Error in Numerical Analysis

Definition 1 (Scientific Notation). Let k be a real number, then k can be written in the following
form
= × 10
where is any real number and the exponent is an integer. This notation is called the scientific
notation or scientific form and sometimes referred to as standard form.
Example 1. Write the following numbers in scientific notation:
1. 0.00000834.
2. 25.45879.
3. 3400000
4. 33
5. 2, 300, 000, 000.

Solution:
1. 0.00000834 = 8.34 × 10−6
2. 25.45879 = 2.545879 × 101
3. 3400000 = 3.4 × 106
4. 33 = 3.3 × 101
5. 2.3 × 109

Floating Point Number:

In the decimal system any real number ≠ 0 can be written in the decimal normalized floating-
point form in the following way
= ±0. ··· · · · × 10 , 1 ≤ ≤ 9, 0 ≤ ≤ 9, for each = 2,· · ·
, and is an integer called the exponent (n can be positive, negative or zero).

In computers we use a finite number of digits in representing the numbers and we obtain the
following form
= ±0. ··· · · · × 10 , 1 ≤ ≤ 9, 0 ≤ ≤ 9,
for each = 2,· · · , . These numbers are called k-digit decimal machine numbers.

Also, the normalized floating-point decimal representation of the number ≠ 0 can be written
in other way as
1
= ± × 10 , ( ≤ < 1),
10
the number is called the normalised mantissa.
The floating-point representation in binary number system can be defined by the same way as in
the decimal number system. If ≠ 0, it can be represented as
= ± × 2 ,( ≤ < 1), ℎ = (0. · · ·) , = 1.
Errors
In the area of scientific computing, errors will always happen. Instead, numerical analysts try to
investigate the possible and best ways to minimize the error. The main goals of error analysis are
to study the error and figure out how to predict it and reduce it as much as possible

Error Analysis
In numerical analysis, we use the numerical method to get close to the accurate answer to the
problem, which means we make an error. The difference between the exact solution and the
approximate solution is the numerical error.

Definition 2 (Numerical Error).

Let x be the exact solution of the underlying problem and x∗ its approximate solution, then the
error (denoted by e) in solving this problem is = − ∗

Sources of Error in Numerical Computations

 Blunders (Gross Errors) These errors, which can also be called "human errors," are
caused by errors and inaccuracies made by people. They can be kept to a minimum by
being careful during scientific studies. Such errors will add to the overall error of the
problem and can have a big effect on how accurate the answer is.
 Modelling Errors During the modeling process, such errors happen when scientists
ignore factors that affect the model in order to make the problem easier to understand.
These errors are also called formulation flaws.
 Data Uncertainty These errors are caused by the fact that the details about the physical
problem aren't always clear. They are also called "data errors."
 Discretisation Errors Computers represent a function of continuous variable by a number
of discrete values. Also, scientists approximate and replace complex continuous problems
by discrete ones and this results in discretisation errors.

Absolute and Relative Errors

Definition 3 (Absolute Error). The absolute error of the error is defined as the absolute
value of the error ; = | − ∗|

Definition 4 (Relative Error). The relative error of the error e is defined as the ratio between
| ∗|
the absolute error and the absolute value of the exact solution = | |= | | , ≠0

Example 2. Let x = 3.141592653589793 is the value of the constant ratio π correct to 15

decimal places and x∗ = 3.14159265 be an approximation of x.
Compute the following quantities:
a. The error.
b. The absolute error.
c. The relative error.
Solution:
a. The error
= – ∗ = 3.141592653589793 − 3.14159265 = 3.589792907376932 − 09
= 3.589792907376932 × 10 = 0.000000003589792907376932.

b. The absolute error

= | − ∗ |= |3.141592653589793 − 3.14159265| = 3.589792907376932 − 09.

c. The relative error

− ∗ | 3.141592653589793 − 3.14159265
|
= = =
| | | | 3.141592653589793
3.589792907376932 − 09
= 1.142666571770530e − 09
3.141592653589793

Roundoff Errors
Computers can only represent numbers with a certain number of digits, so some numbers can't
be represented accurately. The roundoff error is what happens when you replace a number with
its closest machine number, and the process of doing this is called "correct rounding."

