyhedral control-volume is built around each grid vertex, generating a primal-
dual grid. Starting in a primal grid cell, the cell centre is joined to cell face mid-points, cell face
mid-points are joined to cell edge mid-points. As a result the primal grid cells are decomposed
into subcells which are mainly sub-hexahedra cells, four for a tetrahedra, six for a prism, eight for a
hexahedra, the pyramid is the special case and is decomposed into four sub-hexahedra (corresponding
to the four base vertices) and one octahedra corresponding to the summit (top) node (vertex) [24]. In
each case the number of subcells corresponds to the number of vertices defining the primal cell, and
each subcell belongs to the control-volume of the unique vertex to which it is attached. Cell vertex
control-volumes are defined by a local assembly at each primal grid vertex of all subcells attached to
the vertex. The resulting set of polyhedral control-volumes define a dual grid relative to the primal
grid called the primal-dual. Rock permeability and porosity are assumed to be piece-wise constant
over each polyhedral control-volume and flow variables belong to the control-volumes and are vertex
centred, Fig’s. 2 and 3. Therefore discontinuities in rock properties occur over the control-volume
faces, and thus locally over interior subcell faces.

4. Flux-Continuous Schemes with Full Pressure Continuity

We illustrate the procedure with a primary hexahedral cell with local vertices numbered 1, ..., 8,
Fig2, where the vector ΦV of eight primal cell vertex pressures are locally numbered ΦV = (φ1 , φ2 , ...φ8 )t .
4 SPE-142206

First the vector of auxiliary interface pressures ΦA are introduced, which are comprised of a
vector of cell-edge interface pressures ΦE , and a vector of cell-face and (one) cell-volume interface
pressures ΦF , where ΦA = (ΦE , ΦF )t . ΦE are located at the cell edge mid-points. ΦF are located
at the cell face mid-points and at the cell-centre, the circumcentre is an alternate possibility for
tetrahedra. Thus the set of primal vertex pressures ΦV and auxiliary interface pressures ΦA together
define a set of discrete pressures with one discrete pressure value located at each corner of the subcells
that make up a primal grid cell. For example we refer to the primal hexahedra of Fig. 1 and the
subcells indicated by dashed lines and interior subcell corners indicated with letters. Rock properties
belong to control-volumes and can jump by orders of magnitude across the interior subcell faces. By
construction the same interface pressures are shared by adjacent subcells so that pressure continuity
is automatically satisfied. The auxiliary interface pressures are expressed algebraically in terms of
the primary cell vertex pressures in a preprocessing step, through the following set of local equations
imposed over the primary cell. The pressure φ and position vector r = (x, y, z) assume a trilinear
variation over each subcell. Referring to Fig. 1 and subcell 1, then the subcell pressure variation is
defined by the tri-linear variation

e − ηe)(1 − ζ)φ
φ = (1 − ξ)(1 e 1 + ξ(1
e − ηe)(1 − ζ)φ
e b1 + ξeeη (1 − ζ)φ
e b0
e η (1 − ζ)φ
+(1 − ξ)e e b4 + (1 − ξ)(1
e − ηe)ζφe e1 + ξ(1
e − ηe)ζφ e m1 (2)
eη ζφ
+ξe e m0 + (1 − ξ)e
e η ζφ
e m4
e − ηe)(1 − ζ)r
r = (1 − ξ)(1 e 1 + ξ(1
e − ηe)(1 − ζ)r
e b1 + ξeeη (1 − ζ)r
e b0
e η (1 − ζ)r
+(1 − ξ)e e b4 + (1 − ξ)(1
e − ηe)ζre e1 + ξ(1
e − ηe)ζr e m1 (3)
eη ζr
+ξe e m0 + (1 − ξ)e
e η ζr
e m4

where ξ,e ηe, ζe are subcell master element coordinates defined over the unit square 0 ≤ ξ, e ηe, ζe ≤ 1.
Subcell Darcy fluxes are defined by resolving Darcy velocity along surface area outward normal
vectors with F = −K∇φ • ∆S, and approximated using subcell approximations of the form given
by Eq’s.2,3 for subcell 1. The interface pressures are continuous across the respective interfaces by
construction, and as a consequence of the subcell approximation, pressure is bilinear and continuous
over the entire face of each subcell with full pressure support FPS. A local normal flux continuity
condition is imposed for each edge-centred pressure, a surface zero divergence condition is imposed
for each auxiliary cell-face pressure and zero divergence is imposed in each primal cell sub-volume,
to determine the auxiliary pressure at the cell-centre. The earlier tetrahedral pressure support (TPS)
schemes only require one interface pressure corresponding to each edge flux and are consequently
only point-wise continuous in pressure, however TPS requires the minimum number of auxiliary
equations. The bilinear FPS subcell surface support also retains two degrees of freedom in position
of flux continuity on a sub-face. This formulation permits maximum flexibility in quadrature in
3-D, which is achieved by allowing the coordinates of flux quadrature to vary with unequal values in
general. For example referring to face b0, b1, m1, m0 of Fig.1, then we can choose ηe 6= ζe and crucially
different values can be chosen for each interior subcell face (sub-interface) of a given control-volume
Fig.3(a). Such a formulation has already proven beneficial in 2-D [11]. This formulation leads to
new multiple families of 3-D flux-continuous schemes with full pressure support together with the
maximum quadrature range.
5 SPE-142206

8 t3 7

t4 t0 t2

t1 6
5
e4 m3 e3
m0
m4
m2
e1 e2
m1
4 3
b3
b4
b0 b2
2
1
b1

Figure 1: Primal hexahedral cell and subcells. Vertices 1, 2, ...8 are primary variables, : other (letter) vertices are
auxiliary variables. Subcell 1 has corners anti-clockwise (1, b1, b0, b4, e1, m1, m0, m4)

4.1. Flux Continuity Conditions


Primal fluxes are defined over the hexahedral subcell faces that are inside the primal cell, by
resolving the discrete Darcy flux along each normal surface area vector. Interior subcell fluxes are
constructed for each subcell face belonging to a primary grid cell. As a result, a left and right flux will
be defined with respect to each interior subcell face, Fig2. Flux continuity is achieved by equating
subcell fluxes at common faces leading to

Fi = −(Ti1 φξe + Ti2 φηe + Ti3 φζe)|Lσ = −(Ti1 φξe + Ti2 φηe + Ti3 φζe)|R
σ (4)

where suffix i indicates the local subcell coordinate, and e.g. suffix σ1 = σ(e e indicates the
η , ζ)
chosen parameterized flux continuity position on subface b0, b1, m1, m0. On subface b4, b0, m0, m4
the coordinates are σ2 = σ(ξ, e ζ).
e In this formulation the coordinates can be chosen locally per
subface, giving rise to a multiple family of schemes. If they are constrained to have the same
coordinates in parameter space, then this would reduce to a single family of schemes. Note that
the tensor coefficients Tij |Lσ , Tij |R
σ result from approximating flux in a subcell and performing flux
resolution normal to a subcell face, thus the these tensors are approximated using local subcell
permeability and geometry.
We give an expanded example of Eq. 4 using Eq’s. 2,3 with respect to face b0, b1, m1, m0 of
Fig.1, the left and right hand fluxes of Eq. 4 are expanded in the form
e b1 − φ1 ) + ηe(1 − ζ)(φ
Fb1 = −(T111 ((1 − ηe)(1 − ζ)(φ e b0 − φb4 ) + ζ(1
e − ηe)(φm1 − φe1 )
+e e m0 − φm4 )) + T1 ((1 − ζ)(φ
η ζ(φ e b0 − φb1 ) + ζ(φ
e m0 − φm1 ))
12
1
+T13 ((1 − ηe)(φm1 − φb1 ) + ηe(φm0 − φb0 )))
e 2 − φb1 ) + ηe(1 − ζ)(φ
e b2 − φb0 ) + ζ(1
e − ηe)(φe2 − φm1 ) (5)
= −(T211 ((1 − ηe)(1 − ζ)(φ
+e e m2 − φm0 )) + T ((1 − ζ)(φ
η ζ(φ 2 e b0 − φb1 ) + ζ(φ
e m0 − φm1 ))
12
2
+T13 ((1 − ηe)(φm1 − φb1 ) + ηe(φm0 − φb0 )))

The non-symmetric positions of quadrature are illustrated in Fig.3(a).

4.2. Auxiliary Control Volume Divergence Free Conditions


Full (bilinear) pressure continuity over the control-volume subcell faces is achieved by the in-
troduction of further auxiliary pressures at cell centres and cell-face centres (in addition to edge
interface pressures). Therefore, in addition to the flux continuity conditions corresponding to the
6 SPE-142206
8 7 8 7

6 6
5 5

4 3 4 3

2 2
1 1

(a) (b)

Figure 2: Local subcell fluxes and supporting nodes: Local primal cell-vertex numbers indicate primary nodes (d.o.f),
all other nodes are (auxiliary) interface nodes (a) TPS scheme. (b) FPS scheme.

