Interpolation and

Curve Fitting

Pierre Bzier

De Casteljau construction of Bezier curve

From www.wikipedia.org

Administration
Assignment

2 is out; due on April 9 midnight

Chapter 2 Excises 17, 66, 81, 82 (20% each)
Chapter 2 Excises 35 (20%)
write

MATLAB code to solve a tridiagonal system

send

your code to TA
([email protected]), EC229b

Numerical Methods Wen-Chieh Lin

Outline
Overview

of interpolation problem
Polynomial interpolation
Bezier curves and b-splines
Least square approximation
Robust curve fitting

Numerical Methods Wen-Chieh Lin

Interpolation
Basic

problem: for given data

( xi , yi ),

i 1,, m

with x1 < x2 < < xm determine function

f:
such that

f ( xi ) yi ,

i 1,, m

f is interpolating function, or interpolant, for

given data

Numerical Methods Wen-Chieh Lin

Interpolation (cont.)
Additional

data might be specified, such as

slope of interpolation function at given points

Additional

constraints might be imposed, such

as smoothness, monotonicity, or convexity of
interpolation function

could be function of more than one variable

Numerical Methods Wen-Chieh Lin

Purposes of Interpolation
Plotting

smooth curve through discrete data

points
Reading

between lines of table

Differentiating

Replacing

or integrating tabular data

complicated function by simple one

Numerical Methods Wen-Chieh Lin

Interpolation vs Approximation
Interpolation function fits given data points
exactly
Interpolation is inappropriate if data points
subject to significant errors
Approximation is usually preferable for
smoothing noisy data

Numerical Methods Wen-Chieh Lin

Issues in Interpolation
There

are arbitrarily many functions

interpolating given data points

What

form should function have?

How should function behave between data points

Should

function inherit properties of data, such

as montonicity, convexity, or periodicity?

Are

parameters that define interpolating

function meaningful?
Numerical Methods Wen-Chieh Lin

Choosing Interpolant
Choice of function for interpolation based on
How

easy function is to work with

determining

its parameters
evaluating function
differentiating or integrating function
How

well properties of function match

properties of data to be fit (smoothness,
monotonicity, convexity, periodicity, etc.)
Numerical Methods Wen-Chieh Lin

Functions for Interpolation

Families

of functions commonly used for

interpolation include
Polynomials
Piecewise

polynomials
Trigonometric functions
Exponential functions
Rational functions
We

will focus on interpolation by polynomial

and piecewise polynomials for now
Numerical Methods Wen-Chieh Lin

10

Polynomial Interpolation
Simplest

type of interpolation uses

polynomials

Unique

polynomial of degree at most n-1

passes through n data points (xi, yi), i = 1, , n,
where xi are distinct

There

are many ways to represent or compute

polynomial, but in theory all must give same
result
Numerical Methods Wen-Chieh Lin

11

Lagrangian Polynomials
For

a cubic polynomial case,

P3 ( x )
( x x1 )( x x2 )( x x3 )
( x x0 )( x x2 )( x x3 )
y0
y1
( x0 x1 )( x0 x2 )( x0 x3 )
( x1 x0 )( x1 x2 )( x1 x3 )
( x x0 )( x x1 )( x x3 )
( x x0 )( x x1 )( x x2 )

y2
y3
( x2 x0 )( x2 x1 )( x2 x3 )
( x3 x0 )( x3 x1 )( x3 x2 )

Numerical Methods Wen-Chieh Lin

12

Error of Polynomial Interpolation

The

error term of polynomial interpolation can

be expressed as
f
()
E ( x ) ( x x0 )( x x1 ) ( x xn )
( n 1)!
( n 1)

the smallest interval contains {x, x0 , x1 ,, xn }

Not very useful for computing real error as f is
usually unknown
Good for interpretation of some effects in error
analysis
Numerical Methods Wen-Chieh Lin

13

Error of Polynomial Interpolation

Let

E ( x ) f ( x ) Pn ( x ) ( x x0 )( x x1 ) ( x xn ) g ( x )

Define

auxiliary function

W (t ) f (t ) Pn (t ) (t x0 )(t x1 ) (t xn ) g ( x )
n+2 zeros, x0, x1, , xn, and x
W
(t) has n+1 zeros if W is a C1 function
W
(t) has n zeros if W is a C2 function
and so on..
W(n+1)(t) has at least one zero between ( x0 , xn , x )
has

( n 1)

() 0

Numerical Methods Wen-Chieh Lin

14

Error of Polynomial Interpolation

W ( n 1) () 0
n 1

d
n 1
f (t ) Pn (t ) (t x0 )( t x1 ) (t xn ) g ( x )
t
dt

f ( n 1) () 0 ( n 1)! g ( x )
f ( n 1) ()
E ( x ) ( x x0 )( x x1 ) ( x xn )
( n 1)!

