Analysis of numerical dissipation and dispersion

Modied equation method: the exact solution of the discretized equations

satises a PDE which is generally dierent from the one to be solved
Original PDE Modied equation Au
n+1
= Bu
n
u
t
+Lu = 0
u
t
+Lu =

p=1

2p

2p
u
x
2p
+

p=1

2p+1

2p+1
u
x
2p+1
Motivation: PDEs are dicult or impossible to solve analytically but their
qualitative behavior is easier to predict than that of discretized equations
Expand all nodal values in the dierence scheme in a double Taylor series
about a single point (x
i
, t
n
) of the space-time mesh to obtain a PDE
Express high-order time derivatives as well as mixed derivatives in terms
of space derivatives using this PDE to transform it into the desired form
Derivation of the modied equation
Example. Pure convection equation
u
t
+ v
u
x
= 0, v > 0
BDS in space, FE in time:
u
n+1
i
u
n
i
t
+ v
u
n
i
u
n
i1
x
= 0 (upwind)
Taylor series expansions about the point (x
i
, t
n
)
u
n+1
i
= u
n
i
+ t
_
u
t
_
n
i
+
(t)
2
2
_

2
u
t
2
_
n
i
+
(t)
3
6
_

3
u
t
3
_
n
i
+ . . .
u
n
i1
= u
n
i
x
_
u
x
_
n
i
+
(x)
2
2
_

2
u
x
2
_
n
i

(x)
3
6
_

3
u
x
3
_
n
i
+ . . .
Substitution into the dierence scheme yields
_
u
t
_
n
i
+v
_
u
x
_
n
i
=
t
2
_

2
u
t
2
_
n
i

(t)
2
6
_

3
u
t
3
_
n
i
+
vx
2
_

2
u
x
2
_
n
i

v(x)
2
6
_

3
u
x
3
_
n
i
+. . .
original PDE O[t, x] truncation error ()
Next step: replace both time derivatives in the RHS by space derivatives
Derivation of the modied equation
Dierentiate () with respect to t

2
u
t
2
+ v

2
u
xt
=
t
2

3
u
t
3

(t)
2
6

4
u
t
4
+
vx
2

3
u
x
2
t

v(x)
2
6

4
u
x
3
t
+ . . . (1)
Dierentiate () with respect to x and multiply by v
v

2
u
tx
+ v
2
2
u
x
2
=
vt
2

3
u
t
2
x

v(t)
2
6

4
u
t
3
x
+
v
2
x
2

3
u
x
3

v
2
(x)
2
6

4
u
x
4
+ . . . (2)
Subtract (2) from (1) and drop high-order terms

2
u
t
2
= v
2
2
u
x
2
+
t
2
_

3
u
t
3
+ v

3
u
t
2
x
+O(t)
_
+
x
2
_
v

3
u
x
2
t
v
2
3
u
x
3
+O(x)
_
(3)
Dierentiate formula (3) with respect to t

3
u
t
3
= v
2
3
u
x
2
t
+O[t, x] (4)
Dierentiate formula (2) with respect to x

3
u
x
2
t
= v

3
u
x
3
+O[t, x] (5)
Dierentiate formula (3) with respect to x

3
u
t
2
x
= v
2
3
u
x
3
+O[t, x] (6)
Derivation of the modied equation
Equations (4) and (5) imply that

3
u
t
3
= v
3
3
u
x
3
+O[t, x] (7)
Plug (5)(7) into (3)

2
u
t
2
= v
2
2
u
x
2
+ v
2
(vt x)

3
u
x
3
+O[t, x] (8)
Substitute (7) and (8) into () to obtain the modied equation
u
t
+v
u
x
=
v
2
t
2
_

2
u
x
2
+ (vt x)

3
u
x
3
_
+
v
3
(t)
2
6

3
u
x
3
+
vx
2

2
u
x
2

v(x)
2
6

3
u
x
3
+. . .
which can be rewritten in terms of the Courant number = v
t
x
as follows
u
t
+ v
u
x
=
vx
2
(1 )

