You must log in to edit PetroWiki. Help with editing

Content of PetroWiki is intended for personal use only and to supplement, not replace, engineering judgment. SPE disclaims any and all liability for your use of such content. More information

Numerical methods analysis of fluid flow

Jump to navigation Jump to search

Systems of nonlinear partial differential equations (PDEs) are needed to describe realistic multiphase, multidimensional flow in a reservoir. As a rule, these equations cannot be solved analytically; they must be solved with numerical methods. This article provides an overview of these methods.

One-dimensional convection/dispersion equation

To illustrate the mathematics, we discuss the numerical solution of the 1D convection/dispersion (C/D) equation


as introduced in vector analysis of fluid flow. As a reminder, v is velocity, D is dispersion, and C is concentration. Eq. 1 is a good example to use because it illustrates many useful numerical methods that can be compared with the analytical solution given by Eq. 2.


We first introduce the concept of finite differences to convert Eq. 1 to an equation that can be solved numerically. We then present a numerical representation of Eq. 1 and illustrate its solution. For more details, you should consult the reservoir simulation page, as well as sources in the literature.[1][2][3][4][5][6][7]

Finite differences

One way to solve a PDE is to convert the PDE to finite-difference form. The finite-difference form is obtained by replacing the derivatives in the PDE with differences that are obtained from Taylor’s series. To illustrate the procedure, let us suppose that we know the function f(x) at two discrete points x = xi and x = xi + Δx, where Δx is an increment along the x-axis (Fig. 1). We can approximate the derivative df(x)/dx at x = xi by solving the Taylor’s series,


for df(x)/dx. The result is


where ET is the term


If we neglect ET, we obtain the finite-difference approximation of the first derivative.


Eq. 6 is an approximation because we are neglecting all of the terms in ET, which is called the truncation error. In the limit as the increment Δx approaches zero, the truncation error approaches zero, and the finite difference approaches the definition of the derivative.

The finite difference in Eq. 6 is called a forward difference. Other differences are possible. Two that we use next are the backward difference,


and the centered difference,


Eqs. 6 through 8 are all derived from Taylor’s series.

Numerical solution of the 1D C/D equation

We illustrate the application of finite differences in a fluid flow problem by considering a specific finite-difference representation of the 1D C/D equation. For a more detailed discussion of the numerical analysis of Eq. 1, see Chap. 4 of Peacemen.[1] In our example, we choose a backward difference for the time derivative in Eq. 1, a centered difference for the space derivative in the convection term, and a centered-in-time/centered-in-space difference for the dispersion term. Eq. 1 is converted from a PDE to the difference equation


The subscripts of concentration C denote points in space, and the superscripts denote points in time. For example, the present time, tn, is denoted by superscript n and future time tn+1 is denoted by n+1. The time increment is Δt = tn+1 - tn. Similarly, the space increment is Δx = xi + 1 - xi. The concentration at time tn+1 and spatial location xi is denoted by RTENOTITLE.

The future concentration distribution is found from the current concentration distribution by rearranging Eq. 9. We collect terms in Cn+1 on the left-hand side and terms in Cn on the right-hand side, thus


Eq. 10 is now written in the form


where the coefficients are


All values of the variables in the coefficients are known at time tn. If we assume that the spatial subscript is in the range 1 ≤ INX, the system of finite-difference equations becomes


Eq. 13 can be written in matrix form as


where RTENOTITLE  is the NX × NX matrix of coefficients, RTENOTITLE is the column vector of unknown concentrations at time tn+1, and RTENOTITLE is the column vector of right-hand-side terms that depend on known concentrations at time tn. Both column vectors RTENOTITLE and RTENOTITLE have NX elements.

The system of equations in Eq. 14 is called a tridiagonal system because it consists of three lines of nonzero diagonal elements centered about the main diagonal. All other elements are zero. Techniques for solving the tridiagonal system of equations, using the Thomas algorithm, are presented in several sources.[1][2][3][4][8] A solution of the set of equations for physical parameters v = 1 ft/day and D = 0.01 ft2/day and finite-difference parameters Δx = 0.1 ft and Δt = 0.1 day is shown in Fig. 2. The difference between the analytical solution and the numerical solution is because of numerical dispersion,[1][9][10] which is beyond the scope of this chapter. What interests us here is the appearance of matrices in the mathematics of fluid flow. Matrices are the subject of the next section.

Matrices and linear algebra

An example of a matrix was introduced earlier for the 1D C/D equation. It is often easier to work with many fluid flow equations when they are expressed in terms of matrices. Our review follows the presentation in Fanchi.[9] We begin our discussion with an example of a matrix that is used later in this chapter, namely the matrix associated with the rotation of a coordinate system. We then summarize some important properties of matrices and determinants and review the concepts of eigenvalues and eigenvectors from linear algebra.

