3

INTERPOLATION AND CURVE FITTING

 Introduction
 Newton Interpolation: Finite Divided Difference
 Lagrange Interpolation
 Spline Interpolation
 Polynomial Regression
 Multivariable Interpolation
Chapter 3 Interpolation & Curve Fitting / 2

3.1 Introduction
 Interpolation is a technique to estimate the value between a set of data.

 A general approach is to map the data into an n-th order polynomial:

fn (x)  a0 n
 a1x L x ax
n
(3.1)
an i
i
i0

 This chapter covers three types of techniques, i.e. the Newton interpolation,
the Lagrange interpolation and the Spline interpolation.

 The resulting equation can be used for curve fitting.

3.2 Newton Interpolation: Finite Divided Difference
 For the linear interpolation, consider Fig. 3.1 and the following relation:
f1 x   f x0 
f x1   f x0

x  x0 
x1  x0

f x

f
x

1

f
x
 x
f

0


0
x

1
x

FIGURE 3.1 Linear interpolation

Rearranging the above relation leads to

f x 
f x  f x1   f x0  x
1 0 x    (3.2)
0
x1  x0

The term  f x1   f x0  x1  x0 is referred to as the first-order finite


divided difference.

Example 3.1

Calculate an approximate value of ln 2 using the linear interpolation using

the following ranges: (1,4) and (1,6). Use ln 1  0, ln 4  1.3862944 and
ln 6  1.7917595 as the source data. Estimate the relative error if the actual
value is ln 2  0.69314718.
Solution
For the linear interpolation between x0  1 and x1  6:
1.7917595  0
f 2  0   2 1  0.35835190
1
6 1
This produces a relative error of t  48.3%.
 For the linear interpolation between x0  1 dan x1  4:
1.3862944  0
f 2  0   2 1  0.46209812
1
4 1
This produces a relative error of t  33.3%, which smaller than the
previous case.


 For the quadratic interpolation, consider Fig. 3.2 and the following relation:

y
fx 
2
ln x

1
f1
x
0
0 2 4 6 x
FIGURE 3.2 The result for the linear interpolation for ln 2

f 2 x   b0  b1 x  x0   b2 x  x0  x  x1  (3.3)

By taking the values of x equal to x0, x1 dan x2 in Eq. (3.3), the coefficients
bi can be calculated as followed:

b0 
f  x0  (3.4a)
f  x1   f  x0 (3.4b)
b1 

x1  x0

f x2   f x1  f x1   f  x0

x2  x1 
b   (3.4c)
x1  x0
2
x2  x0
Example 3.2

Repeat Example 3.1 using the quadratic interpolation.

Solution
From Example 3.1, the data can be rewritten as followed:
x0  1: f (x0 )  0
x1  4 : f (x1)  1.3862944
x2  6 : f (x2 )  1.7917595
Using Eq. (3.4):
b0  0
1.3862944  0
b  0.46209812
1
4 1
1.7917595 1.3862944 1.3862944  0
6  4 4  1 
b2   0.051873113
6 1
Hence, Eq. (3.3) yields
f2 (x)  0  0.46209812x 1 0.051873113x 1x  4
For x  2:
f 2 2  0  0.462098122  1  0.0518731132  12  4,
 0.56584436.
This produces the relative error t  18.4% as shown below.

 y

fx 
ln x
1
f2
x

0
f1
0 2 4 x x 6
FIGURE 3.3 The result for the quadratic interpolation for ln 2

 The general form representing an n-th order interpolation function (requires

n1 pairs of data) is:

f n  x   b0  b1  x  x0  L bn x  x0 x  x1 L  x 
(3.5)
xn1 

where the coefficients bi can be obtained via

b0 
f  x0 
b1 
f x1 , x0 
b2 
f x2 , x1 , x0 
M
bn 
f xn , xn1 ,K, x1 , x0 

where the terms [] are finite divided differences and are defined as:


f x,x   f x  f x 
i (3.6a)
i j
xi  x j


f x,x,x
  
f xi , x  f xj , x 

i j k
(3.6b)
xi  xk
f x , x ,K, x , x  
f xn , xn1 ,K, x2 , x1   f xn1 , xn2 ,K, x1 , (3.6c)
x0 
n n1 1 0
xn  x0

