You are on page 1of 18

This article was downloaded by: [Fondren Library, Rice University ]

On: 04 October 2012, At: 11:01


Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,
37-41 Mortimer Street, London W1T 3JH, UK

Numerical Heat Transfer, Part B: Fundamentals: An


International Journal of Computation and Methodology
Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/unhb20

AN IMPLICIT FINITE-VOLUME METHOD USING


NONMATCHING BLOCKS OF STRUCTURED GRID
a a a a
Željko Lilek , Samir Muzaferija , Milovan Perić & Volker Seidl
a
Institut für Schiffbau, Uniuersität Hamburg, Lämmersieth 90, Hamburg, D-22 305, Germany

Version of record first published: 23 Mar 2007.

To cite this article: Željko Lilek, Samir Muzaferija, Milovan Perić & Volker Seidl (1997): AN IMPLICIT FINITE-VOLUME METHOD
USING NONMATCHING BLOCKS OF STRUCTURED GRID, Numerical Heat Transfer, Part B: Fundamentals: An International Journal
of Computation and Methodology, 32:4, 385-401

To link to this article: http://dx.doi.org/10.1080/10407799708915015

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions

This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to
anyone is expressly forbidden.

The publisher does not give any warranty express or implied or make any representation that the contents
will be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug doses should
be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims,
proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or indirectly in
connection with or arising out of the use of this material.
AN IMPLICIT FINITE-VOLUME METHOD
USING NONMATCHING BLOCKS OF
STRUCTURED GRID
Downloaded by [Fondren Library, Rice University ] at 11:01 04 October 2012

Zeljko Lilek, Samir Muzaferija, Milovan Peric, and Volker Seidl


Institut fiir Schiffbau, Universitiit Hamburg, Liimmersieth 90,
D-22 305 Hamburg, Germany

This artide presents a finite-uolume method for computing flow problems using block-struc-
lured grids. The grids may move in S011UJ blocks, and they do not have to matcb at block
interfaces. However, in the resulting linear equation systems, all blocks ore implicitly
coupled and the method is fuUy conseroatioe. The discretization is of second order in both
space and ti11UJ, and the solution algorithm is based on the SIMPLE method. The approach
of treating the nonma/ching block interfaces can be applied to other types of sotulion
methods as well.

INTRODUCTION
When flows in complex geometries are calculated, one has to use either a
block-structured or an unstructured grid in order to adapt it properly to domain
boundaries. The block-structured grid uses regular lexicographic data structure
within each block, which allows the use of many efficient solvers developed for
structured grids. Whenever the geometry is not too complicated so that it can be
subdivided into a reasonable number of blocks, it is more efficient to use a
block-structured than an unstructured grid. The latter requires connectivity tables
that identify the neighbors of each node; due to indirect addressing, larger memory .
requirement, and more complicated solvers for linear equation systems, the com-
puting times per iteration are usually longer on unstructured than on block-struc-
tured grids. Unstructured grids, on the other hand, are preferred for very complex
geometries and automation of grid generation.
The block-structured grid or zonal approach is often used, either with
overlapping [1, 2] or nonoverlapping [3] blocks. In most cases the computation is
performed separately in each block, whereas the solution in one block provides
boundary conditions for the neighbor blocks. The convergence of such an approach
is improved if the grids overlap, which is known as Schwartz's method [4]. This

Received 28 April 1997; accepted 27 June 1997.


This work was sponsored by the German Science Foundation (Deutsche Forschunsgemeinschaft);
the authors greatly appreciate this support.
Samir Muzaferija is on leave from the Faculty of Mechanical Engineering, University of
Sarajevo. Bosnia and Herzegovina.
Address correspondence to Dr. Milovan Peric, lnstitut fiir Schiflbau, Universitat Hamburg,
Lammersieth 90, D-22 305, Hamburg, Germany. E-mail: peric@schiffbau.uni-hamburg.de
Downloaded by [Fondren Library, Rice University ] at 11:01 04 October 2012
FINITE·VOLUME METHOD FOR BLOCK·STRUCTURED GRIDS 387

The following two sections describe the discretization and the solution
method. Finally, conclusions and future steps are presented.

