Numerical Differentiation

C. S. Mathpati
Institute of Chemical Technology
Mumbai
Taylor series expansion
Inclusion of the 2
nd
derivative term improves the accuracy to O(h
2
).
Similar improved versions can be developed for the backward and centered
formulas as well as for the approximations of the higher derivatives
) (
) ( ) ( ) (
) (

) (
) ( ) ( ) (
2
1
2
1
2
2
h O h
x f
h
x f x f
x f
h
x f
h x f x f x f
i i i
i
i
i i i
+
' '

=
'
+
' '
+
'
+ =
+
+

) (
) ( ) ( ) (
) ( h O
h
x f x f x f
x f
i i i
i
+
+
= ' '
+ +
2
1 2
2
) (
) ( ) ( ) (
) (
) (
) ( ) ( ) ( ) ( ) (
) (
2
1 2
2
2
1 2 1
2
3 4
2
2
h O
h
x f x f x f
x f
h O h
h
x f x f x f
h
x f x f
x f
i i i
i
i i i i i
i
+
+
= '
+
+

= '
+ +
+ + +
High Accuracy Differentiation Formulas
Forward finite-divided-difference formulas
( )
( ) ( )
( )
( ) ( ) ( )
( )
( ) ( ) ( )
( )
( ) ( ) ( ) ( )
1
2 1
2 1
2
3 2 1
2
'
4 3
'
2
2
''
4 5 2
''
i i
i
i i i
i
i i i
i
i i i i
i
first derivative
f x f x
f x
h
f x f x f x
f x
h
second derivative
f x f x f x
f x
h
f x f x f x f x
f x
h
+
+ +
+ +
+ + +

=
+
=
+
=
+ +
=
( )
( )
( )
( )
2
2
O h
O h
O h
O h
4
Backward finite-divided-difference formulas
5
Centered finite-divided-difference formulas
Numerical differentiation
x
y y
x x
y y
mB
A

=
2 3
2 3
2 3
x
1
x
2
x
3
y
1
y
2
y
3
x x
y=f(x)
True slope
A

B

C

x
y y
x
y y
x
y y m m
m
B A
C
A

=
|
.
|

\
|
A

+
A

=
+
=
2 2
1
2
1 3 2 3 1 2
x
y y
x x
y y
mA
A

=
1 2
1 2
1 2
lim
0
dy y
x dx x
A
A
=

: backward
difference
: central difference
: forward difference
The diff function
The diff Function
d= diff (x)

Backward difference & central difference method



example x=[5, 7, 12, -20];
d= diff(x)

d = 2 5 -32
First, create a file called fx1.m which contains y=f(x):
function y = fx1(x)
y = 1.2 - .25*x - .5*x.^2 - .15*x.^3 -.1*x.^4 ;

Command window:
>> x=0:.25:1
0 0.25 0.5 0.75 1
>> y = fx1(x)

1.2 1.1035 0.925 0.6363 0.2

>> d = diff(y) ./ diff(x)

d = -0.3859 -0.7141 -1.1547 -1.7453

Forward: x = 0 0.25 0.5 0.75 1
Backward: x = 0.25 0.5 0.75 1
x f(x)
i-2 0 1.2
i-1 0.25 1.1035
i 0.50 0.925
i+1 0.75 0.6363
i+2 1 0.2
Differentiation Using MATLAB
Forward difference of accuracy O(h
2
) is computed as:



Backward difference of accuracy O(h
2
) is computed as:



9
Example : High Accuracy Differentiation Formulas
f(x) = -0.1x
4
0.15x
3
0.5x
2
0.25x + 1.2
At x = 0.5 True value for First Derivative = -0.9125
Using finite divided differences and a step size of h = 0.25

t
0.2 4(0.6363) 3(0.925)
(0.5) 0.8593 5.82%
2(0.25)
f c
+
' = = =
Forward
O(h)
Backward
O(h)
Estimate -1.155 -0.714

t
(%) 26.5 21.7
x f(x)
i-2 0 1.2
i-1 0.25 1.1035
i 0.50 0.925
i+1 0.75 0.6363
i+2 1 0.2
t
3(0.925) 4(1.1035) 1.2
(0.5) 0.8781 3.77%
2(0.25)
f c
+
' = = =
Polynomial derivatives
n n
n n n
a x a x a x a x a x f + + + + + =

1
2
3
1
2 1 ) (
1
2
2
1
1 ) 1 (

+ + + = n
n n
a x a n x na
dx
df

1
2
2
1
1

+ + + = n
n n
b x b x b
Command Description
d=diff(x) Returns a vector d containing the differences between adjacent elements
in the vector x.
b=polyder(p) Returns a vector b containing the coefficients of the derivative of the
polynomial represented by the vector p.
b=polyder(p1,p
2)
Returns a vector b containing the coefficients of the polynomial that is
the derivative of the product of the polynomials represented by p1 and
p2. cf. equivalent comment: b=polyder(conv(p1,p2))
[num, den]=
polyder(p2,p1)
Returns the vector num and den containing the coefficients of the
numerator and denominator polynomials of the derivative of the
quotient p
2
/p
1
, where p1 and p2 are polynomials.
Numerical differentiation functions
b = polyder (p)
p = [a1,a2,,an]
b = [b1,b2,,bn-1]
b = polyder (p1,p2)
[num, den] = polyder(p2,p1)