2
u
x
2
. .
numerical diusion
+
v(x)
2
6
(3 2
2
1)

3
u
x
3
. .
numerical dispersion
+. . .
Remark. The CFL stability condition 1 must be satised for the discrete
problem to be well-posed. In the case > 1, the numerical diusion coecient
vx
2
(1 ) is negative, which corresponds to a backward heat equation
Signicance of terms in the modied equation
Exact solution of the discretized equations
Au
n+1
= Bu
n

u
t
+Lu =

p=1

2p

2p
u
x
2p
+

p=1

2p+1

2p+1
u
x
2p+1
Even-order derivatives

2p
u
x
2p
cause numerical dissipation
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0.2
0.4
0.6
0.8
1
smearing (amplitude errors)
Odd-order derivatives

2p+1
u
x
2p+1
cause numerical dispersion
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0.2
0.4
0.6
0.8
1
wiggles (phase errors)
u
t
+ v
u
x
= 0
Qualitative analysis: the numerical behavior of the discretization scheme largely
depends on the relative importance of dispersive and dissipative eects
Stabilization by means of articial diusion
Stability condition (necessary but not sucient)
The coecients of the even-order derivatives in the modied equation must
have alternating signs, the one for the second-order term being positive
If this condition is violated, it can be enforced by adding articial diusion:
Stabilized methods +(v )
2
u streamline diusion
Nonoscillatory methods +(v )
2
u + (u)u shock-capturing viscosity
Remark. In the one-dimensional case both terms are proportional to

2
u
x
2
Free parameters =
c

h
1+|v|
, (u) = c

h
2
R(u)
where h is the mesh size
and R(u) is the residual
Problem: how to determine proper values of the constants c

and c

???
Alternative: use a high-order time-stepping method or ux/slope limiters
Lax-Wendro time-stepping
Consider a time-dependent PDE
u
t
+Lu = 0 in (0, T)
1. Discretize it in time by means of the Taylor series expansion
u
n+1
= u
n
+ t
_
u
t
_
n
+
(t)
2
2
_

2
u
t
2
_
n
+O(t)
3
2. Transform time derivatives into space derivatives using the PDE
u
t
= Lu,

2
u
t
2
=

t
_
u
t
_
=

t
(Lu) = L
u
t
= L
2
u
3. Substitute the resulting expressions into the Taylor series
u
n+1
= u
n
tLu
n
+
(t)
2
2
L
2
u
n
+O(t)
3
4. Perform space discretization using nite dierences/volumes/elements
Lax-Wendro scheme for pure convection
Example. Pure convection equation
u
t
+ v
u
x
= 0 (1D case)
Time derivatives L = v

x

u
t
= v
u
x
,

2
u
t
2
= v
2
2
u
x
2
Semi-discrete scheme u
n+1
= u
n
vt
_
u
x
_
n
+
(vt)
2
2
_

2
u
x
2
_
n
+O(t)
3
Central dierence approximation in space
_
u
x
_
i
=
u
i+1
u
i1
2x
+O(x)
2
,
_

2
u
x
2
_
i
=
u
i+1
2u
i
+u
i1
(x)
2
+O(x)
2
Fully discrete scheme (second order in space and time)
u
n+1
i
u
n
i
t
+ v
u
n
i+1
u
n
i1
2x
=
v
2
t
2
u
n
i+1
2u
n
i
+ u
n
i1
(x)
2
+O[(t)
2
, (x)
2
]
Remark. LW/CDS is equivalent to FE/CDS stabilized by numerical dissipation
due to the second-order term in the Taylor series (no adjustable parameter)
Forward Euler vs. Lax-Wendro (CDS)
Modied equation for the FE/CDS scheme
u
t
+ v
u
x
=
vx
2

2
u
x
2

v(x)
2
6
(1 + 2
2
)

3
u
x
3
+ . . . where = v
t
x
unconditionally unstable since the coecient
vx
2
=
v
2
t
2
is negative
Modied equation for the LW/CDS scheme
u
t
+v
u
x
=
v(x)
2
6
(1
2
)

