MSA / L-1 CSE-4745 - Numerical Methods July 25, 2022

Introduction to Numerical Computing

Numerical method is an approach to find an approximate solution to complex

mathematical problems using only simple arithmetic operations namely addition,
subtraction, multiplication, division and functional evaluations.
 Numerical methods are usually employed for problem for which an analytical result
either cannot be obtained or would require too much work.
 Numerical analysis treats the subject matter of error analysis extensively. On the
other hand, in Numerical methods, focus is given more towards the techniques of
solving non-analytical problems rather than on the errors that accompany these
methods.
 Numerical computations invariably involve a large number of arithmetic calculations
and, therefore, require fast and efficient computing devices. The microelectronics
revolution and the subsequent development of high power, low cost, personal
computers have had a profound impact on the application of numerical computing
methods to solve scientific problems.
 The traditional numerical computing methods usually deals with the following topics:
 Finding roots of equations.
 Solving system of linear algebraic equations
 Interpolation & Regression Analysis
 Numerical Integration
 Numerical Differentiation
 Solving of Differential Equation
 Boundary value Problems
 Solution of matrix problems

Numeric Data
Numerical computing may involve two types of data: discrete data and continuous data.
 Data that are obtained by counting are called discrete data. Examples of discrete data
are total number of students in a class, total number of items in a box etc.
 Data that are obtained through measurement are called continuous data. Example of
continuous data is the temperature of a patient as measured by a thermometer.

Process of Numerical Computing

Numerical computing involves formulation of mathematical models of physical problems
that can be solved using basic arithmetic operations. The process of numerical computing
can be roughly divided into the following four phases which are illustrated in Fig-1.1
[Page-4, Balagurusamy]:
1. Formulation of a mathematical model
2. Construction of an appropriate numerical method
3. Implementation of the method to obtain a solution
4. Validation of the solution

Page 1 of 4
MSA / L-1 CSE-4745 - Numerical Methods July 25, 2022

 The formulation of a suitable mathematical model is critical to the solution of the

problem. A mathematical model can be broadly defined as a formulation of certain
mathematical equation that expresses the essential features of a physical system or
process. Models may range from a simple algebraic equation to a complex set of
differential equations. [Fig-1.2, Balagurusamy, Page-5]
 Once a mathematical model is available, our first step would be to try to obtain an
explicit analytical solution. In most cases, the mathematical models may not be
amenable to analytical solutions or they may not be solved efficiently using analytical
techniques. In such cases, we have to construct appropriate numerical methods to
solve mathematical models. For a given problem, there might be several alternative
numerical methods. We must consider different factors or trade-offs before selecting
a particular method- such as type of equation, type of computer available, accuracy,
speed of execution, and programming and maintenance efforts required.
 The third phase of the numerical computing process is the implementation of the
method selected. This phase is concerned with the following three tasks:
i. design of an algorithm
ii. writing of a program
iii. executing it on a computer to obtain the results.
 Once we are able to obtain the results, the next step is the validation of the process.
Validation means the verification of the results to see that it is within the desired
limits of the accuracy. If it is not, then we must go back and check each of the
following
i. mathematical model itself
ii. numerical method selected
iii. computational algorithm used to implement the method.
This may mean modification of the model, selection of an alternative numerical
method or improving the algorithm (or a combination of them). Once a modification
is introduced, the cycle begins again. [Fig-1.3, Balagurusamy, Page-7]

Characteristics of Numerical Computing

Numerical methods exhibit certain computational characteristic during their
implementation. It is important to consider these characteristics while choosing a
particular method for implementation. The characteristics that are critical to the success
of implementation are:
1. Accuracy
2. Rate of convergence
3. Numerical stability
4. Efficiency
Accuracy: Every method of numerical computing introduces errors. Different type of
errors like truncation errors or round off errors occurs which affect the accuracy of the
results. The results we obtain must be sufficiently accurate to serve the purpose for which
the mathematical model was built.
Rate of convergence: Many numerical methods are based on the idea of an iterative
process. This process involves generation of a sequence of approximations with the hope
that the process will converge to the required solution. Certain methods converge faster