Polynomial derivatives
Result: der2 = [10, 4, -3]
prod = [150, 80, -7]
num = [50, 40, 23]
den = [25, 20, 4]
Example p1= 5x +2
p2=10x2+4x-3
12
Richardson Extrapolation
There are two ways to improve derivative estimates when employing finite divided
differences:
Decrease the step size, or
Use a higher-order formula that employs more points.
A third approach, based on Richardson extrapolation, uses two derivative estimates (with O(h
2
) error) to
compute a third (with O(h
4
) error) , more accurate approximation. We can derive this formula following
the same steps used in the case of the integrals:


) (
3
1
) (
3
4
2 /
1 2 1 2
h D h D D h h ~ =
Example: using the previous example and Richardsons formula, estimate the first
derivative at x=0.5 Using Centered Difference approx. (with error O(h
2
)) with
h=0.5 and h=0.25 :

D
h=0.5
(x=0.5) = (0.2-1.2)/1 = -1 [
t
=(-.9125+1)/-.9125 = -9.6% ]
D
h=0.25
(x=0.5) = (0.6363-1.103)/0.5=-0.9343 [
t
=(-.9125+0.9343)/-.9125 = -2.4% ]

The improved estimate is:
D = 4/3(-0.9343) 1/3(-1) = -0.9125 [
t
=(-.9125+.9125)/-.9125 = 0% perfect!]
f(x) = -0.1x4 0.15x3 0.5x2 0.25x + 1.2
At x = 0.5 True value for First Derivative = -0.9125
13
Derivatives of Unequally Spaced Data
Data from experiments or field studies are often collected at unequal intervals. One
way to handle such data is to fit a Lagrange interpolating polynomial, and then
calculate the 1
st
derivative. As an example, second order Lagrange interpolating
polynomial is used below:








*Note that any three consecutive points, x
i-1
x
i
and x
i+1
can be used to calculate the
derivative. The points do not need to be spaced equally.
( )( )
( )( )
( )( )
i i i i
i i
i
i i i i
i i
i
i i i i
i i
i
x x x x
x x x
x f
x x x x
x x x
x f
x x x x
x x x
x f x f


+


+


=
'
+ +

+
+
+
+
+

1 1 1
1
1
1 1
1 1
1 1 1
1
1
2
) (
2
) (
2
) ( ) (
[

=
=
=

=
=
n
i j
j
j i
j
i
n
i
i i n
x x
x x
x L
x f x L x f
0
0
) (
) ( ) ( ) (
14
( )( )
( )( )
( )( )
2 7 -
0
W/m 56 . 70 ) 333 . 1 )( 840 )( 1800 ( 3.5x10 - 0) q(z
: compute to used be can which
/ 333 . 1 333 . 1 4 . 14 4 . 14
25 . 1 75 . 3 0 75 . 3
25 . 1 0 ) 0 ( 2
10
75 . 3 25 . 1 0 25 . 1
75 . 3 0 ) 0 ( 2
12
75 . 3 0 25 . 1 0
75 . 3 25 . 1 ) 0 ( 2
5 . 13 ) 0 (
= = =
= + =


+


+


= = '
cm C
z f
0
) 0 (
=
= =
z
dz
dT
C k z q
A temperature gradient can be measured down into the soil as shown in the figure below.
The heat flux at the soil-air interface can be computed with Fouriers Law:
q = heat flux
k = coefficient of thermal diffusivity in soil (3.5x10
-7
m
2
/s)
= soil density( 1800 kg/m
3
)
C = soil specific heat( 840 J/kg . C
o

*Positive flux value means heat is transferred from the air to the soil

Calculate dT/dz (z=0) and determine the heat flux.
Example
Tutorial - 5
The velocity v (m/s) of air flowing past a flat
surface is measured at several distances y (m) away
from the surface. Determine the shear stress,
(N/m2) at the surface (y=0). Assume a value of
dynamic viscosity =1.8 x 10
-5

dv
dy
t =
y m

0

0.002

0.006

0.012

0.018

0.024

v m/s

0

0.287

0.899

1.915

3.048

4.299

Tutorial - 6
For a certain chemical reaction AB time vs
concentration data is given.
Find out values of k& n
-r
A
= -dC
A
/dt =k (C
A
)
n
t

10

20

30

40

50

60

C

3.52

2.48

1.75

1.23

0.87

0.61

Solution
% numerical integration
clc;
clear all;
close all;

t=[10 20 30 40 50 60];
C=[3.52 2.48 1.75 1.23 0.87 0.61];

difft=diff(t);
diffC=diff(C);

dCdt=diffC./difft;

% plot(log(C(2:end)),log(-dCdt),'LineWidth',2);
% figure;plot((C(2:end)),(dCdt),'LineWidth',2);

a=polyfit(log(C(1:end-1)),log(-dCdt),1);
n=ceil(a(1))
k=a(2);

b=polyfit((C(1:end-1)),(-dCdt),0);
n=b
n = 1
k = 0.0296