Nonlinear
ConvectionDominated
Problems
10.1 Burgers Equation
Onedimensional Burgers equation
Conservative form
0
x
u
x
u
u
t
u
2
2
=
c
c
c
c
+
c
c
v
2
2
2
2
2 2
u 0.5 F ; 0
x
u
x
F
t
u
0
x
u
2
u
x t
u
= =
c
c
c
c
+
c
c
=
c
c


.

\

c
c
+
c
c
v
v
Inviscid Burgers Equation
Onedimensional inviscid Burgers equation
Larger values of convect faster and overtake
slower
Multivalued solution may occur
Postulate a shock to allow the development of
discontinuous solutions
0
x
u
u
t
u
=
c
c
+
c
c
u
u
Inviscid Burgers Equation
Formation of multivalued solution
The nonlinearity allows discontinuous
solutions to develop
Shockfitting
a
b
shock
t = t
0
t = t
1
t = t
2
Viscous Burgers Equation
Viscous term reduces the amplitude in
high gradient regions
Prevents multivalued solutions from
developing (second derivative increases
faster than first derivative)
t = t
0
t = t
1
t = t
2
10.1.2 Explicit Schemes
FTCS scheme (nonconservative)
FTCS (conservative form)
0
x 2
u u 2 u
x 2
u u u
t
u u
n
1 j
n
j
n
1 j
n
1 j
n
1 j
n
j
n
j
1 n
j
=
A
+
+
A
+ +
+
) ( ) ( v
2
n
1 j
n
j
n
1 j
n
1 j
n
1 j
n
j
1 n
j
u 5 0 F
0
x 2
u u 2 u
x 2
F F
t
u u
.
) (
=
=
A
+
+
A
+ +
+
v
Explicit Schemes
Fourpoint Upwind Scheme
Truncation errors
O(Ax
2
) if q = 0.5
O(Ax
3
) if q = 0.5
A
+
+
A
= <
A
+
+
A
= >
+ + +
+ +
x 3
F F 3 F 3 F q
x 2
F F
F L 0 u
x 3
F F 3 F 3 F q
x 2
F F
F L 0 u
2 j 1 j j 1 j 1 j 1 j
(4)
x
1 j j 1 j 2 j 1 j 1 j
(4)
x
) (
,
) (
,
LaxWendroff Scheme
Inviscid Burgers equation for unsteady one
dimensional shock flows
Replace temporal derivative by equivalent spatial
derivative (more complicated for nonlinear case)
2
u 0.5 F ; 0
x
F
t
u
= =
c
c
+
c
c
u
F
A
x
F
A
x t
F
x t
u
x
F
A
x
F
u
F
t
u
u
F
t
F
t
F
x x
F
t t
u
2
2
2
2
c
c
= 
.

\

c
c
c
c
= 
.

\

c
c
c
c
=
c
c
c
c
=
c
c

.

\

c
c
=
c
c

.

\

c
c
=
c
c

.

\

c
c
c
c
= 
.

\

c
c
c
c
=
c
c
;
Chain rule
LaxWendroff Scheme
Centraldifference discretization
For Burgers equation
2
n
1 j
n
j 2 1 j
n
j
n
1 j 2 1 j
2 1 j 2 1 j
1 j j
n
1 j
n
j
2 1 j
j 1 j
n
j
n
1 j
2 1 j
2 1 j 2 1 j
2 1 j
2 1 j
2 1 j
2 1 j
x
F F A F F A
x x
x x
F F
A
x x
F F
A
x x
x
F
A
x
F
A
x
F
A
x
A
=


.

\



.

\


.

\

c
c

.

\

c
c
= 
.

