CFD Notes 2

Classification of Partial Differential Equations
(Using Cramers rule)

The governing equations in the form of partial
differential equations are by far the most prevalent form
used in computational fluid dynamics.
Consider the system of quasilinear equations given

below.
(65 a)
(65 b)
Where: u and v are the dependent variables, functions of

x and y.
coefficients a1, a2, b1, b2, c1, c2, d1, d2, f1 and f2 can be
functions of x, y, u and v.
 Consider any point in the xy-plane.
 Let us seek the lines (or directions) through this point
along which the derivatives of u and v are indeterminant,
and across which may be discontinuous. Such lines are
called characteristic lines.
 To find such lines, we assume that u and v are

continuous, and hence
(66 a)
(66 b)
Equations (65 a and b) and (66 a and b) constitute a

system of four linear equations with four unknowns
(∂u/∂x,∂u/∂y,∂v/∂x, and ∂v/∂y).
These equations can be written in matrix form as

(67)
Let [A] denote the coefficient matrix.
 Moreover, let ;|A| be the determinant of [A].
 From Cramer’s rule, if |A| is not equal to 0, then

unique solutions can be obtained for ∂u/∂x, ∂u/∂y,
∂v/∂x, and ∂v/∂y.
 On the other hand, if |A| = 0, then ∂u/∂x, ∂u/∂y, ∂v/∂x

and ∂v/∂y are, at best, indeterminant.
 We are seeking the particular directions in the xy-
plane along which these derivatives of u and v are
indeterminant.
 Therefore, let us set |A| = 0, and see what happens.
Hence:
(68)
Divide eqn (68) by (dx)2
(69)
 Equation (69) is a quadratic equation in dy/dx.
 For any point in the xy-plane, the solution of Eq. (69)

will give the slopes of the lines along which the
derivatives of u and v are indeterminant.
 Why? Because Eq. (69) was obtained by setting |A| =

0, which from the matrix Eq. (67) insures that the
solutions for the derivatives ∂u/∂x, ∂u/∂y, ∂v/∂x and
∂v/∂y are, at best, indeterminant.
 These lines in the xy space along which the

derivatives of u and v are indeterminant are called
the characteristic lines for the system of equations
given by Eq. (65 a and b).
In eqn (69), let
Then Eq. (69) can be written as
(70)
Hence, from the quadratic formula:
(71)
 Equation (71) gives the direction of the
characteristic lines through a given xy point.
 These lines have a different nature, depending on

the value of the discriminant in Eq. (71).
 Denote the discriminant by D.
(72)
The characteristic lines may be real and distinct, real

and equal, or imaginary, depending on the value of D.
Specifically:
 If D > 0: Two real and distinct characteristics exist

through each point in the xy-plane. When this is the
case, the system of equations given by Eqs. (65 a and
b) is called hyperbolic.
 If D = 0: One real characteristic exists. Here, the
system of Eqs. (65 a and b) is called parabolic.
 If D < 0: The characteristic lines are imaginary. Here,

the system of Eqs. (65 a and b) is called elliptic.
 The classification of quasilinear partial differential

equations as either elliptic, parabolic or hyperbolic is
common in the analysis of such equations.
 These three classes of equations have totally different

behavior, as will be discussed shortly.
 The origin of the words elliptic, parabolic or

hyperbolic used to label these equations is simply a
direct analogy with the case for conic sections.
 The general equation for a conic section from analytic
geometry is
where, if
b2 −4ac > 0, the conic is a hyperbola
b2 −4ac = 0, the conic is a parabola
b2 −4ac < 0, the conic is an ellipse
If we return to Eq. (67), and actually attempt to solve for,

say ∂u/∂y, using Cramer’s rule, we have
where the numerator determinant is

(73)
 The reason why |N| must be zero is that ∂u/∂y is

indeterminant, of the form 0/0.
 Since |A| has already been made equal to zero, then

|N| must be zero to allow ∂u/∂y to be indeterminant.
 The expansion of Eq. (73) will lead to equations

involving the flowfield variables which are ordinary
differential equations, and in some cases are
algebraic equations; these equations obtained from
Eq. (73) are called the compatibility equations.
 They hold only along the characteristic lines. This is

the essence of solving the original hyperbolic partial
differential equation: simply integrate simpler,
ordinary differential equations (the compatibility
equations) along the characteristic lines in the xy-
plane. This is called the method of characteristics.
Classification of Partial Differential Equations