3
u
x
3

v(x)
3
8
(1
2
)

4
u
x
4

v(x)
4
120
(1+5
2
6
4
)

5
u
x
5
+. . .
conditionally stable for
2
1 in 1D,
2

1
8
in 2D,
2

1
27
in 3D
the second-order derivative (leading dissipation error) has been eliminated
the negative dispersion coecient corresponds to a lagging phase error i. e.
harmonics travel too slow, spurious oscillations occur behind steep fronts
the leading truncation error vanishes for
2
= 1 (unit CFL property)
Forward Euler vs. Lax-Wendro (FEM)
Galerkin FEM
_
u
t
_
i
M
u
n+1
i
u
n
i
t
, where Mu
i
=
u
i+1
+4u
i
+u
i1
6
Modied equation for the FE/FEM scheme
u
t
+ v
u
x
=
vx
2

2
u
x
2

v(x)
2
3

2
3
u
x
3
+ . . . where = v
t
x
unconditionally unstable since the numerical diusion coecient is negative
the leading dispersion error due to space discretization has been eliminated
Modied equation for the LW/FEM scheme
u
t
+v
u
x
=
v(x)
2
6

2
3
u
x
3

v(x)
3
24
(1 3
2
)

4
u
x
4
+
v(x)
4
180
(1
15
2

2
+9
4
)

5
u
x
5
+. . .
conditionally stable for
2

1
3
in 1D,
2

1
24
in 2D,
2

1
81
in 3D
the positive dispersion coecient corresponds to a leading phase error i. e.
harmonics travel too fast, spurious oscillations occur ahead of steep fronts
the truncation error does not vanish for
2
= 1 (no unit CFL property)
Lax-Wendro FEM in multidimensions
Pure convection equation
u
t
+v u = 0 in (0, T) v = v(x)
Boundary conditions u = g on
in
= {x : v n < 0} inow boundary
Time derivatives L = v
u
t
= v u streamline derivative

2
u
t
2
= (v )
2
u streamline diusion (second derivative in the ow direction)
Semi-discrete scheme u
n+1
= u
n
t v u
n
+
(t)
2
2
(v )
2
u
n
+O(t)
3
Weak formulation for the Galerkin method
_

w(u
n+1
u
n
) dx = t
_

wv u
n
dx +
(t)
2
2
_

w(v )
2
u
n
dx
Integration by parts using the identity (ab) = a b +b a yields
_

wv v udx =
_

(wv) v udx +
_

out
wv n v uds
=
_

v w v udx
_

w v v udx +
_

out
wv n v uds
Taylor-Galerkin methods
Donea (1984) introduced a family of high-order time-stepping schemes which
stabilize the convective terms by means of intrinsic streamline diusion
Convection-dominated PDE
u
t
+Lu = 0 in (0, T)
Taylor series expansion up to the third order
u
n+1
= u
n
+ t
_
u
t
_
n
+
(t)
2
2
_

2
u
t
2
_
n
+
(t)
3
6
_

3
u
t
3
_
n
+O(t)
4
Time derivatives
u
t
= Lu,

2
u
t
2
=

t
_
u
t
_
=

t
(Lu) = L
u
t
= L
2
u

3
u
t
3
= L
2 u
t
= L
2 u
n+1
u
n
t
+O(t) to avoid third-order space derivatives
Substitution u
n+1
= u
n
tLu
n
+
(t)
2
2
L
2
u
n
+
(t)
2
6
L
2
(u
n+1
u
n
)+O(t)
4
Remark. The Lax-Wendro scheme is recovered for u
n+1
= u
n
(steady state)
Euler Taylor-Galerkin scheme
Semi-discrete FE/TG scheme
_
I
(t)
2
6
L
2
_
u
n+1
u
n
t
= Lu
n
+
t
2
L
2
u
n
Space discretization: Galerkin FEM (nite dierences/volumes also feasible)
The third-order term results in a modication of the consistent mass matrix
Example. Pure convection in 1D
u
t
+ v
u
x
= 0, L = v

x
Modied equation for the FE/TG scheme (Galerkin FEM, linear elements)
u
t
+ v
u
x
=
v(x)
3
24
(1
2
)

