cater
Errors Analysis and Numerical
Computations
ia, INTRODUCTION _
In the area of applied mathematics there is a need of numerical results
cause there are several instant in a practical application when we use the
results. For example, a set of tabulated data is given and from it the
rences have to be drawn, or a system of algebraic equations is given and one is
required to solve them. The aim of numerical analysis is to provide efficient
methods for determining numerical answers to such problems.
Sometimes the methods and process by which the desired result is to be found
are approximate. An approximate is one which consists of approximate data,
approximate methods or both. It is therefore evident that the error in a computed
result may be due to the error in the data or the errors in the method or both. In
this chapter we shall describe some basis concepts or ideas regarding numerical
computations, errors and their analysis.
1.2. | APPROXIMATE NUMBERS AND SIGNIFICANT FIGURES
(1) Approximate Numbers. In the discussion of computation there are two
kinds of numbers that are used :
() Absolutely exact numbers
(i) Approximate numbers
ner
‘The numbers such as 1, 2, 3, .-- ete, are exact numbers because there
‘sno approximation or uncertainty associated with them. Although such numbers
s \2, m,¢ ete, are also exact numbers but they cannot be expressed exactly by a
finite number of digits, When we express these numbers in a digital form, we must
write them as 1.4142, 3.1416, 2.7183 etc. Thus such numbers are only
“proximations to the true values. Therefore, such numbers are called the
‘proximate numbers.
An approximate number is therefore defined as a number which is used as an
“?broximation to its true value and differs only slightly from the true value.
(2) Significant Figures. The digits 1, 2, 3, .
"umber are called significant digits or signifi
“nificant figure except when it is used to fix the di
“unknown or discarded digits.
For example : The number 0.00123 has only three signi
ae 3. On the other hand the numbers 3.1416, 0.66753 and
'Ye significant digits. oe
.. 9 that are used to express a
icant figures, and 0 is also ;
lecimal point or to fill the pla
es Numerical Analysis
Tn order to find the significant digits, we keep the following concepts in our
in
a Ifthe number is in positional notation, then the significant figures in the
number consists
(i) all non-zero digits il :
(ii) the zero digits which lie between significant digits and lie to the
right of decimal point and at the same time, te the right of a
non-zero digit. j
(2) Ifthe number is in Scientific notation (i.e., k x 10"), then the significant:
digits are all digits explicitly in hk. : 3
A list of numbers and their significant digits 1s given which a general idea to
the students that how to calculate the significant digits :
1 "Numbers Significant Digits _|
| 0.00263 2, 6 3
| 4253 4, 2/7508
| 4200 42
| 35-06 35,0, 6
| 0.0350 Septal
| 26x 108 3 8)
3.909 x 104 | eros
5x10 5
55.00 5, ‘5, 0, 0
== REMARK :
© Significant digits :re counted from left to right starting with the left most
| 1.3. | ROUNDING OF NUMBERS
If we consider a number 21° — 2.061068702, ..., which never
in order to use this number in a practical computation, must e
manageable number of digits such as 2.06 or 2.061 or 2,06107, ete,
‘This process of cutting off unwanted digits and retaining as m
called rounding-off,
To round-off a number to n significant digits, discard all dij
n‘" digit according to the following rule:
(Ifthe discarded number
digits as such.
For example : ok
(i) aap numbers are rounded off c
is less than 5 at the (2 +1)
1.963 becomes 1.96
2.354 becomes 2.35
Gi) If the discarded number is great ir
3
ly 6 at (n + 1)" place, leave the n" digit
108 48,36
is exactly 6 at (n +1) place, add 1 to the n‘*
44995001 becomes 4.500
ers rounded-off to three significant digits is given as :
7.894 becomes 7.89
12.865 becomes 12.9
6.4356 becomes 6.44
3.4567 becomes 34600
3.8254 becomes 3.82
= REMARK :
ided-off to n significant digits (or n decimal places) are said to be correct
ignificant
1.4. | MATHEMATICAL PRELIMINARIES
I section, we state, without proof, certain mathematical results which
ld be useful in the sequel.
