ME – 361 NUMERICAL METHODS FOR ENGINEERS

Numerical Integration

Given a complicated f(x) or a set of tabulated values, xi’s and corresponding f(xi)’s:

b
Question: For a given range, [a,b]  f (x) dx = ?
a

Answer: Replace f (x) or the tabulated data with a simple function g(x) and
operate on g(x), instead

b b
f (x)  g (x) ,  f (x) dx
a
  g(x) dx
a

Prof. Dr. Faruk Arınç Fall 2020

 ME – 361 NUMERICAL METHODS FOR ENGINEERS

x i+1 x i+1

 f(x) dx
xi
  P (x) dx
xi
1

h
 h f(x i ) + f(x i+1 ) - f(x i ) 
2
h
 f(x i ) + f(x i+1 ) 
2
h
 f i + f i+1 
2

0
Trapezoidal Rule

Δx = h
h
 f 0  2 f1  2 f 2  ...  2 f n-1  f n 
xn
x
0
f (x) d x 
2
Prof. Dr. Faruk Arınç Fall 2020
 ME – 361 NUMERICAL METHODS FOR ENGINEERS

Second-degree
Polynomials, P2(x) x2 x2

 f(x) dx
x0
  P (x) dx
x0
2

s (s - 1) 2
P2 (x) = f0 + s Δf0 + Δ f0
2!

x - x0
s=
h

Simpson's One-third Rule

h
f 0  4 f1  2 f 2  ...  2 f n-2  4 f n-1  f n 
xn

x0
f (x) d x 
3
Prof. Dr. Faruk Arınç Fall 2020
 ME – 361 NUMERICAL METHODS FOR ENGINEERS

Newton - Cotes Formulae

Trapezoidal Rule (set of first-degree polynomials)

h - (x n - x 0 )3 ' '
 f  2 f1  2 f 2  ...  2 f n-1  f n 
xn
x 0
f (x) d x 
2 0
E trap 
12 n 2
f ( )

Simpson's One-third Rule (set of second-degree polynomials)

f (x) d x  f 0  4 f1  2 f 2  ...  2 f n -2  4 f n -1  f n 
xn h - (x n - x 0 )5 (4)
x0 3
E1/3 
180 n 4 f ( )

Simpson's Three-eights Rule (set of third-degree polynomials)

3h
 f0  3 f1  3 f 2  2 f3  ...  2 f n-3  3 f n-2  3 f n-1  f n 
xn
x 0
f (x) d x 
8

Prof. Dr. Faruk Arınç Fall 2020

 ME – 361 NUMERICAL METHODS FOR ENGINEERS

Thomas Simpson FRS

Fellow of the Royal Society

1710 - 1761

Prof. Dr. Faruk Arınç Fall 2020

 ME – 361 NUMERICAL METHODS FOR ENGINEERS

Difference Table for f (x) = e-x

x f (x) Δf Δ2 f Δ3 f Δ4 f
0 1
- 0.632120559
1 0.367879441 0.399576401
- 0.232544158 - 0.252580458
2 0.135335283 0.146995943 0.1596613
- 0.085548215 - 0.092919158
3 0.049787068 0.054076785 0.05873611
- 0.031471429 - 0.034183048
4 0.018315639 0.019893738 0.021607807
- 0.011577692 - 0.012575241
5 0.006737947 0.007318497
- 0.004259195
6 0.002478752
Prof. Dr. Faruk Arınç Fall 2020
 ME – 361 NUMERICAL METHODS FOR ENGINEERS

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
0 1 2 3 4 5 6
Prof. Dr. Faruk Arınç Fall 2020
 ME – 361 NUMERICAL METHODS FOR ENGINEERS

