Numerical Differentiation

Example Suppose we are given a table below and want to compute f′(2). Approximating f′(2) by (f(3) − f(2))/(3 − 2) = 19 results in overestimation and approximating f′(2) by (f(2) − f(1))/(2 − 1) = 7 yields an underestimated result.
x f(x)
1 1
2 8
3 27
4 64
numdiff.gif
Introducing E as the shift operator as
E f(x) = f(x + h),
where h is a step size and ∆ is the difference operator defined as
f(x) = f(x + h) − f(x),
the relationship between ∆ and E is
∆ = E −1,
as
f(x)
=
f(x + h) − f(x)
=
Eff
=
(E − 1) f.
Using Taylor's series expansion for f(x + h),
f(x + h)
=
f(x) + h f′(x) + h2

2!
f"(x) + h3

3!
f"′(x) +…
=
f + h D f + h2 D2

2!
f + h3 D3

3!
f + …
=
(1+ hD + (hD)2

2!
+ (hD)3

3!
+ …) f
=
ehD f,
where D is the differential operator defined as
D f = f′(x).
Therefore, it follows that
E = ehD,
or
1+∆ = ehD.
This is equivalent to
h D = ln (1 + ∆),
or
h D = ∆− 2

2
+3

3
4

4
+ …
or
D = 1

h

∆− 2

2
+3

3
4

4
+ …
.
Thus, one can associate the differential operator, D, with the difference operator, ∆. A table of differences on f(x) can be easily created as
x f(x) f 2 f 3 f 4 f
1 1 7 12 6 0
2 8 19 18 6 0
3 27 37 24 6 0
4 64 61
Therefore,
f′(2)
=
Df|2
=
1

h

f|22 f|2

2
+3 f|2

3
− …
=
1

1
(19 − 9 + 2 − 0)
=
12.
(this is not part of the textbook, something extra)



File translated from TEX by TTH, version 4.03.
On 21 Aug 2022, 22:51.