4
u
x
4
+
v(x)
4
180
(1 5
2
+ 4
4
)

5
u
x
5
+ . . .
conditionally stable for
2
1 in 1D,
2

1
8
in 2D,
2

1
27
in 3D
the leading dispersion error is of higher order than that for LW/FEM
the leading truncation error vanishes for
2
= 1 (unit CFL property)
Leapfrog Taylor-Galerkin scheme
Taylor series u
n1
= u
n
t
_
u
t
_
n
+
(t)
2
2
_

2
u
t
2
_
n

(t)
3
6
_

3
u
t
3
_
n
+O(t)
4
It follows that u
n+1
u
n1
= 2t
_
u
t
_
n
+
(t)
3
3
_

3
u
t
3
_
n
+O(t)
4
Time derivatives
u
t
= Lu,

3
u
t
3
= L
2 u
t
= L
2 u
n+1
u
n
t
+O(t)
Semi-discrete LF/TG scheme
_
I
(t)
2
6
L
2
_
u
n+1
u
n1
2t
= Lu
n
Modied equations for leapfrog schemes with L = v

x
LF/CDS
u
t
+ v
u
x
=
v(x)
2
6
(1
2
)

3
u
x
3
+
v(x)
4
120
(1 10
2
+ 9
4
)

5
u
x
5
+ . . .
LF/FEM
u
t
+ v
u
x
=
v(x)
2
6

2
3
u
x
3
+
v(x)
4
360
(2 27
4
)

5
u
x
5
+ . . .
LF/TG
u
t
+ v
u
x
=
v(x)
4
360
(2 + 5
2
7
4
)

5
u
x
5
+ . . .
fourth-order accurate, non-dissipative and conditionally stable for
2
1
the truncation error shrinks as compared to that for 2nd-order LF schemes
the unit CFL property is satised for phase angles in the range 0

2
Crank-Nicolson Taylor-Galerkin scheme
Taylor series expansions up to the fourth order
u
n+1
= u
n
+ t
_
u
t
_
n
+
(t)
2
2
_

2
u
t
2
_
n
+
(t)
3
6
_

3
u
t
3
_
n
+O(t)
4
u
n
= u
n+1
t
_
u
t
_
n+1
+
(t)
2
2
_

2
u
t
2
_
n+1

(t)
3
6
_

3
u
t
3
_
n+1
+O(t)
4
It follows that u
n+1
= u
n
+
t
2
_
_
u
t
_
n
+
_
u
t
_
n+1
_
+
(t)
2
4
_
_

2
u
t
2
_
n

2
u
t
2
_
n+1
_
+
(t)
3
12
_
_

3
u
t
3
_
n
+
_

3
u
t
3
_
n+1
_
+O(t)
4
Time derivatives
u
t
= Lu,

2
u
t
2
=

t
_
u
t
_
=

t
(Lu) = L
u
t
= L
2
u
_

3
u
t
3
_
n
+
_

3
u
t
3
_
n+1
= L
2
_
_
u
t
_
n
+
_
u
t
_
n+1
_
= 2L
2 u
n+1
u
n
t
+O(t)
Fourth-order accurate Crank-Nicolson time-stepping
u
n+1
= u
n

t
2
L(u
n
+ u
n+1
) +
(t)
2
4
L
2
(u
n
u
n+1
) +
(t)
2
6
L
2
(u
n+1
u
n
)
Crank-Nicolson Taylor-Galerkin scheme
Semi-discrete CN/TG scheme
_
I +
t
2
L +
(t)
2
12
L
2
_
u
n+1
u
n
t
= Lu
n
Modied equations for Crank-Nicolson schemes with L = v

x
CN/CDS
u
t
+ v
u
x
=
v(x)
2
6
_
1 +

2
2
_

3
u
x
3
+
v(x)
4
120
(1 + 5
2
+
3
2

4
)