Numerical Methods Wen-Chieh Lin

15

Interpretation of Error Term

f
()
E ( x ) ( x x0 )( x x1 ) ( x xn )
( n 1)!
( n 1)

Interpolation

Error

works better for x within xi

s

is smaller if x is centered within xi

Interpolationoutside

(x0 ,xn), called

extrapolation, will have large errors than
interpolation
Numerical Methods Wen-Chieh Lin

16

Disadvantages of Polynomial Interpolation

Need

to re-compute the interpolation function

if adding or removing a data point

Divided-differences

method avoids this

problem using fewer arithmetic operations

Divided-differences

gives the same polynomial

as Lagrangian interpolation
Numerical Methods Wen-Chieh Lin

17

Divided Differences
Given

data points (xi, yi), i = 0,,n, divided

differences, denoted by f[], defined recursively
by
f [ x1 , x2 , xn ] f [ x0 , x1 , xn 1 ]
f [ x0 , x1 , xn ]
xn x0
Some examples:
f [ x0 ] y0 f 0
f [ x1 ] f [ x0 ] f1 f 0
f [ x0 , x1 ]

x1 x0
x1 x0

f [ x1 , x2 ] f [ x0 , x1 ]
f [ x0 , x1 , x2 ]
x2 Wen-Chieh
x0
Numerical Methods
Lin

18

Divided differences (cont.)

Pn ( x ) a0 ( x x0 )a1 ( x x0 )( x x1 )a2
( x x0 )( x x1 ) ( x xn 1 )an
Let

a0 f 0 f [ x0 ]

then Pn ( x0 ) a0 f 0

a1 f [ x0 , x1 ]
then Pn ( x1 ) a0 ( x1 x0 )a1

Let

f 0 f1
f 0 ( x1 x0 )
f1
x0 x1
Numerical Methods Wen-Chieh Lin

19

Divided differences (cont.)

Let

then

f [ x1 , x2 ] f [ x0 , x1 ]
a2 f [ x0 , x1 , x2 ]
x2 x0

Pn ( x2 ) a0 ( x2 x0 )a1 ( x2 x0 )( x2 x1 )a2
f1 f 0
f 0 ( x2 x0 )
x1 x0
f 2 f1 f1 f 0

x2 x1 x1 x0
( x2 x0 )( x2 x1 )
x2 x0
f 2
Numerical Methods Wen-Chieh Lin

20

Divided differences (cont.)

One

can show in similar fashion that

Pn(xi) = fi if ai = f[x0, x1, , xi]

We

can then rewrite

Pn ( x ) a0 ( x x0 )a1 ( x x0 )( x x1 )a2

( x x0 )( x x1 ) ( x xn 1 )an
Pn ( x ) f [ x0 ] ( x x0 ) f [ x0 , x1 ]

( x x0 )( x x1 ) f [ x0 , x1 , x2 ]
( x x0 )( x x1 ) ( x xn 1 ) f [ x0 ,, xn ]
Numerical Methods Wen-Chieh Lin

21

Example: Divided Differences

Numerical Methods Wen-Chieh Lin

22

Example: Divided Differences (cont.)

Pn ( x ) f [ x0 ] ( x x0 ) f [ x0 , x1 ]
( x x0 )( x x1 ) f [ x0 , x1 , x2 ]
( x x0 )( x x1 ) ( x xn 1 ) f [ x0 ,, xn ]
Numerical Methods Wen-Chieh Lin

23

Example: Divided Differences (cont.)

If only first four points are used,

P3 ( x ) 22.0 ( x 3.2)
8.400
( x 3.2)( x 2.7)
2.856
( x 3.2)( x 2.7)( x 1.0)( 0.528)
Numerical Methods Wen-Chieh Lin

24

Example: Divided Differences (cont.)