Theorem 1. (Intermediate theorem). If fx) is a continuous function in a
ed interval [a, 61, ie, ax
(ay)
then the truncation error
Thus truncation errors is a type of algorithm error,
mple : If we using a decimal computer having a fixed word length of 4
unding-off of 13.658 Sives 13.66 whereas the truncation error is 13.65.
lute error. Absolute error is defined as the numerical difference
tue value ofa quantity and its approximate value
If X be the true value of a quantity and X’ be its approximate value, then
X-X" | is the absolute error. It is denoted by H,, that is,
Ey=|X-X"|
5. Relative error. The relative error B, is defined by
X=X
gE, = |4=x
5. Percentage error. Percentage error Hy is defined by
ah 3 499)
E, = 100 E, = 100 ics
Absolute accuracy. If there is a number X such that Ey $X, then Xis an
‘on the magnitude of absolute error and measures the absolute accuracy.
Theorem 1. If the first significant figure of a number is k, and the number is
di
(kx 10"
to n significant figures, then the relative error is less than
Proof:
Let N = Exact value of any number
n= Number of correct significant digits
m = Number of correct decimal places.
Now we shall take three cases :
Case I. Suppose m 4x 10"~ forall values off except k= 1 and n = 1. But this ig gue
trivial caso when = 1, that is, when NV contains only one digit different ftom nan
and this digit is 1 to 9 which would never occur in practice.
1
“hx 10
Case II. Suppose m = n, in this case Vis a decimal number and A is supposed
to be at first decimal figure, then we have
Hence, E,
I
perce
‘
eed te ee
2kx10"-¥— 1 2h 10"-2— 7 “kx 10"=
Case III. Suppose m>n. In this case k occupies then (m-n+ 1)" decimal
place, therefore, —
1 - (mn +) =
p N2kx10-% Day
a 10" zs zi ana
2kx10-"x10"~1-10-™ 9k x10"-2-4 ax 10*=-1
1
kx 10"
Hence the thcorem is true in all cases.
"= REMARKS :
© The relative and percentage errors are independent of the units used while
‘expressed in terms of these units.
E,<
‘fa number is correct to n decimal places then the error =da0-%,
I the fist significant digit of number isk and the number is correct ton
then the relative error is less than 1.
(kx 10-")
SOLVED EXAMPLES a emer
(ii) 87.46238
significant digits = 37.46
53 and X" = 37.46
|X-x |
87.46 | = 0.00285
0.00235. _ 5
87,40288 4! a9"
ne E, = 100,
100 x 6.27 x 10°5 = 6,27 x 10-9
Example 2, // 0.383 is the approximate value fs, find absolute, relative and
age errors
Solution. Here X=4=0, 333333 and X’ = 0.333
8, =| x-¥)|
= | 0.833333 - 0.333 | = 0.000333
rg, =|X=X| _ 0.000833 _
r< |X. |= 0.333833
nd = 100£, = 100 x 0.000999 = 0.0999%.
Example 3. Round-~ ee the number 75462 to four significant digits and then
late the absolute error and percentage error.
Solution. 75462 rounded-off to four significant digits = 75460.
Here X (True value) = 75462, X’ + 75460
Ey =|X-X’ | = | 75462-75460 |
Absolute error = 2
Now
1.000999
Se ae
id ae
Ey = 100E, = 100 x 7-5 = 100 x 0.000265
me
Percentage error = 0.00265.
Example 4, Find the absolute, relative and percentage errors if x is
rounded-off to three decimal digits. Given x = 0.005998.
Solution. Rounded-off to three decimal digits of given number = 0.006.
Thus X = 0.005998, X’ = 0.006.
Absolute error B, = | X~X’ |
= | 0.005998 - 0.006 | = 0.000002
x=X"
x
_ 9.000002
0.005998
and Percentage error Ey = 100B,
= 100(0.00033344) = 0.033344,
‘xample 5. Evaluate the sum S = Me ss doe
e and relative errors.