8 7
m4 a8 m0

a5 a6
6 a3
a4
5
e1
a1 m1 a2

4 3 b4 b0

1 2 1 b1

Figure 3: (a) Multi-family fluxes on subcell with non-symmetric positions of quadrature (b) Auxiliary fluxes in a
subcell. Red arrow: surface flux. Green arrow: volume flux
7 SPE-142206

5 4 3 4 n 3 4 n 3
N

W E
6 2 w e w e
1 m m

S
7 8 9 1 s 2 1 s 2

(a) (b) (c)

Figure 4: (a) 9-point scheme support nodes and primal control-volume (dashed) for node 1:(b) Primal fluxes solid ar-
rows N,S,E,W auxiliary-fluxes hollow arrows (c) range of auxiliary control-volumes (dot-dashed) surrounding auxiliary
interface pressure m

edge mid-point auxiliary pressures ΦE , zero divergence equations are imposed over surface and vol-
ume auxiliary control-volumes, that are constructed to surround the respective additional auxiliary
variables ΦF . The additional divergence equations add the extra constraints needed to close the
system and thus express the local degrees of freedom ΦA in terms of the primal cell-vertex pressures
ΦV . Thus the discrete integral of divergence is formed on specified cell face areas and over an interior
specified cell-volume with
X
− (K∇Φ) · n̂∆s = 0 (6)
∂ΩAU X

generalising [36] in 2-D. Each discrete divergence involves construction of two types of Auxiliary
Control Volume (ACV) approximation for 3D FPS schemes: A surface ACV surrounding each cell-
face centre and a sub-volume ACV surrounding the center of the primal cell, each of a specified size.
Auxiliary fluxes are then defined over the faces of the resulting ACV’s (we will refer to a primary
control-volume flux as a primal flux, there is one primal flux for each primal cell edge, though it
may act at any chosen point of the corresponding control-volume subface). For simplicity we first
illustrate key points in 2-D, referring to the standard 9-point support Fig. 4(a) and the cell defined
by nodes 1, 2, 3, 4. Edge interface pressures are defined by s, e, n, w and the cell centre interface
pressure is m. Four subcells are indicated in Fig. 4(b), the first being 1, s, m, w. A two-dimensional
array of ACV’s (dot-dashed) are also illustrated in Fig. 4(c), indicating the variability in the size of
the ACV. An ACV is constructed as a scaled cell that can be expanded to the size of the primary cell
surface/volume or contracted to within a tolerance of the auxiliary surface/volume interface pressure.
The Fig. 4 shows that the ACV overlaps the four subcells, leading to sub-subcells. i.e. subcells of
the surface ACV or volume ACV as illustrated in Fig.3(b), top right corner m0 is primal cell centre,
Fig.1. Pressure gradients are defined by interpolation of the subcell gradients and since permeability
is piecewise constant over the sub-subcells that comprise the ACV, the auxiliary flux approximation
is naturally built using a control-volume finite element (CVFE) framework, with associated auxiliary
fluxes resolved over the ACV faces, hollow arrows Fig. 4(b). Both the primal fluxes and auxiliary
fluxes are linear functions of ΦV and ΦA and the auxiliary divergence approximations are built from
local CVFE approximations.
8 SPE-142206

4.2.1. Local Flux and Divergence Conditions


The local auxiliary pressures for each cell are expressed as a total vector of edge-interface pressures
(ΦE ) and cell-face/volume pressures (ΦF ) where ΦA = (ΦE , ΦF )T with dimension NA = NE + NF .
The primal cell-vertex pressures are ΦV = (φ1 , φ2 , ..., φNV )T , with dimension NV . The system of flux
continuity and auxiliary divergence conditions are written as

AL ΦA + BL ΦV = AR ΦA + BR ΦV

where AL , AR are NA XNA matrices and BL , BR are NA XNV matrices.


The continuous fluxes of the families of FPS schemes are defined by the first NE rows of
¢
F = (AL (AL − AR )−1 (BR − BL ) + BL )ΦV (7)

where F is a vector of dimension NE . The primal divergence equations are then assembled as the
sum of outward normal primal fluxes of all subcell interfaces that comprise the net primal polyhedral
control-volume surface. Four equation sets are described below according to the element (or cell)
type.

4.2.2. Full Pressure Support: Hexahedral Element


The hexahedral element is divided into 8 sub-hexahedra corresponding to each primal cell vertex.
Each sub-hexahedra has three interior faces that form part of the respective vertex control-volume.
There are a total of twelve interior interfaces, one per cell edge. Consequently twelve local flux
continuity conditions are imposed in order to determine the the twelve edge interface pressures. Seven
zero divergence conditions are imposed, six on the cell faces for the six surface auxiliary pressures
and one sub-volume condition for the cell-centre auxiliary pressure. NE = 12, NF = 7, NV = 8.

4.2.3. Full Pressure Support: Prism Element


The prism element is divided into 6 sub-hexahedra corresponding to each primal cell vertex. The
prism has nine edges, consequently nine local flux continuity conditions are imposed. In this case FPS
requires six zero divergence conditions are imposed, five on the cell faces for the five surface auxiliary
pressures and one sub-volume condition for the cell-centre auxiliary pressure. NE = 9, NF = 6, NV =
6

4.2.4. Full Pressure Support: Tetrahedral Element


The tetrahedral element is divided into 4 sub-hexahedra corresponding to each primal cell vertex.
The tetrahedra has six edges, consequently six local flux continuity conditions are imposed. In this
case FPS requires five zero divergence conditions, four on the cell faces for the four surface auxiliary
pressures and one sub-volume condition for the cell-centre auxiliary pressure. NE = 6, NF = 5, NV =
4

4.2.5. Full Pressure Support: Pyramid Element


Special treatment is required for the pyramid cell. The pyramid element is divided into 5 subcells
corresponding to each primal cell vertex. The five subcells are comprised of four sub-hexahedra
corresponding to the base nodes and an octahedral subcell corresponding to the summit node [24].
The octahedral subcell has four interior interfaces inside the pyramid. The continuity conditions are
treated naturally by the sub-bilinear formulation. The pyramid has eight edges and eight local flux
9 SPE-142206

continuity conditions are imposed. In this case FPS requires six zero divergence conditions, five on
the cell faces for the five surface auxiliary pressures and one sub-volume condition for the cell-centre
auxiliary pressure. The degrees of freedom of the five equation system Eq.6 are the five interface
pressures. NE = 8, NF = 6, NV = 5

5. Relationship between FPS and CVFE for a spatially constant tensor


The relationship between respective double family FPS, TPS and double family CVFE schemes
is given in [11] for a spatially constant tensor field in two dimensions. The quadrilateral flux-
continuous schemes are mapped onto the more transparent control-volume finite element CVFE
nine-point framework by relating the schemes through their respective quadrature parameterizations
that are defined with respect to the master cell coordinates. While CVFE is not flux-continuous
the scheme is locally conservative [10], and the relationship (for a spatially constant tensor) leads
to understanding of the instability induced by the earlier TPS schemes, which are shown to yield
decoupled solutions for high full-tensor anisotropy ratios. Here 0 ≤ ξ, η ≤ 1 define master element
coordinates over the primal cell. Bilinear expansions of φ, r in ξ, η are used in defining the CVFE
double-family over the primal cell, Fig.4b, where e.g.
φ = (1 − ξ)(1 − η)φ1 + ξ(1 − η)φ2 + ξηφ3 + (1 − ξ)ηφ4 (8)
Derivatives of the bilinear expansions of φ, r in ξ, η are then used in defining the CVFE double-
family fluxes over the primal cell, Fig.4b. The resulting flux at S c.f. Fig.4(b), contributing to
control-volume 1 is given by
FS = − 12 (T11 ((φ2 − φ1 )(1 − η) + (φ3 − φ4 )η) + 12 T12 ((φ4 − φ1 ) + (φ3 − φ2 )) (9)
where 0 ≤ ξ, η < 1/2 and T denotes the cell-wise constant tensor. Quadrature points are chosen such
that they do not both coincide with the decoupled point ξ = η = 1/2 at the cell centre (discussed
below). Crucially here these double family schemes also permit quadratures with ξ 6= η. Assembly
of normal fluxes over a primal control-volume on a structured grid Fig. 4(a) leads to a double family
of 9-point schemes [11].
Note 0 ≤ ξ, η < 1/2 for fluxes defined in their respective control-volumes. The CVFE double-family
framework is a natural extension of the single family [6] and includes all possible double-parameter,
consistent, locally conservative, nine-point diagonal and full-tensor schemes, for spatially constant
tensor coefficients. By construction c.f. Eq.’s 8, 9, for spatially constant tensor coefficients the CVFE
fluxes and schemes are exact for linear and bilinear solutions of the respective pressure equations as
defined in [11].
The equivalent 3-D multi-family CVFE schemes are derived in an analogous fashion from trilinear
expansions of φ, r in ξ, η, ζ, where e.g.
φ = (1 − ξ)(1 − η)(1 − ζ)φ1 + ξ(1 − η)(1 − ζ)φ2 + ξη(1 − ζ)φ3
+(1 − ξ)η(1 − ζ)φ4 + (1 − ξ)(1 − η)ζφ5 + ξ(1 − η)ζφ6 (10)
+ξηζφ7 + (1 − ξ)η ζφ8
over a primal cell and the equivalent multi-family 3-D CVFE flux at S is defined by
FS = − 41 (T11 ((1 − η)(1 − ζ)(φ2 − φ1 ) + η(1 − ζ)(φ3 − φ4 ) + ζ(1 − η)(φ6 − φ5 )
+ηζ(φ7 − φ8 )) + T12 ((1 − ζ)(φ3 − φ2 ) + ζ(φ7 − φ6 )) (11)
+T13 ((1 − η)(φ6 − φ2 ) + η(φ7 − φ3 )))
10 SPE-142206

where FS is calculated over the subcell face b0, b1, m1, m0. Here T is a cell-wise constant tensor.
Assembly of the 3-D outward normal fluxes over the primal control-volume of a structured grid
leads to a multi-family of 27-point schemes. The above 2-D schemes can be obtained from the 3-D
formulation by setting T13 = 0 and ζ = 0.

