Finite Difference Methods9
Note thatD
0
x
u=(1/2)(D
+
x
+D
−
x
)u. More generally,
one can use the so-called “θ-method”
u
η
(x)γ[θD
+
x
+(1−θ)D
−
x
]u(x) ( 7)
whereθis a method parameter. Thus,
u
η
(x)=[θD
+
x
+(1−θ)D
−
x
]u(x)+
h
2
(1−2θ)u
ηη
(x)
−
h
2
6
[(1−θ)u
(3)
(x−η
2
h)+θu
(3)
(x+η
1
h)]
where 0<η
i
<1,i=1,2, so
u
η
(x)=[θD
+
x
+(1−θ)D
−
x
]u(x)
+
O(h
2
)if|1−2θ|=O(h)
O(h)otherwise
If 0≤θ≤1, we get
u
η
(x)=
1
h
[θu(x+h)+(1−2θ)u(x)−(1−θ)u(x−h)]
+
h
2
(1−2θ)u
ηη
(x)−
h
2
6
u
(3)
(x+η
3
h),−1<η
3
<1
Note that forθ=1/2, we get (6), forθ=1, we get (4),
and forθ=0, we get (5). Hence, theθ-method generalizes
the previous methods.
An approximation for the second derivative is obtained
as
u
ηη
(x)γD
+
x
D
−
x
u(x)≡D
0
xx
u(x),
thecentral difference of second order(8)
We have thenD
0
xx
u(x)=D
+
x
D
−
x
u(x)=h
−2
[u(x+h)−
2u(x)+u(x−h)], thus
u
ηη
(x)=D
0
xx
u(x)−
h
2
12
u
(4)
x
(x+η
4
h),−1<η
4
<1
or
u
ηη
(x)=D
0
xx
u(x)+O(h
2
), h→0
Similar expressions hold forD
+
y
,D
−
y
,D
0
y
, and so on. In
particular, ifu
(4)
x
,u
(4)
y
∈C(
ρ),
D
+
x
D
−
x
u(x,y)+D
+
y
D
−
y
u(x,y)
=h
−2
x
[u(x+h
x
,y)+u(x−h
x
,y)−2u(x,y)]
+h
−2
y
[u(x, y+h
y
)+u(x, y−h
y
)−2u(x,y)]
=u
xx
(x, y)+u
yy
(x, y)+O(h
2
x
)+O(h
2
y
), h
x
,h
y
→0
(9)
Forh
x
=h
y
=h,wehave
(5)
u:=[D
+
x
D
−
x
+D
+
y
D
−
y
]u(x,y)=h
−2
[u(x+h, y)
+u(x−h, y)+u(x, y+h)+u(x, y−h)−4u(x,y)]
(5)
is called the5-point difference operator.
Various difference methods adjusted to the type of the
problem, with discretization error estimates based on trun-
cation errors, are presented.
When deriving truncation error estimates for difference
approximations, one uses Taylor expansion (assuming suf-
ficient regularity ofu)
u(x
i
+h)=u(x
i
)+hu
η
(x
i
)+···+
1
(k−1)!
×h
k−1
u
(k−1)
(x
i
)+R(x
i
,h,k) (10)
where the remainder termR(x
i
,h,k)can be written as
R(x
i
,h,k)=(1/k!)h
k
u
(k)
(ξ
i
),ξ∈(x
i
,x
i
+h)or in the
alternative formR(x
i
,h,k)=
xi+h
x
i
[1/(k−1)!](x
i
+h−
s)
k−1
u
(k)
(s)ds.
2 TWO-POINT BOUNDARY VALUE
PROBLEMS
The most common among problems of applied mathemat-
ics type that appear in physics, engineering, and so on are
boundary value problems for partial differential equations.
As an introduction to difference methods for such problems,
we consider here the corresponding problem in one dimen-
sion, the two-point linear differential equation problem:
Findu∈C
2
[a,b] such that
Lu≡−(k(x)u
η
)
η
+p(x)u=f(x), a<x<b ( 11)
with boundary conditions
r
0
(u)≡γ
0
u(a)−δ
0
k(a)u
η
(a)=α (12)
r
1
(u)≡γ
1
u(b)+δ
1
k(b)u
η
(b)=β
Here,u
η
=du/dx,k(x)≥k
0
>0,a≤x≤bandk∈
C
1
(a, b),andp, f∈C(a,b)are given real-valued func-
tions andγ
0
,δ
0
,γ
1
,δ
1
,α,βare given real numbers. The
operator
Lis self-adjoint, that is,
b
a
Luvdx=
b
a
Lvudx
The solutionuwill then be a twice continuously differen-
tiable function.
Such problems arise, for instance, if we letube the
displacement of a (thin) elastic string subjected to forces
with distribution defined byf. In the simplest model,k
is constant,p(x)≡0, and the string is fixed at(a,α),