GOVERNING EQUATIONS AND THEIR DISCRETIZATION

Governing Equations
The starting point in a finite-volume (FV) method are the conservation
Downloaded by [Fondren Library, Rice University ] at 11:01 04 October 2012

equations for mass, momentum, and scalar quantities (enthalpy, species concentra-
tion, turbulent kinetic energy, etc.) in their integral form, which read:

d
-fpdV+ fp(v - Vb)' n dS = 0
dt v s
d
dt f/ U i dV + IsPu/v - Vb) • n dS = Is Tijij • n dS - IsPi;' n dS + f/b; dV (2)

d
- f pep dV + fpep(v - vb) . n dS
dt v s
= frs grad ep' n dS + f vpb.p dV (3)

In these equations, p is the fluid density, V is the control volume (CV)


bounded by a closed surface S, V is the fluid velocity vector whose Cartesian
components are U i , Vb is the velocity of the CV surface, t is time, r is the diffusion
coefficient and b", is the volumetric source of the conserved scalar quantity ep, P is
the pressure, b, is the body force in the direction of the Cartesian coordinate Xi' n
is the unit vector normal to S and directed outward, and T;j are the components of
the viscous stress tensor defined (for Newtonian incompressible fluids considered
here) as

aui aUj)
T-- =
'J ( aX +ax;-
J.L -
j
(4)

with J.L being the dynamic viscosity of the fluid.


When the control volume moves, the so-called space conservation law (SCL),
which is expressed by the following relation between the rate of change of CV
volume and its surface velocity [5], also has to be satisfied:

d
- f dV - fv b • n dS = 0 (5)
dt v s

In order to solve these equations using the FV method, the solution domain
has to be subdivided into a finite number of CVs that can in principle be of any
shape. In this study only two-dimensional (2D) problems and quadrilateral CVs will
be considered; however, the method extends in a straightforward way to CVs and
grids of any kind. Also, only implicit time-integration schemes will be considered,
since the method is meant to be efficient for solving steady- as well as unsteady-flow
__ ~Ll~_~
Downloaded by [Fondren Library, Rice University ] at 11:01 04 October 2012

and
FINITE·VOLUME METHOD FOR BLOCK·STRUCTURED GRIDS 389
Downloaded by [Fondren Library, Rice University ] at 11:01 04 October 2012

Figure 2. A typical CV and the notation


used.

CV volume, which is for plane geometries calculated as


V = H(rne - r sw) X (rnw - r se)] (6)
where rne is the position vector of the cell comer ne; d. Figure 2. For
axisymmetric geometries, the CV volume is defined by the rotation of the
front surface by 1 rad around the axis of symmetry and can be calculated as
1 4
V= 6" L (Xl - XI+I)(Y,z + y,z+ 1 + YIYI+l) (7)
I-I
where X denotes the axial and Y the radial directions, and denotes CV
vertices numbered counterclockwise.
Components of the surface vector Sn = S'j + SYj:

S: = (Yne - Yse)re S[ = (x se - xne)re


(8)
S~ = (Ynw - Yne)rn S~ = (x ne - xnw)rn
where re and r n are the distances of a cell-face center from the axis of
symmetry (for plane 2D problems, r is set to unity). Note that (Sn), for a CV
centered around node P equals -(So)n of the CV centered around node S;
thus, only surface vector components for the east and north faces are
computed and stored.
Interpolation factors Ae and An' which are defined as

(9)

The above quantities related to the east and north faces of a CV are
calculated once and stored at a location associated with that CV. However, this is
done for inner faces only; if the east or north side of a CV coincides with either a
domain boundary or a block interface, the memory locations reserved for these
quantities remain empty. For the boundary and block-interface faces, a separate
data structure is introduced.
In the interior of each block, one CV has four neighbors with which it shares
common faces. However, as the grid may not match at block interfaces, CVs along
Downloaded by [Fondren Library, Rice University ] at 11:01 04 October 2012
FINITE·VOLUME METHOD FOR BLOCK·STRUcrURED GRIDS 391

approximations are necessary since the values of the integrand, f = PtP(v - Vb) in
convective and f = r grad tP in diffusive fluxes [see Eq. (3)] are not known at the
cell-face center. Therefore, interpolation and numerical differentiation have to be
used to express the cell-face values of variables and their derivatives through the
nodal values. Details on various options can be found in Ferziger and Peric [6];
here only the methods used in the present calculations will be briefly described.
Cell-face values of variables are approximated using linear interpolation:
Downloaded by [Fondren Library, Rice University ] at 11:01 04 October 2012