If 5th point is added

P4 ( x ) P3 ( x )
( x 3.2)( x 2.7)( x 1.0)( x 4.8)(0.256)
Numerical Methods Wen-Chieh Lin

25

Example: Divided Differences

for a Polynomial

Numerical Methods Wen-Chieh Lin

26

Identical Polynomials
Every

polynomial of degree n that has the

same value at n+1 distinct data points is
exactly the same
Proof
Suppose

there are two different polynomials

Their difference D(x) = Pn(x) Qn(x) is a
polynomial of at most degree n
P and Q match at all n+1 points D(x) = 0 at
these n+1 points, i.e. D(x) has n+1 distinct zeros
However, D(x) is of at most degree n n zeros at
most
D(x) should be identically zero
Numerical Methods Wen-Chieh Lin

27

Different Forms of
Polynomial Interpolation
Polynomials

of degree n that interpolate n+1

data points are identical

Their

error terms are also identical

Only

need to derive the error term for one

form of interpolating polynomial
Numerical Methods Wen-Chieh Lin

28

Error EstimationNext Term Rule

It

can be shown in Chap 5 that

En(x) = the value of the next term that would be

added to Pn(x)

Numerical Methods Wen-Chieh Lin

29

Function Differences Method

If x-values of data points are evenly spaced,
divided differences method can be expressed as
Pn (x )

f [ x0 ] ( x x0 ) f [ x0 , x0 h ]
( x x0 )( x x0 h ) f [ x0 , x0 h, x0 2h ]
( x x0 )( x x0 h ) ( x x0 ( n 1)h ) f [ x0 ,, x0 nh ]

Numerical Methods Wen-Chieh Lin

30

Definition of Function Differences

First-order

differences
f i f i 1 f i
Second-order differences
2
f i ( f i ) ( f i 2 f i 1 ) ( f i 1 f i )

f i 2 2 f i 1 f i
Binomial coefficients
nth-order differences
n(n 1)
n
f i f i n nf i n 1
f i n 2 f i
2!
i 0,, ( N n )
Numerical Methods Wen-Chieh Lin

31

f1 f 0 f 0
f [ x0 , x0 h ]

h
h
f [ x0 , x0 h, x0 2h ]

f 2 f1
h

f1 f 0
h

2h

f0
2
2h

f 2 2 f1 f 0

2
2h

f [ x0 , x0 h, x0 2h, x0 3h ]

f 3 2 f 2 f1
2h2

f 2 2 f1 f 0
2h2

3h

3 f 0
3
3! h
n
f0
f [ x0 , x0 h,, x0 ( n 1)h ] n
Numerical Methods
n! h Wen-Chieh Lin

32

Function Differences (cont.)

Pn (x )
f [ x0 ] ( x x0 ) f [ x0 , x0 h ]
( x x0 )( x x0 h ) f [ x0 , x0 h, x0 2h ]
( x x0 )( x x0 h ) ( x x0 ( n 1)h ) f [ x0 ,, x0 nh ]

f 0
f 0 ( x x0 )

h
f0
( x x0 )( x x0 h )

2
2! h
2

n f 0
( x x0 )( x x0 h ) ( x x0 ( n 1)h )
n! h n
Numerical Methods Wen-Chieh Lin

33

Function Differences (cont.)

f 0

Pn (x ) f 0 ( x x0 )
h
2 f 0
( x x0 )( x x0 h )

2
2! h
n f 0
( x x0 )( x x0 h ) ( x x0 ( n 1)h )
n! h n
x x0
s
h
s( s 1) 2
s( s 1) ( s (n 1)) n
f 0 sf 0
f 0
f0
2!
n!
Numerical Methods Wen-Chieh Lin

34

Example: Function Difference

s( s 1) 2
s( s 1)( s 2) 3
P( x ) f 0 sf 0
f0
f 0
2!
3!
s( s 1)
s( s 1)( s 2)
x 0
3
0.25s
1 .5
1.5 0.25s 0.25

2
6
0.5
3

Numerical Methods Wen-Chieh Lin

35

Interpolating Data with a High-Degree

Polynomial is Bad!

Often causes undesirable wiggling in flat region!

Original Function

Fitting with P2(x)

Fitting with P4(x)

Fitting with P6(x)

Numerical Methods
Wen-Chieh
Fitting
with P (x) Lin
8

36

Remedy to High-Degree Polynomial

Break

up an interval into subintervals

Employ piecewise polynomial interpolation

Numerical Methods Wen-Chieh Lin

37

Spline Curves
Linear

spline

Slope is discontinuous!