Hence, Eq. (3.5) forms the Newton interpolation polynomial:

f n x   f x0   x  x0  f x1 , x0   x  x0 x  x1  f x2 , x1 , x0 
L x  x  L(x  (3.7)
n1
)f , ,K, x 
0
x0 x1 x x n xn1
 TABLE 3.1 Newton finite divided difference interpolation polynomial

i xi fxi  I II III
0 x0 f x0 f x1x0 f x2x1x0 f x3x2x1x0
1 x1 f x1 f x2x1 f x3x2x1
2 x2 f x2 f x3x2
3 x3 f x3

 The relation for error for the n-th order interpolation function can be
estimated using:

Rn 
f xn1 , xn , xn1 ,K, x0  x  x0  x  x1 Lx  (3.8)
xn 

Example 3.3

Repeat Example 3.1 with an additional fourth point of x3  5,

fx3  1.6094379 using the third order Newton interpolation polynomial.
Solution
The third order Newton interpolation polynomial can be written as
f 3 x   b0  b1 x  x0   b2  x  x0  x  x1   b3 x  x0 x  x1 x  x2 

i xi fxi  I II III
0 1 0 0.46209813 0.0597386 0.00786553
1 4 1.3862944 0.22314355 0.0204110
2 5 1.6094379 0.18232156
3 6 1.7917595

Thus,
f 3  x  0  0.46209812x  1  0.0597386x  1x  4
 0.00786553x  1x  4x  5,
f 3 2  0  0.462098122  1  0.05973862  12  4
 0.007865532  12  42  5,
 0.6287691.
and producing a relative error of t  9.29%.

y

2
fx  l n x

f3x

0
0 2 4 6 x
TABLE 3.4 The result for the cubic interpolation for ln 2
 3.3 Lagrange Interpolation
 The formula for the Lagrange interpolation is:
n

f n  x    Li  x  f  x i (3.9)

i 0

where the parameter Lix is defined as

L x  
xxj
n
i
(3.10)
j 0
j i
xi  x
j

 For n  1, the first order Lagrange interpolation function can be written as:

f x 
x  x1 f  x  x  x 0 f x 
1



0 1
x 0  x1 x1  x0

 For n  2, the second order Lagrange interpolation can be written as:

 x  x1 x  x2  x  x0 x 
xf 2x   f x   f x 
2  0  1

x0  x1 x0  x2  x0 x  x2 

 x1 1

x  x0 x  x1 
 f x2 
x2  x0 x 2  x1 

 The error term for the Lagrange interpolation can be estimated using:

Rn 
f xn1 , xn , xn1 ,K, x0   x  xi (3.11)
n


i0

Example 3.5

Repeat Examples 3.1 and 3.2 using the first and second order Lagrange
interpolation functions, respectively.
Solution
From Examples 3.1:
x0  1 : f (x0 )  0
x1  4 : f (x1)  1.3862944
x2  6 : f (x2 )  1.7917595
 Chapter 3 Interpolation & Curve Fitting / 10

For the first order Lagrange interpolation:

f x  x 
f x  x  f x 
1 x1 
x0
0 1
x0  x1 x1  x0
24 2 1
 f 2  0  1.3862944
1


1 4 4 1
 0.4620981
For the second order Lagrange interpolation:
f x 
2 x   x  x0 x  x2 f x 
 x  x1 x  x2  x  1
f
0  x1 x0  x2 0
x1  x0 x  x2 
1

x  x0 x  x1

 x2  x0 x2  f x2 
 x1 
f (2)  2  42  6 2  12  6
0  1.3862944
2
1  41  6 4  14  6
2  12  4
 1.7917595
6  16  4
 0.5658444

3.4 Spline Interpolation
 The polynomial interpolation can cause oscillation to the function and this
can be remedied by using the spline interpolation.

fx f
 x


x
x
(a) polynomial (b) Linear spline
FIGURE 3.5 Comparison between polynomial and spline interpolation

 The formula for the linear spline interpolation for n1 data between points
x0 and xn (n intervals) is
f11 x  
f x0   m0  x  x0 untuk x0  x  x1

f12 x  f x1   m1  x  x1  untuk x1  x  x2
M M (3.12)
f1n x  f xn1   mn1  x  xn1 untuk xn1  x  xn

dengan mi is the gradient for the (i1)-th interval:

f xi1   f xi
m  (3.13)
 i

xi1  xi

Example 3.6

Use the following data to estimate the function fx at x  5 using the linear
spline interpolation.
x f(
x
)
3 2
. .
0 5
4 1
. .
5 0
7 2
 . .
0 5
9 0
. .
0 5
Solution
The value of x  5 is in the range 4.5  x  7 , where the gradient is

m1 2.5 1.0
 7.0  4.5 0.60
From Eq. (3.12):
f12 x  f  x1   m1  x  x1  ,
f12 5  1.0  0.605  4.5  1.3.