(10)

where Ae was introduced in Eq. (9). This is a second-order approximation at the


location e' on the straight line connecting nodes P and E. If the cell-face center e
does not coincide with the location e', as in Figure 2, the integral approximation

f f· n dS "" FXS' + trs:


S,
Je e Je e
(11)

will not be second-order-accurate. As long as e' is close to e, as is the case in


Figure 2, the first-order error term is small; however, if e' comes close to cell
corners, the first-order term will dominate. Thus, the grid quality can be analyzed
by comparing Ire - r~1 and Irne - rs.J. The second-order accuracy can be restored
by adding a correction term as follows:

tPe "" tPe' + (grad tP )e' . (r, - re,) (12)

where the gradient at e' can be obtained by interpolating the cell-center gradients
using Eq. (10).
Linear interpolation leads in an FV method to the same result as the use of
central differences for the first derivatives in finite-difference (FD) methods, which
is why it is usually referred to as a central-differencing scheme (CDS).
It is advantageous to use the so-called deferred correction approach when
implementing this scheme in implicit solution algorithms [7]. In that case only the
first-order-upwind approximation (UDS: tPe "" tPP if the flow is from P to E,
otherwise tPe "" tPE) is fused to calculate the elements of the coefficient matrix,
while the explicitly calculated difference between the CDS and UDS approxima-
tions is added to the right-hand side of the equation system:

(13)

The term in parentheses can be multiplied by a factor 0 .,; 'Y .,; 1, thus blending the'
two schemes; this may be necessary in order to suppress the oscillations near
discontinuities (e.g., shocks in compressible flows) or peaks in profiles (e.g., kinetic
energy of turbulence or its dissipation rate near solid walls) when the grid is too
coarse. Note that the same approximation is used both at the block interfaces and
at regular inner cell faces.
In order to calculate the diffusive fluxes, the gradient vector at the cell face is
required. To this end one can use coordinate transformation. as is most often done
392 Z. LlLEK ET AL.

in conjunction with structured grids [8]. However, a much simpler approach,


developed by Muzaferija [9], is possible. The gradient is first calculated explicitly at
CY centers using midpoint-rule approximation based on the Gauss theorem:

(k = e.w;n.s, ... ) (14)

Here, cf>k is calculated in the same way as in convective fluxes (although one could
Downloaded by [Fondren Library, Rice University ] at 11:01 04 October 2012

use a different approximation; however, to preserve the accuracy of the gradient


approximation, the interpolation should be of at least second order).
The above approximation of the gradient is valid for CYs of arbitrary shape;
note that for CYs along block interfaces the contribution of all subfaces must be
included. Also, in the case of axisymmetric geometries, one' has to take into
account the contribution of the front and back faces when calculating the deriva-
tive with respect to r, since they do not cancel out as in the case of plane
geometries. The additional term to be added on the right-hand side of Eq. (14) is
- cf>pS, where S is the area of the front face.
The cell-center derivatives can now be interpolated to the cell-face centers
using the same interpolation technique as for convective terms. However, oscilla-
tory solutions may develop in this case; see Ferziger and Peric [6] for detailed
discussion. Muzaferija [9] suggested the following approximation, which prevents
oscillations and retains the second-order accuracy:

(15)

The underlined term is calculated using prevailing values of the variables and
treated as another deferred correction; see above. If the line connecting nodes P
and E is orthogonal to the cell face, the underlined term is zero (since the vectors
r E - r p and n, are then co-linear) and the usual central-difference approximation
of the derivative is recovered. The explicitly calculated gradient at the cell face
(denoted by overbar in the above expression and obtained by linear interpolation)
is used only to account for the cross-derivative. Thus, the implicit part of the
approximation involves only the nearest neighbors, resulting in a compact coeffi-
cient matrix. Note that in the momentum equations, not only the normal derivative
of the dominant velocity component, but also tangential derivatives of the other
components are needed at the cell-face center. These can be obtained by interpo-
lating cell-center values, as they cause no problems.
The convective fluxes are nonlinear terms, so linearization is necessary. Here
the simple Picard iteration is employed. The mass flux through the cell face is
taken from the previous iteration:

(16)

The calculation of mass fluxes is described below. Source terms may also be
nonlinear; the same approach can be used to linearize them.
FINITE-VOLUME METHOD FOR BLOCK-STRUcruRED GRIDS 393

