9/11/2012
WEEK 1
Introduction to Numerical Methods
Mathematical modeling
Approximation and round off errors
Truncation errors and Taylor Series
LESSON OUTCOMES
At the end of this topic, the students will be able:
To
describe
numerical
techniques
as
compared to analytical methods
To use Taylor series expansion to approximate
a function
To perform error analysis associated with
numerical methods
9/11/2012
Why Numerical Method?
Could handle large systems of equations,
nonlinearity and complex geometries that is not
common
It provide approximate solutions to many of the
engineering problems.
What is Numerical Method
Powerful analysis tool in problem solving and
understanding problem in mathematical language
Techniques by which mathematical problems are
formulated, so that they can be solved with
arithmetic operations
The role of numerical method in solving engineering
problem:
PROBLEM
FORMULATION
Fundamental laws are used to
develop mathematical
equations that can represent
the specific problem
SOLUTION
Suitable numerical methods
are then selected to solve
the mathematical equations
INTERPRETATION
The results obtained can
then be used to
predict/analyze/understand
the specific problem better
5
Mathematical modeling
Is the use of mathematics to
Describe real world phenomena
Investigate important questions about the observed world
Explain real world phenomena
Test ideas
Make predictions about the real world
The real world refers to
Engineering
Physiology
Wildlife management
Economics
Etc
Physics
Ecology
Chemistry
Sports
9/11/2012
A mathematical model is represented as a functional
relationship of the form
Dependent variable
Observed behaviour/state/phenomenon of a system
Characteristic that reflects behaviour or state of the system
i.e. y, f(x), f(t)
Independent variable
Dimension that determine a system i.e. time, t , x
Parameter
Quantity that serves to relate to functions and variables
Reflective of the systems properties or composition
Forcing functions
External influence that acts on system i.e. acceleration gravity, g
8
Example
Apply Newtons second law,
And also can write as
F = ma
dv
m F
dt
or
m
d 2x
F
dt 2
Eq. relates a linear position x
(dependent variable) to the applied
force, F (forcing function) and the
time, t (independent variable). The
mass, m is the only parameter in
the above model.
9
9/11/2012
Example of mathematical modeling
10
Example
Assume that interested to predict the velocity of the falling parachutist with time
Use fundamental knowledge to find a
mathematical equation correlates the velocity
to the various forces acting on the parachutist
Newtons 2nd law of Motion
the time rate change of momentum of a
body is equal to the resulting force acting on
it.
The model is formulated as
F = ma
F=net force acting on the body (N)
m=mass of the object (kg)
a=its acceleration (m/s2)
11
acceleration a
dv
dt
dv
dt
F FD FU
FD downwardforceof gravity
FU upward forceof air resistance
FD mg
FU cv
Model relates acceleration of falling
object to the forces acting on it,
dv
dt
mg cv
(differential equation)
12
9/11/2012
Exact or analytical solution: it exactly satisfies the original equation
dependent variable
forcing functions
t,s
v,(m/s)
0.00
16.40
27.77
35.64
41.10
10
44.87
12
47.87
53.39
parameter
independent variable
Analytical solution of the parachutist
13
Unfortunately, there are many mathematical
models that cannot be solved exactly.
Numerical solution that approximates the exact solution
14
dv
mg cv
dt
dv
dt
v( t
i 1 )
v( t
v( t )
i
ti 1 ti
m
i 1 )
v( t )
i
ti 1 ti
g
c
m
v(t )
i
v(ti1 ) v(ti ) g v(ti ) (ti 1 ti )
m
The use of finite difference to approximate the
first derivative of v with respect to t
Approximate or
numerical solution
15
9/11/2012
Numerical solution
t,s
v,(m/s)
0.00
19.60
32.00
39.85
44.82
10
47.97
12
49.96
53.39
v(ti1 ) v(ti ) g v(ti ) (ti 1 ti )
m
Comparison between the exact and
numerical solution
16
Approximation and Roundoff Errors
Significant figures
Numbers to be used in confidence
98
2 significant figures
98.09
4 significant figures
0.0098
2 significant figures
17
Important of significance figures in numerical methods:
Numerical methods yield approximate results, therefore,
need to develop criteria to specify the confident in
approximate result.