\

c
c
c
c
+ +
+
+
+
+
+
+
+
) ( ) (
/ /
/ /
/ /
/ /
/
/
/
/
u u 5 0 A & u u 5 0 A u A
1 j j 1/2 j 1 j j 1/2 j
) ( . ) ( .
+ +
+ = + = =
LaxWendroff Scheme
Temporal derivative
Inviscid Burgers equation
Rearrange
2
n
1 j
n
j 2 1 j
n
j
n
1 j 2 1 j
n
j
1 n
j
n
j
1 n
j
2
2
n
j
1 n
j
x
F F A F F A
2
t
t
u u
x
F
A
x 2
t
t
u u
t
u
2
t
t
u u
t
u
A

.

\

A
+
A
=
c
c
c
c A
+
A
~
c
c A
+
A
~
c
c
+ +
+
+ +
) ( ) (
) (
/ /
0
x 2
F F
x
F F A F F A
2
t
t
u u
x
F
t
u
n
1 j
n
1 j
2
n
1 j
n
j 2 1 j
n
j
n
1 j 2 1 j
n
j
1 n
j
=
A
+
A

.

\
 A
+
A
=
c
c
+
c
c
+ + +
+
) ( ) (
/ /
  ) ( ) ( . ) ( .
/ /
n
1 j
n
j 2 1 j
n
j
n
1 j 2 1 j
2
n
1 j
n
1 j
n
j
1 n
j
F F A F F A
x
t
5 0 F F
x
t
5 0 u u
+ + +
+

.

\

A
A
+
A
A
=
LaxWendroff Scheme
Linear pure convection equation
Nonlinear  inviscid Burgers equation
Equivalent twostage algorithm (more economical)
  ) ( ) ( . ) ( .
/ /
n
1 j
n
j 2 1 j
n
j
n
1 j 2 1 j
2
n
1 j
n
1 j
n
j
1 n
j
F F A F F A
x
t
5 0 F F
x
t
5 0 u u
+ + +
+

.

\

A
A
+
A
A
=
) ( . ) ( .
) ( . ) ( .
n
1 j
n
j
n
1 j
2 n
1 j
n
1 j
n
j
1 n
j
n
1 j
n
j
n
1 j
2
n
1 j
n
1 j
n
j
1 n
j
T T 2 T C 5 0 T T C 5 0 T T
T T 2 T
x
t u
5 0 T T
x
t u
5 0 T T
+ +
+
+ +
+
+ + =
+

.

\

A
A
+
A
A
=
A
A
=
A
A
+ =
+
+
+ + +
) (
) ( . ) ( .
*
/
*
/
*
/
2 1 j 2 1 j
n
j
1 n
j
n
j
n
1 j
n
1 j
n
j 2 1 j
F F
x
t
u u
F F
x
t
5 0 u u 5 0 u
Burgers Equation
Thommens extension of LaxWendroff
scheme for viscous flow problems
Error in textbook
Stability limit
2
n
1 j
n
j
n
1 j 2 1 j 2 1 j
n
j
1 n
j
n
2 j
n
1 j
n
j
n
1 j
n
j
n
1 j
n
j
n
1 j
n
1 j
n
j 2 1 j
x
t
s where
u u 2 u s F F
x
t
u u
u u 2 u 5 0 u u 2 u 5 0 s 5 0
F F
x
t
5 0 u u 5 0 u
A
A
=
+ +
A
A
=
+ + + +
A
A
+ =
+ +
+
+ + +
+ + +
v
) ( ) (
)] ( . ) ( . [ .
) ( . ) ( .
*
/
*
/
*
/
) /( ) ( x A 2 x t or x 2 t A t
2 2 2
A + A s A A s + A A v v
10.1.3 Implicit Schemes
Burgers equation (viscous)
CrankNicolson implicit formulation
Thomas algorithm for tridiagonal matrices
cannot be used directly due to the appearance
of nonlinear implicit term
Use Taylor series expansion of at n
th
time
level to convert to tridiagonal form
2
xx x
n
j
1 n
j
1 n
j
1 n
j
n
j xx
1 n
j
n
j x
1 n
j
x 1 2 1 L x 2 1 0 1 L u u u
u u L 5 0 F F L 5 0
t
u
A = A = = A
+ + + =
A
A
+ +
+ +
+
/ ) , , ( ), /( ) , , ( ,
) ( . ) ( . v
1 n
j
F
+
1 n
j
F
+
CrankNicolson Scheme
Taylorseries expansion (linearlization of F)
Linear tridiagonal system (in terms of Au or u)
n
j
n
j
2 1 n
j
n
j
1 n
j
n
j
2
2
2
n
j
n
j
1 n
j
u
u
F
A t O u A F F
or
t
F
t 5 0
t
F
t F F
=
(
c
c
= A + A + =
+
(
c
c
A +
(
c
c
A + =
+ +
+
), (
.
 
