Download as pdf or txt
Download as pdf or txt
You are on page 1of 14

CE 205

Numerical Methods
Dr. Charisma Choudhury
Lecture 1
March 30, 2009

Objective
Some of the analysis methods you have
used so far..
Algebra
Calculus
Differential Equations etc.

Often not possible to determine analytical


solution
Complex calculation
Sheer size of problem

Why Numerical Methods?


Example:
No closed form solution
Length of one half of the curve y = sin(x):

1 + cos 2 ( x) dx

Large matrices

Computer is not so smart


Need to break the problem to simple mathematical problems
Add, subtract, multiply, divide, compare

Practical considerations
Example: Length of a board derived by solving equation:
x 3 - x 2 - 3x + 3=0
How accurately can we measure for practical applications?

Course Outline
Numerical solution of algebraic and
transcendental equations
Solution of systems of linear equations
Linear and non-linear curve-fitting by least
square regression
Finite differences
Divided differences
Interpolation
Numerical differentiation and integration
Numerical solution of differential equations

Course Outline
Numerical solution of algebraic and
transcendental equations
Solution of systems of linear equations
Linear and non-linear curve-fitting by least
square regression
Finite differences
Divided differences
Interpolation
Numerical differentiation and integration
Numerical solution of differential equations

Lecture Plan
Algebraic and transcendental
equations
Curve-fitting
Differential equations
CE applications and review

5 classes
3 classes
4 classes
1 class

References
Any standard undergraduate textbook on
Numerical Methods
Some examples:

Numerical Analysis: Goel & Mittal


Applied Numerical Analysis: Gerald & Wheatley
Numerical Methods for Engineers: Chapra & Canale
Introductory Methods of Numerical Analysis: Sastry

Grading Policy
3 quizzes
Tentatively on 5th, 9th and 12th class
Or maybe one computer programming
assignment?
The last one is only a make-up
quiz/assignment

Approach
Often multiple methods for solving same
problem
Which is the most relevant method?

Examine the problem


What inputs do we have?
What accuracy do we need?
What is the computational burden?
What is the rate of convergence?

Errors in Numerical Methods

Error in original data/measurement error


Truncation error: ex=1+ x/1! + x 2/2!+
Round-off error: 1/3, etc.
Calculations errors
significant digits

Errors
Abs error = |true-apprx|
Relative= abs error /true
5000+/-0.10
0.005+/-0.1

True=10/3, apprx = 3.333


Abs=1/3000
Rel=1/10000

Significant digits 4

Algebraic and Transcendental


Equations
Algebraic: y =ax+b
Transcendental: y = a sinx+b cosx

Solution Methods

Bisection/ Half-interval Search


Method of false position/Regula Falsi
Secant Method
Newton Raphson
Iteration Method
Many more

Choice of Method Depends on ..


Required accuracy
Rate of convergence
Inputs
How does initial approximation affect the
computation?

Often combination of multiple methods is


the optimum

Bisection/ Half-interval Search


Background:
Lessons from Graphical approach
Solve: f(x)=0
Let y=f(x)
Take a set of rectangular coordinates within a
range (say xL and xU)
Plot
Root(s): the point(s) where y crosses x

Background
Different variations (Examples shown
in next page)
Single root
Multiple roots
No root within specified range

1 root

no roots

3 roots

2 roots

Background
Observations
If f(xl) and f(xU) have opposite signs
Odd number of roots in between

If f(xl) and f(xU) have same signs


Zero/even number of roots in between

Exceptions
Discontinuous function

Discontinuous Function

Deductions
If f(x) is
continuous between xL and xU and
f(xl) * f(xU) <0
(i.e. f(xl) and f(xU) have opposite signs)
There is at least 1 real root f(x)=0 between
xL and xU

10

Example
Tangential functions
can have any number of roots even
irrespective of sign change/unchange

Deductions
If f(x) is
continuous and strictly monotonic between
xL and xU and
f(xl) * f(xU) <0
(i.e. f(xl) and f(xU) have opposite signs)
There is atmost 1 real root f(x)=0 between
xL and xU

11

Basis of Bisection Method


Narrow down interval [f(xL) and f(xU) ]to
locate where the sign change occurs
Divide into equal sub-intervals to search
the point where the sign change occurs

Steps
Choose xL and xL such that f(xL)*f(xU)<0
Initial estimate of root xR=(xL + xU)/2
f(xR)*f(xL)<0, root is in lower interval,
replace xU by xR
f(xR)*f(xL)>0, root is in upper interval,
replace xL by xR
Terminate when f(xR)=f(xL) or
f(xU)-f(xL)<tolerance

12

Example
f(x)=x3-x-1
x L =1,x U =2
x R=1.5

Solution: x=1.324
xL

xR
1.0000
1.0000
1.2500
1.2500
1.3125
1.3125
1.3125
1.3203
1.3242

xU
1.5000
1.2500
1.3750
1.3125
1.3438
1.3281
1.3203
1.3242
1.3262

2.0000
1.5000
1.5000
1.3750
1.3750
1.3438
1.3281
1.3281
1.3281

f(xL)
-ve
-ve
-ve
-ve
-ve
-ve
-ve
-ve
-ve

f(xR)
+ve
-ve
+ve
-ve
+ve
+ve
-ve
-ve
+ve

f(xU)
+ve
+ve
+ve
+ve
+ve
+ve
+ve
+ve
+ve

DifferenceRemark
1.0000 replace upper
0.5000 replace lower
0.2500 replace upper
0.1250 replace lower
0.0625 replace upper
0.0313 replace upper
0.0156 replace lower
0.0078 replace lower
0.0039 OK at 0.01 tolerance

Note:
OK at 0.01 tolerance. Not ok at 0.001 tolerance, need more iterations.

13

Practice Problems
Find roots of the following in bisection
method
a. e x = Sin(

x
2

b. xe x = 2

14

You might also like