5.1. Full Pressure Support FPS and CVFE Mapping


For a spatially constant tensor the FPS schemes that are defined over subcells and parameterized
by master (subcell) coordinates 0 ≤ ξ,e ηe < 1, are related to the CVFE schemes (parameterized by
e ηe, c), η = η(ξ,
master (cell) coordinates 0 ≤ ξ, η < 1/2) through a mapping of the form ξ = ξ(ξ, e ηe, c),
where 0 ≤ c < 1 parameterizes the size of the auxiliary control-volume surrounding the primal cell
mid-point, [11]. The direct correspondence between the subcell parameter volume and cell parameter
volume with
ξ = 21 ξe
(12)
η = 12 ηe
and 0 ≤ ξ, η < 1/2, with maximum quadrature bounded above by 1/2, only results in the limit as
c → 0, which occurs as the auxiliary control-volume size tends to zero. Thus the FPS flux integration
e ηe < 1 map onto the corresponding CVFE intervals with 0 ≤ ξ, η < 1 , for a sufficiently
intervals 0 ≤ ξ, 2
small auxiliary control-volume. Consequently with c ' 0, an analysis of CVFE applies directly to
FPS [11]. This relationship aids in an M-matrix analysis, a quasi-positive M-matrix (QM-matrix)
analysis in the region where M-matrices are violated, and in a decoupled mode analysis of the FPS
schemes in 2-D, which is also presented in [10],[11].
Triangular Pressure Support TPS and CVFE Mapping The original point-wise continuous
schemes with triangle pressure support (TPS) also map on to a double family CVFE nine-point
formulation with corresponding mapping

(pE+q−pq) (p + qE − pq)
ξ= 2(p+q−pq)
, η= (13)
2(p + q − pq)
2
[11], where E = TT11T
12
22
(ellipticity measure) and 0 < p, q ≤ 1 are local TPS quadrature coordinates
measured from the cell centre to the cell-face mid-points [9]. This shows that an M-matrix analysis
of the CVFE double-family applies to the TPS double-family. However as for the original single
parameter family, for high full-tensor anisotropy ratio where E → 1, from Eq. 13 it follows that
ξ → 1/2 and η → 1/2 leading to a decoupled formulation, discussed further below.

6. M-matrix Conditions

Well known conditions for a matrix A with elements ai,j to be an M-matrix (and for the resulting
discrete solution to have a local discrete maximum principle (DMP) as defined in [11]) are that
the diagonal coefficients be positive ai,i > 0 and the matrix be strictly diagonally dominant or
weakly diagonally dominant with strict inequality for at least one row, A must also be irreducible
and ai,j ≤ 0, for i 6= j [4]. In the analysis that follows, row inequality follows from the Dirichlet
condition, and we assume that flux coefficients are non-vanishing so that the connectivity required
for the matrices to be irreducible holds true. A summary of M-matrix conditions are given below,
details are given in [14].
11 SPE-142206

6.1. Hexahedral M-matrix Conditions


The 2-D analysis for a constant tensor field [10] shows that as the auxiliary control-volume tends
to zero, the family of FPS schemes can be expressed in terms of a CVFE family. This is also shown
for double parameter families [11].
Hexahedral schemes and M-matrix conditions are presented in [14] for single family and multi-
family planar schemes. The hexahedral-scheme formulation yields a 27-point operator in 3-D on
a structured grid, 19-point and 13-point schemes are also discussed for structured grids, while the
formulations are generalised for unstructured grids. The M-matrix conditions are summarized below.
The relationships between FPS, TPS and CVFE for spatially constant tensor coefficients in 2-D
and 3-D is also discussed in [14]. We consider the single parameter family of 3-D CVFE schemes
constructed by assembling normal cell-wise fluxes, e.g. Eq. 11 over the primal control-volume, as
a representation of FPS for spatially constant tensor coefficients, where ξ, η, ζ are master element
(primal hexahedra) parametric coordinates defined over the unit square 0 ≤ ξ, η, ζ ≤ 1, symmetry
is seen by inspection of the table. For a single parameter family parameterised by η, the mapping
between FPS and CVFE for a spatially constant tensor is given by η = ηe/2 and 0 ≤ η < 1/2 for the
single family flux over the lower left hand subcell Fig.1(as in 2-D).
Hexahedral M-matrix conditions with positive diagonal dominance and non-positive off diagonals
are
| T12 | + | T23 | + | T13 | ≤ η(T11 + T22 + T33 ) ≤ min(T11 , T22 , T33 ) (14)

− 12 η(1 − η)(T11 + T22 ) + 12 η 2 T33 + 12 (1 − η) | T12 |≤ 0


− 12 η(1 − η)(T11 + T33 ) + 12 η 2 T22 + 12 (1 − η) | T13 |≤ 0 (15)
− 12 η(1 − η)(T22 + T33 ) + 12 η 2 T11 + 12 (1 − η) | T23 |≤ 0
The above conditions ensure all off-diagonals are non-positive by symmetry and equality between the
diagonal M1 > 0 and sum of absolute off-diagonals. The two-dimensional inequalities are recovered
in each plane from Eq’s.14,15. Also the inequality resulting from summation of Eq. 15 is satisfied by
the left hand side inequality of Eq.14. The M-matrix conditions can also be derived from a cell-wise
analysis, analogous to that presented below for tetrahedra.
Reduced support may be deduced from both of Eq.’s 14,15,and the table scheme coefficients with
e.g.
η = (| T12 | + | T23 | + | T13 |)/(T11 + T22 + T33 ) (16)
However, optimal support in 3-D requires the introduction of multiple families. The multi-family
mapping between FPS and CVFE for a spatially constant tensor is given by ξ = ξ/2, e η = ηe/2, ζ =
e
ζ/2. Therefore 0 ≤ ξ, η, ζ < 1/2 for multi-family fluxes defined over the lower left hand subcell Fig.1,
equivalent bounds are enforced over the other subcells to ensure that approximations remain inside
their respective control-volumes.

6.1.1. M-matrix Conditions and Optimal Support 19 to 13 Point Schemes


Here the term optimal support refers to schemes that remain consistent while using reduced
support, e.g. in 2-D a reduction from 9-point to 7-point schemes is achieved by right (or left)
triangulation of a structured grid. The FPS mapping onto CVFE for planar 19 point schemes,
defined by collapsing quadratures to the three cross-intersecting planes, is established directly from
correspondence with the 2-D analysis [11], and it also follows that the planar families are SPD for a
12 SPE-142206

constant elliptic tensor. The M-matrix conditions for families of 19 point schemes are given by

2 | T12 |≤ (ζ2 T11 + ζ1 T22 ) ≤ min(T11 , T22 )


2 | T13 |≤ (η2 T11 + η1 T33 ) ≤ min(T11 , T33 ) (17)
2 | T23 |≤ (ξ2 T22 + ξ1 T33 ) ≤ min(T22 , T33 )

where 0 ≤ ξ1 , ξ2 , η1 , η2 , ζ1 , ζ2 < 1/2. The conditions of Eq. 17 are deduced by inspection of scheme
coefficients and choosing ξ, η, ζ to satisfy positive diagonal dominance and non-positive off-diagonal
conditions, thus ensuring sufficient conditions for an M-matrix. The resulting conditions can also
be deduced from the net cell flux contribution to a node. We emphasize that the M-matrix analysis
is for a cell-wise constant tensor. The optimal points per plane that reduces each planar 9-point
scheme to a 7-point scheme, resulting in the net 19 point scheme reducing to a net 13-point scheme
are defined by the left limit of Eq.17 where

(ζ2 T11 + ζ1 T22 ) = 2 | T12 |, (η2 T11 + η1 T33 ) = 2 | T13 |, (ξ2 T22 + ξ1 T33 ) = 2 | T23 | (18)