n
j xx
n
j
1 n
j xx
1 n
j
n
j x
1 n
j
1 n
j
n
j xx
1 n
j
n
j
n
j x
1 n
j
u L 5 0 u u L u u L t 5 0 u
or u u L 5 0 u u F 2 L 5 0
t
u
v v
v
. ) ( .
) ( . ) ( .
+ = A +
+ + A + =
A
A
+ + +
+ +
+
CrankNicolson Scheme
Thomas algorithm
The matrix coefficients must be reevaluated at every
time step (to recover nonlinearity of the equation)
Truncation error O(At
2
, Ax
2
)
Unconditionally stable in Von Neumann sense (linear)
+ + =
A
A
=
+ =
A
A
=
= + +
+
+
+
+
+ +
n
1 j
n
j
n
1 j
n
j
n
1 j
n
j
n
j
n
1 j
n
j
n
j
1 n
1 j
n
j
1 n
j
n
j
1 n
1 j
n
j
su 5 0 u s 1 su 5 0 d
s 5 0 u
x
t
25 0 c
s 1 b
s 5 0 u
x
t
25 0 a
d u c u b u a
. ) ( .
. .
. .
Generalized CrankNicolson
Mass operator and fourpoint upwind
Truncation error O(At
2
, Ax
2
)
A
+
=
A
+
+
A
= >
=
+ + A + =


.

\

A
A
+
+ +
+ +
+
2
n
1 j
n
j
n
1 j n
j xx
1 j j 1 j 2 j 1 j 1 j (4)
x
x
1 n
j
n
j xx
1 n
j
n
j
n
j
4
x
1 n
j
x
x
u u 2 u
u L
x 3
F F 3 F 3 F q
x 2
F F
F L 0 u
2 1 M
u u L 5 0 u u F 2 L 5 0
t
u
M
) (
,
), , , (
) ( . ) ( .
) (
o o o
v
Generalized CrankNicolson
Quadridiagonal system of equations can be
solved using generalized Thomas algorithm
+ + + + =
+
A
A
=
+
A
A
+ =
+
A
A
+ =
A
A
=
= + + +
+
+
+
+
+ +
n
1 j
n
j
n
1 j
n
j
n
1 j
n
j
n
j
n
j
n
1 j
n
j
n
2 j
n
j
n
j
1 n
1 j
n
j
1 n
j
n
j
1 n
1 j
n
j
1 n
2 j
n
j
u s 5 0 u s 2 1 u s 5 0 d
s 5 0 u
x
t
6
q
25 0 c
s 2 u
x
t
q 5 0 1 b
s 5 0 u
x
t
q 5 0 25 0 a
u
x
t
6
q
e
d u c u b u a u e
) . ( ) ( ) . (
. ) . (
.
. ) . . (
o o o
o
o
o
Artificial Dissipation
CrankNicolson with additional dissipation
For small values of viscosity (highRe), it is
desirable to add some artificial dissipation
Modified CrankNicolson
Choose v
a
empirically
) ( .
1 n
j
n
j xx a
F F tL 5 0
+
+ A v
 