6 6
Example: Find the integral 
0

I = e  x dx = f(x) dx
0
using forward-difference table

I True Relative Error

Exact 0.997521
Trap Rule 1.07929 8%

Simpson’s 1/3 Rule 1.00247 0.5 %

Simpson’s 3/8 Rule 1.00757 1%

Note that Simpson’s 1/3 rule requires odd number of data points (even number of
panels), multiples of three, and that Simpson’s 3/8 rule requires even number of
data points (odd number of panels), multiples of four. Otherwise, use a combination
of the rules.
Prof. Dr. Faruk Arınç Fall 2020
 ME – 361 NUMERICAL METHODS FOR ENGINEERS

Richardson’s Extrapolation

Trap-rule formula:

b - a b - a    b - a  ''
3
1 1 n 1

 2 f a  2 f b   f  a + n i   - 12 n 2 f ( )
b
I=  a
f(x) dx 
n  i 1  

Note that the error is roughly proportional to n2, where n is the number of panels.

Let’s denote the true integral by I*. Find I with trap rule using first with n1 number of
panels and then with n2 number of panels

b - a  b - a 
3 3
*
I = I n1 + E n1 = I n1 - f (1 )
''
I* = I n2 + E n2 = I n2 - f '' ( 2 )
12 n12 12 n 22

Prof. Dr. Faruk Arınç Fall 2020

 ME – 361 NUMERICAL METHODS FOR ENGINEERS

I* - I n2 = E n2 I* - I n2
 
E n2 * *
E n2
=> *
= => I - I n2 = I - I n1
*
I - I n1 = E n1 I - I n1 E n1 E n1

b - a 
3

- f '' ( 2 ) 2 2
Take the ratio of E n2 12 n 2
 n1  f '' ( 2 )  n1 
= 2
=     
 b - a  '' f (1 )
3 ''
two errors. E n1  n2   n2 
- 2
f (1 )
12 n1

2
 n1 
 
I n2 - I n1
I* - I n2  I* - I n1   Solve for I*: I  I n1 +
*
2
 n2  n 
1-  1
 n2 

Prof. Dr. Faruk Arınç Fall 2020

 ME – 361 NUMERICAL METHODS FOR ENGINEERS

3 3

 x  
3
Example: Find the integral I= - 2 x 2 + 7 x - 5 dx = f(x) dx
1 1

2
Exact solution: I = 20
3

Richardson’s extrapolation using n1 = 2 and n2 = 4:

3 - 1 1 1  1 25 
In1 =  f(1) + f(3) + f(2)  =  + + 9  = 22
2 2 2  2 2 

3 - 1 1 1 
In1 = f(1) + f(3) + f(1.5) + f(2) + f(2.5)  = 21
4  2 2 

21 - 22 2
I*  22  2
= 20 Why did we get the exact answer?
2 3
1-  
4

Prof. Dr. Faruk Arınç Fall 2020

 ME – 361 NUMERICAL METHODS FOR ENGINEERS

Romberg Integration Table

1  b-a 2N - 1
 b-a 
TN , 1   N-1 , 1
2 
T 
2 N-1

i=1
f a + N
 2
i

, i = 2

4 j - 1 TN+1 , j-1 - TN , j-1

TN , j = j-1
, j = 2, 3, ...
4 -1

where T N , j : Estimate of an integral

n = 2 N panels

Prof. Dr. Faruk Arınç Fall 2020

 ME – 361 NUMERICAL METHODS FOR ENGINEERS

1
n
Example: I =  e- x dx N j=1 j=2 j=3
(Panels)
0

Exact I = 0.63212 1 0 0.6839 0.6323 0.6321

2 1 0.6452 0.6321 0.6321

4 2 0.6354 0.6321 0.6321
8 3 0.6329 0.6321
16 4 0.6323

Accuracy T-T  

Prof. Dr. Faruk Arınç Fall 2020

 ME – 361 NUMERICAL METHODS FOR ENGINEERS

Monte Carlo Integration

Example:

The error function is defined by

x
2
e
t 2
Erf(x) = dt
 0

The integral must be obtained, numerically. It is given in tabular form for different
values of x, from which Erf(1) = 0.84270

• Use Monte-Carlo method to evaluate the integral. Choose N = 100 and N =

500, and compare with the true value.