Approximation of Time Integral


In the case of unsteady flows, the integration also needs to be performed in
time. The equations are divided by lit for the sake of simplicity. The simplest
implicit scheme is the first-order Euler scheme: The surface and volume integrals
are calculated at the new time level, so the solution from the previous time step
appears only in the unsteady term,
Downloaded by [Fondren Library, Rice University ] at 11:01 04 October 2012

1 f' l/J~+! - l/J~


- n+'l/Jdt"" - - - - - (17)
lit ~ lit

where the superscript n denotes the time levels. Here, l/J stands for the volume
integral in the unsteady term of Eqs. (1)-(3); it involves only the variable values at
the CV center. This scheme is simple and unconditionally stable, but the first-order
error term introduces serious numerical diffusion effects, so the scheme is useful
only when time accuracy is not important (e.g., when marching toward a steady
solution). When the time history of flow is important, a second-order scheme is
necessary. One of the simplest implicit second-order schemes is the three-time-
levels scheme. It implies integration over a time interval lit centered around the
time level t n + 1. The surface and volume integrals at time t; + 1 are then centered
approximations with respect to time, so multiplying them by lit is a second-order
approximation of the time integral. A second-order approximation of the time
derivative at time t n + 1 is achieved by fitting a parabola through solution at three
time levels, thus:

1 f,.+1+111/2 al/J 1 (al/J)n+! 3l/J~+1 - 4l/J~ + l/J~-1


- -- dt "" - lit - (18)
lit '.+ 1 -111/2 at lit at 2M

This scheme is as simple as the implicit Euler scheme (one only needs to store the
solution at one more level; there is no computational overhead). It is also less
prone to oscillations than the Crank-Nicolson scheme, with almost the same
accuracy. More details about various time-integration techniques can be found in
[6] and other textbooks.

The Effects of Grid Movement


In many cases the grid has to move, either as a result of a predefined
movement of the solution domain boundary (e.g., piston-driven flows) or in the
course of grid adaptation to the newly calculated shape of the boundary (e.g., in
free-surface flows). The space conservation law (SCL) then has to be taken into
account. Why this is important can be seen by considering the mass conservation
equation (1) for a fluid of constant density: It can then be written as

d
-At f.dV - fV . n dS + i:
",. b ...
.n dS =0 (19)
Downloaded by [Fondren Library, Rice University ] at 11:01 04 October 2012
FINITE-VOLUME METHOD FOR BLOCK-STRUcrURED GRIDS 395

Algebraic Equation Systems


Upon separation of the terms involving unknown variables and the explicitly
calculated ones, an algebraic equation of the form

I=E,W,N,S, ... (24)

is obtained for each CV. The coefficients A E, A w , AN' and As contain contribu-
Downloaded by [Fondren Library, Rice University ] at 11:01 04 October 2012

tions from surface integrals over faces common to the cell around node P and the
corresponding neighbor; A p contains in addition contributions from the unsteady
term and possibly from source terms (volume integrals). Qp contains all terms that
are treated as known (source terms, parts of surface integrals treated explicitly as
deferred corrections, part of the unsteady term). These coefficients are stored as
cell data; i.e., they are associated with the cell index. However, for CVs next to a
block interface, more than four neighbors are involved. Since this does not fit into
the regular data structure within a block, the coefficients resulting from flux
approximations at interface faces are associated with the face and included in the
interface data list. This is possible since in the present method the coefficient AI
contains only terms resulting from surface integral over cell faces separating CVs
around P and around neighbor I; the contributions of a particular neighbor node to
integrals over other cell faces are treated explicitly. Thus, if the east side of a CV is
a block interface, the coefficient A E is set to zero, as the cell face e is not included
in the regular data structure. On the other hand, in the list of interface cell faces,
there are two storage locations for the coefficients A L and A R associated with
that face; see Figure 3. The algebraic equation at node L receives the contribution
ALtPR' while at node R, the contribution is ARtPL' Since for all conservative
schemes [6]

Ap = - L:Ai + L:mk k = e,w,n,s ... (25)


I k

where the superscripts c and d denote contributions from convective and diffusive
fluxes, respectively, there is no need to store at interfaces the contributions to the
coefficient A p of the cells around Land R nodes, since these will come indirectly
through the neighbor contributions; see [6] for more details.
For the solution domain as a whole, the algebraic equation system can be
written as
Acl> = Q (26)

where A is a square N X N coefficient matrix, cl> is the vector of unknowns, Q is


the vector of right-hand sides, and N is the number of CVs. The global matrix A is
irregular; however, since an iterative method will be used to solve the linearized
equation systems, the irregularity can be easily dealt with: One simply splits the
global matrix into a sum of the decoupled local matrices within each block
(remember: AI = 0 if 1 refers to a location on the other side of the interface) and
the coupling matrix (which includes contributions from interface cell faces). The
necessary modifications to the linear equations solver are described below.
396 Z. LILEK ET AL.

