3.5.

ADI METHODS

3.5 ADI methods

Solve the above as
� � � �
k k
I − A1h v n+1/2 = I + A2h v n
2 2
� � � �
k k
I − A2h v n+1 = I + A1h v n+1/2
2 2
This is the Peaceman–Rachford Algorithm, which is an ADI method—alternating direction implicit
method. Meaning that the two­dimensional problem has been reduced to two one­dimensional implicit
problems by factoring the scheme.
Let’s now perform stability analysis of this scheme.
� � � �
k n+1 k
I − A2h v = I + A1h v n+1/2
2 2
� � � �
k ihξ2 −ihξ2 n+1 k ihξ1 −ihξ1
1 − b2 2 (e −2+e ) v̂ = 1 + b1 2 (e −2+e ) v̂ n+1/2
2h 2h
1 + 2µb2 sin2 (hξ2 /2) v̂ n+1 = 1 − 2µb1 sin2 (hξ1 /2) v̂ n+1/2
� � � �

1 − 2µb1 sin2 (hξ1 /2) n+1/2

v̂ n+1 = vˆ .
1 + 2µb2 sin2 (hξ2 /2)
1 − 2µb2 sin2 (hξ2 /2) n
Similarly v̂ n+1/2 = v̂
1 + 2µb1 sin2 (hξ1 /2)
1 − 2µb1 sin2 (hξ1 /2) 1 − 2µb2 sin2 (hξ2 /2) n
Meaning v̂ n+1 = · · v̂ .
1 + 2µb1 sin2 (hξ1 /2) 1 + 2µb2 sin2 (hξ2 /2)
� �
Since � 1−x
1+x � ≤ 1 for any x ≥ 0, we conclude |v̂
n+1
| ≤ |v̂ n |.
� �
Now using Parseval
� � π/h � π/h
n+1 2
|vml | · h2 = |v̂ n+1 (ξ1 , ξ2 )|2 dξ1 dξ2
m,l −π/h −π/h
� π/h � π/h
≤ |v̂ n (ξ1 , ξ2 )|2 dξ1 dξ2
−π/h −π/h
�
n 2
= |vml | · h2 .
m,l

Therefore we have stability for all values of m and l, and the order of accuracy of the scheme is O(k 2 +h2 ).
Thus we can take k = h and the scheme is efficient and accurate.
The Douglas–Rachford method starts with the backward­time, central­space scheme for ut = A1 u+A2 u
n+1 n
(I − kA1h − kA2h )vml = vml ,

to obtain (after dropping an O(k 3 ) term)

n+1
(I − kA1h )(I − kA2h )vml = (I + k 2 A1h A2h )vml
n
.

The method is

(I − kA1h ) v n+1/2 = (I + kA2h ) v n

(I − kA2h ) v n+1 = v n+1/2 − kA2h v n .
 CHAPTER 3. SYSTEMS OF PDES IN HIGHER DIMENSIONS

The Mitchell–Fairweather method is second­order accurate in time and fourth­order accurate in space.
Recall the operator δ 2 :
f (x + h) − 2f (x) + f (x − h)
δ 2 f (x) ≡
h2
and n n n n n n
vm+1,l − 2vml + vm−1,l vm,l+1 − 2vml + vm,l−1
δx2 vml
n
= and δ 2 n
y vml = .
h2 h2
Then
h2
δ 2 f = f �� + f ���� + O(h4 ) = f �� + O(h2 ),
12
so f �� = δ 2 f + O(h2 ). Then

h2 ����
δ2 f = f �� + f + O(h4 )
12
h2
= f �� + (f �� )�� + O(h4 )
12
�� h2 2 ��
= f + (δ f + O(h2 )) + O(h4 )
12
2
h
= f �� + δ 2 f �� + O(h4 )
12
h2
� � 2
d
= 2
δ f = 1 + δ2 f + O(h4 ). (3.1)
12 dx2

Now start with Peaceman–Rachford idea for ut = A1 u + A2 u:

� �� � � �� �
k k n+1 k k
I − A1 I − A2 u = I + A1 I + A2 un + O(k 3 )
2 2 2 2

Multiply both sides by

h2 2 h2
� �� �
1+ δ 1 + δy2
12 x 12
and replace
h2 2 ∂2
� �
1+ δ
12 x ∂x2
by δx2 + O(h4 ) (see (3.1)). Similar changes are made for the derivatives with respect to y. The result is

h2 2 k h2
� �� �
k
1+ δ − b1 δ 2 1 + δy2 − b2 δy2 un+1
12 x 2 x 12 2
h2 h2
� �� �
k k
= 1 + δx2 + b1 δx2 1 + δy2 + b2 δy2 un + O(k 3 ) + O(kh4 ).
12 2 12 2

We obtain the Mitchell–Fairweather scheme:

� � � � � � � �
1 1 1 1
1− b1 µ1 − 2 2
h1 δx v n+1/2 = 1+ b2 µ2 + 2 2
h2 δ y v n
2 6 2 6
� � � � � � � �
1 1 1 1
1− b2 µ2 − 2 2
h2 δy v n+1 = 1+ b1 µ1 + 2 2
h1 δx v n+1/2 .
2 6 2 6
3.6. BOUNDARY CONDITIONS FOR ADI METHODS

3.6 Boundary conditions for ADI methods

One can obtain the boundary conditions for the intermediate step v n+1/2 by solving for it using the boundary
conditions at steps n and n + 1.
For example by subtracting the equations of the Peaceman–Rachford method
� � � �
k n+1/2 k
I − A1 u = I + A2 u n
2 2
� � � �
k n+1 k
I − A2 u = I + A1 un+1/2
2 2

(note that we use u’s and not v’s and keep everything is operator form for the moment) we get
� � � �
n+1/2 1 k n 1 k
u = I + A2 u + I − A2 un+1 .
2 2 2 2

n n+1/2 n+1/2
Now if the indices i and j in vij range from 1 to m, the desired boundary conditions at v1i and vmi
for i = 2, 3, . . . , m − 1 are computed as
� � � �
n+1/2 1 k n 1 k n+1
v1i = 1 + A2 v1i + 1 − A2 v1i
2 2 2 2
n n n n+1 n+1 n+1
b2 µv1,i−1 + 2(1 − b2 µ)v1i + b2 µv1,i+1 −b2 µv1,i−1 + 2(1 + b2 µ)v1i − b2 µv1,i+1
= + .
4 4
n+1/2
And similarly for vmi for i = 2, 3, . . . , m − 1.
The boundary condition
n+1/2
vij = u(tn+1/2 , xi , yj )
is only first order accurate and if used with the Peaceman–Rachford method (or other similar second order
accuracy) will result in the overall accuracy being only first order.
Boundary conditions for the Mitchell–Fairweather scheme are obtained as follows. First we eliminate
δx2 v n+1/2 terms by multiplying the first equation by 1+ 12 (b1 µ1 + 16 )h21 δx2 and the second by 1− 21 (b1 µ1 − 16 )h21 δx2
to obtain:
� n+1
b1 µ1 + 16 1 + 12 (b2 µ2 + 61 )h22 δy2 v1i + b1 µ1 − 16 1 − 21 (b2 µ2 − 16 )h22 δy2 v1i
� �� � n � ��
n+1/2
v1i =
2b1 µ1
for i = 2, 3, . . . , m − 1, and similarly for the other boundary.