• Use Gauss-Legendre quadrature method to evaluate the same integral.

Choose N = 3 and N = 6, and compare with the true.

Prof. Dr. Faruk Arınç Fall 2020

 ME – 361 NUMERICAL METHODS FOR ENGINEERS

1.10

1.00

0.90

N =100
0.80

I = 0.73
0.70

0.60

0.50

0.40
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

Prof. Dr. Faruk Arınç Fall 2020

 ME – 361 NUMERICAL METHODS FOR ENGINEERS

1.10

1.00

0.90

0.80 N = 500

0.70
I = 0.81

0.60

0.50

0.40
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

Prof. Dr. Faruk Arınç Fall 2020

 ME – 510 NUMERICAL METHODS FOR ME II

Gauss(ian) Quadrature Methods

b n
Newton-Cotes Formulae: I   f (x) dx  w
i0
i f (x i )
a

The range, (a,b), is divided into equal spacings (panels).

A polynomial of small degree passes through these equally-spaced points.

Why equal spacing?

Does the function has to be a polynomial?
b N b

Error in Newton-Cotes Formulae: I =  f(x) dx = w i 1

i f(x i ) +  g(x) h(x) dx
a a

Exact Approx. How to minimize?

Prof. Dr. Faruk Arınç Fall 2020
 ME – 361 NUMERICAL METHODS FOR ENGINEERS

Gauss(ian) Quadrature Methods

b n
I   f (x) dx   wi f (xi )
a i0

n + 1 values of wi There are 2 n + 2 parameters which define a

=>
n + 1 values of f (xi) polynomial of degree 2 n + 1

n b
Choose xi’s such that, the sum  w i f (x i ) gives the integral  f (x) dx
i0 a

exactly when f (x) is already a polynomial of degree 2 n + 1.

Prof. Dr. Faruk Arınç Fall 2020

 ME – 361 NUMERICAL METHODS FOR ENGINEERS

Johann Carl Friedrich Gauss

1777 – 1855

Prof. Dr. Faruk Arınç Fall 2020

 ME – 361 NUMERICAL METHODS FOR ENGINEERS

Example f(x) = 3 x 2 + 2 x + 1

4 4
Iex =  f(x) dx = 81 Itrap =  f(x) dx = 81.375 with h = 0.5 => 6 panels
1 1

60

50

40

30

20

10

0
Prof. Dr. Faruk Arınç
1.0 1.5 2.0 2.5 3.0 3.5 4.0 Fall 2020
 ME – 361 NUMERICAL METHODS FOR ENGINEERS

60

50

40
f(x) = 3 x 2 + 2 x + 1
30

f(x2)
20

10
f(x1)
0
1.0 1.5 2.0 2.5 3.0 3.5 4.0

4 N
Iex =  f(x) dx = 81 I= w
i 1
i f(x i ) = Iex N=2 xi = ? wi = ?
1
Nodes and weights

Prof. Dr. Faruk Arınç Fall 2020

 ME – 361 NUMERICAL METHODS FOR ENGINEERS

Infinite Set of Orthogonal Functions

Two functions, f(x) and g(x), both real-valued and continous, are called orthogonal
to each other with respect to a third (weighting) function, w(x), in the range [a,b] if
b
I   w(x) f(x) g(x) dx
a
 0

1
Examples:  (1) sin(n  x) sin(m  x) dx
0
 0 when n  m

{sin(n x) , n = 1, 2, ... }

1 Fourier
 (1) cos(n  x) cos(m  x) dx
0
 0 when n  m Series

{cos(n x) , n = 0, 1, ... }
Prof. Dr. Faruk Arınç Fall 2020
 ME – 361 NUMERICAL METHODS FOR ENGINEERS

