6.1 Tbc Scalar Advection Equation
307
In the case of the unforced scalar advection equation, it is helpful to perforrn the
Taylor series expansions about the point (xj , tn) because this isolates the errors in
the trajectory calculations from those generated by the interpolation of the tracer
field. Let 1/fd = 1/f(xj, r") ; then the error produced by linear interpolation is
Since the wind speed is constant, the backward trajectory is exact and 1/fj+1 = 1/Id.
This can be verified by expanding 1/fj+I in a Taylor series. Defining s = x j - xj ,
= 1/Id + Ilt 81/1 I + s 81/f I
J
8t d
8x d
+ (1lt)2 8
2 1/1 I + s Ilt 8
2 1/f I + s2 8
2 1/f I +.. .
(6.8)
2 8t 2 d
8t8x d 2 8x 2 d
'
and since s = U Si, the preceding reduces to
1/f n+1 .
J
= 1/fd + Ilt ( 8
-
8t
+ U - 8) 1/f I + -
Bx
(llt)2 (
- - + U -
8
8)2 I
1/f + ...
d
2
8t
8x
d
Substituting (6.7) and (6.9) into (6.6) yields
1/f'r
1
- [(I - cx)1/fj_p + cx1/f'j_P _I] <>< _cx(l - cx) (llx)2 8 2 1/f I
.(6.10)
If the Courant number is held constant as Ilt
0 and Ilx
0 and Ilt
0, the truncation
0, the crror is
crror is clearly O(llx). If Sx] Ilt
0 as Ilx
no larger than O(llx). It may appear that the semi-Lagrangian scheme could be
inconsistent in the limit I'::1t I Ilx
unity, cx = U I'::1t I Ilx, and using
0, but once the Courant number drops below
2
1/f
2 82 1/f
8
- - U -
8t 2 -
8x 2 '
the leading-ordcr truncation error reduces to
This is identical to the leading-ordcr truncation error in Eulerian upstream differencing (2.12). The global truncation error of the semi-Lagrangian scheme (6.5) is
therefore of first order in space and time .
Consistent with (6.10) , the preceding semi-Lagrangian scheme is exact whenever the Courant number is an integer, i.e., whenever the departure point exactly
307
In the case of the unforced scalar advection equation, it is helpful to perforrn the
Taylor series expansions about the point (xj , tn) because this isolates the errors in
the trajectory calculations from those generated by the interpolation of the tracer
field. Let 1/fd = 1/f(xj, r") ; then the error produced by linear interpolation is
Since the wind speed is constant, the backward trajectory is exact and 1/fj+1 = 1/Id.
This can be verified by expanding 1/fj+I in a Taylor series. Defining s = x j - xj ,
= 1/Id + Ilt 81/1 I + s 81/f I
J
8t d
8x d
+ (1lt)2 8
2 1/1 I + s Ilt 8
2 1/f I + s2 8
2 1/f I +.. .
(6.8)
2 8t 2 d
8t8x d 2 8x 2 d
'
and since s = U Si, the preceding reduces to
1/f n+1 .
J
= 1/fd + Ilt ( 8
-
8t
+ U - 8) 1/f I + -
Bx
(llt)2 (
- - + U -
8
8)2 I
1/f + ...
d
2
8t
8x
d
Substituting (6.7) and (6.9) into (6.6) yields
1/f'r
1
- [(I - cx)1/fj_p + cx1/f'j_P _I] <>< _cx(l - cx) (llx)2 8 2 1/f I
.(6.10)
If the Courant number is held constant as Ilt
0 and Ilx
0 and Ilt
0, the truncation
0, the crror is
crror is clearly O(llx). If Sx] Ilt
0 as Ilx
no larger than O(llx). It may appear that the semi-Lagrangian scheme could be
inconsistent in the limit I'::1t I Ilx
unity, cx = U I'::1t I Ilx, and using
0, but once the Courant number drops below
2
1/f
2 82 1/f
8
- - U -
8t 2 -
8x 2 '
the leading-ordcr truncation error reduces to
This is identical to the leading-ordcr truncation error in Eulerian upstream differencing (2.12). The global truncation error of the semi-Lagrangian scheme (6.5) is
therefore of first order in space and time .
Consistent with (6.10) , the preceding semi-Lagrangian scheme is exact whenever the Courant number is an integer, i.e., whenever the departure point exactly