n
j xx
n
j x
1 n
j
n
j xx a
1 n
j xx
1 n
j
n
j
4
x
1 n
j x
u tL 5 0 u M
u u tL u L u u L t 5 0 u M
A + =
A A +
+ + + +
v
v v
.
) ( ) ( .
) (
10.1.4 BURG:
Numerical Comparison
Propagation of a shock wave governed by
viscous Burgers equation
Exact solution
0 t x u 0 1 t x u s B.C.
x x 0 , 0
0 x x , 1
0 x u x u
max
0
= =
s <
s s
= =
) , ( , . ) , (
) , ( ) (
max max
max
v
1
t
x 5 0
d u t x G
d e
d e
t
x
u
0
2
0
G 5 0
G 5 0
=
+ ' ' =
=
}
}
}
Re ;
) ( .
) ( ) , ; (
Re .
Re .
Burgers Equation
BURG: Numerical Comparison
ME = 1, FTCS scheme
ME = 2, twostage LaxWendroff scheme
ME = 3, Explicit fourpoint upwind scheme
ME = 4, CrankNicolson (CNFDM): o = 0, q = 0
ME = 4, CrankNicolson (CNFEM): o = 1/6, q = 0
ME = 4, CrankNicolson, Mass Operator (CNMO): o = 1/12, q = 0
ME = 4, CrankNicolson, 4pt. Upwind (CN4PU): o = 0, q = 0.5
ME = 5, CrankNicolson plus additional dissipation
Note: Optimum o and q (locally freezing nonlinear coefficients)
2
j
opt
2
j
opt
x
t u
4
1
2
1
q
x
t u
12
1
6
1


.

\

A
A
+ =


.

\

A
A
+ = o
Burgers Equation
Burgers Equation
Propagating Shock Solution
R
cell
= 1.0, C = 0.25
Burgers Equation: Propagating Shock
R
cell
= 100, C = 1.0
R
cell
= 3.33, C = 1.0
Velocity distribution at t = 2.0; R
cell
= 100
10.2 Systems of Equations
Continuity equation
Momentum equations
Energy equation
Equation of state (compressible flows)
Turbulent kinetic energy equation
Rate of turbulent energy dissipation equation
Reynolds stresses equations
Multiphase flows
Chemical reactions
Systems of Equations
1D unsteady compressible inviscid flow
Continuity equation, xmomentum equation,
energy equation
v
p
2
2
2
C
C
;
u u 5 0
1
p
p
u
u
F ;
u 5 0
1
p
u q
0
x
F
t
q
=







.

\

+
+ =






.