Jean-Baptiste Joseph Fourier

French Mathematician

1768 -1830

Prof. Dr. Faruk Arınç Fall 2020

 ME – 361 NUMERICAL METHODS FOR ENGINEERS

Expansion of f(x) in terms of infinite set of orthogonal functions

Suppose the functions, Qn(x), forms an orthogonal set, n -> inf.

f(x) = c 0 Q0 (x) + c1 Q1(x) + c 2 Q 2 (x) + ... + cn Qn (x) + ....

 w(x) Q (x) f(x) dx =

a
n

 w(x) Q (x)  c
a
n 0 Q0 (x) + c1 Q1(x) + c 2 Q2 (x) + ... + c n Qn (x) + ....  dx

=  w(x) Qn (x)  c n Qn (x)  dx

a

 w(x) Q (x) f(x) dx

a
n

cn = b

 w(x) Q (x) dx
2
n
Prof. Dr. Faruk Arınç a Fall 2020
 ME – 361 NUMERICAL METHODS FOR ENGINEERS

Example

Expand the function f(x) = 1 in Fourier sine series in the range [0,1], i.e., find the
coefficients, an’s, in the orthogonal expansion

f(x) = a
0
n sin(n x)

The orthogonality property:

1
1 2
 sin  n x  dx = - [ cos n x  0 ] =
1
if n is odd
0
n n
=0 if n is even

Prof. Dr. Faruk Arınç Fall 2020

 ME – 361 NUMERICAL METHODS FOR ENGINEERS

4 4 4
Solution: f(x) = 1 = sin( x) + sin(2 x) + sin(3 x) + ....
 3 5

Prof. Dr. Faruk Arınç Fall 2020

 ME – 361 NUMERICAL METHODS FOR ENGINEERS

Infinite Set of Orthogonal Polynomials

Legendre polynomials, which form an infinite set, have this property when w(x) = 1
and the range [-1,+1]

 (1) P (x) P (x) dx

-1
n m  0 , n  m

1
P0(x) = 1 P1(x) = x P2 (x)  (3 x 2 - 1)
2

1 1 1
P3 (x)  (5 x 3 - 3 x) P4 (x)  (35 x 4 - 30 x 2  3) P5 (x)  (63 x 5 - 70 x 3  15 x)
2 8 8

{Pn (x) , n = 0, 1, ... }

Prof. Dr. Faruk Arınç Fall 2020

 ME – 361 NUMERICAL METHODS FOR ENGINEERS

Adrien-Marie Legendre

French mathematician

1752 – 1833

Prof. Dr. Faruk Arınç Fall 2020

 ME – 361 NUMERICAL METHODS FOR ENGINEERS

Legendre Polynomials

1
P0(x) = 1 P1(x) = x P2 (x)  (3 x 2 - 1)
2

1 1 1
P3 (x)  (5 x 3 - 3 x) P4 (x)  (35 x 4 - 30 x 2  3) P5 (x)  (63 x 5 - 70 x 3  15 x)
2 8 8

Chebyshev Polynomials

T0(x) = 1 T1(x) = x T2(x) = 2 x2 - 1

T3(x) = 4 x3 - 3 x T4(x) = 8 x4 - 8 x2 + 1 T5(x) = 16 x5 - 20 x3 + 5 x

Prof. Dr. Faruk Arınç Fall 2020

 ME – 361 NUMERICAL METHODS FOR ENGINEERS

Pafnuty Lvovich Chebyshev

Russian mathematician

1821 - 1894

Prof. Dr. Faruk Arınç Fall 2020

 ME – 361 NUMERICAL METHODS FOR ENGINEERS

Laguerre Polynomials

1 2
L (x)  1
0
0 L (x)  - x  1
0
1
L02 (x)  (x  4 x  2)
2

1 1 4
L (x)  (- x 3  9 x 2  18 x  6)
0
3
L04 (x)  (x  16 x 3  72 x 2  96 x  24)
6 24