Cubic

spline

Numerical Methods Wen-Chieh Lin

38

Cubic Spline
gi-1(x)

x0

x1

xi-1

gi(x)

xi

gi+1(x)

xi+1

knots

xn-1

xn

For each segment

gi ( x ) ai ( x xi ) bi ( x xi ) ci ( x xi ) d i
3

Unique solution exists if 4 constraints are specified

solving a system of 4 equations for 4 unknowns
Need additional constraints for g0(x) and gn-1(x)
Numerical Methods Wen-Chieh Lin

39

Cubic Spline
gi-1(x)

x0

x1

xi-1

gi(x)

xi

xi+1

gi+1(x)

knots

xn-1

xn

gi ( x ) ai ( x xi )3 bi ( x xi ) 2 ci ( x xi ) d i
gi ( xi ) yi
Continuous position
gi ( xi 1 ) gi 1 ( xi 1 )
g 'i ( xi 1 ) g 'i 1 ( xi 1 )
Continuous slope

Continuous curvature

g ' 'i ( xi 1 ) g ' 'i 1 ( xi 1 )

Numerical Methods Wen-Chieh Lin

40

Sketch for Computing Spline Coefficients

Define

hi xi 1 xi
Si g"i ( xi ), i 1,..., n 1
Sn g"n 1 ( xn )

We
ll compute

coefficients, ai, bi, ci, di by

Representing

ai, bi, ci, di as Si

Forming a matrix equation by stacking constraint
equations of all segments

HS Y
Numerical Methods Wen-Chieh Lin

41

Computing di , ai and bi
gi ( x ) ai ( x xi )3 bi ( x xi ) 2 ci ( x xi ) d i
From

gi ( xi ) yi , we know d i yi

From

gi ( xi 1 ) gi 1 ( xi 1 ) , we have

ai ( xi 1 xi ) bi ( xi 1 xi ) ci ( xi 1 xi ) d i yi 1
3

Using

hi xi 1 xi and d i yi , we obtain

ai hi bi hi ci hi yi yi 1
3

Numerical Methods Wen-Chieh Lin

42

Computing di , ai and bi (cont.)

gi ( x ) ai ( x xi ) bi ( x xi ) ci ( x xi ) d i
3

g"i ( x ) 6ai ( x xi ) 2bi ,

Si g"i ( xi ) 2bi

From

g"i ( xi 1 ) g"i 1 ( xi 1 ) , we get

Since

Si 1 g"i 1 ( xi 1 ) g"i ( xi 1 ) 6ai hi 2bi

Hence

Si
bi
2

Si 1 Si
ai
6hi
Numerical Methods Wen-Chieh Lin

43

Computing ci
Substitute

so far into
Then

the relations for ai, bi, ci, di obtained

ai hi bi hi ci hi yi yi 1
3

solve for ci
Si 1 Si 3 Si 2
hi hi ci hi yi yi 1
6hi
2

yi 1 yi 2hi Si hi Si 1
ci

hi
6
Numerical Methods Wen-Chieh Lin

44

Relating S and h
gi ( x ) ai ( x xi ) bi ( x xi ) ci ( x xi ) d i
2
g 'i ( x ) 3ai ( x xi ) 2bi ( x xi ) ci
3

From

g 'i ( xi 1 ) g 'i 1 ( xi 1 ) , we know that

g 'i 1 ( xi ) g 'i ( xi )

Plug

x = xi into gi
(x)

g 'i ( xi ) 3ai ( xi xi ) 2 2bi ( xi xi ) ci ci

Similarly,

we have

22
g 'i 1 ( xi))
33aai
(
(
x
x

x
x
)

22bibi1(1x( x
xi
x1i)1
)
ci
c1i 1
i
11
i
i
i1
1)
i

3ai 1hi21 2bi 1hi 1 ci 1

Numerical Methods Wen-Chieh Lin

45

Relating S and h (cont.)

Equating

results obtained from previous slide

ci 3ai 1hi21 2bi 1hi 1 ci 1

Substituting

for a, b, c, d their relationship in

terms of S and y

yi 1 yi yi yi 1

hi 1Si 1 ( 2hi 1 2hi ) Si hi Si 1 6

h
i 1
i

6f [ xi , xi 1 ] f [ xi 1 , xi ]
Numerical Methods Wen-Chieh Lin

46

Matrix for Computing Cubic Spline Coefficients

h
0

2(h0 h1 )
h1
h1
2(h1 h2 )
h2

h2
2( h2 h3 ) h3

f [ x , x ] f [ x , x ]
1
2
0
1

f [ x2 , x3 ] f [ x1 , x2 ]

6 f [ x3 , x4 ] f [ x2 , x3 ]

f
[
x
,
x
]

f
[
x
,
x
]
n 2
n 1
n 1 n

hn 2

S0

S2

S3

Sn 1
2( hn 2 hn 1 ) hn 1

Sn

hi xi 1 xi
Si g"i ( xi ), i 1,..., n 1
Sn g"n 1 ( xn )
Numerical Methods Wen-Chieh Lin

47