2

fx
4

2
6

FIGURE 3.6 Linear spline interpolation for Example 3.6

8

 For the quadratic spline, the general polynomial for each interval is
f x  a x  b x  c
2
x

(3.14)
2i i i i

The coefficient ai, bi and ci for each interval can be evaluated using:

1. Continuity at vertices (2n2 equations):

ai1
 bi xi1  ci1  f  xi1
2
xi1 (3.15)

1

2
ax
 b x  c  f x  (3.16)
i i1 i i1 i1

2. Conditions at end points (2 equations):

a x  b x  c  f x 
2
 (3.17)
1 0 1 0 1 0

a x  b x  c  f x 
2
(3.18)
n n n n n n

3. Differentiability at vertices (n1 equations):

f x   2ax  b
2ai1xi1  bi1  2aixi1  (3.19)
bi

4. Assumption at the first interval (1 persamaan):

(3.20)
a1  0

fx

0
2 4 6 x 8

FIGURE 3.7 The quadratic spline interpolasi for example 3.7

 For the cubic spline, the general polynomial for each interval is
f x  a x  b x  c x  d
3 2

(3.21)
3i i i i i

 The formulation for the cubic spline can be derived as followed:

For interval xi1, xi, the second derivative can be written as:
x
f x   f  x x  f  x  (3.22)
x xi1
i
i i i1
xi1  xi i i xi  xi1

The integration of Eq. (3.22) twice produces two integration constants,

which can be evaluated using f  x   f  xi1  at xi1 and f x   f xi  at xi,
thus produces
f x 
f   xi1 x  x 
3
f   xi   x  x 3
3
i

6 xi  xi1  6xi  xi1 
 i
i1
⎡ f xi1  f   xi1  xi  xi1  x
 
⎤
x
⎢
x  xi1 
⎣ i 6
⎥
⎦i (3.23)

⎡ f xi  f  xi  xi  xi1 ⎤

⎢  x  x 
x  x  6
⎥ i1

⎣ i i1 ⎦
 The differentiability can be maintained troughout the function via

f i1 xi   
f i  xi  (3.24)

Eq. (3.23) can be differentiated for both the i1-th and i-th intervals and
following Eq. (3.24):

xi  xi1  f  xi1   2xi1  xi1  f  xi   xi1  xi  f  xi1 

6 6
 xi1  xi  f xi1  f xi    f xi1 f xi (3.25)
  
xi

  xi1 

Example 3.8

Repeat Example 3.6 using the cubic spline interpolation.

Solution
Using Eq. (3.25) for the first interval (i  1):
4.5  3 f  3  27  3 f  4.5  7  4.5 f  7
6
2.5  1 
  2.5  1

 6 
7  4.5 4.5  3
It is known that f  3  0 .
Hence
8 f  4.5 2.5 f  7  9.6
For the second interval (i  2):
2.5 f  4.5 9 f  7  9.6
Both equations is solved simultaneously to yield
f  4.5  1.67909
f  7  1.53308
Using Eq. (3.23) for the first interval:

f31 1.67909
x x  33  2.5 4.5  x
64.5  4.5 

3 3
⎡  1 1.679094.5  3⎤
⎢  x  3,
⎣ 4.5  3 6 ⎥
⎦
 0.186566x  3  1.6666674.5  x   0.246894x  3.
3
For the second and third intervals:
f32
x   0.1119397  x 3  0.102205x  4.53  0.2996217  x 
 1.638783x  4.5
f 33  x   0.1277579  x   1.7610279  x   0.25x  7 
3

At x  5, use the function for the second interval:

f32
5  0.1119397  53  0.1022055  4.53  0.2996217  5
 1.6387835  4.5,
 1.102886 .


fx

2
e
Splin
c
Cubi

0
2 4 6 8 x
Interpolation
Lagrange
Cubic Newton/

FIGURE 3.8 The cubic spline interpolation for Example 3.8

 3.5 Polynomial Regression
 A “best fit” polynomial of an order lower than the number of intervals is
sometimes required to represent the data and can be evaluated via
polynomial regression.