Hermite Polynomials

H0(x) = 1 H1(x) = 2 x H2(x) = 4 x2 - 2

H3(x) = 8 x3 - 12 x H4(x) = 16 x4 - 48 x2 + 12 H5(x) = 32 x5 - 160 x3 + 120 x

Prof. Dr. Faruk Arınç Fall 2020

 ME – 361 NUMERICAL METHODS FOR ENGINEERS

Admond Nicolas Laguerre

French mathematician

1834 – 1886

Prof. Dr. Faruk Arınç Fall 2020

 ME – 361 NUMERICAL METHODS FOR ENGINEERS

Charles Hermite

French mathematician

1822 - 1901

Prof. Dr. Faruk Arınç Fall 2020

 ME – 361 NUMERICAL METHODS FOR ENGINEERS

ORTHOGONAL POLYNOMIALS

Name Range of Weighting Integral Property

Orthogonality Function
1
2
Legendre [- 1 , + 1] w(x) = 1  w Pk Pk dx 
-1
2 k 1
1

1 w T k Tk dx   if k  0
Chebyshev [- 1 , + 1] w(x) = -1
2
1-x 1

 w Tk Tk dx 
-1
2
if k  0


 (  k  1)

 
Laguerre [0 , ∞ ] w (x) = xα e-x w L k L k dx 
0
k!



w (x)  e 
-x2
Hermite [- ∞ , +∞ ] w H k H k dx   2 k
k!
-

Prof. Dr. Faruk Arınç Fall 2020

 ME – 510 NUMERICAL METHODS FOR ME II

Gauss - Legendre Quadrature Method

b +1 N
I   f(x) dx =  F(z) dz  w
i1
i F(z i ) , N = 2, 3, ....
a -1

N
I  w
i1
i F(z i ) if F  z i  is a polynomial of degree 2 N -1 or LESS

x=
b - a  z+b+a
, dx =
b-a
dz
Change of Variable:
2 2

zi’s are the roots of the Legendre polynomial, PN(z) = 0

zi’s and wi’s are given in mathematical handbooks as tabulated values

Prof. Dr. Faruk Arınç Fall 2020

 ME – 361 NUMERICAL METHODS FOR ENGINEERS

2x-b-a 2
Define a new independent variable: z = dz = dx
b-a b-a
b 1

Find dz and F(z) I =  f(x) dx =  F(z) dz

a 1

Choose N

Find nodes (xi) and weights (wi) from tabulated data

The nodes (xi) are the zeros of Legendre orthogonal polynomials, which are
orthogonal in the range, (-1, +1)

b 1 N
I =  f(x) dx =  F(z) dz  w
i 1
i F(zi )
a 1

Prof. Dr. Faruk Arınç Fall 2020

 ME – 361 NUMERICAL METHODS FOR ENGINEERS

Gauss-Legendre Nodes and Weights

N ± zi wi
N ± zi wi
2 0.5773502692 1.0000000000
0.1834346425 0.3626837834
0.0000000000 0.8888888889
3 0.5255324099 0.3137066459
0.7745966692 0.5555555556 8
0.7966664774 0.2223810345
0.3399810436 0.6521451549
4 0.9602898565 0.1012285363
0.8611363116 0.3478548451
0.0000000000 0.3302393550
0.0000000000 0.5688888889
0.3242534234 0.3123470770
5 0.5384693101 0.4786286705
9 0.6133714327 0.2606106964
0.9061798459 0.2369268850
0.8360311073 0.1806481607
0.2386191861 0.4679139346
0.9681602395 0.0812743884
6 0.6612093865 0.3607615730
0.1488743390 0.2955242247
0.9324695142 0.1713244924
0.4333953941 0.2692667193
0.0000000000 0.4179591837
10 0.6794095683 0.2190863625
0.4058451514 0.3818300505
7 0.8650633667 0.1495513492
0.7415311856 0.2797053915
0.9739065285 0.0666713443
0.9491079123 0.1294849662
Prof. Dr. Faruk Arınç Fall 2020
 ME – 361 NUMERICAL METHODS FOR ENGINEERS

