Professional Documents
Culture Documents
LONG CHEN
The finite volume method (FVM) is a discretization technique for partial differential
equations, especially those that arise from physical conservation laws. FVM uses a volume
integral formulation of the problem with a finite partitioning set of volumes to discretize
the equations. FVM is in common use for discretizing computational fluid dynamics equa-
tions.
(a) Mesh and dual mesh of cell-centered FVM (b) Mesh and b.
dual mesh of vertex-centered FVM
a. Cell-centered b.a.Vertex-centered
Cell-centered Vertex-centered
Figure 1. in
Control volume
F IGUREFigure 1. Control
1. Mesh andvolume
dual variants
mesh used
of two the finite
FVMs.
(a) cell-centered
(a) cell-centered and (b) vertex-centered controland
Thevariants
volumeunknownsused in the finite volume method:
method:
(b) vertex-centered
volume tessellation.
are asso-
control volume tessellation.
ciated to black nodes.
1 depicts
1 depicts a 2-D triangle complex and atwo2-Dtypical
triangle complex
control and tessellations
volume two typical control
(amongvolume
many tessellations (among many
others) used in the3.finite others)
volume
V ERTEX used inInthe
-method.
CENTERED thefinite volume
cell-centered
FINITE method. In themethod
finite volume
VOLUME METHOD cell-centered
shownfinite
in volume method shown in
Fig. serve
Fig. 1a, the triangles themselves 1a, theastriangles themselves
control volumes withserve as control
solution volumes
unknowns with of
(degrees solution unknowns (degrees of
freedom)
We now stored on another
discuss freedom)
a per triangle stored
basis.
popular onvertex-centered
In the
choice a per
of V triangle
and B. basis.
finite In the vertex-centered
volume
To simplify method shown finite
the notation, in wevolume
con- method shown in
Fig. 1b, control volumes are Fig. 1b, as
formed control volumesdual
a geometric are to
formed as a geometric
the triangle complexdual
and to the triangle complex and solution
solution
sider two dimensional triangular grids
unknowns
unknowns stored on a per triangulation
and homogenous
stored
vertexonbasis.
Dirichlet
a per triangulation vertexboundary
basis. condition. We
refer to [16] for the general treatment in high dimensional simplicial grid and [15] for
rectangle grids.
Encyclopedia Encyclopedia
of Computational Mechanics. of Computational
Edited Mechanics.
by Erwin Stein, Edited
Rene de Borst by Thomas
and Erwin Stein, Rene de Borst and Thomas J.R. Hughes.
J.R. Hughes.
c 2004 John Wiley
! 2 c 2004 John Wiley & Sons, Ltd.
& Sons, Ltd.!
Let R be a polygon and let T be a triangular grid of . Denoted by VT be the
linear finite element spaces of H01 () based on T :
VT = {v H01 () : v| P1 ( ), T },
where P1 ( ) is the linear polynomial space on . We shall choose V = VT . The dimension
N is the number of interior vertices of T .
FINITE VOLUME METHODS 3
F IGURE 2. Three types of grids and dual grids. The gray areas are the
control volumes of interior nodes. Type A: The point c is the barycenter
of . Type B: The points c is the middle point of the longest edge. Type
C: The point c is the circumcenter of .
1
4 LONG CHEN
VT such that
and formulate the linear finite volume method as: find u
(7) a u, v) = (f, v) for all v VB .
(
Remark 4.1. For Neumann boundary condition, we shall choose
VT = {v H 1 () : v| P1 ( ), T }, and
VB = {v L2 () : v|b = constant, bi B}.
i
For e bi , the flux (Ku) ne will be given by the boundary condition and can
be moved to the right hand side. All algorithms and analysis in this notes can be applied
to Neumann boundary condition in a straightforward way.
Since the trial space VT and the test space VB are different, the weak formulation (7) is
known as Petrov-Galerkin method.
The nodal basis of linear finite element space VT is the standard hat function:
i VT , i (xj ) = ij , xj N (T ), i = 1, , N.
PN j j . Choosing v = i , i = 1, , N in (7), we obtain a linear algebraic
Let u
= j=1 U
equation
(10) AU
= F ,
with Z Z
Aij = (Kj ) n, Fi = f dx.
bi bi
PN
Let uL = j=1 Uj j . Choosing v = i , i = 1, , N in (9), we obtain another linear
algebraic equation
(11) AU = F,
with Z Z
Aij = (Kj ) i , Fi = f i dx.
We shall prove that when K(x) is piecewise constant on each triangle, then A = A;
See [1, 10, 16]. The solution vectors are point values for uL and u at vertices.
R The only
difference is the different way to compute the right hand side. For FEM, Fi = i f i dx,
is a weighted average over the star of a vertex. For FVM, Fi = bi f dx is the average over
R
the control volume bi . When we choose type A control volume, i.e. choosing c to be the
barycenter of , Fi can be thought as an approximation of Fi using mass lumping. This
modification enables the solution of linear FVM to satisfy the conservation property. It is
interesting to note that on uniform grids, three methods (FDM, FEM and FVM) result the
same matrix (up to a scaling). The right hand side are chosen from very different perspec-
tive. And amazingly for all three choices of right-hand side, the resulting approximation
converges to the same solution with the same order.
Lemma 5.1. Given a polyhedron with k-sides in Rn , let |Fi | denote the (n1)-measure
of the face Fi and ni the outward normal of the i-th side. Then
k
X
(12) |Fi |ni = 0.
i=1
Proof. We pick up any interior point x and define i (x) as the simplex spanned
by Fi and x. Note that |i (x)| is a linear function of x and |i (x)| = 0 for x Fi .
Thus the direction of the vector i is the normal direction of the plane i (x) = 0. The
magnitude can be computed by a simple difference formula of directional derivatives to
yield |i (x)| = |Fi |ni .
