You are on page 1of 8

Mark Scheme

Final Mark Scheme


1

(i) 1+D = exp(hD)


ln(1+D) = hD
2/2 + 3/3
hence result

2624/01

June 2004
[M1A1]
[A1]
[A1]

[subtotal 4]
x

(ii)

0
0.2
0.4
0.6
0.8

f(x)
0.480453
0.621665
0.766446
0.913002
1.060116

0.141212
0.144781
0.146556
0.147114

0.003569
0.001775
0.000558

4
0.4804530
0.6216650
0.7664455
0.9130021
0.000577 1.0601161

-0.001794
-0.001217

f '(0) = (0.141212 - 0.003569/2 - 0.001794/3 - 0.000577/4 ) / 0.2


=
0.706060
1st order
0.697138
2nd order
0.694148
3rd order
0.693426
4th order
f '(0) = 0.69 or 0.693 is justified

(iii)

f '(0)
0.2 0.697138
0.1 0.694300

errors of order h2, so


(4 x 0.694300 - 0.697138)/3
= 0.693354
Considering part (ii), 0.693 is justified

[M1A1A1]

[M1A1]
[M1A1]
[A1]
[A1]
[A1]
[subtotal 10]

[M1]
[M1A1]
[A1]
[M1A1]
[subtotal 6]
[TOTAL 20]

(i)

f(x)
1
x
x2
x3
x4

I
2h
0
2h3/3
0
2h5/5

S
h/3 (1 + 4 + 1) = 2h
h/3 (-h + 0 + h) = 0
h/3 (h2 + 0 + h2) = 2h3/3
h/3 (-h3 + 0 + h3) = 0
h/3 (h4 + 0 + h4) = 2h5/3

S - I = 4h5/15

In Taylor expansion of f(x) the quartic term is x4 f iv(0) / 4!


Hence the (leading) error is (approximately) h5 f iv(0) / 90

(ii)

The local error applies when the range of integration is covered by a single
application of the rule. The global error applies when the the range is subdivided
into n subintervals and the rule is applied to each.
In practice n will be increased and h will be reduced (with 2nh=b-a)
so the global error is the important one.
local: O(h5)

(iii)

(A = 1/90)

global: O(h4)

global errors of order h4


so an improved estimate is (16 x 2.375916 - 2.376601)/15
= 2.375870
(or rounded as far as 2.376)

[M1A1]
[A1]
[A1]
[A1]
[A1A1]
[M1A1]
[A1]
[subtotal 10]

[E1]
[E1]
[E1]
[E1]
[B1B1]
[subtotal 6]
[M1]
[M1A1]
[A1]
[subtotal 4]
[TOTAL 20]

Final Mark Scheme


3

(ii)

(iii)

(iv)

2624/01

(i) y" NOT a write down: some evidence of chain rule reqd.
y''' = ex +y"/(2y0.5) - (y' )2/(4y1.5)

y(0) = 1
y'(0) = 2
y''(0) = 2
y'''(0) = 1
T3 = 1 + 2x + x2 + x3/6
T3(0.1) = 1.2101667
T3(0.2) = 1.4413333

June 2004
[M1A1]
[M1A1A1A1]
[subtotal 6]

[B2]
[-1 per error]
[M1A1]
[A1]
[A1]
[subtotal 6]

y(0.1) =
1.210167
y'(0.1) = 2.205247
y''(0.1) = 2.107487
y'''(0.1) = 1.149810
T3(0.1 + x) = 1.210167 + 2.205247x + 2.107487 x2 / 2 +1.149810 x3 / 6
T3(0.2) = 1.441420

The second is likely to be (much) more accurate as the errors are of order h4
and two such with h=0.1 amount to (much) less than one with h=0.2.

[B1]
[B1]
[B1]
[M1]
[A1]
[subtotal 5]

[E1E1E1]
[subtotal 3]
[TOTAL 20]

(ii)

(i) A problem is ill conditioned if small changes in the input values produce large
changes in the output values (the solution)
Ill conditioning produces problems for numerical methods because, in general,
it is not possible to store input values without error

x0

y0

1.5

0.1

1.09

[E1E1]
[E1E1]
[subtotal 4]

y1

4.793703
7
1.5
0.1
1
1.1
4.262277
6
1% change in y0 produces 11% change in y1: ill conditioned.

[M1A1]
[M1A1]
[M1E1]
[subtotal 6]

(iii)

x0

y0

y1

0.5
0.1
1
1.09
1.4233333
0.5
0.1
1
1.1
1.4162277
1% change in y0 produces 0.5% change in y1: well conditioned.
k

x0

y0

[M1A1]
[A1]

y1

2.5
0.1
1
1.09
42.242263
2.5
0.1
1
1.1
32.722776
1% change in y0 produces 23% change in y1: very ill conditioned.

[M1A1]
[A1]
[subtotal 6]

(iv)

x0

y0

y1

1.5
0.1
1
1.09
4.7937037
1.515
0.1
1
1.09
4.9299237
1% change in k produces 3% change in y1: (mildly) ill conditioned.
(or other small change)

(or fairly well conditioned)

[M2A1]
[A1]
[subtotal 4]
[TOTAL 20]

Examiners Report

Report on the Units taken in June 2004

2624 Numerical Analysis


General Comments
There were just a dozen candidates for this paper, and the standard they set was high. All
appeared well prepared, with a good understanding of the nature of this subject.

Comments on Individual Questions


Q.1

The algebra in part (i) and the numerical evaluations in part (ii) caused few problems.
In part (iii), however, a number of candidates became confused about the order of
the method. A second order formula for a derivative has errors of order h2.

Q.2

The easiest way to tackle part (i) is to integrate successively f(x) = 1, x, x2, x3, x4. The
last integral gives an error in Simpsons rule of magnitude 4h2/15. Then since the
general Taylor expansion of f(x) has a term x4fiv(x)/4!, the general error will be of
magnitude h2/90. Many candidates chose a more difficult approach, comparing the
Taylor expansions of Simpsons rule and the integral of f(x). To their credit, however,
they were generally successful. In part (iii) a few fell into the trap of extrapolating
using the local rather than the global error.

Q.3

Almost everyone managed to use the chain rule to obtain the given result for the
second derivative. Many then promptly forgot the chain rule when it came to the third
derivative. The numerical work that followed was done well. The explanations
required in part (iv) were rarely complete. The point here is that the sum of two errors
of the form A(0.1)4 will be much less than a single error of the form A(0.2)4.

Q.4

This question was very well done by most candidates. The numerical evaluations
were correct and the comments on conditioning were accurate.

Coursework: Numerical Analysis


Only 12 candidates in total from two different centres submitted work for this module.
Nevertheless, a variety of tasks was attempted and the majority were completed to a very
high standard.

You might also like