Example 3. Approximate the following decimal numbers to three digits by using rounding
rules:
x1 = 1.34579.
Solution: x1 = 1.35.

Truncation Error
It's much harder to figure out what went wrong with truncation, discretization, or
approximation mistakes. We'll look at them as part of solving differential equations. Taking a
small number of terms from a Taylor series can lead to many standard numerical methods,
such as the trapezoidal rule for quadrature, Euler's method for differential equations, and
Newton's method for nonlinear equations. The truncation error is the result of the making up
of the terms that were left out. For many methods, the size of this error relies on a parameter
(often called the step size), whose right value is a compromise between getting a small error
and doing the calculation quickly. For a lot of scientific problems, these close estimates are
needed.

Uncertainty
Uncertainty in the data is always a possibility when we are solving real world problems. It can
be caused by errors in measuring real quantities, errors in storing the data on the computer
(rounding errors), or errors in an earlier calculation if the data is the answer to another problem.

Accuracy
Accuracy is the ability of an instrument to measure the correct value. In other words, it is the
closeness of the measured value to a standard or true value. Accuracy is the degree of closeness
between a measurement and its true value.
 Example:
Consider a thermometer having the scale range up to 500ºC. The thermometer has an accuracy
of ±0.5% (percent) of scale range i.e. 0.005 x 500 = ± 2.5 ºC. Therefore, the reading will have
a maximum error of ± 2.5 ºC.

Precision
The precision of something is how close two or more measures are to each other. If you weigh
the same thing five times and get the same result each time—3.2 kg—then your measurement
is very accurate, but not always accurate. Accuracy has nothing to do with precision. Precision
is how many times you can take the same readings and get the same results. It shows how close
two or more values are to one another.

Example: If the weather temperature reads 28 °C outside and it is 28 °C outside, then the
measurement is said to be accurate. If the thermometer continuously registers the same
temperature for several days, the measurement is also precise.

Note that the accuracy and precision are different and they are not related. The problem maybe
very accurate but imprecise and vice versa.

Stable and Unstable Computations: Conditioning

Stability is one of the most important characteristics in any efficient and robust numerical scheme.

Definition 5 (Numerical Stability). The numerical algorithm is said to be stable if the final result
isn't too affected by changes that happen during the process of computing. If not, it is said to be
unstable. The idea of stability is similar to and closely linked to the idea of conditioning.

Definition 6 (Conditioning). Conditioning is a measure of how sensitive the output to small

changes in the input data. In literature conditioning is also called sensitivity.
• The problem is called well-conditioned or insensitive if small changes in the input data
lead to small changes in the output data.
• The problem is called ill-conditioned or sensitive if small changes in the input data lead to
big changes in the output data.

Definition 7 (Condition Number of a Function). If is a differentiable function at in its domain

( )
then the condition number of f at is ( ( )) = | ( )|
, ( )≠0
Note: Condition number of a function at in its domain sometimes denoted by ( ).

Definition 8 (Condition Number of a Matrix). If A is a non-singular × matrix, the condition

‖ ‖
number of is defined by ( ) = ‖ ‖‖ ‖ , where ‖ ‖ = max ≠ 0 ‖ ‖ is a m ×
1 column vector

Remark 1. Note that:

1. The problem is ill-posed or sensitive if ≫ 1.
2. The problem is well-posed or insensitive if < 1.
Example 4. Find the condition number of the function ( ) = √

Solution:
( ) = √ ⇒ ( )= , ≠0
√
( ) √
( ( )) = | ( )|
,= = |
√
This indicates that the small changes in the input date lead to changes in the output data of half
size the changes in the input data.

Definition 9 (Well-Posed Problem). The problem is well-posed if satisfies the following three
conditions:
a. The solution exists.
b. The solution is unique.
c. The solution depends continuously on problem data.
Otherwise, the problem is called ill-posed.

Ref: (Sabawi, n.d.)

https://cedw.tu.edu.iq/images/Departments/math/mohammed_abd/Errors_Analysis_and_Basic_
Definitions_in_Numerical_Analysis.pdf