Relativeerror Ey =
= 0,00083344
Examy
Hts absolun Numerical Analy sig
Solution. Since
732 + 2.236 + 2.646 = 6.614.
Since VS, V5, V7 each has 3 decimal places so each has the error
= ; (10° 8) + 0.0006. Thus the total absolute error
‘x = 0.0005 + 0.0005 + 0,0005 = 0.0016.
This absolute error shows that the sum is correct to 2 decimal places 80 $ ig
correct to 8 significant digits only,
we take s
So, relative error E, =
Example 6. Suppose 1.414 is used as an approximation. to VB. Find the:
absolute and relative errors.
Solution. Here X (True value) = V2 = 1414213562
and X’ (Approximate value) = 1.414 = | 1.414213562 - 1.414 |
0.0002135637 = 0.21356237 x 10° 3.
Eq _ 0.21356237 x 107?
Relative error 5-3 ° Taapisee
Example 7. Find the relative error of the number 8.6 if both of ite
correct.
Solution. The given number 8.6 has one decimal place therefore,
Absolute error (10-4) =0.05.
= 0.151011 x 10-3
Then, relative error £, = 2% = 0.00581,
Example 8, Find the percent if 625.483 i i
= ntage error if 625.483 ig approximated
Solution. Here _X = 625,483 and X’ = 625
Fa=|X-X | =| 625.483 - 625 | = 0.483
X-X’| __ 0.483
E, = uses
pag | $25,493 = 9.000772
¥,=100B,, = 100(0.000772) = 0.0772
Given that :
@= 10.00 +0.05
5 = 0.0356 + 0.0002
= 15300 + 100
i y absolute error in as
‘olute error ina = | +0.05 1-005
Absolute error in b = |0,0008-4.« 0.0009,
‘ine=| +100 | = 100
Absolute error ind = | +500 | =500 ysis and Numerical Computations :
hus the maxim
m absolute error ina +b +¢ 4d
= 0.05 + 0.0002 + 100 + 500 = 600.0502.
Example 10. Three approximate value of number of © are given as 0.30, 0.33
3
1 0.34. Which of these three is the best approximation ?
Solution. Here X (True value) ie 0.38833
Case I. If the approximate value
=X (say)
Absolute error (E,) = | X= X’ | = | 0.33333 - 0.30 | = 0.00333
33 = X’ (say)
= | 0.83333 - 0,33 | = 0.00333
Case IIT. If the approximate value = 0.34 = X’ (say)
Absolute error (E,) = | X-X’ | = | 0.33333 - 0.34 |
= | - 0.0067 | = 0.0067.
n second case absolute error is least, therefore 0.33 is the best approximation.
Example 11. An approximate value of n is given by 3.1428571 and its true
1415926. Find absolute and relative errors.
Solution. Here X (True value) = 3.1415926
and X’ (approximate value) = 3.1428571
Absolute error E, = | X-X’ |
= | 3.1415926 - 3.1428571 |
= | - 0.0012645 | = 0.0012645.
E, _ 0.0012645
Relative error E, x = 31415926 > 0.000402502.
Case II. If the approximate value
Absolute error (Eq) = | X-X’
EXERCISE 1
1 Iftrue value = 2 approximate value = 3.33 then find the absolute and relative errors.
2. Round-off the following numbers to two decimal places : 48.21416, 2.3742, 52.275,
2.375, 2.385, 81.255. ;
8. Found-off the following numbers correct to four significant digits : 3.26425, 95.46735,
4985561, 0.70035, 0.00032217, 16583, 30.0567, 0.859378, 3.141559.
4.Ifn= a is approximated as 3.14, find the absolute error, relative error and percentage
error,
5. Given that : a = 10.00 + 0.05, b = 0.0356 + 0.0002, c = 15300 + 100, d = 6200 + 500.
a se
Find the maximum value of the absolute error in (i) a +5e~d, Gi)d® ‘
6. \29 = 5.385 and Vm = 3.317 correct to 4 significant figures. Find the relative errars in.
their sum and difference.