Calculation of Pressure
The pressure-velocity coupling is achieved using the wel1-known SIMPLE
algorithm [10]. The solution process starts with a guessed pressure field. Each time
the linearized momentum equations are solved, mass conservation is imposed on
the new velocities (to within a certain tolerance) by applying a velocity correction
that is proportional to the gradient of the pressure correction. The dependence of
velocity on the pressure gradient is obvious from the momentum equation. Special
care is needed in a co-located variable arrangement to avoid pressure-velocity
Downloaded by [Fondren Library, Rice University ] at 11:01 04 October 2012

decoupling. Detailed discussion of this issue and the measures to solve the problem
is available in [6]; here only the main steps are given.
The velocity field obtained by solving the linearized momentum equations is
denoted by ui. The normal velocity component at cel1 face e is calculated by
interpolating neighbor nodal values and subtracting a correction term that should
detect oscil1ations and help smooth them out [9, 11]:

The overbar denotes interpolation from neighbor nodal values (here linear; see
Lilek and Peric [12] for details about using higher-order interpolation and integra-
tion techniques); the double overbar denotes arithmetic averaging. The correction
term in square brackets vanishes when the pressure variation is linear or quadratic;
it is proportional to the square of mesh spacing and a third derivative of pressure
[6]. Thus, this term is a second-order correction that goes consistently toward zero
as the grid is refined; it is large only when the pressure variation is not smooth.
The normal velocity component is proportional to the normal derivative of
pressure, which can be approximated at the cel1-face center by the fol1owing
central-difference approximation:

PE' + pp'
(28)
(r E - r p) • n

Here, an approximation is made in that the pressure values at locations E' and P'
are replaced by pressure values at E and P. This neglecting of the nonorthogonality
does not affect the convergence of the procedure if the nonorthogonality is not
severe (i.e., when the angle between vectors r E - r p and n is smal1er than 45°).
The use of the correct normal derivative would complicate the resulting algebraic
equation; the error can be eliminated by applying another correction, as described
in [6].
The mass fluxes calculated using the above normal velocity do not-in
general-satisfy the mass conservation law. They need to be corrected by invoking
a pressure correction, in the spirit of the SIMPLE method [10]. From the above
equation one can derive a relation between the corrections of the normal velocity
Downloaded by [Fondren Library, Rice University ] at 11:01 04 October 2012
Downloaded by [Fondren Library, Rice University ] at 11:01 04 October 2012
Downloaded by [Fondren Library, Rice University ] at 11:01 04 October 2012
400 Z. LILEK ET AL.

tion algorithm, the only difference being that domain decomposition for a struc-
tured grid corresponds to a block-structured grid with matching interfaces. Al-
though multigrid acceleration was not tried in the present method with nonmatch-
ing grids, experience with parallel computing and structured grids suggests that the
multigrid method would be little affected by the modification of the iteration
matrix on block-structured nonmatching grids [16].
Downloaded by [Fondren Library, Rice University ] at 11:01 04 October 2012

CONCLUSIONS
In the preceding sections a finite-volume method was presented that uses
nonoverlapping block-structured grids with nonmatching interfaces. It is based on
second-order discretization in space and time (midpoint rule integration, linear
interpolation, central differences) and uses the SIMPLE algorithm for
pressure-velocity coupling and an ILU-type linear equation solver. The main
features of the method can be summarized as follows.

The method is easy to implement and represents a compromise between the


effectiveness of geometrically inflexible structured grids and the flexibility of
unstructured grids. No coordinate transformation is used, which is especially
attractive for the implementation of complex turbulence models.
The ability to use nonmatching block interfaces allows blockwise local grid refine-
ment, which is especially useful when flows around bodies are studied. Since
the block interfaces are treated implicitly and in a fully conservative manner,
the convergence properties are almost the same as in the case of a structured
grid.
The method is applicable to problems that require moving and sliding grids, since
the grid may move in some and stay fixed in other blocks.