This provides a 3-D generalisation of the reduction of 9-point schemes to 7-point schemes via quadra-
ture, with analogous conditions to M-matrices on quadrilaterals and triangles [11]. For constant coef-
ficients the scheme will self-adapt the discretization support locally according to the local orientation
of the tensor field.
This is a multi-family (planar) generalization of the methods presented in [36] leading to families
of optimal schemes. As before the point of flux continuity on each (planar) subface determines
the scheme. For example in the ζ =constant plane, if the same parametric coordinate is chosen in
each direction then ζ1 = ζ2 and the flux continuity positions are placed symmetrically leading to a
single family, otherwise we obtain a double family [11], repeated in each plane generates a planar
multi-family formulation.
These conditions now establish the following Conditional M-matrix theorems:
(1) A family of consistent locally conservative schemes applied to the pressure equation on a structured
or unstructured hexahedral grid with a cell-wise constant full-tensor field has an M-matrix if Eq’s. 14,
15 hold. Note: This result applies to a structured or unstructured hexahedral grid since the analysis
can also be performed cell-wise with a cell-wise constant approximation of T .
(2) Families of consistent locally conservative schemes with reduced support, between 19 to 13 nodes
on a structured grid, applied to the pressure equation with a cell-wise constant full-tensor field have
M-matrices if the conditions of Eq. 17 hold. 13-point schemes are derived from Eq. 18 and the
general 19-point operator, and optimal 13-point M-matrix conditions are defined by Eq. 25, which
follows from the tetrahedra M-matrix analysis, summary below. Note: FPS fluxes are exact for
piece-wise linear and trilinear fields in 3-D since the pressure basis functions are piecewise trilinear,
c.f. Eq. 2. In these cases FPS schemes are exact for the respective pressure equation and poisson
equation that results when the divKgrad operator acts on a trilinear field. For a spatially cell-wise
constant full-tensor field the FPS schemes reduce to the schemes in the theorem, which are exact for
linear and trilinear fields as qualified above.
Both FPS and TPS schemes are exact for piecewise linear pressure fields since the basis functions
are trilinear and linear respectively. Unstable TPS solutions occur when M-matrix conditions are
violated and the solution is non-linear. Alternative optimal scheme developments in 3-D are presented
in [24, 1].
13 SPE-142206

7. Decoupled Solution Modes


Decoupling is shown to occur in two dimensions when the flux is calculated at the cell centre in
a cell vertex formulation [11]. In this case ξ = η = 1/2 defines the point of decoupling where the
flux of Eq.9 then reduces to

FS = − 21 (T11 ((φ2 − φ1 ) 21 + (φ3 − φ4 ) 12 ) + 12 T12 ((φ4 − φ1 ) + (φ3 − φ2 )) (19)

and permits the decoupled solution mode

φ1 = +1, φ2 = −1, φ3 = +1, φ4 = −1

over the cell, resulting in FS = 0, the same holds true for the other cell fluxes i.e. FE = FN = FW = 0,
Fig. 5(a). Thus the repeated decoupled field satisfies the resulting discrete 9-point pressure equation
approximation. As shown in [11] and above, crucially TPS schemes calculate the flux at, or in the

-1 +1 -1
4 3
3

2
1
-1
1 2 +1
+1 -1

(a) (b)

Figure 5: (a) Decoupled field, quadrilateral cell (b) No decoupling on triangle cell

neighborhood, of the cell mid-point for problems involving full-tensors with high anisotropy ratios,
which explains the instability introduced by the earlier point-wise continuous methods. The 3-D
schemes share an analogous property. The 3-D schemes are decoupled if ξ1 = ξ2 = η1 = η2 = ζ1 =
ζ2 = 1/2, resulting in fluxes being calculated at the cell-centres. This is seen by substitution of these
values into the 3-D flux of Eq. 11 which leads to

FS = − 41 (T11 ( 14 (φ2 − φ1 ) + 41 (φ3 − φ4 ) + 14 (φ6 − φ5 )


+ 41 (φ7 − φ8 )) + T12 ( 12 (φ3 − φ2 ) + 12 (φ7 − φ6 )) (20)
+T13 ( 21 (φ6 − φ2 ) + 12 (φ7 − φ3 )))
which defines the 3-D cell-centred flux. Substitution of the decoupled hexahedra solution mode of
Fig. 7(a)

φ1 = +1, φ2 = −1, φ3 = +1, φ4 = −1, φ5 = −1, φ6 = +1, φ7 = −1, φ8 = +1

into Eq.20 yields


FS = 0 (21)
As in 2-D all other cell-wise fluxes with cell-midpoint quadrature point are also equal to zero when
operating on a decoupled field. Therefore a repeated ±1 decoupled field satisfies the resulting discrete
14 SPE-142206

pressure equation approximation for a general grid of unstructured quadrilaterals (2-D) hexahedra
in 3-D whenever discrete divergence is approximated using fluxes with quadrature at the cell mid-
point, all fluxes are zero hence divergence is zero. Thus 27-point and 19-point are only particular
examples where decoupling can occur, the corresponding decoupled field is illustrated in the table
below for a structured 27-point operator, with decoupled 19-point operator acting on values in the
three intersecting planes.

Table 4. Decoupled solution.


φ value φ value φ value
φi,j,k−1 −1 φi,j,k 1 φi,j,k+1 −1
φi+1,j,k−1 1 φi+1,j,k −1 φi+1,j,k+1 1
φi+1,j+1,k−1 −1 φi+1,j+1,k 1 φi+1,j+1,k+1 −1
φi,j+1,k−1 1 φi,j+1,k −1 φi,j+1,k+1 1
φi−1,j+1,k−1 −1 φi−1,j+1,k 1 φi−1,j+1,k+1 −1
φi−1,j,k−1 1 φi−1,j,k −1 φi−1,j,k+1 1
φi−1,j−1,k−1 −1 φi−1,j−1,k 1 φi−1,j−1,k+1 −1
φi,j−1,k−1 1 φi,j−1,k −1 φi,j−1,k+1 1
φi+1,j−1,k−1 −1 φi+1,j−1,k 1 φi+1,j−1,k+1 −1
A 2-D unstructured quadrilateral segment shown in Fig.6 illustrates unstructured grid decoupling.
Unstructured hexahedral decoupling is illustrated by adding a top layer of the same unstructured

+1

-1
+1

-1
+1
+1
-1

-1

+1 -1
+1

(a)

Figure 6: unstructured grid decoupling

quadrilateral segment to Fig.6, only now with all signs of the illustrated decoupled mode reversed,
to form an unstructured hexahedra segment with decoupled solution mode. Solution decoupling
occurs when all quadrature parameters are set equal to a 1/2, i.e. all flux quadrature points are
at the cell-centre. There is a direct correspondence between the 2-D schemes and 3-D schemes,
both in terms of the respective mappings between CVFE and the triangular pressure support TPS
schemes and between CVFE and full pressure support FPS schemes [10, 11]. This in turn provides
a link with the 3-D TPS formulation, which for constant permeability coefficients can be rewritten
15 SPE-142206

+1 -1
7 -1
8 +1
+1 6
-1 6 5
5

+1
-1 3
4 2
-1 3
2 1 +1
-1 4 -1
1
+1 +1

(a) (b)

+1 +1
5 3

-1
4
+1
-1 3
4
2 2
1
-1 1
-1
+1 +1

(c) (d)

Figure 7: Cell-Type (a) Hexahedra - decoupling (b) Prism - partial decoupling on base and r.h.s quad face (c) Pyramid
- partial decoupling on base (d) Tetrahedra NO decoupling
16 SPE-142206

as an effective CVFE scheme with quadrature that lies in the neighborhood of the cell mid-point for
strong full-tensor fields, leading to a decoupled solution as in 2-D. The extended support of the 3-D
FPS schemes enables quadrature points to be chosen such that this neighborhood is avoided. Note
that the above (flux) analysis is for a cell-wise constant tensor. Other possible decoupled solution
modes are illustrated for the general cell or element types in Fig. 7, in particular we note that only
the hexahedra will permit decoupling in all three dimensions. The prism is confined to two planes
Fig. 7(b), the pyramid is confined to the base plane Fig. 7(c) while the tetrahedra does not permit
decoupling on any face consistent with the triangle.
The problem of decoupling can occur for problems with spatially constant high anisotropy ratio
full-tensors, as illustrated in the results section. Therefore in these cases the above analysis and
observations are directly applicable.
This observation indicates that the flux and therefore velocity field will not see the decoupled
modes of a pressure field and consequently flux is expected to converge.
Flux convergence presented in [14, 15] and in the results section confirm the decoupling theory.
Flow results presented in [11] also confirm the observation.