Although quantities such as , e, or 7 represent specific
quantities, they cannot be expressed exactly by a limited
number of digits. Computers retain only a finite number of
significant figures.
= 3.141592653589793238462643
18
9/11/2012
Accuracy and Precision
Increasing precision
Increasing accuracy
How closely individual
computed or measured
values agree with each
other
How closely a computed
or measured value agree
with the true value
Inaccurate & imprecise
accurate & imprecise
Inaccurate & precise
accurate & precise
19
Error definitions
Error
Truncation errors ~ result
when approximations are
used to represent exact
mathematical procedures
Round-off errors ~ result
when numbers having a
limited significant figures
are used to represent
exact numbers
True value
Approximation
value
True value = Error + Approximation value
20
True value = error + approximation value
(1)
True error (Et)= true value approximation
(2)
(3)
t designates true percent relative error
21
9/11/2012
Calculation of errors
True value of length of a bridge is 10,000 cm.
When you measure, the length recorded is
9,999 cm. Compute the true error and true
percentage relative error of the bridge.
22
Answer: 1 cm and 0.01%
However, in actual situation, true value is rarely available.
Therefore, need to estimate the true value approximation
(4)
In numerical method, iterative approach is used to
compute answer, in which error is estimated as the
difference between previous and current approximations.
(5)
The signs of error can be negative or positive,
Absolute value of error, |a| need to be lower than
prespecified percent tolerance, s
n is significant figures.
23
Error estimates for iterative method
Suppose that we have exponential function as,
Starting with the simplest version, ex=1, add
terms to estimate e0.5. Compute true (t) and
approximate error (a) after each term is added
until |a| falls below s , conforming to 3
significant figures. Note that true value of e0.5
is 1.648721.
24
9/11/2012
Answer
25
Round-off errors
Results when numbers having limited significant
figures are used to represent exact numbers
ln 2 = 0.693 147 180 559 945 309 41...
A device only shows 8
significant numbers, so
round-off error is discrepancy
introduced by omission of
significant figures.
Round-off error for this case is
0.00000000055994530941
26
Other example of roundoff error
27
9/11/2012
Truncation errors and Taylor series
Truncation errors
Truncation error is the discrepancy introduced by the fact that
numerical methods may employ approximations to represent exact
mathematical operations and quantities.
Truncation error are errors resulted from using an approximation
in place of an exact mathematical procedure.
The difference between the calculated value using exact
mathematical equation and approximation mathematical equation.
28
Taylor series
Provides a means to predict a function value at one point in terms of the
function value and its derivative at another point
Zero order
First order
Second
order
nth order
29
Taylor series by defining a step size h = xi+1 - xi
(6)
Rn
f ( n 1) ( ) n 1
h
(7)
(n 1)!
30
10
9/11/2012
31
Remainder for the Taylor series Expansion
Remainder term, Rn, accounts for all terms from (n+1) to infinity.
f ( n 1) ( ) n 1
h
(n 1)!
Rn
It also usually expressed as:
Rn O(h n1 )
Let say we truncated the Taylor series expansion after zero-order term to
yield
f ( xi1 ) f ( xi )
Remainder term, Rn for zero order version
R0 f '( xi ) h
f ' '( xi )
2!
h2
f 3( xi ) 3
h ...
3!
Let truncate the remainder itself,
R0 f '( xi ) h
This result is still inexact because neglected second and higher order terms.
32
Alternative simplification that tranforms the approximation into an
equivalence based on graphical insight
derivative mean-value theorem states that if a function f(x) and its
derivative are continous over interval from xi to xi+1,
there exist at least one point on the function that has a slope, designated by
f(), parallel to line joining f (xi) and f(xi+1)
Thus,
So,
Zero order version
First order version
33
11
9/11/2012
Numerical
differentiation
Forward finite divided difference
Backward finite divided difference
Centered finite divided difference
34
Forward finite divided difference approximation of first derivative
(14)
Where,
(15)
Backward finite divided difference approximation of first derivative
f(xi ) f(xi 1 ) R1
h
h
f
f'(xi ) i O(h)
h
f'(xi )
Where,
(16)
h xi xi 1
Centered finite divided difference approximation of first derivative
(17)
Where,
h xi 1 xi xi xi 1
35
Example
Use forward and backward difference approximations of O(h) and a centered
difference approximation of O(h2) to estimate the first derivative of
f(xi ) 0.1x 4 0.15x3 0.5x 2 0.25x 1.2
at x = 0.5 using a step size h = 0.5. Repeat using h = 0.25. Also calculate the
true percent relative error for each approximation.