The method can be easily extended to three-dimensional problems. The


treatment of nonmatching block interfaces, when parts of grid have to move, is
applicable to unstructured grids as well.
In a companion article [17], results of computations for three unsteady-flow
problems are presented, demonstrating the applicability, accuracy, and efficiency of
the method.

REFERENCES
1. B. J. Hubbard and H. C. Chen, Calculations of Unsteady Flows around Bodies with
Relative Motion Using a Chimera RANS Method, Proc. 10th ASCE Engineering Mechan-
ics Conf., vol. II, pp. 782-785, University of Colorado at Boulder, Boulder, CO, 21-24
May 1995.
2. C. Y. Perng and R. L. Street, A Coupled Multigrid-Domain-Splitting Technique for
Simulating Incompressible Flows in Geometrically Complex Domains, Int. J. Numer.
Meth. Fluids, vol. 13, pp. 269-286 1991.
FINITE-VOLUME METHOD FOR BLOCK·STRUcrURED GRIDS 401
C;

3. P. Coelho, J. C. F. Pereira, and M. G. Carvalho, Calculation of Laminar Recirculating


Flows Using a Local Non-Staggered Grid Refinement System, Int. 1. Numer. Meth.
Fluids, vol. 12, Pl? 535-557, 1991.
4. H. A. Schwartz, Uber einige Abbildungsaufgaben. Gesammelte Mathematische Abhand-
lungen, Vierteljahrzeitschrift der Naturforschenden Gesellschaft in Zurich, vol. 15, pp.
272-286, 1870.
5. I. Demirdsic and M. Peric, Space Conservation Law in Finite Volume Calculations of
Fluid Flow, Int. J. Numer. Meth. Fluids, vol. 8, pp. 1037-1050, 1988.
6. J. H. Ferziger and M. Peric, Computational Methods for Fluid Dynamics, Springer-Verlag,
Downloaded by [Fondren Library, Rice University ] at 11:01 04 October 2012

Berlin, 1996.
7. P. K. Khosla and S. G. Rubin, A Diagonally Dominant Second-Order Accurate Implicit
Scheme, Comput. Fluids, vol. 2, pp. 207-209, 1974.
8. I. Demirdzic and M. Perle, Finite Volume Method for Prediction of Fluid Flow in
Arbitrarily Shaped Domains with Moving Boundaries, Int. J. Numer. Meth. Fluids, vol.
10,pp. 771-790, 1990.
9. S. Muzaferija, Adaptive Finite Volume Method for Flow Predictions Using Unstruc-
tured Meshes and Multigrid Approach, Ph.D. thesis, University of London, 1994.
10. S. V. Patankar and D. B. Spalding, A Calculation Procedure for Heat, Mass and
Momentum Transfer in Three-Dimensional Parabolic Flows, Int. 1. Heat Mass Transfer,
vol. 15, pp. 1787-1806, 1972.
11. C. M. Rhie and W. L. Chow, A Numerical Study of the Turbulent Flow Past an Isolated
Airfoil with Trailing Edge Separation, AlAA J., vol. 21, pp. 1525-1532, 1983.
12. Z. Lilek and M. Perle, A Fourth-Order Finite Volume Method with Colocated Variable
Arrangement, Comput. Fluids, vol. 24, pp. 239-252, 1995.
13. I. Demirdzic, Z. Lilek, and M. Peric, A Colocated Finite Volume Method for Predicting
Flows at all Speeds, Int. J. Numer. Meth. Fluids, vol. 16, pp. 1029-1050, 1993.
14. H. L. Stone, Iterative Solution of Implicit Approximations of Multi-Dimensional Partial
Differential Equations, SlAM J. Numer. Anal., vol. 5, pp. 530-558,1968.
15. Z. Lilek, S. Muzaferija, and M. Peric, Efficiency and Accuracy Aspects of a Full-Multi-
grid SIMPLE Algorithm for Three-Dimensional Flows, Numer. Heat Transfer, Part B,
vol. 31, pp. 23-42, 1997.
16. E. Schreck and M. Peric, Computation of Fluid Flow with a Parallel Multigrid Solver,
Int. 1. Numer. Meth. Fluids, vol. 16, pp. 303-327 1993.
17. Z. Lilek, S. Muzaferija, M. Peric, and V. Seidl, Computation of Unsteady Flows Using
Nonmatching Blocks of Structured Grid, Numer. Heat Transfer, Part B, vol. 32, pp.
403-418, 1997.

You might also like