Rotation of a Cartesian coordinate system

Fig. 3 illustrates a rotation of Cartesian coordinates from one set of orthogonal coordinates {x1, x2} to another set {y1, y2} by the angle θ. The equations relating the coordinate systems are


The set of equations in Eq. 15 has the matrix form


which can be written as


The column vectors RTENOTITLE and RTENOTITLE are


with two elements each, and the rotation matrix RTENOTITLE is the 2 × 2 square matrix,


Properties of matrices

In general, a matrix with m rows and n columns has the order m × n and is referred to as a m × n matrix. The entry in the ith row and jth column of the matrix is the ijth element of the matrix. If the number of rows equals the number of columns so that m = n, the matrix is called a square matrix. On the other hand, if mn, the matrix is a rectangular matrix.

If the matrix has a single column so that n = 1, it is a column vector as in Eq. 18. If the matrix has a single row so that m = 1, it is a row vector. A row vector can be created from a column vector by taking the transpose of the column vector. For example, the transpose of the column vector RTENOTITLE in Eq. 18 may be written as  


where the superscript T denotes the transpose of the matrix. In general, we can write a m × n matrix RTENOTITLE with a set of elements {aij: i = 1, 2, ... n; j = 1, 2, ... m} as


The transpose of matrix RTENOTITLE is


The conjugate transpose of matrix RTENOTITLE is obtained by finding the complex conjugate of each element in RTENOTITLE and then taking the transpose of the matrix RTENOTITLE. This operation can be written as


where * denotes complex conjugation. Recall that the conjugate z* of a complex number z is obtained by replacing the imaginary number RTENOTITLE with RTENOTITLE wherever it occurs. If all the elements of matrix RTENOTITLE are real, the conjugate transpose of matrix RTENOTITLE is equal to the transpose of matrix RTENOTITLE.

If the matrix RTENOTITLE is a square matrix and the elements of matrix RTENOTITLE satisfy the equality aij = aji, the matrix RTENOTITLE is called a symmetric matrix. A square matrix A¯¯ is Hermitian, or self-adjoint, if RTENOTITLE= RTENOTITLE+ (i.e, the matrix equals its conjugate transpose).

The set of elements {aii} of a square matrix RTENOTITLE is the principal diagonal of the matrix. The elements {aji} with ij are off-diagonal elements. The matrix RTENOTITLE is a lower triangular matrix if aij = 0 for i < j, and RTENOTITLE is an upper triangular matrix if aij = 0 for i > j. The matrix RTENOTITLE is a diagonal matrix if aij =0 for ij.

Matrix operations

Suppose the matrices RTENOTITLE, RTENOTITLE, and RTENOTITLE with elements {aij}, {bij}, and {cij} have the same order m × n. We are using double underlines to denote matrices. Other notations are often used, such as boldface. The addition or subtraction of two matrices may be written as


The product of a matrix RTENOTITLE with a number k may be written as


The product of matrix RTENOTITLE with order m × n and matrix RTENOTITLE with order n × p is


where matrix RTENOTITLE has order m × p. Notice that matrix multiplication is possible only if the number of columns in RTENOTITLE equals the number of rows in RTENOTITLE. This requirement is always satisfied for square matrices.

The transpose of the product of two square matrices RTENOTITLE and RTENOTITLE is


and the adjoint of the product of two square matrices is


Notice that the product of two matrices may not be commutative (i.e., RTENOTITLE RTENOTITLERTENOTITLE RTENOTITLE in general).

The identity matrix, RTENOTITLE, is a square matrix with all off-diagonal elements equaling zero and all diagonal elements equaling one. The identity matrix preserves the identity of a square matrix RTENOTITLE in matrix multiplication, thus


By contrast, a null matrix RTENOTITLE is a matrix in which all elements are zero. In this case, the product of the null matrix with a matrix RTENOTITLE is


The matrix, RTENOTITLE, is singular if the product of matrix RTENOTITLE with a column vector RTENOTITLE that has at least one nonzero element yields the null matrix; that is, RTENOTITLE is singular if


The concepts of identity matrix and matrix singularity are needed to define the inverse matrix. Suppose we have two square matrices RTENOTITLE and RTENOTITLE that satisfy the product


Notice that the matrices RTENOTITLE and RTENOTITLE commute. The matrix RTENOTITLE is nonsingular, and the matrix RTENOTITLE is the inverse of RTENOTITLE, thus RTENOTITLE = RTENOTITLE-1, where RTENOTITLE-1 denotes the inverse of RTENOTITLE. Eq. 32 can be written in terms of the inverse as


The inverse matrix is useful for solving systems of equations. For example, suppose we have a system of equations that satisfies


where the column vector RTENOTITLE and the matrix RTENOTITLE are known, and the column vector RTENOTITLE contains a set of unknowns. Eq. 13 is an example for the 1D C/D equation. We can solve for RTENOTITLE in Eq. 34 by premultiplying Eq. 34 by RTENOTITLE-1. The result is


Of course, we have to know RTENOTITLE-1 to find RTENOTITLE. This leads us to a discussion of determinants.