5
u
x
5
+ . . .
CN/FEM
u
t
+ v
u
x
=
v(x)
2
12

2
3
u
x
3
+ . . .
CN/TG
u
t
+ v
u
x
=
v(x)
4
720
(4 5
2
+
4
)

5
u
x
5
+ . . .
fourth-order accurate, non-dissipative and unconditionally stable
cannot be operated at
2
> 1 since the matrix becomes singular
the phase response is far superior to that for 2nd-order CN schemes
the leading truncation error vanishes for
2
= 1 (unit CFL property)
Remark. Both LF/TG and CN/TG degenerate into the unstable Galerkin
discretization if the solution reaches a steady state so that u
n+1
= u
n
Multistep Taylor-Galerkin schemes
Fractional step algorithms of predictor-corrector type lend themselves to the
treatment of (nonlinear) problems described by PDEs of complex structure
Purpose: to avoid a repeated application of spatial dierential operators to
the governing equation and/or enhance the accuracy of time discretization
Taylor series u
n+1
= u
n
+ t
_
u
t
_
n
+
(t)
2
2
_

2
u
t
2
_
n
+O(t)
3
Factorization I + t

t
+
(t)
2
2

2
t
2
= I + t

t
_
I +
t
2

Richtmyer scheme (two-step Lax-Wendro method)

u
n+1/2
= u
n
+
t
2
_
u
t
_
n
u
n+1
= u
n
+ t
_
u
t
_
n+1/2

u
n+1/2
= u
n

t
2
Lu
n
u
n+1
= u
n
tLu
n+1/2
second-order RK method (forward Euler predictor + midpoint rule corrector)
stability and phase characteristics as for the single-step Lax-Wendro scheme
Multistep Taylor-Galerkin schemes
Taylor series u
n+1
= u
n
+ t
_
u
t
_
n
+
(t)
2
2
_

2
u
t
2
_
n
+
(t)
3
6
_

3
u
t
3
_
n
+O(t)
4
Factorization I + t

t
+
(t)
2
2

2
t
2
+
(t)
3
6

3
t
3
= I + t

t
_
I +
t
2

t
+
(t)
2
6

2
t
2
_
= I + t

t
_
I +
t
2

t
_
I +
t
3

t
_
no high-order derivatives
Three-step Taylor-Galerkin method (Jiang and Kawahara, 1993)
u
n+1/3
= u
n
+
t
3
_
u
t
_
n
u
n+1/2
= u
n
+
t
2
_
u
t
_
n+1/3
u
n+1
= u
n
+ t
_
u
t
_
n+1/2

u
n+1/3
= u
n

t
3
Lu
n
u
n+1/2
= u
n

t
2
Lu
n+1/3
u
n+1
= u
n
tLu
n+1/2
third-order time-stepping method, conditionally stable for
2
1 (optimal)
no improvement in phase accuracy as compared to the two-step TG algorithm
lagging phase error at intermediate and short wavelengths, unit CFL property
High-order Taylor-Galerkin schemes
Multistep TG methods involving second time derivatives oer high accuracy
and an isotropic stability domain for nonlinear multidimensional problems
Two-step third-order TG scheme (Selmin, 1987)
u
n+1/2
= u
n
+
t
3
_
u
t
_
n
+ (t)
2
_

2
u
t
2
_
n
predictor
u
n+1
= u
n
+ t
_
u
t
_
n
+
(t)
2
2
_

2
u
t
2
_
n+1/2
corrector
is chosen so as to obtain the desired stability/accuracy characteristics
excellent phase response of the FE/TG method is reproduced for =
1
9
stable for
2