7.1. Multi-Family Anisotropic Quadrature

When applying the methods to general heterogeneous cases a locally upscaled tensor can be
used to define the unique ξ, η, ζ quadrature points over a cell. Strictly the local upscaling involves
performing three mini simulations subject to periodic boundary conditions over the cell with ”grid”
defined by the subcells that comprise the cell. This is only required in regions where the tensor
varies. In this work this is confined to the interface between the jumps in permeability of the
problem presented, and a harmonic based operator has been used. Note the locally upscaled tensor
is only used to define the quadrature, once the quadrature is defined, the flux continuous method is
then used to solve the original fine scale problem (with the original permeability field). However, we
can only expect to obtain an approximate optimal quadrature point and therefore the full support
of the scheme will typically be retained in such cases.
For a single family formulation a bound is required on the upper quadrature limit in order to
avoid the decoupled zone. The multi-family FPS formulation offers further advantages over the single
family schemes. First the reduction from 27 to 19 nodes in the planes can be achieved directly via
quadratures in the planes. Secondly the multi-family enables maximum flexibility in the definition
of quadrature. For example referring to Eq.18, instead of a single optimal quadrature point, there
are multiple families of optimal schemes. E.g. we may choose ζ1 to be a Gauss point and define ζ2
such that optimal quadrature is obtained. However, we have observed that solution resolution is also
sharpened by increasing the quadrature value multiplying the larger diagonal tensor coefficient [11].
This motivates the selection of extreme quadrature values for the family parameters multiplying the
larger diagonal tensor coefficients and reduced quadrature for the accompanying family parameters,
leading to anisotropic quadrature which is outside of the decoupled zone. For example if T11 > T33
then ζ3 → 1/2 while ζ1 → 0, the choice for ζ2 depends on relative ratios e.g. T11 : T22 . This strategy
is unique to the multi-family formulation and has proven to be highly effective in two and three
dimensions. While the leading quadrature parameter is chosen according to strength of anisotropy,
crucially the values of quadrature are independent of the tensor coefficients. For a spatially varying
tensor field this is an important advantage and simplification when compared to the tensor dependent
optimal quadrature points.
17 SPE-142206

8. Tetrahedral M-matrix Conditions


Cell-wise M-matrix conditions are derived for a tetrahedron. Discrete cell-vertex fluxes in a
tetrahedron with discontinuous coefficients can always be expressed as a linear combination of edge
differences in pressure [8]. Now working in a local coordinate system with origin at vertex 1 of the
tetrahedron (with vertices 1, 2, 3, 4), we express the three local components of flux as
¡ ¢
Fi = − Ti1 (φ2 − φ1 ) + Ti2 (φ3 − φ1 ) + Ti3 (φ4 − φ1 ) (22)
for i = 1, 2, 3 where Tij (j = 1, 2, 3) are approximate effective cell-wise constant tensor coefficients
derived from flux continuity conditions, and are functions of contributing subcell geometry, perme-
ability and quadrature that arise in deriving the subface fluxes with respect to subcells, as in e.g.
either side of Eq.5. The net flux contribution for a given tetrahedron vertex (index 1) from the three
interior subcell faces is given by F = F1 + F2 + F3 , and is written as
¡ ¢
F = − (T11 + T21 + T31 )(φ2 − φ1 ) + (T12 + T22 + T32 )(φ3 − φ1 ) + (T13 + T23 + T33 )(φ4 − φ1 ) (23)
Gathering coefficients of the diagonal pressure φ1 and off-diagonal pressures φ2 , φ3 , φ4 respectively,
it follows that cell-wise M-matrix conditions with diagonal dominance and positive diagonal, here
with a11 > 0 and non-positive off-diagonals a1j ≤ 0, j 6= 1 will be obtained if
| T21 + T31 | < T11
| T12 + T32 | < T22 (24)
| T13 + T23 | < T33
where symmetry may be lost in physical space. If the left hand side of Eq.24 is replaced with the
moduli of cross-terms and symmetry is assumed, then Eq.24 is satisfied if
| T12 | + | T13 | < T11 , | T12 | + | T23 | < T22 , | T13 | + | T23 | < T33 (25)
We now state a third Conditional M-matrix theorem: (3) Families of consistent locally conservative
schemes applied to the pressure equation on a tetrahedral grid with a cell-wise constant full-tensor
field have an M-matrix if Eq. 24 holds. The conditions of Eq. 25, which follow from Eq. 24, also
hold for a 13-point scheme as discussed above.

9. Positive Definite Approximation


The local discrete tensor flux approximations in this work are constructed from (a) discrete flux
continuity, (b) linear interface pressure continuity and (c) zero divergence over the cluster (dual-cell).
The flux coefficients are discrete approximations of the local tensor components defined according to
orientation of the local frame of reference at each interface and result from the constraint equations of
7, and are functions of tetrahedra subcell geometry, permeability tensor coefficients and quadrature
point.
While the discrete matrix is not generally symmetric positive definite, conditions for a positive
definite discrete matrix are deduced from the local tetrahedra contribution to inner product of
potential with the fully discrete operator.
The cell energy inner product involves cell-vertex pressures on cell fluxes of Eq. 22 and leads to
the positive definite condition (with positive quadratic form) for discrete ellipticity of the physical
space scheme, that the eigenvalues of the symmetric matrix
18 SPE-142206

 
T11 (T12 + T21 )/2 (T13 + T31 )/2
(T12 + T21 )/2 T22 (T23 + T32 )/2  , (26)
(T13 + T31 )/2 (T23 + T32 )/2 T33
are positive. The latter condition defines an average ellipticity condition with respect to the mean
symmetric tensor (T + Tt )/2 (symmetric part of T) and provides a means of testing the physical
space formulation for positive-definiteness, (c.f. [7] for cell-vertex schemes in 2-D, also cell centred
triangle SPD schemes are presented in [13]), from which solution existence and uniqueness follow,
convergence again requires exploiting the link with the mixed method.
The result is stated as a Positive Definite Theorem: (3) Families of consistent locally conservative
schemes applied to the pressure equation on a tetrahedral grid with a cell-wise effective constant full-
tensor field are positive definite if the eigenvalues of the symmetric matrix of Eq. 26 are positive.
A number of computational results involving challenging cases presented in the next section clearly
demonstrate the robustness and benefits of the new FPS formulation for three dimensional grids.

9.1. Cell-Vertex Discretization on Triangles and Tetrahedra Prevent Decoupling


Calculating flux at the centre of a quadrilateral cell in 2-D, quadrilateral face or hexahedral
cell in 3-D is shown above to lead to decoupling, where the discrete flux operator acting on the
±1 solution mode is zero. The above quadrilateral and hexahedral analysis applies to structured
and unstructured cell-vertex formulations and structured cell-centred formulations. Decoupling on
unstructured cell-centered meshes is discussed in [15]. For a cell vertex formulation decoupling does
not occur on a triangle cell in 2-D or a triangle face or a tetrahedra in 3-D. This is seen by considering
fluxes on a triangle cell, defined with respect to a local frame of reference at local vertex 1, Fig. 5(b),
the flux component F1 = −(T11 (φ2 −φ1 )+T12 (φ3 −φ1 )) and operating on a possible ±1 field indicated
in Fig. 5b leads to
F1 = 2(T11 + T12 ) (27)
Decoupling requires that all fluxes vanish. Thus for decoupling we must have T11 +T12 = 0, T21 +T22 =
0, consequently the flux is non-zero for an elliptic tensor. Also with respect to a local frame at vertex
3 of Fig. 5(b) F1 = −2T22 and therefore any non-vanishing tensor does not permit decoupling. In
the case of a tetrahedra

F1 = −(T11 (φ2 − φ1 ) + T12 (φ3 − φ1 ) + T13 (φ4 − φ1 ))

and operating on the field indicated in Fig. 7(d) leads to

FS = 2(T11 + T12 ) (28)

and φ4 − φ1 = 0. Again decoupling requires that all fluxes vanish. Thus for decoupling here we must
have T11 + T12 = 0, T21 + T22 = 0 and T31 + T32 = 0 for the three fluxes belonging the local frame of
vertex 1 to vanish. Similarly for other vertices, therefore as in 2-D the flux is non-zero for an elliptic
tensor. Therefore spurious oscillations on the mesh scale cannot occur for a cell vertex formulation
on a triangular grid in 2-D or triangular cell face or tetrahedra in 3-D, and TPS can be used in this
case [26]. The same cannot be said for a cell-centred formulation [15], where it is shown that TPS on
unstructured cell-centred triangles can also lead to decoupling, while FPS will prevent decoupling.
This observation provides a new and important motivation for choosing a cell vertex formulation
over a cell-centred formulation, in addition to mesh efficiency issues.
19 SPE-142206

9.2. Corollary: A Monotone Discretization Matrix Avoids Decoupling

An obvious feature of a decoupled solution is the oscillation between positive and negative values.
Thus we may conclude that while a scheme with a monotone discretization matrix may not be able
to yield solutions that are free of spurious oscillations, the monotone property is sufficient to prevent
decoupling, since a monotone matrix ensures that any problem with non-negative boundary data
yields a positive solution. This observation motivates the idea of constructing a monotone matrix
scheme for high full-tensor anisotropy ratios as presented in [19, 18]. An alternative approach is
presented in [27]