\

+
=
=
c
c
+
c
c
) . (
.
) (
TwoStage LaxWendroff
Single equation
System of equations
A
A
=
A
A
+ =
+
+
+ + +
) (
) ( . ) ( .
*
/
*
/
*
/
2 1 j 2 1 j
n
j
1 n
j
n
j
n
1 j
n
1 j
n
j 2 1 j
F F
x
t
q q
F F
x
t
5 0 q q 5 0 q
A
A
=
A
A
+ =
+
+
+ + +
) (
) ( . ) ( .
*
/
*
/
*
/
2 1 j 2 1 j
n
j
1 n
j
n
j
n
1 j
n
1 j
n
j 2 1 j
F F
x
t
u u
F F
x
t
5 0 u u 5 0 u
LaxWendroff Scheme with
Artificial Viscosity
Continuity equation
Xmomentum equation
Energy equation
CrankNicolson Scheme
System of equations
Linearization
33 block tridiagonal system (solved by block
Thomas algorithm)
  ) ( ) ( .
1 n
1 j
1 n
1 j
n
1 j
n
1 j
n
j
1 n
j
F F F F
x
t
25 0 q q
+
+
+ +
+
+
A
A
=
q
F
A
q A F F
1 n n 1 n
c
c
=
+ A + =
+ +
(33 matrix)
1 n
j
n
j
1 n
j
n
1 j
n
1 j
1 n
1 j 1 j
1 n
j
1 n
1 j 1 j
q q q
F F
x
t
5 0 q A
x
t
25 0 q I q A
x
t
25 0
+ +
+
+
+ +
+ +
A + =
A
A
= A
A
A
+ A + A
A
A
) ( . . .
CrankNicolson Scheme
Use Von Neumann analysis for the
linearized equation
Amplification matrix
Numerical Stability
q
F
A ; 0
x
q
A
t
q
c
c
= =
c
c
+
c
c
u
u
sin .
sin .
A
x
t
i 5 0 1
A
x
t
i 5 0 1
G
A
A
+
A
A
=
m all for 0 1 satisfy G of igenvalues E
m
. s
10.3 Group Finite Element Method
Conventional finite element method introduces
a separation approximate solution (trial
function, interpolation function) for each
dependent variable
Galerkin method produces large numbers of
products of nodal values of dependent
variables, particularly from the nonlinear
convective terms
Inefficient, timeconsuming
Group finite element formulation is effective
in dealing with convective nonlinearities
Group Finite Element Method
Group finite element formulation
1. The equations are cast in conservative form
2. A single approximation solution is used for
the group of terms in the differential terms
(i.e., approximate F directly instead of the
nonlinear convective term ucu/cx)
Onedimensional Group Formulation
F F & u u
l
l l
l
l l
= =  
Group Finite Element Method
Onedimensional Group Formulation
Conventional finite element
) (
) ( .
2
1 j
2
1 j
1 j 1 j
1 j 1 j j x
j xx j x
j
x
u u
x 4
1
x 2
u u
u u 5 0 F L
0 u L F L
dt
du
M
+
+
+
A
=
)
`
+ =
= +
(
v
)
`
A
+ +
=
c
c
+ +
x 2
u u
3
u u u
x
u
u
1 j 1 j 1 j j 1 j
) (
Conservative
form
Nonconservative
form
Onedimensional Burgers equation
Conventional and group FEMs
10.4 2D Burgers Equation
Twodimensional Burgers equation
Equivalent to 2D momentum equations
for incompressible laminar flow with
zero pressure gradient


.

\

c
c
+
c
c
=
c
c
+
c
c
+
c
c


.

\

c
c
+
c
c
=
c
c
+
c
c
+
c
c
2
2
2
2
2
2
2
2
y
v
x
v
y
v
v
x
v
u
t
v
y
u
x
u
y
u
v
x
u
u
t
u
v
v
2D Burgers Equation
Exact solution
Use ColeHopf transformation
Transform the 2D Burgers equation into one
single equation 2D diffusion equation
u
c
u c
=
u
c
u c
=
y
2
v ,
x
2
u
v
v
0
y x t
2
2
2
2
=


.

\

c
u c
+
c
u c
c
u c
2D Burgers Equation
Steady 2D Burgers equation
Exact solution
0
y x
2
2
2
2
=
c
u c
+
c
u c
{ }
{ }
) cos( ] [
) sin( ] [
) cos( ] [
) cos( ] [
) cos( ] [
) ( ) (
) ( ) (
) ( ) (
) ( ) (
) ( ) (
y e e a xy a y a x a a
y e e a x a a 2
v
y e e a xy a y a x a a
y e e a y a a 2
u
y e e a xy a y a x a a
0 0
0 0
0 0
0 0
0 0
x x x x
5 4 3 2 1
x x x x
5 4 3
x x x x
5 4 3 2 1
x x x x
5 4 2
x x x x
5 4 3 2 1
v
v
+ + + + +
+ +
=
+ + + + +
+ +
=
+ + + + + = u
Exact solution for 2D Burgers equation
2D Burgers Equation Exact u
a
1
= a
2
= 1.3*10
13,
a
3
= a
4
= 0, a
5
= 1, = 25, x
0
=1, v = 0.04
2D Burgers Equation Exact v
a
1
= a
2
= 1.3*10
13,
a
3
= a
4
= 0, a
5
= 1, = 25, x
0
=1, v = 0.04
Multidimensional Group FEM
Twodimensional Burgers equation
Approximate solutions for (u,v), and groups (u
2
, uv, v
2
)
and the components of S
For example (bilinear for rectangular elements)
)
`
+ +
=
= = =
=


.

