You are on page 1of 9

Mark Scheme 2623

June 2005

MEI Numerical Methods (2623)


1 (a) 2/3 stored as 0.666 666 7
mpe is 0.000 000 05
mpre is greatest when x is least
mpre is 0.000 000 05 / 0.1 =

Absolute error 0.000 000 033...

5 * 10^ -7

5E-07

[A1A1]
[A1]
[M1]
[M1A1]
[subtotal 6]

(b) Maximum possible, in theory, is 1p per call: 1000 per day


This would only arise if every call rounded downwards under tariff A
In practice, about half would round up and half would round down under tariff A
So likely benefit is 500 per day

[B1]
[E1]
[M1]
[A1]
[subtotal 4]

(c) Computations of this type contain rounding errors


The rounding errors will be different when the two sums are computed
Adding from large to small loses precision (the small number is lost)
Adding from small to large allows each number to contribute to the sum
Hence the second sum is likely be more accurate

[E1]
[E1]
[E1]
[E1]
[E1]
[subtotal 5]
[TOTAL 15]

2 (i) x
1
2
(A)

f(x)
-4
24 root in the interval (1, 2)
r

Xr

f(Xr)

1
2
1.142857
1.251374
1.618711

-4
24
-3.47891
-2.68556
4.257268

[B1]
[subtotal 1]

(ii)
secant 0
method 1
2
3
4
(B)
fixed
point

r
0
1
2
3
4

Xr
1
1.37973
1.437547
1.44558
1.446682
root at 1.45 seems
secure
Fixed point method (B) is much faster.

[M1]
[M1A1]
[M1A1]
[A1]

[M1A2]
[A1]
[E1]
[subtotal 11]

(iii)

30
25
20
15
10
5
0
-5 0

0.5

1.5

2.5

-10

[G1]
The non-linearity of the function around the
root
makes the secant method slow

[E1]
[E1]
[subtotal 3]
[TOTAL 15]

3 (i) x
2
3
4

f(x)
0.832555
1.048147
1.177410

f(x)

2.5
3.5
2.25
2.75
3.25
3.75

0.957231
1.119269
0.900517
1.005784
1.085659
1.149676

M1 =
T1 =
S1 =

(2*M1 + T1) / 3 =

2.096294 [M1A1]
2.009965 [A1]
2.067518 [M1A1]
[subtotal 5]

(ii)
T2 =
M2 =
S2 =
T4 =
M4 =
S4 =

(M1 + T1) / 2 =
(2*M2 + T2) /3 =
(M2 + T2) / 2 =
(2*M4 + T4) / 3 =

2.053129 [A1]
2.076500 [A1]
2.068710 [A1]
2.064815 [A1]
2.070818 [A1]
2.068817 [A1]
[subtotal 6]

2.067518 diffs
ratio
(iii) S1
S2
2.068710 0.001192 of diffs
S4
2.068817 0.000107 0.089854 ( = 1/16 approx)
I = (16*S4 - S2) / 15 or equivalent extrapolation =
2.068824
2.0688 seems secure

[M1A1]
[M1A1]

[subtotal 4]
[TOTAL 15]

4 (i)

x
1

f(x)
4

f(x)

f(x)

f(x)

3
2

6
3

76

a 13
a7

a4

87 3a
80 2a

76 a

(ii)

87 3a = a 13 gives a = 25

[M1]

f(x) = 4 - 3(x-1) + 6(x-1)(x-2)/2 + 12(x-1)(x-2)(x-3)/6


= 4 - 3x + 3 + 3x2 - 9x + 6 + 2x3 - 12x2 + 22x -12
= 2x3 - 9x2 + 10x + 1

(iii)

f '(x) = 6x - 18x + 10 = 0
x = 2.26
(2.26376)
f(2.26...) = 0.718

[A4]
(-1 each
error)
[subtotal 4]

Algebra may appear in (iii)


rather than (ii) for full credit

[M1A1A1A1A1]
[A1]
[A1]
[subtotal 8]
[M1]
[A1]
[A1]
[subtotal 3]
[TOTAL 15]

2623 - Numerical Methods

Comments on Individual Questions


1)

Errors and approximations


In part (a) there was some confusion between absolute and relative error. Many
candidates did not appreciate that to maximize relative error it is necessary to
make the denominator as small as possible.
Many of the explanations in part (b) were not as clear as they might have been.
The calculations were mostly accurate, though some candidates confused pounds
and pence.
Part (c) was worth 5 marks, but many answers made only one or two points. The
question contains several quite distinct requests and, as a matter of examination
technique, candidates would be advised to respond carefully to each one in turn.
There were a small number of excellent solutions to this part.

2)

Secant and fixed point methods to solve an equation


Parts (i) and (ii) were generally well done, though some candidates did not use the
methods specified. There can be no credit for using an alternative method even if
it gives the correct numerical solution.
In part (iii), many of the sketches were of poor quality and the accompanying
explanation was frequently inadequate. The point here is that the curve has a
shallow gradient to the left of the root but is steep to the right of the root. This
makes the secant method slow to converge.

3)

Numerical integration
The numerical work was very well done in parts (i) and (ii), with many candidates
appearing to be comfortable using the relationships between the trapezium rule,
the mid-point rule and Simpsons rule to minimize labour. In part (iii) the
extrapolation defeated some, but for many it proved no problem.

4)

Difference table, Newtons forward difference method


The missing values in the difference table were found correctly by most, though
some made sign errors. Demonstrating the value of a presented little difficulty.
The algebra required to obtain the cubic was more of a challenge, however, and
there were many errors. In part (iii), a significant number did not think to find the
minimum by differentiation.

You might also like