10. Quasi-Positive QM-matrices

While the FPS schemes presented here do not have M-matrices or monotone matrices for high
full-tensor anisotropy ratios, for quadrature points outside the decoupled zone the schemes yield re-
sults with relatively few spurious oscillations. These schemes can be categorized in terms of another
matrix. We have defined a Quasi-M-matrix or QM-matrix as a matrix with the minimum of only
one unique positive off-diagonal coefficient that violates the M-matrix conditions in the case of a
symmetric 2-D cell-wise constant matrix [10], and provided the quadrature is below the decoupled
zone. FPS schemes are shown to possess a range of QM-matrices when the fundamental M-matrix
(and monotone) conditions are violated in 2-D, when | T1,2 |> min(T1,1 , T2,2 ) [10]. The fundamental
M-matrix (and monotone) violation occurs in 3-D when Eq.24 is violated, which is the least constrain-
ing M-matrix condition in 3-D, but can be more constraining than 2-D. The QM-matrix definition
is extended in 3-D to matrices with up to three unique positive off-diagonal coefficients that violate
the M-matrix conditions, where each plane can permit one unique contribution. For example η = 0
has a QM-matrix in 2-D, and from table 1 the particular 3-D scheme corresponding to η = 0 will
have three unique coefficients that violate the M-matrix conditions, with one contribution from each
plane if the three cross-terms are non-zero, so in this case the 3-D scheme has a QM-matrix. The
schemes resulting from Eq.18 appear optimal with respect to a QM-matrix.

11. Numerical Results

A number of numerical tests have been performed to test the methods for various grids. The
earlier TPS schemes are contrasted with the FPS schemes for a range of problems in 2-D and 3-D
including hexahedral and prismatic elements. Here we present results for four test cases.

11.1. Case 1: Highly Anisotropic Discontinuous Full-Tensor

In this case the permeability tensor field has anisotropy ratio of 3000:1 at 25 degrees to the x-axis
and is defined by
µ ¶
2464.36 1148.68
K= (29)
1148.68 536.64
Dirichlet boundary conditions are imposed with pressure set to zero on external boundaries. A source
of strength 1 is specified at (0.1,0.1) and a sink of strength -1 at (0.9,0.9).
20 SPE-142206

11.1.1. Quadrilateral Element


The effect grid type on the solution is also considered. The first set of results are for a Cartesian
32x32 grid Fig.8, while the second set correspond to an anisotropy favoring quadrilateral mesh, Fig.9
with middle region at 30 degrees to the horizontal. The grid is deliberately chosen to only favor
anisotropy so that conclusions drawn are not based on assumption of perfect alignment with principal
axes.
The TPS result shown in Fig.8(b) clearly shows strong spurious oscillations on the cartesian grid,
consistent with earlier 2-D TPS results. The spurious oscillations are consistent with decoupling
discussed above. The FPS optimal support scheme yields good resolution of the pressure field and
is free of spurious oscillations Fig.8(c), as in earlier tests.
The anisotropy favoring quadrilateral mesh of Fig.9(a) proves to be highly effective. The TPS
result, while still oscillatory, shows a significant reduction in oscillations and much improvement in
pressure field resolution Fig.9(b). The FPS scheme result also has increased resolution compared to
the cartesian grid result. The anisotropy-favoring FPS triangle (optimal support) scheme result and
grid are shown in Fig. 10.

11.1.2. Quasi-Positivity Scheme Comparison


The tables 1 and 2 below present a comparison of the schemes in terms of quasi-positivity. The
FPS optimal scheme, is by definition of quasi-positivity the scheme with minimum violation of the
M-matrix conditions [11], for this constant coefficient case the scheme has 2 positive off-diagonal
coefficients of minimum size for the M-matrix violating tensor considered. A measure of scheme
quasi-positivity is defined by R which is the ratio of a positive off-diagonal coefficient for a given
scheme to that of the equivalent optimal scheme positive off-diagonal coefficient on a uniform grid. So
that if R > 1 the scheme coefficient is not optimal. The effect of the anisotropy favoring quadrilateral
grid is then determined from tables 1 and 2 below by the values of R produced for the grid. If R < 1
quasi-positivity is improved. Here FPS(OS) is FPS optimal support scheme. Iso Quad refers to the
cartesian grid. Ani Quad is the anisotropy favoring quadrilateral mesh. FPS X1 is the extreme 1
ξ = 0, η = 0.49, FPS X2 is the extreme 2 ξ = 0.49, η = 0.0.

1 1 1

0.8 0.8 0.8

0.6 0.6 0.6


y

0.4 0.4 0.4

0.2 0.2 0.2

0 0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x x x

(a) (b) (c)

Figure 8: Highly anisotropic case with cartesian grid. Plot range (-0.001, 0.001), contour number 30 (a) mesh (b)
TPS q=1. (c) FPS optimal support
21 SPE-142206

1 1 1

0.8 0.8 0.8

0.6 0.6 0.6


y

y
0.4 0.4 0.4

0.2 0.2 0.2

0 0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x x x

(a) (b) (c)

Figure 9: Highly anisotropic case with cartesian grid. Plot range (-0.001, 0.001), contour number 30 (a) mesh (b)
TPS q=1. (c) FPS optimal support

1 1

0.8 0.8

0.6 0.6
y

0.4 0.4

0.2 0.2

0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x x

(a) (b)

Figure 10: FPS on Anisotropy-favoring triangular mesh (a)mesh (b) contour

Scheme 1 +ve R>1 R=1 R<1


FPS(OS) Iso Quad 110 0 110 0
FPS Ani Quad 118 0 0 118
TPS ISO Quad 58 58 0 0
TPS Ani Quad 58 58 0 0
FPS(OS) Iso Triangle 58 0 58 0
FPS Ani Triangle 68 7 0 61
FPS X1 ISO QUAD 58 58 0 0
FPS X2 ISO QUAD 110 0 0 110
FPS X1 Ani QUAD 58 57 0 1
FPS X2 Ani QUAD 118 0 0 118
Table 1: No of rows with 1 off-diagonal positive - and R category
22 SPE-142206

Scheme 2 +ve 1 R > 1 2 R > 1 1 R = 1 2R<1


FPS(OS) Iso Quad 729 0 0 729 0
FPS Ani Quad 707 0 0 0 707
TPS ISO Quad 783 0 783 0 0
TPS Ani Quad 777 143 258 0 376
FPS(OS) Iso Triangle 783 0 0 783 0
FPS Ani Triangle 739 7 5 0 727
FPS X1 ISO QUAD 783 0 783 0 0
FPS X2 ISO QUAD 729 0 0 0 729
FPS X1 Ani QUAD 783 59 383 0 341
FPS X2 Ani QUAD 708 0 0 0 708
Table 2: Number of rows with 2 off-diagonal positives - and R category

The tables clearly indicate the benefit of using the anisotropy-favoring mesh in terms of improve-
ment in quasi-positivity.

11.1.3. Pressure and Flux Convergence


We have noted above that decoupled modes are not seen by the TPS flux, i.e. the TPS flux
is zero when operating on decoupled modes. Numerical evidence consistent with this observation
is presented in Fig. 11 below, where the TPS flux convergence rates show convergence, while the
oscillatory TPS pressure field is not converging. In contrast FPS pressure and flux are converging for
this problem. Note the errors are computed with respect to a fine scale numerical reference solution.
Effects of flux on flow are presented in [11] that also confirm computation of a physical flow is possible
while pressure is decoupled and oscillatory, though such a method is highly questionable for practical
use and certainly not recommended, particularly for compressible flow where pressure with spurious
oscillations will be damaging. Further discussion of this point is given in the reference.
L2 Pressure Error L2 Velocity Error
−12 0
TPS q=1 Error = 0.0894 TPS scheme q=1 Error = 1.8
−12.5 FPS DB xi=0, eta2=0.4995 Error = 0.634 FPS DB xi=0, eta2=0.4995 Error = 1.87
−1
FPS scheme eta= (OS) Error = 0.689 FPS scheme eta= (OS) Error = 1.89
−13
−2

−13.5
−3
−14
2

log2 L2
log L

−4
2

−14.5
−5
−15

−6
−15.5

−16 −7

−16.5 −8
4 4.5 5 5.5 6 6.5 7 7.5 8 4 4.5 5 5.5 6 6.5 7 7.5 8
log2 N log2 N

(a) (b)

Figure 11: Convergence test case: Pressure and Flux Convergence


23 SPE-142206

11.2. 3-D Unstructured Hexahedra and Prismatic Grids


Each of the cases that follow in 3-D involve a source at the domain centre and Dirichlet boundary
conditions with pressure set to zero on all external boundaries. The grids used are shown in Fig. 12,
and are comprised of a multi-block domain structure with 5 sub-domains (blocks), where the middle
sub-domain is oriented at 30o to the horizontal and contains the point source.