7. The discharge @ over a notch for head H is calculated by the formula =
is a given constant. If the head is 75 cm. and an error of 0.15 cm. is
measurement, estimate the percentage error is computing the discharge.
. The height of an observation tower was estimated to be 47 m. whereas.
was 45 m. Calculate the percentage relative error in the meas! - Numerical Analysis
9. Two numbers are 3,5 and 47.279 both of which are correct to the significant figures
given. Find their product
10, If the number p is correct to 8 significant digits, what will be the maximum relative
, B, = 0.000999 2, 48.21, 2.97, 52.28, 2.38, 2.98, 81.26
47, 4986000, 0.7004, 0,0003222, 1.668, 30,06, 0.8594, 3.142
0.002857, B= 0.0009, B, = 0.09%
5.) 1000.08 Gi) 6.766 x 10"
6. 1.149x10"4, 4.836x10"4 7.0.5 8. 4.44% 9. 165.55 10. 0.0050,
1.6. | A GENERAL FORMULA FOR ERRORS
Let y=/lty %, ¥g, «,%,) be a function of several independent variables
Ee See
Tf&xy, Sta, .., Sx, be the errors in-xy, %9, ...&%q respectively, then an error 6y in
the function y given by the relation
+ By = fly + Bx, 9 + Btpy 1 ay + Bt)
Now expanding the R.H.S. of (1) by Taylor's theorem for a function of st
variables, we get
¥ + by = fly, mp, «..
+ terms having higher powers of 8, 8,
Since, the errors &r,, Sr, .., &, are relatively small, so neglectin
squares, products and higher powers, we get
This is the general formula for computing the error of a func!
observed that the right-hand of 2) is merely the total differential of th
_ For the relative error of the function y, we have
= 2. dy 81 | ay be,
Bitte bx: 5 tewagh
; halbMch
My = age then from (3),
Bon Bap kid
function y, we hav
n|®| 0 as n>, and henee if fx) is
approximated will be first terms, then the maximum error will be obtained by
remainder term R,,(x).
where, R,,(x)
i
If the accuracy required in specific in advance, it would be possible to find ny,
the number of terms such that the finite series yields the required accuracy.
SOLVED EXAMPLES
Example 1. If w= ux*y*/z4 and error in x,y,z be 0.001 compute the
maximum error in u when x =y =2=1,
Solution. Since, we know that
2
eiPSle ele ele ele e
Now
es.
Rie sin alysis and Numerical Com, f
Analysis Computations a
Example 2. If r
ntage error inh
Solution. Since, we known that
or. -
r= 81 = ano 34 [sao]
S3h(h° —2), find the percentage error in r at h=1, if the
21né
Now, relative error = =
r ~ ah(h®— 2)
Percentage error = | 100 x relative error |
Since the percentage error in h = 5%
ee
x 100
6
Percentage error in r = | 244° —6) _ ee 8
3(n8 — 2)
21-6 “
aan 5x5 | = 25%.
Example 3. Jn a AABC, 6cm, c= 15cm. ZB = 90°. Find the possible error
uted ZA if errors in measurements of a and c are 1mm and 2mm
tively
Solution. Since the AABC is a right angle triangle, with base c= 15 em. and
titude a = 6 em. Then,
tanA=2 = Actan'(f) €
e c
Now wa -4 ge
6 cm.
Op
4 15cm.
Fe 0] +
= 0.0108 radians.
Thus possible error in A = 0.0103 radians f
+ (58 2) [8a = 1mm, d= 2 mm]
2
Example 4, If V= 5 (e+ | and the error in V is at the most 0.4%, find the
A
Percentage error allowable in r and h sake + = al h=58em.
Solution. Since we have 8V =2 5" br =9 2V an a. sor Percentage error in
1((6.)
-$5f 43s]
este 0.0206.
(i) Percentage error in r= © x 109 = 208 te
r (av
aor
_ 100 _8v
i
2
50(5.8)
ae ae 6.1?