Page 1 of 4
MSA / L-1 CSE-4745 - Numerical Methods July 25, 2022

than others. Some methods may not converge at all. It is, therefore, important to test for
convergence before a method is used.
Numerical Stability: Errors introduced into computation, from whatever source,
propagate in different ways. In some cases these errors tend to grow exponentially, with
disaster computational results. A computing process that exhibits such exponential error
growth is said to be numerically unstable. We must choose methods that are not only fast
but also stable.
 Numerical instability may also arise due to ill-conditioned problems. There are many
problems which are inherently sensitive to round off errors and other uncertainties.
Thus, we must distinguish between sensitivity of the methods and sensitivity inherent
in problems. When the problem is ill-conditioned, there is nothing we can do to make
a method to become numerically stable.
Efficiency: Efficiency in numerical method means the amount of effort required by both
human and computer to implement the method. A method that requires less of computing
time and less of programming effort and yet achieves the desired accuracy is always
preferred.

New trends in Numerical computing

In recent years, the increasing power of computer hardware has affected the approach of
numerical computing in several ways. It has forced scientists and engineers to search for
algorithms that computationally fast and efficient. An important new trend is the
construction of algorithms to take advantage of specialized computer hardware such as
vector computers and parallel computers. Another trend is the use of sophisticated
interactive graphics, in which the user can view the results graphically and advice the
computer, graphically, on how to proceed further. Object oriented numerical computing
is gaining importance due to the popularity of languages like C++ and java.

Page 1 of 4
MSA / L-1 CSE-4745 - Numerical Methods July 25, 2022

Representation of Numbers

All modern computers are designed to use binary digits to represent numbers and other
information. The memory is usually organized into strings of bits called words.
The largest number a computer can store depends on its word length. For example, the
largest binary number a 16 bit word can hold is 16 bits of 1. This binary number is
equivalent to a decimal value of 65535. The following relation gives the largest decimal
number that can be stores in a computer:
Largest number = 2n – 1
where n is the word length in bits. Thus, we see that the greater number of bits, the larger
the number that may be stored.

Integer representation
Decimal numbers are first converted into the binary equivalent and then represented in
either integer or floating-point form.
For integers, the decimal or binary point is always fixed to the right of the least
significant digit and therefore, fraction are not included. The magnitude of the number is
restricted to 2n –1 where n is the word length in bits.
Again, negative numbers are stored by using the 2’s complement. This is achieved by
taking the 1’s complement of the binary representation of the positive number and then
adding 1 to it. [Example 3.8, Balagurusamy, Page-50]
If we reserve one bit to represent the sign of the number, called sign bit, we have only n-1
bits to represent the number. Thus a 16bit word can contain numbers –2 15 to 215 – 1
(i,e. –32768 to 32767). Generally, if the sign bit is 1, the number is negative and if the
sign bit is 0, the number is positive.

Floating point Representation

Floating-point numbers are stored and processed differently. The entire memory location
is divided into three fields called sign , exponent and mantissa as shown in Fig 3.1 [page-
51, Balagurusamy]. Typically, floating numbers use a field width of 32 bits where 24 bits
are used for the mantissa, 7 bits for the exponent and 1 bit for sign.
For example, 35.7812 can be expressed 0.357812 * 10 2. Similarly, the number
987654321 can be expressed as 0.987654 * 10 9. By writing a large number in exponential
form, we may lose some digits. If x is a real number, its floating-point form
representation is x = f * 10E where the number f is called mantissa and E is the exponent.
Example- 3.10: Page 52, Balagurusamy.
 The shifting of decimal points to the left of the most significant digit is called
normalization and the numbers represented in normalized form are known as
normalized floating point numbers.
 Mantissa should satisfy the following conditions:
For positive numbers: less than 1.0 but greater than or equal to 0.1.
For negative numbers: greater than -1.0 but less than or equal to -0.1.
That is 0.1  | f | < 1.

Page 1 of 4