Determinants, eigenvalues, and eigenvectors

The determinant (det) of a square matrix RTENOTITLE is denoted by det RTENOTITLE or | RTENOTITLE |. Two examples of determinants are the determinants of a 2×2 matrix and a 3×3 matrix. The determinant of a 2×2 matrix is


and the determinant of a 3×3 matrix is

RTENOTITLE RTENOTITLE....................(37)

Determinants are useful for determining if an inverse matrix RTENOTITLE-1 exists. Inverse matrices are needed to solve finite-difference equations representing fluid flow. The condition det RTENOTITLE  says that an inverse matrix RTENOTITLE-1 exists, even though we may not know the elements of the inverse matrix. Determinants are also useful for determining eigenvalues and eigenvectors.

Eigenvalues and eigenvectors are useful for understanding the behavior of physical quantities that may be represented by a matrix. An example in fluid flow is permeability, which we discuss in more detail later in this chapter. First, we need to define the concepts of eigenvalue and eigenvector.

Eigenvalues are the values of λ in the eigenvalue equation


where RTENOTITLE is an n × n square matrix and RTENOTITLE is a column vector with n rows. The eigenvalue equation may be written as


where RTENOTITLE is the n × n identity matrix. Eq. 39 has nonzero solutions, RTENOTITLE, if the eigenvalue, λ, is a characteristic root of RTENOTITLE, that is, λ must be a solution of


Eq. 40 is the characteristic equation of RTENOTITLE, and the n values of λ are the characteristic roots of the characteristic equation. The characteristic roots, λ, are obtained by expanding the determinant in Eq. 40 into an nth-degree polynomial and then solving for the n values of λ. These concepts are illustrated in the next section.


ai, bi, ci, di = finite-difference coefficients, Eq. 11
aij, bij, cij = elements of matrices, Eq. 24
RTENOTITLE = matrices, Eq. 24
RTENOTITLE = rotation matrix, Eq. 17
RTENOTITLE = column vector of unknown concentrations at tn +1, Eq. 14
RTENOTITLE = column vector of terms that depend on known concentrations at tn +1, Eq. 14
ET = truncation error, Eq. 5
i = imaginary operator
RTENOTITLE = identity matrix, Eq. 29
Jx, Jy, Jz = fluid flux in x-, y-, z-directions
(Jx)x = fluid flux in x-direction at location x
(Jy)y = fluid flux in y-direction at location y
(Jz)z = fluid flux in z-direction at location z
RTENOTITLE = matrix of coefficients, Eq. 14
n = exponent
R = region
S = surface
t = time
tn = present time
tn+1 = future time
v = velocity, L/t, ft/sec
x,y,z = space dimensions
xi = discrete point in x-direction, Eq. 3
Δt = time interval
Δx = length
Δy = width
Δz = thickness
θ = angle, Eq. 15
λ = eigenvalues, Eq. 38


  1. 1.0 1.1 1.2 1.3 Peaceman, D.W. 1977. Fundamentals of Numerical Reservoir Simulation. Oxford, UK: Elsevier Publishing.
  2. 2.0 2.1 Aziz, K. and Settari, A. 1979. Petroleum Reservoir Simulation. Essex, UK: Elsevier Applied Science Publishers.
  3. 3.0 3.1 Mattax, C.C. and Dalton, R.L. 1990. Reservoir Simulation, Vol. 13. Richardson, Texas: Monograph Series, SPE.
  4. 4.0 4.1 Ertekin, T., Abou-Kassem, J.H., and King, G.R. 2001. Basic Applied Reservoir Simulation, Vol. 7. Richardson, Texas: Textbook Series, SPE.
  5. Munka, M. and Pápay, J. 2001. 4D Numerical Modeling of Petroleum Reservoir Recovery. Budapest, Hungary: Akadémiai Kiadó.
  6. Fanchi, J.R. 2006. Principles of Applied Reservoir Simulation, third edition. Burlington, Massachusetts: Gulf Professional Publishing/Elsevier.
  7. Fanchi, J.R. 2000. Integrated Flow Modeling, No. 49. Amsterdam, The Netherlands: Developments in Petroleum Science, Elsevier Science B.V.
  8. Chapra, S.C. and Canale, R.P. 2002. Numerical Methods for Engineers, fourth edition. Boston, Massachusetts: McGraw-Hill Book Co.
  9. 9.0 9.1 Fanchi, J.R. 2006. Math Refresher for Scientists and Engineers, third edition. New York: Wiley Interscience. Cite error: Invalid <ref> tag; name "r11" defined multiple times with different content
  10. Lantz, R.B. 1971. Quantitative Evaluation of Numerical Diffusion (Truncation Error). SPE J. 11 (3): 315–320. SPE-2811-PA.

Noteworthy papers in OnePetro

Use this section to list papers in OnePetro that a reader who wants to learn more should definitely read

External links

Use this section to provide links to relevant material on websites other than PetroWiki and OnePetro

See also

Mathematics of fluid flow

Vector analysis of fluid flow

Diagonalizing the permeability tensor