11.3. Highly Anisotropic Locally Aligned Tensor


The first 3-D case involves a homogeneous domain with planar high anisotropic permeability
tensor at an angle of 30 degrees to the x-axis of the global reference frame, with point source at
the domain centre and Dirichlet boundary conditions specified with zero pressure on the boundary
surfaces. The anisotropic ratio in the X-Z plane is 1000:1 along the 30 degree principal direction.
The permeability tensor is:  
750.25 0 432.58
K= 0 250.75 0  (30)
432.58 0 1
In this case the middle sub-domain of the multi-block grid is therefore aligned with principal axes
of the tensor. Therefore the area of interest in terms of scheme performance is in the surrounding
sub-domains that are not aligned with the tensor principal axes.
The nature of the decoupling in the TPS pressure contours is clear Fig 13(a). We note that
even though the tensor is essentially locally diagonal in the middle sub-domain and consequently
TPS is able to locally resolve the solution, decoupling occurs in regions around the perimeter of the
central sub-domain. This result confirms that decoupling occurs on unstructured quadrilateral and
hexahedral elements as predicted by the decoupling theory above.
The FPS extreme quadrature scheme with ξ = 0, η = 0, ζ = 0.49 is used. The extreme scheme
yields the pressure field in Fig. 13(b). The FPS scheme yields a well resolved pressure solution free
of spurious oscillations.

11.4. Case 2b: Highly Anisotropic Non-Aligned Tensor


The second case involves a homogeneous domain with planar highly anisotropic diagonal perme-
ability tensor with respect to the global reference frame, with point source at the domain centre and
Dirichlet boundary conditions specified with zero pressure on the boundary surfaces. The anisotropic
ratio in the X-Z plane is 1000:1. The permeability tensor is:
 
1000 0 0
K= 0 1 0  (31)
0 0 1
The same multi-block domain comprised of 5 sub-domains or blocks is used, again the middle sub-
domain containing the point source is at 30o to the horizontal and is this time non-aligned with
principal axes of the tensor by −30o . In this case there is a full planar tensor varying over the
whole domain, with variation according to the local coordinate system per sub-domain, providing a
different challenge to the methods.
The nature of the decoupling in the TPS pressure field is now seen to occur throughout the field,
particularly over the central most non-aligned domain Fig. 14(a). This result adds further confir-
mation that decoupling occurs on unstructured quadrilateral and hexahedral elements as predicted
by the decoupling theory above.
24 SPE-142206
Z

Z Y X

Y X
0.6

0.4

Z
0.2

0
0 0.25 0.5 0.75 10.5
1
0 0.50
1 0.2 0.4 0.6 0.8 1
Y

X Y X

(a) (b)

X
Y

0.6

X
0.4 Y
Z

0.2
0

0.2 0
0
0.4
0.25
Y
0.6
0.5
X
0.8
0.75

1 1

(c) (d)

Figure 12: Unstructured Hexahedral Mesh Comprised of 5 sub-domains (a) Hex cross-section (b) Prism cross-section
(c) Hex-Mesh (d) Tetra-Mesh
25 SPE-142206

Z Z

X Y X Y

(a) (b)

Figure 13: Plot of High anisotropic case with Anisotropic Hex. Slice Y=0.5 (a)TPS (b) FPS ξ = 0, η = 0, ζ = 0.49

The FPS extreme quadrature scheme with ξ = 0, η = 0, ζ = 0.49 is used. The extreme scheme
yields the pressure field in Fig. 14(b) and yields a well resolved pressure solution free of spurious
oscillations.
Z Z

X Y X Y

(a) (b)

Figure 14: Plot of High anisotropic case with Anisotropic Hex. Slice Y=0.5 (a)TPS (b) FPS ξ = 0, η = 0, ζ = 0.49

12. Anisotropic Tensor 50:1 on Hex and Prism

In the next case the permeability tensor has principal coefficient ratio of 50:1 oriented at 30
degrees to the x-axis in the x − y plane and ratio of 37.75 in x − z with permeability tensor given by:
 
37.75 21.22 0
K =  21.22 13.25 0  (32)
0 0 1
26 SPE-142206

The TPS hexahedra scheme is used on the grid of Fig. 12(a). While the anisotropy is relatively
low 50 : 1 in x − y, decoupling still occurs in the x − y and y − z planes, Fig. 15. The TPS Prism

Z Z Z

X Y X Y X Y

(a) (b) (c)

Figure 15: Anisotropic 50:1, 30 degrees. TPS Hex scheme (a)Slice of X=0.5 (b) Slice Y=0.5 (c) Slice Z=0.3

scheme is also applied to this case using the prism grid of Fig. 12(b) which is anisotropy favoring
in x − z. The triangulated faces are in the x − z plane and consequently while improving planar
resolution in x − z, cannot prevent TPS decoupling that arises in the other planes where the prisms
have quadrilateral faces, Fig. 16, as predicted by the decoupling analysis above.

Z Z Z

X Y X Y X Y

(a) (b) (c)

Figure 16: Anisotropic 50:1, 30 degrees. TPS Prism scheme (a)Slice of X=0.5 (b) Slice Y=0.5 (c) Slice Z=0.3

In contrast the anisotropic FPS Prism scheme yields a well resolved pressure field, that benefits
from anisotropy favoring triangulation grid in x − z and with no indication of spurious oscillations
in this case, Fig. 17.
27 SPE-142206

Z Z Z

X Y X Y X Y

(a) (b) (c)

Figure 17: Anisotropic 50:1, 30 degrees. FPS anisotropic Prism scheme (a)Slice of X=0.5 (b) Slice Y=0.5 (c) Slice
Z=0.3

13. Highly Anisotropic Tensor on Tetrahedral mesh


The permeability field is defined by the 3-D planar version of the 2-D test case above with
 
2464.36 1148.68 0
K =  1148.68 536.64 0  (33)
0 0 536.64
The same boundary conditions are applied, zero pressure on all exterior boundaries with point source
in the domain centre. The TPS tetrahedra scheme is applied to this case on a grid comprised of
a uniform array of tetrahedra Fig. 12(d). The tetrahedra prevent TPS inducing decoupling in the
solution Fig. 18, also predicted by the decoupling analysis above. The FPS tetrahedra scheme yields
Z Z Z

X Y X Y X Y

(a) (b) (c)

Figure 18: Highly anisotropic case : 3000:1,30 degree. TPS scheme on Tetra (a)Slice of X=0.5 (b) Slice Y=0.5 (c)
Slice Z=0.5

a well resolved pressure field with no indication of spurious oscillations on the grid scale Fig. 19.
FPS and TPS produce comparable solutions in this case.
28 SPE-142206

Z Z Z

X Y X Y X Y

(a) (b) (c)

Figure 19: Highly anisotropic case 3000:1, 30 degree. FPS on Tetra (a)Slice of X=0.5 (b) Slice Y=0.5 (c) Slice Z=0.5

14. Highly Anisotropic Tensor on Pyramid Mesh

The same problem and tensor field is used to test the pyramid methods. The TPS Pyramid
scheme is applied to this case on a grid comprised of a uniform array of pyramids. The triangulated
faces are in the x − z and y − z planes and cannot prevent TPS inducing decoupling in the base x − y
plane where the pyramids have quadrilateral faces, Fig. 20, again as predicted by the decoupling
analysis above. In contrast the anisotropic FPS Prism scheme yields a well resolved pressure field

Z Z Z

X Y X Y X Y

(a) (b) (c)

Figure 20: High anisotropy case 3000:1, 25 degree. TPS on pyramid (a)Slice of X=0.5 (b) Slice Y=0.5 (c) Slice Z=0.5

with no indication of spurious oscillations Fig. 21.

15. Conclusions

Properties and performance of anisotropic Darcy-flux approximations for general elements are
presented. The schemes belong to the multi-family locally conservative flux-continuous, finite-volume
framework for solving the general tensor pressure equation on tetrahedra, prisms, pyramids and
29 SPE-142206

Z Z Z

X Y X Y X Y

(a) (b) (c)