 Consider the following form of polynomial:

f x  a0
 a1x  a2 x2 L a xnn (3.26)

where the error for the i-th data (xi,yi) is

ey
i i  f x   y   a 
2
a x L a x
n

a i i 0
x1 i 2 i n i

The sum of the squares of error for N data is


N N
xn
S   ei   yi a2 x2 L  a
2
 a0  a1x  (3.27)
i1 i1 i
i n 2
i

These errors can be minimised as followed:

S
 0  2 y   a  a x 2 L a xn   1
N a x
a0 i1
i 0 1 i 2 i n i

S
 0  2 y   a  a x 2 L a xn   x 
N a x
a1 i1
i 0 1 i 2 i n i i

M
S
0 2 y a  a x 2 L a x  x 
n n

N a x
an i1
i 0 1 i 2 i n i i

producing the following system of equations:

a0 N  a1   a2  x 2 L a  xn   y
xi
i n i i
a  x  a  x   x L 2 3
x
n1
xy
0 i
a a
1 i3 2 i n i i i
a  x  a  x  a  x 4 L
2
 xn2   x2 y
a (3.28)
0 i 1 i 2 n i i i
i
 M
a  xn  a  xin1  2  xi n2 L n  xi2n   xni y i
0 i 1
a a
 or, in a form of matrix equation:
⎡ N  x
2
L
n
 x ⎤ ⎧a 0 ⎧  yi ⎫
x i ⎫
⎢ i i ⎥ ⎪
L x
n1
a ⎪ ⎪
⎢  xi
2 3
x i x i ⎪  xi yi

⎥⎥ ⎪ ⎪
i 1
⎢ x 2  x
4
L x
n2 2
⎬ ⎪⎨ x y⎪ (3.29)
3
x ⎨a ⎬
i i ⎥ ⎪ 2⎪ ⎪ i i⎪
⎢ Mi M ⎥ M M
⎢ i
M O M
⎪ ⎪ ⎪ ⎪
⎢ x n
 xi ⎥⎦ ⎪a
n
⎩ n ⎪ ⎪
⎩ xi yi ⎪
2n
⎣ i  xi
n1
 xi
n2
L ⎭ ⎭
 The standard deviation  for this case can be evaluated as followed:


e
2

 S  i1
(3.30)
N N
n 
 The case for n  1 is referred to as1the linear regression:
ni

y  a0  a1x
1
⎡⎢ N  x 
⎤ ⎧ 0 ⎫ ⎨ ⎧ yi ⎫
a
2i ⎥ ⎨ ⎬
⎬ x  x  a  x y (3.31)
⎣ i i ⎦ ⎩ 1⎭ ⎩ ii ⎭

Example 3.9

Determine a best-fit quadratic equation which can represent the following

experimental data:
xi 0.0 0.1 0.15 0.3 0.4 0.5
5 1 1 6 2
yi 0.9 0.8 0.832 0.7 0.5 0.5
56 90 17 71 39

xi 0.7 0.7 0.82 0.9 1.1

0 4 8 7
yi 0.3 0.3 0.306 0.2 0.1
78 70 42 04

Solution
For n  2, the quadratic polynomial can be written as
2
 f x  a0  a1x  a2 x
From Eq. (3.28):
⎡N 2
 x ⎤ ⎧a0  yi ⎫
⎢  ⎧
⎪ ⎪
 ⎥ ⎫
i
xi 3 ⎪
x ⎪
2 
⎢  xi  xi i⎥
a
⎬  ⎨ xi yi ⎬
⎨ 1
⎢ x 2  x 3  x 4 ⎥ ⎪a ⎪ ⎪ x 2 y ⎪
⎣ i i i⎦ ⎩ 2⎭ i i⎭

⎩
From the following table:
 xi  6.01 N  11
2
 x  4.6545
i   5.905
3
 x  4.1150 yi  2.1839
i
4
 x  3.9161  xi  1.3357
i
yi
Hence the matrix equation becomes: 2
 xi yi
⎡ 11 6.01 4.6545⎤ ⎧a0 ⎧ 5.905 ⎫
⎢ ⎫ ⎥ ⎪ ⎪ ⎪ ⎪
6.01 4.6545 4.1150  a  2.1839
⎨ 1⎬ ⎨ ⎬
⎢4.6545 4.1150 3.9161⎥⎥ ⎪a ⎪ ⎪1.3357⎪
⎣ ⎦ ⎩ 2⎭ ⎩ ⎭
a0  0.998 a1  1.018 a2  0.225
Hence, the quadratic function which can represent the data is
f x  0.998 1.018x  0.225x
2

y
1

0.8
fx

0.6

0.4

0.2
 0 x
0 0.2 0.4 0.6 0.8 1 1.2
FIGURE 3.9 The fitting of data using a quadratic function
3.6 Multivariable Interpolation
 The Newton, Lagrange and spline interpolation method can be applied for
multivariable cases, e.g. for two and three variables:
   q      
p