(Eigen method)
 The eigenvalue method is based on a display of the
system of partial differential equations written in
column vector form.
 For example, let us assume that f1 and f2 in Eqs. (65

a) and (65 b) are zero for simplicity, such that the
equations become
0 (74 a)
0 (74 b)
Defining W as the column vector

the system of equations given by Eqs. (74 a) and
(74 b) can be written as
(75)
Or
(76)
where [K] and [M] are the appropriate 2 x 2 matrices in

Eq. (75).
Multiplying Eq. (76) by the inverse of [K], we have
(77)
Or
(78)
Where by definition
With the system of equations written in the form of

eqn (78), the eigenvalues of [N] determine the
classification of the system.
If the eigen values are all real, the equations are

hyperbolic. If the eigen values are all complex, the
equations are elliptic.
Hyperbolic Equations (see theory part from text
book)
Fig: Domain and boundaries for the solution of

hyperbolic equations.Two-dimensional steady flow.
hyperbolic equations. Three-dimensional steady flow.
hyperbolic equations: one-dimensional unsteady flow.
hyperbolic equations: two-dimensional unsteady flow
Parabolic Equations (see theory part from text book)

parabolic equations in two dimensions
Elliptic Equations (see theory part from text book)

elliptic equations in two dimensions.
Discretization of Partial Differential Equations

 Analytical solutions of partial differential equations
involve closed-form expressions which give the
variation of the dependent variables continuously
throughout the domain.
 In contrast, numerical solutions can give answers at
only discrete points in the domain, called grid
points.
 The grid points are identified by an index i which

runs in the x-direction, and an index j which runs in
the y-direction.
 Hence, if (i, j) is the index for point P in Figure, then

the point immediately to the right of P is labeled as
(i+1, j), the immediately to the left is (i−1, j), the point
directly above is (i, j+1), and the point directly below
is (i, j−1).
Fig: Grid points
Derivation of Elementary Finite Difference Quotients
 Finite difference representations of derivatives are

based on Taylor’s series expansions.
 For example, if ui,j denotes the x-component of

velocity at point (i, j), then the velocity ui+1, j at point
(i+1, j) can be expressed in terms of a Taylor’s series
expanded about point (i, j), as follows:
(79)
 For numerical computations, it is impractical to

carry an infinite number of terms in Eq. (79).
 Therefore, Eq. (79) is truncated.
 For example, if terms of magnitude (Δx)3 and higher

order are neglected, Eq. (79) reduces to
(80)
 We say that Eq. (80) is of second-order accuracy,

because terms of order (Δx)3 and higher have been
neglected.
 If terms of order (Δx)2 and higher are neglected,
we obtain from Eq. (79)
(81)
 where Eq. (81) is of first-order accuracy.
 In Eqs. (80) and (81), the neglected higher-order

terms represent the truncation error in the finite
series representation.
 For example, the truncation error for Eq. (80) is
and the truncation error for Eq. (81) is

The truncation error can be reduced by:
(a) Carrying more terms in the Taylor’s series, Eq.

(79). This leads to higher-order accuracy in the
representation of ui+1,j.
(b) Reducing the magnitude of Δx.
Let us return to Eq. (79), and solve for (∂u/∂x)i,j
or
(82)
 In Eq. (82), the symbol O(Δx) is a formal mathematical

notation which represents ‘terms of-order-of Δx’.
 Eq. (82) as a first-order forward difference.

Let us now write a Taylor’s series expansion for
ui−1,j, expanded about ui,j.
Or
(83)
Solving for (∂u/∂x)i,j, we obtain
(84)
Equation (84) is a first order rearward difference

expression for the derivative (∂u/∂x) at grid point
(i, j).
Let us now subtract Eq. (83) from (79).
(85)
Solving Eq. (85) for (∂u/∂x)i,j, we obtain
(86)
Equation (86) is a second order central difference for

the derivative (∂u/∂x) at grid point (i, j).
To obtain a finite-difference expression for the

second partial derivative (∂2u/∂x2)i,j, first recall that
the order-of-magnitude term in Eq. (86) comes from
Eq. (85), and that Eq. (86) can be written
(87)
Substituting Eq. (87) into (79), we obtain
(87.1)
Solving Equation for (∂2u/∂x2)i,j, we obtain
(88)
Equation (88) is a second-order central second

difference for the derivative (∂2u/∂x2) at grid point
(i, j).
Difference expressions for the y-derivatives are

obtained in exactly the same fashion.
The results are directly analogous to the previous
equations for the x derivatives. They are:
Forward
Difference
Rearward
Difference
Central
Difference
Central
Second
Difference
The same philosophy can be used to quickly

generate a finite difference quotient for the mixed
derivative (∂2u/∂x∂y) at grid point (i, j).
For example,
(89)
In Eq. (89), write the x - derivative as a central

difference of the y - derivatives, and then cast the y-
derivatives also in terms of central differences.
Or
(90)
What happens at a boundary?
Let grid point 1 be on the boundary, with points