Figure 21: High anisotropy case 3000:1, 25 degree. FPS on pyramid (a)Slice of X=0.5 (b) Slice Y=0.5 (c) Slice Z=0.5

hexahedra elements in 3-D [14]. These schemes have full pressure continuity imposed across control-
volume faces, in contrast to the earlier point-wise continuous schemes and have a larger quadrature
range.
When applying the earlier point-wise continuous TPS schemes to problems with non-linear so-
lutions and full-tensor fields with high anisotropy ratios, the schemes violate the maximum principle,
and have a limited quadrature range which is shown to cause decoupling, leading to strong spurious
oscillations in the numerical solution. Three dimensional decoupling is shown to occur when calcu-
lating fluxes at quadrilateral and hexahedral cell mid-points. The earlier point-wise TPS schemes
are confined to the neighborhood of the decoupled zone at the cell centre. Decoupling analysis shows
that a decoupled TPS flux will not detect the decoupled mode. This is confirmed experimentally,
the TPS flux is found to converge at high anisotropy ratio where TPS pressure fields are oscillatory,
adding further evidence of decoupling.
Tensor-coefficient dependent family M-matrix limits for computing locally bounded solutions
free of spurious oscillations and relation to optimal support were presented in [14, 37] and are
summarized here. The schemes are shown to be positive definite, for the general tensor case. Cell-
vertex triangle and tetrahedral schemes are shown to prevent decoupling. Element types that permit
decoupling with TPS are identified and test case results using each element type confirm the predicted
properties.
The schemes are tested on a range of problems involving strong full-tensor anisotropy where both
M-matrix and monotone matrix conditions are violated. Results of improved resolution are obtained
either absent of spurious oscillations in the discrete pressure field or significantly reduced when using
the FPS schemes. Both FPS and TPS schemes are tested on anisotropy favoring grids and yield
improved resolution. FPS optimal support schemes consistently yield well resolved pressure fields.
Optimal support is also obtained by anisotropy favoring FPS schemes on prism grids which also
improve solution resolution. The multi-family formulation permits anisotropic quadratures to be
chosen that also lead to improved solution resolution. The anisotropic quadrature points remain
outside of the decoupled zone ensuring robustness of the FPS formulation.
30 SPE-142206

References
[1] I. Aavatsmark, G.T. Eigestad, B.T. Mallison, and J.M. Nordbotten, A new finite
volume approach to efficient discretization on challenging grids, in SPE Reservoir Simulation
Symposium, Houston, Texas, USA, 2007, p. paper SPE 106435.

[2] , A compact multipoint flux approximation method with improved robustness, Num. Methods
Partial Diff. Eqns, 24 (2008), pp. 1329–1360.

[3] T. Arbogast, M. F. Wheeler, and I. Yotov, Mixed finite elements for elliptic problems
with tensor coefficients as cell centered finite differences, SIAM Journal on Numerical Analysis,
34 (1997), pp. 828–852.

[4] O. Axelsson, Iterative Solution Methods, Cambridge University Press, 1994.

[5] L.J. Durlofsky, A triangle based mixed finite element finite volume technique for modeling
two phase flow through porous media, Journal of Computational Physics, (1993), pp. 252–266.

[6] M.G. Edwards, Symmetric, flux continuous, positive definite approximation of the elliptic full
tensor pressure equation in local conservation form, in 13th SPE Reservoir Simulation Sympo-
sium, San Antonio, Texas, USA, 1995.

[7] , Unstructured, control-volume distributed, full-tensor finite volume schemes with flow based
grids, Comput.Geo., 6 (2002), pp. 433–452.

[8] , Higher-resolution hyperbolic - coupled - elliptic flux-continuous cvd schemes on structured


and unstructured grids in 3-d, Int. J. Numer Meth in Fluids, 51 (2006), pp. 1079–1095.

[9] M.G. Edwards and C.F. Rogers, Finite volume discretization with imposed flux continuity
for the general tensor pressure equation, Comput. Geo., 2 (1998), pp. 259–290.

[10] M.G. Edwards and H. Zheng, A quasi-positive family of continuous darcy-flux finite volume
schemes with full pressure support, Journal of Computational Physics, 227 (2008), pp. 9333–
9364.

[11] , Double-families of quasi-positive darcy-flux approximations with highly anisotropic tensors


on structured and unstructured grids, Journal of Computational Physics, 229 (2010), pp. 594–
625.

[12] C. L. Farmer, D. E. Heath, and R. O. Moody, A global optimization approach to grid


generation, in 11th SPE Reservoir Simulation Symposium, Anaheim CA, USA, 1991, pp. 341–
350.

[13] H. A. Friis, M. G. Edwards, and J. Mykkeltveit, Symmetric positive definite flux-


continuous full-tensor finite-volume schemes on unstructured cell centered triangular grids, Siam
J. Sci. comput, 31 (2008), pp. 1192–1220.

[14] Edwards M. G. and Zheng H., Quasi m-matrix multi-family continuous darcy-flux approxi-
mations with full pressure support on structured and unstructured grids in 3-d: to appear, SIAM
Journal on Scientific Computing.
31 SPE-142206

[15] Friis H.A. and Edwards M G, A family of mpfa finite volume schemes with full pressure
support for the general tensor pressure equation on cell-centred triangular grids, Journal of
Computational Physics, 230 (2011), pp. 205–231.

[16] J.M. Hyman, M. Shashkov, and S. Steinberg, The numerical solution of diffusion prob-
lems in strongly heterogeneous non-isotropic materials, Journal of Computational Physics, 132
(1997), pp. 130–148.

[17] I.Aavatsmark, Introduction to multipoint flux approximation for quadrilateral grids, Com-
put.Geo., 6 (2002), pp. 405–432.

[18] C. Le Potier, Schema volume finis pour des operator operateurs de diffusion fortement
anisotropes sur des maillages de triangle nonstructures, C. R., Math., Acad. Sci. Paris, Ser.
I, 340 (2005), pp. 921–926.

[19] K. Lipnikov, M. Shashkov, D. Svyatskiy, and Yu. Vassilevski, Monotone finite volume
schemes for diffusion equations on unstructured triangular and shape-regular polygonal meshes,
Journal of Computational Physics, 227 (2008), pp. 492–512.

[20] M. J. Mlacnic and L. J. Durlofsky, Unstructured grid optimization for improved mono-
tonicity of discrete solutions of elliptic equations with highly anisotropic coefficients, Journal of
Computational Physics, 216 (2006), pp. 337–361.

[21] J.M. Nordbotten, I. Aavatsmark, and G.T. Eigestad, Monotonicity of control volume
methods, Numerische Mathematik, 106 (2007), pp. 255–288.

[22] M. Pal and M.G. Edwards, Family of flux-continuous finite-volume schemes with improved
monotonicity, in 10th European Conference on the Mathematics of Oil Recovery, Amsterdam,
Netherlands, 2006.

[23] , Flux-splitting schemes for improved montonicity of discrete solution of elliptic equation
with highly anisotropic coefficients, in ECOMASS CFD-2006 Conference, Egmond aan Zee, The
Netherlands, 2006.

[24] , Quasi-monotonic continuous darcy-flux approximation for general 3-d grids of any element
type, in SPE Reservoir Simulation Symposium, Houston, Texas, USA, 2007, p. paper SPE
106486.

[25] M. Pal, M.G. Edwards, and A.R. Lamb, Convergence study of a family of flux-continuous,
finite-volume schemes for the general tensor pressure equation, International Journal for Numer-
ical Methods in Fluids, 51 (2006), pp. 1177–1203.

[26] M. Pal and M. G. Edwards, Anisotropy Favoring Triangulation CVD(MPFA) Finite-


Volume Approximations. Int. J. Numer Meth, Fluids. 20 Oct. 2010, DOI:10.1002/fld.2412.

[27] , Non-linear flux-splitting schemes with imposed discrete maximum principle for elliptic
equations with highly anisotropic coefficients. Int. J. Numer Meth, Fluids. doi:10.1002/fld.2258,
2010.
32 SPE-142206

[28] R. A. Raviart and J. M. Thomas, A mixed finite element method for second order prob-
lems, in Mathematical Aspects of the Finite Element Method, Lectures Notes in Math. 606,
I. Galligani and E. Magenes, eds., Springer-Verlag New York, 1977, pp. 292–315.

[29] B. Riviere, Discontinuous Galerkin method for solving the miscible displacement problem in
porous media, PhD thesis, University of Texas at Austin, 2000.

[30] B. Riviere, M.F. Wheeler, and K. Banas, Discontinuous galerkin method applied to a
single phase flow in porous media, Comput.Geo., 49 (2000).

[31] T. F. Russel and M. F. Wheeler, Finite element and finite difference methods for contin-
uous flows in porous media, in Frontiers in Applied Mathematics, R.E. Ewing, ed., SIAM, 1980,
pp. 35–106.

[32] T.F. Russell, Relationships among some conservative discretization methods, in Numerical
Treatment of Multiphase Flows in Porous Media, Z. Chen et al., ed., Springer, Heidelberg,
2000, pp. 267–282.

[33] S.H.Lee, P.Jenny, and H.A.Tchelepi, A finite-volume method with hexahedral multiblock
grids for modeling flow in porous media, Comput.Geo., 6 (2002), pp. 353–379.

[34] S.Verma, Flexible grids for reservoir simulation, PhD thesis, Stanford University, 1996.

[35] M.F. Wheeler and I. Yotov, A multipoint flux mixed finite element method, SIAM J. Num
Anal., 44 (2006), pp. 2082–2106.

[36] H. Zheng and M.G. Edwards, Continuous darcy-flux approximations with full pressure sup-
port for structured and unstructured grids in 3-d, 11th European Conference on the Mathematics
of Oil Recovery (ECMOR XI), (2008).

[37] , Quasi-positive multi-family continuous darcy-flux approximations with full pressure sup-
port on structured and unstructured grids in 3-d; paper spe-119186, in proc: SPE Reservoir
Simulation Symposium, The Woodlands, Texas, USA, 2-4 February, 2009.

You might also like