Examples:

Exact G-L nodes G-L

Integral Comments
Solution & weights Solution

4 2 4 Exact result

81 2 81 Exact result

0.9975212 2 0.8706496 Low accuracy

0.9975212 4 0.9971697 More accurate

Prof. Dr. Faruk Arınç Fall 2020

 ME – 361 NUMERICAL METHODS FOR ENGINEERS

Example

6 6 N 2
I =  e dx =  f(x) dx 
-x
w
i 1
i F(zi ) = w 1 F(z1) + w 2 F(z 2 )
0 0

2x-b-a 2x-6 x-3

z= = = => x = 3 (z + 1) dx = 3 dz
b-a 6 3

6 1

I =  e dx = 3
-x
 e -3(z+1)
dz  3 
 e -3(z1 +1)
+ e -3(z 2 +1)

0 1

1
z1,2 =  =  0.5773502692 I  0.8706496
3
Iexact  0.9975212

12.7 % error with only 2 nodes and weights

Prof. Dr. Faruk Arınç Fall 2020

 ME – 361 NUMERICAL METHODS FOR ENGINEERS

Error expression for Gauss-Legendre quadrature method:

22 N  1 N !
4

E f (2N)   , - 1   1
(2 N  1) 2 N !
3

where N is the number of nodes and weights used.

Prof. Dr. Faruk Arınç Fall 2020

 ME – 361 NUMERICAL METHODS FOR ENGINEERS

GAUSS QUADRATURE METHODS

Name Quadrature Formula

 1 N
Gauss-Legendre  f (x) dx  w
i1
i f (x i )
-1

 1 N
1
Gauss-Chebyshev  2
f (x) dx  w i f (x i )
-1 1-x i1

 N
Gauss-Laguerre e
-x
f (x) dx  w
i1
i f (x i )
0

 N
Gauss-Hermite e
- x2
f (x) dx  w
i1
i f (x i )
-

Prof. Dr. Faruk Arınç Fall 2020

 ME – 361 NUMERICAL METHODS FOR ENGINEERS

GAUSS QUADRATURE METHODS

Name Nodes and Weights

N  xi wi

2 0.577350269 1.00000000
0 0.888888889
Gauss-Legendre 3
0.774596669 0.555555556

0.339981043 0.652145155
4
0.861136312 0.347854845

 1 
i-  
Gauss-Chebyshev x i  cos  2  i  1, 2, ... N wi = for all i
 N  N
 
 
Prof. Dr. Faruk Arınç Fall 2020
 ME – 361 NUMERICAL METHODS FOR ENGINEERS

N xi wi

0.58578 64376 0.85355 33906

2
3.41421 35624 0.14644 66094
0.41577 45568 0.71109 30099
Gauss-Laguerre 3 2.29428 03603 0.27851 77336
6.28994 50829 0.01038 92565
0.32254 76896 0.60315 41043
1.74576 11012 0.35741 86924
4
4.53662 02969 0.03888 79085
9.39507 09123 0.00053 92947

Prof. Dr. Faruk Arınç Fall 2020

 ME – 361 NUMERICAL METHODS FOR ENGINEERS

N  xi wi
2 0.70710 67811 0.88622 69255
0. 1.18163 59006
3
1.22474 48714 0.29540 89752
Gauss-Hermite
0.52464 76233 0.80491 40900
4
1.65068 01239 0.08131 28354
0. 0.94530 87205
5 0.95857 24646 0.39361 93232
2.02018 28705 0.01995 32421

Prof. Dr. Faruk Arınç Fall 2020

 ME – 361 NUMERICAL METHODS FOR ENGINEERS

Prof. Dr. Faruk Arınç Fall 2020