2 and 3 a distance Δy and 2Δy above the boundary
respectively.
We wish to construct a finite difference

approximation for ∂u/∂y at the boundary.
It is easy to construct a forward difference as
(91)
which is of first-order accuracy.
 How do we find a second-order accurate finite

difference at the boundary?
 The answer is simple, and it illustrates another

method of deriving finite-difference quotients.
 Assume that at the boundary u can be expressed by

the polynomial
(92)
Applied to the grid points in Fig., Eq. (92) yields
u1 = a
Solving this system for b:
(93)
Returning to Eq. (92), and differentiating:
(94)
Equation (94), evaluated at the boundary where y = 0,

yields
(95)
Combining Eqs. (94) and (95), we obtain
(96)
It remains to show the order-of-accuracy of Eq. (96).

Consider a Taylor’s series expansion about the point
1.
(97)
 Compare Eqs. (97) and (92). Our assumed

polynomial expression in Eq. (92) is the same as
using the first three terms in the Taylor’s series.
 Hence, Eq. (92) is of O(Δy)3. In forming the

derivative in Eq. (96), we divided by Δy, which
then makes Eq. (96) of O(Δy)2. Thus, we can write
from Eq. (96)
(98)
This is our desired second-order-accurate difference

quotient at the boundary. Eqs. (98) are called one-sided
differences, because they express a derivative at a point
in terms of dependent variables on only one side of the
point.

CFD Notes 2

Uploaded by

Copyright:

Available Formats

CFD Notes 2

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

CFD Notes 2

Uploaded by

Copyright:

Available Formats

Classification of Partial Differential Equations

(Using Cramers rule)

Consider the system of quasilinear equations given

Where: u and v are the dependent variables, functions of

 To find such lines, we assume that u and v are

Equations (65 a and b) and (66 a and b) constitute a

These equations can be written in matrix form as

Let [A] denote the coefficient matrix.

 Moreover, let ;|A| be the determinant of [A].

 From Cramer’s rule, if |A| is not equal to 0, then

 On the other hand, if |A| = 0, then ∂u/∂x, ∂u/∂y, ∂v/∂x

 Therefore, let us set |A| = 0, and see what happens.

 Equation (69) is a quadratic equation in dy/dx.

 For any point in the xy-plane, the solution of Eq. (69)

 Why? Because Eq. (69) was obtained by setting |A| =

 These lines in the xy space along which the

In eqn (69), let

Then Eq. (69) can be written as

Hence, from the quadratic formula:

 These lines have a different nature, depending on

 Denote the discriminant by D.

The characteristic lines may be real and distinct, real

 If D > 0: Two real and distinct characteristics exist

 If D < 0: The characteristic lines are imaginary. Here,

 The classification of quasilinear partial differential

 These three classes of equations have totally different

 The origin of the words elliptic, parabolic or

b2 −4ac > 0, the conic is a hyperbola

b2 −4ac = 0, the conic is a parabola

b2 −4ac < 0, the conic is an ellipse

If we return to Eq. (67), and actually attempt to solve for,

where the numerator determinant is

 The reason why |N| must be zero is that ∂u/∂y is

 Since |A| has already been made equal to zero, then

 The expansion of Eq. (73) will lead to equations

 They hold only along the characteristic lines. This is

Classification of Partial Differential Equations

 For example, let us assume that f1 and f2 in Eqs. (65

Defining W as the column vector

where [K] and [M] are the appropriate 2 x 2 matrices in

Multiplying Eq. (76) by the inverse of [K], we have

With the system of equations written in the form of

If the eigen values are all real, the equations are

Fig: Domain and boundaries for the solution of

Fig: Domain and boundaries for the solution of

Fig: Domain and boundaries for the solution of

Discretization of Partial Differential Equations

 The grid points are identified by an index i which

 Hence, if (i, j) is the index for point P in Figure, then

Derivation of Elementary Finite Difference Quotients

 Finite difference representations of derivatives are

 For example, if ui,j denotes the x-component of

 For numerical computations, it is impractical to

 Therefore, Eq. (79) is truncated.

 For example, if terms of magnitude (Δx)3 and higher

 We say that Eq. (80) is of second-order accuracy,

 where Eq. (81) is of first-order accuracy.

 In Eqs. (80) and (81), the neglected higher-order

 For example, the truncation error for Eq. (80) is