36
12
9/11/2012
Exercise
Use forward and backward and a centered difference to estimate the first
derivative of the function
f(xi ) 0.5x3 0.1x 2 0.8x 0.7
at x = 0.5 using a step size h=0.5. Repeat using h = 0.25. Also calculate the
true percent relative error for each approximation.
Ans:
h=0.5
FDM:1.525
BDM:0.875
CDM:1.200
h=0.25
FDM:1.26875
BDM:0.94375
CDM:1.10625
41.9%
18.60%
11.63%
18.02%
12.21%
2.91%
37
Higher derivatives
Second forward finite difference approximation of higher derivatives
Second backward finite difference approximation of higher derivatives
Second centred finite difference approximation of higher derivatives
38
Error propagation
This section is to study how errors in numbers can
propagate through mathematical functions. If we multiply
two numbers that have errors, we would like to estimate
the error in the product.
If a function f is dependent on
(a) a single independent variable x : f(x)
(b) two independent variables x and y : f(x, y)
(c) several independent variables x1, x2, x3, ... ,xn : f(x1 , x2 ,..., xn ).
39
13
9/11/2012
Function of a single variable.
Let
x be the true value and
x* be an approximate value of x
Then, TSE for f(x) computed near f(x*) is given by
f(x) f(x*) f'(x * )( x x*)
f ' ' ( x*)
( x x*)2 ...
2
Truncating after the first derivative term and rearranging the remaining terms
to give
f(x) f(x*) f ' (x*)( x x*)
f(x*) f ' (x*) x *
(21)
where
f ' (x*) f ( x) f ( x*) is an estimate of the error of the function
x* x x * is an estimate of the error of independent variable x
Eq.(21) provides 2 capabilities:
1. to approximate the error in f(x) knowing its derivative.
2. to approximate the error in the independent variable x.
40
Example
Given a value of x* = 2.5 with an error of x* = 0.01, estimate the
resulting error in the function, f(x)=x3.
Solution
f(x*) f ' (x*) x *
So,
f(x*) 3(2.5) 2 (0.01) 0.1875
Because f (2.5) 15.625, it can predict that
f( 2.5 ) 15.625 0.1875
Or the true value lies between 15.4375 and 15.8125. In fact, if x ~2.49,
f(x) could be 15.4382 and if x ~ 2.51, it would be 15.8132.
The first order error analysis provides a fairly close estimate of the true error.
41
Exercise
Knowing a value of x* = 2.0 with an error of x* = 0.01, estimate the resulting
error in the function
f(x) = 0.5x3-0.1x2+0.8x-0.7
Ans: f(2.0)=4.5 0.064
42
14
9/11/2012
Function of a more than One variable.
(22)
(23)
Refer section 4.2.2
for examples
43
Relative error
Refer section 4.2.3
for example
Condition number
=
=
44
Condition no equals 1 indicates that functions
relative error is identical to the relative error in
x
Condition no greater than 1 indicates relative
error is amplified.
Condition no less than 1 indicates relative
error is attenuated.
Function with very large values are said to be
ill-conditioned.
45
15
9/11/2012
Total Numerical Error
Total numerical errors = truncation error + round off error
Roundoff error by increase no. of significant figures or
reduce no. of computation in analysis
Truncation error by decreasing step size (h) or increase
no. of computation in analysis
46
Control numerical error
avoid subtract 2 nearly equal numbers to
avoid loss of significance
Use Taylor series for truncation and roundoff
error analysis
Perform numerical experiments
- repeat computation with different step size or method
and compare results
47
16
