Approximate Integration and Simpson’s Rule

To approximate
Z b
f (x) dx,
a
b−a
let [a, b] be subdivided into n subintervals each of length hn = .
n
The subdivision points are x0 = a, x1 = a + h, . . . , x j = a + jh, . . . , xn = b with values y j = f (x j ).

1. Trapezoid Rule
 
1 1
R0,n = Tn = hn y0 + y1 + y2 + · · · + yn−1 + yn
2 2
1 1


This is obtained by repeating the area of the trapezoid (degree 1 polynomial) approximation hn 2 y0 + 2 y1 over
successive subintervals. It is an exact approximation for linear f (x) (degree 1 polynomials).

2. Simpson’s Rule
 
1 4 2 4 2 4 1
R 1,n = Sn = hn y0 + y1 + y2 + y3 + · · · + yn−2 + yn−1 + yn , where n is even.
3 3 3 3 3 3 3
This is obtained by repeating the area of the quadratic (degree 2 polynomials) approximation hn 13 y0 + 43 y1 + 13 y2

over successive pairs of subintervals. It is an exact approximation for cubics (degree 3 polynomials).
Note
   
2 4 4 1 2
Sn = hn y0 + y1 + y2 + · · · − h n y0 + y2 + · · ·
3 3 3 3 3
   
4 1 1 1
= hn y0 + y1 + y2 + · · · − (2hn ) y0 + y2 + · · ·
3 2 3 2
4 1 4Tn − T n
2
= Tn − T 2n = .
3 3 3

3. Improvements
R2,n = Bn , Boole’s Rule, is obtained using the area of a quartic (degree 4 polynomial) approximation, is an exact
approximation for quintics (degree 5 polynomials), requires n to be a multiple of 4, and uses the coefficients:
14 64 24 64 28 64 24 64 28 28 64 24 64 14
, , , , , , , , , ..., , , , , .
45 45 45 45 45 45 45 45 45 45 45 45 45 45
One can show
16Sn − S 2n
R2,n = .
15
The pattern can be used to define the Richardson’s extrapolates (used in Romberg Integration by taking m → ∞, with
n = 2m ):
4m Rm−1,n − Rm−1, 2n
Rm,n = .
4m − 1
 Page 2 of 2

4. Derivation of Simpson’s Rule via Interpolating Polynomials

p q r
The Lagrange interpolating polynomial which is a polynomial that passes through the same points as f at x = p, x = q
and x = r is
(x − q)(x − r) (x − p)(x − r) (x − p)(x − q)
g(x) = f (p) + f (q) + f (r).
(p − q)(p − r) (q − p)(q − r) (r − p)(r − q)
Z r
f (x) dx is approximately
p
Z r Z r Z r Z r
(x − q)(x − r) (x − p)(x − r) (x − p)(x − q)
g(x) dx = dx · f (p) + dx · f (q) + dx · f (r).
p p (p − q)(p − r) p (q − p)(q − r) p (r − p)(r − q)

In Simpson’s rule we are interested in the case that q − p = r − q = h, that is, q = p + h and r = p + 2h. We show the
x− p x− p x − q 2hu + p − q
last integral is 1/3. Use substitution u = = so 2hu + p = x. Then 2h du = dx and = =
r− p 2h r−q r−q
2hu − h
= 2u − 1. The last integral becomes
h
Z 1 Z 1  3 1  
2 u 2 4 1
u(2u − 1)2h du = h 4u − 2u du = h 4 − u =h −1 = h .
u=0 u=0 3 0 3 3

By similar computations we could get the other two integrals, but there is an easier way. By symmetry the first one is
also h 31 . The case f (p) = f (q) = f (r) shows the three integrals must add to 2h so the middle one is h 43 .

5. Derivation of Simpson’s Rule by attempting to cancel errors

1 1 02 12 1 n2 1 12 n2 1 n2
     
1 1 2 2 1 2
Tn = + + · · · + = + · · · + − = 1 + · · · + n − n
n 2 n2 n2 2 n2 n n2 n2 2 n2 n n2 2
 
sum 1 1 n(n + 1)(2n + 1) 1
= − n2
n n2 6 2
1 1
1 1
= 2 (n + 1)(2n + 1) − 3n2 = 2 2n2 + 1 = + 2


6n 6n 3 6n
1 1 1 1
Similarly T 2n = + = +4 2.
3 6(n/2)2 3 6n
We want to average Tn = 31 + E and T 2n = 31 + 4E to cancel or reduce the term E which appears, where E = 6n12 . Since
4Tn − T 2n = 1 we see 43 Tn − 13 T n2 gives the exact value of 13 for the integral. This expression, which is Sn , is a better
approximation than the Trapezoid rule in this case.