Pk
We then differentiate i=1 |i (x)| = || to get
k
X k
X
|Fi |ni = |i (x)| = 0.
i=1 i=1
The following theorem is critical which says FVM and FEM have the stiffness matrix.
Let us introduce an isomorphism between linear spaces G : VB VT by mapping i
PN PN
i , 1 i N . Then for any u = i=1 ui i VB , Gu = i=1 ui i VT . Note that
6 LONG CHEN
u and Gu share the same vector representation U = (u1 , , uN )T . We also use a simple
notation u
to denote Gu.
Theorem 5.2. Assume K(x) is piecewise constant on each T and bi consists
of middle points of edges, then
a(u, v) = a
(u, v), u, v VT .
Proof. Since we assume K(x) is piecewise constant, we need only show the local stiffness
for Poisson equation on one triangle coincides. That reduces to prove
a
(i , Gj ) = a(i , j ).
To be specific, let us take 1 , 2 as an example. Since 1 is a constant over the triangle
, we can pull it out the integral and apply Lemma 12 twice to get
Z
1 (n1 + n2 )dS = 1 (|e1 |ne1 + |e2 |ne2 )
e1 e2
Z
1 1
= 1 (|l1 |n1 + |l3 |n3 ) = 1 |l2 |n2 = 1 2 .
2 2
6. E RROR ANALYSIS
The error analysis of vertex centered linear FVM (7) relies on the close relation between
linear finite element method and the linear finite volume method. From previous section,
linear FVM approximation u can be thought as a perturbation of the FEM approximation
uL . First order optimal convergence rate in the energy norm can be obtained using this
relation.
Note that the right hand sides may be quite different for type B dual mesh. For example,
let f = 1 and consider the control volume in Figure 2(b). Then Fi = |i |/3 while Fi =
|i |/4. Nerveless optimal first order convergence in H 1 norm can be still derived by
comparing them in H 1 norm [10].
Furthermore if we choose type A dual mesh, then optimal second order convergence in
L2 -norm can be also derived [10, 4, 7, 8]. Note that for general choice of control volumes,
finite volume approximation may not lead to optimal L2 -norm convergent rate. See [11]
for such an example on type B dual mesh. Optimal L norm estimate can be also obtained
by treating it as a perturbation of finite element methods; See [7, 8]. We are not going to
discuss L2 or L error estimate.
Theorem 6.1. Assume K(x) is piecewise constant on each T , the solution u of the
diffusion equation is in the space H01 () H 2 () and the mesh is quasi-uniform with
mesh size h, then the finite volume approximation uh has optimal approximation order
(13) ku uh k1, . h(kuk2, + kf k).
Proof. For any f L2 (), we define h f V0T as
hh f, vh i = (f, Gvh ), for all vh VT .
and Qh f V0T as
hQh f, vh i = (f, vh ), for all vh VT .
FINITE VOLUME METHODS 7
R EFERENCES
[1] R. E. Bank and D. J. Rose. Some error estimates for the box method. SIAM J. Numer. Anal., 24:777787,
1987.
[2] Z. Cai. On the finite volume element method. Numer. Math., 58(1):713735, 1990.
[3] Z. Cai, J. Mandel, and S. F. McCormick. The finite volume element method for diffusion equations on
general triangulations. SIAM J. Numer. Anal., 28:392402, 1991.
[4] P. Chatzipantelidis. Finite volume methods for elliptic PDEs: a new approach. M2AN Math. Model. Numer.
Anal., 36(2):307324, 2002.
[5] L. Chen. A new class of high order finite volume methods for second order elliptic equations. SIAM J.
Numer. Anal., 47(6):40214043, 2010.
[6] S.-H. Chou, D. Y. Kwak, and Q. Li. Lp error estimates and superconvergence for covolume or finite volume
element methods. Numer. Methods Partial Differential Equations, 19(4):463486, 2003.
[7] S.-H. Chou and Q. Li. Error estimates in L2 , H 1 and L in covolume methods for elliptic and parabolic
problems: a unified approach. Math. Comp., 69(229):103120, 2000.
[8] R. E. Ewing, T. Lin, and Y. Lin. On the accuracy of the finite volume element method based on piecewise
linear polynomials. SIAM J. Numer. Anal., 39(6):18651888, 2002.
[9] R. Eymard, T. Gallouet, and R. Herbin. Finite volume methods. In Handbook of numerical analysis, Vol.
VII, Handb. Numer. Anal., VII, pages 7131020. North-Holland, Amsterdam, 2000.
[10] W. Hackbusch. On first and second order box schemes. Computing, 41(4):277296, 1989.
[11] J. Huang and S. Xi. On the finite volume element method for general self-adjoint elliptic problems. SIAM J.
Numer. Anal., 35(5):17621774, 1998.
[12] R. Li, Z. Chen, and W. Wu. Generalized difference methods for differential equations, volume 226 of Mono-
graphs and Textbooks in Pure and Applied Mathematics. Marcel Dekker Inc., New York, 2000. Numerical
analysis of finite volume methods.
[13] R. H. Li. On the generalized difference method for elliptic and parabolic differential equations. In Proc. of
the Symposium on the Finite Element Method between China and France, Beijing, China, 1982.
[14] T. Russell and M. Wheeler. Finite element and finite difference method for continuous flows in porous
media. Frontiers in Applied Mathematics, 1:35, 1983.
[15] E. Suli. Convergence of finite volume schemes for Poissons equation on nonuniform meshes. SIAM J.
Numer. Anal., 28(5):14191430, 1991.
[16] J. Xu and Q. Zou. Analysis of linear and quadratic simplicial finite volume methods for elliptic equations.
Numer. Math., 111(3):469492, 2009.