)
2h 100
are
=
oul
x 0.0206 = 0.22968%
(ii) Percentage error inh =
l= Im. Afthe error in the measurement of lis 0.01.
Here, Ts on
g of both sides, we get
log T=), a
‘ 8 2h +S log. reo, li
4
rors Analysts and Numerical Computations 5 a
Example 6. The error in the mea
0.1%. How accurately shou
Solutjon. Let r be the
easurement of the area of a circle is not allowed
Ud be the diameter be measured ?
dius of @ circle, then its area is
Aan?
> 8a = 2erbr
js By differentiation!
et d= 2r (diameter) be the circle, then (By differentiation]
oA =na8(2
{2
aa
io
? ADs 2
al
x(¢
()
reentage error in d = ad. 100
1
=3 oa x 100
1
= 5 (Percentage error in A)
a a (0.1) = 0.05%.
imal places.
Solution. The roots of the equation x” — 400x + 1=0
00 + ¥159996
wen by 4005159996
2
The smaller root = 400 = S1S8806 = SO0= EUSP — 0.0025.
Example 8, \29 = 5.385 and Vm = 3.317 correct to four significant figures, find
7g error in their sum and difference ?
Solution. y29 = 5.385 and Vit = 3.317 are correct to four significant so that the
ximum error in each case is given by
Maximum error = x10-? E ; %/10;", as,n-=inOnee daa rlae
= 0.0005.
Let xx, = \29 = 5.385, x2 = Vm = 3.317, then
xy = bx = 0.0005.
‘) 2 Relative error in their sum
fel=[e| +
0.0005 | ¢ | OOO sees
8.702 | 8. ants , 4
‘Thus the relative error in (xy +g) < 1.149 x 10
(ii) Relative error in (& ~ 2)
[Se] + |
Now X= xy - xp = 5.985 - 3.317 = 2.068.
Relative error in their difference
JAX] < | 0.0005) |, | 0.0005
x.
4
| Thus the relative error in (1, —x9) < 4.835 x 107 4, o%
Example 9. Find the number of terms of the exponential series such that th
sum gives the value of e* correct to six decimal places at x = 1.
Solution. We know that
o
where Ry) = "09, 0<02x 10-6,
10! = 3.6088 x 108
gives n = 10, and 9! = 3.6288 x 10 31
Hexadecimal to Octal
octal number, we first convert it into fees to convert hex:
equivalent to octal number, ry number and the
For example : Convert (B43:D21),6 to octal
Solution. Bm (1011)
4= (0100,
3= (oon)
D~ (4101)
2= | 0010
1= (0001;
(B43, Dig = (101101000011.1:
adecimal number to
n convert this binary
number system.
Daly 10100100001),
Making groups of S-bits binary equivalent, we got
(702701000014), =( 1101 | |101 | {0001 [o12 |),
= (5503),
(011010010001), =(0: 1110 | 100 | | 100 | {001 1)
= (06441),
(101101000011.110100100001), = (6503-6441),
Hence (B43.D21); = (6503-6441),
4.12. | FLOATING POINT ARITHMETIC
There are two types of arithmetic operations available in a computer.
i) Integer arithmetic. This arithmetic only deals with integer operands
which are used in counting and are used as subscripts.
(ii) Real or floating point arithmetic. This arithmetic uses the numbers
with fractional parts as operands and it is used in most computations.
Due to economic consideration the digital computers are designed such that its
memory has separate cells which are called ‘words’. Each word contains the same
number of binary digits, called ‘bits’ and having only a finite number of digits.
‘These number of digits which can be stored in a computer is known as its ‘word
length’
‘The numbers are stored in a computer in two forms :
(i) Fixed point.
(ii) Floating point. ‘
The fixed point mode is used to represent integers while the floating
mode is uded to represent real number.
An n-digit floating point number in base B has the form
x= (dy dy ...dy)p BY
where dy, do, ..., d, are all digits in the base § and all d’s lie between 0
“is an integer called the exponent and it is such that m se