\

c
c
+
c
c
c
c
+
c
c
+
c
c
v v
v
) ( .
,
) ( .
} , { }, , { }, , {
2 2 2 2
2 2
2
2
2
2
v u v 5 0 v u u 5 0
S
v uv G uv u F v u q
0 S
y
q
x
q
y
G
x
F
t
q
=
=
4
1 l
l l
uv uv ) , ( ) ( q 
Galerkin Finite Element
Linear (Chapter 9)
Nonlinear (Group FE formulation)
The equations are treated as linear at the level at which
the discretization take place (but indeterminate)
Substitution for the nodal groups in terms of the unknown
nodal variables introduces the nonlinearity but also makes
the system determinate
S M M q L M L M G L M F L M RHS
RHS
t
q
M M
y x yy x xx y y x x y
k j
y x
+ + + =
=
(
c
c
) (
,
v
k j yy x y xx y x y x x y
k j
y x
T L M L M L vM L uM
t
T
M M
,
,
) [ + + =
(
c
c
o o
Split Schemes
Twodimensional Burgers equations
Similar to those used in Chapters 8 and 9
Additional complication due to nonlinearity
Generalized FEM/FEM with mass
operators M
x
and M
y
T
y y y y
x x x x
2 1 M
2 1 M
) , , (
) , , (
o o o
o o o
=
=
PseudoTransient Formulation
Use pseudotransient formulation (sect 6.4) for
steadystate solution
For steadystate problems, unsteady formulation
provides an equivalent underrelaxation parameter
for steady iterative schemes
For steadystate solutions, it is desirable to use a
simple time discretization (such as twolevel fully
implicit scheme with  = 1) to simplify the
formulation
PseudoTransient Formulation
Twolevel fully implicit scheme ( = 1)
Linearize the nonlinear terms F, G, and S in
(RHS)
n+1
1 n
1 n
k j
y x
RHS
t
q
M M
+
+
=
(
A
A
+ A + =
+ A + =
+ A + =
+ +
+ +
+ +
q C S S
q B G G
q A F F
1 n n 1 n
1 n n 1 n
1 n n 1 n
PseudoTransient Formulation
Linearization (Jacobian matrices A, B, C)
Approximate Factorization
(
+
+

.

\

=
c
c
=
(
=
c
c
=
(
=
c
c
=
)
`
+ +
=
= = =
) (
) ( .
) ( .
,
) ( .
}, , { }, , { }, , {
2 2
2 2
2 2 2 2
2 2
v 3 u uv 2
uv 2 v u 3 5 0
q
S
C
v 2 0
u v
q
G
B ,
u v
0 u 2
q
F
A
v u v 5 0 v u u 5 0
S
v uv G uv u F v u q
v
v v
*
, ,
*
,
)] . ( [
) ( )] . ( [
k j
1 n
k j y yy y y
n
k j x xx x x
q q C M 5 0 L B L t M
RHS t q C M 5 0 L A L t M
A = A A +
A = A A +
+
v
v
PseudoTransient Formulation
Further simplification to reduce CPU time
Use the same lefthandside for each scalar
component
Perform only one factorization (BANFAC) for
different components
Does not affect the steadystate solution since
(RHS)
n
= 0 in the steady state limit
(


.

\

+
~
(
~
(
~
1 0
0 1
v u
C
,
1 0
0 1
v B ,
1 0
0 1
u A
2 2
v
TWBURG: Numerical Solution
Twodimensional Burgers equations
Steady state solution with the following split algorithm
Solution domain
1s x s 1 , 0 s y s y
max
, y
max
= t/6
Use exact solution for the boundary conditions
Initial conditions obtained from linear interpolation of
the boundary condition in the xdirection
*
, ,
*
,
)] . ( [
) ( )] . ( [
k j
1 n
k j y yy y y
n
k j x xx x x
q q C M 5 0 L B L t M
RHS t q C M 5 0 L A L t M
A = A A +
A = A A +
+
v
v
Computer Program  TWBURG
Approximate Factorization
Error Distributions at y/y
max
= 0.4