y
f pq
x,  Lx i L y f xi , y (3.32)
y x
j

  
i0 j0

  p
q r    
y z

f pqr
x, y,  L xi x L y j
L k
z f xi , y j , (3.33)
z zk
i0 j0 k 0

where

L x 
q


x L y  y L z 
r
z  zl
 
p
xl yl
xi
l 0 xi  xl yj
l 0 y j  yl zk
l 0 zk  zl
l i lj l k

 For example, Eq. (3.32) can be expanded as

x  x1 y  y1 
f x, y   f x x  x0 y  y1 
,y     f1 x 0, y 
x
x0  x1  y0  y  y  y1
1,1
0 0
1
 x0
1 0

x  x1 y  y0 f ,y x  x0 y  y0 f x , y 


 x 

x  x y x  0 1 x y  1 1

x y
0 1 1 0 1 0 1 0

Example 3.10

The following table show the topology of a 3-D non-planar surface:

y
z  f(x,y)
0.2 0.3 0.4 0.5
1 0.6 1.0 1.3 1.7
. 40 03 59 03
0
 x 1 0.9 1.5 2.0 2.5
. 90 24 45 49
5
2 1.5 2.3 3.1 3.9
. 68 84 77 43
0
Use the Newton interpolation to estimate the value of z at the coordinat
(1.6, 0.33). As a comparison, the above data follows the function
f x, y   e sin y  y  0.1, thus f(1.6, 0.33)  1.8350.
x
Chapter 3 Interpolation & Curve Fitting / 20

Solution
At x  1.0:

j yj f1.0, yj I II III

0 0.2 0.640 0.363 0.007 0.005
1 0.3 1.003 0.356 0.012
2 0.4 1.359 0.344
3 0.5 1.703
f1 1.0, y   0.640  0.363 y  0.2   0.007  y  0.2  y  0.3
 0.005y  0.2y  0.3y  0.4,
f1 1.0, 0.33  1.1108 .
At x  1.5:
j yj f1.5, yj I II III
0 0.2 0.990 0.534 0.013 0.004
1 0.3 1.524 0.521 0.017
2 0.4 2.045 0.504
3 0.5 2.549

f1 1.5, y   0.990  0.534 y  0.2   0.013 y  0.2  y  0.3

 0.004y  0.2y  0.3y  0.4,
f1 1.5, 0.33  1.6818 .
At x  2.0:
j yj f2.0, yj I II III
0 0.2 1.568 0.816 0.023 0.004
1 0.3 2.384 0.793 0.027
2 0.4 3.177 0.766
3 0.5 3.943

f1 2.0, y   1.568  0.816 y  0.2   0.023 y  0.2  y  0.3

 0.004y  0.2y  0.3y  0.4,
f1 2.0,0.33  2.6245 .
Next, at y  0.33:

i x fxi 0.33 I II
i

0 1 1.1108 0.57 0.37

. 10 17
0
1 1 1.6818 0.94
. 27
5
2 2 2.6245
.
0
f1,1  x, 0.33  1.1108  0.5710 x  1.0   0.3717 x  1.0 x  1.5 ,
f1,1 1.6, 0.33  1.8406 .
This produces a relative error of t  0.305%.

Exercise

1. The fuel consumption of an engine has been recorded as shown in the following table.
Time, Fuel,
hour liter
1.2 0.3320
1
1.7 0.5473
9
1.8 0.6049
6
2.0 0.7389
1
If a user runs the engine for 1.55 hours, determine the estimated fuel consumption using
the Newton and Lagrange interpolation methods.

2. The following data shows the height function of a hill a a distance x from a reference.
Form a cubic polynomial via regression.
xi 0 1 2 3 4 5 6 7 8
hi 4 5 1 1 2 1 1 3 1
0 7 1 6 1
Also, calculate the corresponding standard deviation.

3. The following data represents temperature distribution T(x,y) in a metal plate:

Temperat y coordinate
ur e 0.5 1 1.5 2
T(x,y) . .
0 0
0.5 7.51 10.05 12.7 15.67
x coordinate

0
1.0 10. 10.00 10.0 10.00
00 0
1.5 12. 9.95 7.32 4.33
51
2.0 15. 10.00 5.00 0.00
00
Estimate the temperature at the coordinate (x, y) = (1.15, 1.42) using:
a. the Newton linear interpolation,
b. the Newton cubic interpolation.