3
4
in 1D/2D/3D (no loss of stability in multidimensions)
Underlying factorization vs. Taylor series expansion
I +t

t
+
(t)
2
2

2
t
2
h
I +
t
3

t
+ (t)
2
2
t
2
i
= I +t

t
+
(t)
2
2

2
t
2
+
(t)
3
6

3
t
3
+
(t)
4
2

4
t
4
Remark. A fourth-order accurate time-stepping method is recovered for =
1
12
Two-step fourth-order TG schemes
TTG-4A scheme (Selmin and Quartapelle, 1993)
u
n+1/2
= u
n

t
3
Lu
n
+
(t)
2
12
L
2
u
n
predictor
u
n+1
= u
n
tLu
n
+
(t)
2
2
L
2
u
n+1/2
corrector
fourth-order accurate in time, isotropic stability condition
2
1
poor phase response at intermediate and short wavelengths as || 1
TTG-4B scheme 0.1409714, 0.1160538, 0.3590284
u
n+1/2
= u
n
tLu
n
+ (t)
2
L
2
u
n
predictor
u
n+1
= u
n
tLu
n+1/2
+ (t)
2
L
2
u
n+1/2
corrector
fourth-order accurate in time, isotropic stability condition
2
0.718
excellent phase response in the whole range of Courant numbers
Semi-implicit Taylor-Galerkin schemes
Problem: fully explicit schemes are doomed to be conditionally stable
Semi-implicit Lax-Wendro method (Hassan et al., 1989)
u
n+1
= u
n
tLu
n
+
(t)
2
2
L
2
u
n+1
+O(t)
3
unconditionally stable
High-order multistep TG schemes (Safjan and Oden, 1993)
[I (t)
2
L
2
]u
n+
i
= u
n
+
i1

j=0
[
ij
tL +
ij
(t)
2
L
2
]u
n+
j
, i = 1, . . . , s
Here 0 =
0
. . .
s
= 1, the free parameter is to be chosen from stability
considerations and the coecients
i
,
ij
,
ij
must satisfy the order conditions

k
i
k
s

j=1
[
ij

k1
j
+
ij
(k 1)
k2
j
] =
_

i0
, i = 1
2
i0
, i = 2
0, otherwise
i = 1, . . . , s
k = 1, . . . , p
for an s-step scheme to be of pth order (p = 2s is the highest possible accuracy)
Pade approximations
Taylor series expansion (Donea et al., 1998)
u
n+1
=
_
1 + t

t
+
(t)
2
2

2
t
2
+
(t)
3
6

3
t
3
+ . . .
_
u
n
= exp
_
t

t
_
u
n
Pade approximations of order p = m + n to the exponential of x = t

t
R
n,m
(x) :=
P
n
(x)
Q
m
(x)
exp(x) multistage Taylor-Galerkin methods
Example. R
2,0
= 1 + x +
x
2
2
(second order)
u
n+1
=
_
1 + x
_
1 +
x
2
__
u
n
= u
n
+ t
_
u
t
_
n+1/2
where u
n+1/2
= u
n
+
t
2
_
u
t
_
n
R
2,0
Richtmyer scheme
R
3,0
Jiang-Kawahara
R
1,1
Crank-Nicolson
R
2,2
CNTG scheme
Pade approximations
m, n 0 1 2 3
0 1 1 + x 1 + x +
1
2
x
2
1 + x +
1
2
x
2
+
1
6
x
3
1
1
1x
1+
1
2
x
1
1
2
x
1+
2
3
x+
1
6
x
2
1
1
3
x
1+
3
4
x+
1
4
x
2
+
1
24
x
3
1
1
4
x
2
1
1x+
1
2
x
2
1+
1
3
x
1
2
3
x+
1
6
x
2
1+
1
2
x+
1
12
x
2
1
1
2
x+
1
12
x
2
1+
3
5
x+
3
20
x
2
+
1
60
x
3
1
2
5
x+
1
20
x
2
3
1
1x+
1
2
x
2

1
6
x
3
1+
1
4
x
1
3
4
x+
1
4
x
2

1
24
x
3
1+
2
3
x+
1
20
x
2
1
3
5
x+
3
20
x
2
+
1
60
x
3
1+
1
2
x+
1
10
x
2
+
1
120
x
3
1
1
2
x+
1
10
x
2

1
120
x
3
m = 0 explicit TG schemes, m > 0 implicit TG schemes