lecture 1a (MA 214)

interpolation theory

Harsha Hutridurga
(special thanks to vrinda jindal for the corrections)

IIT Bombay

Harsha Hutridurga (IIT Bombay) MA 214 1 / 24

 a problem from data analysis

♣ Take a physical experiment

I suppose a quantity f depends on a parameter x
I quantity f is measured for various values of x
x x0 x1 x2 x3 ... xn
f f0 f1 f2 f3 ... fn

I goal: Deduce an experimental law from these measurements

I Experimental law consistent with the measurements

Harsha Hutridurga (IIT Bombay) MA 214 2 / 24

 interpolation problem

♣ Given with
I n + 1 distinct real points: x0 , x1 , . . . , xn
I real numbers: f0 , f1 , . . . , fn
♣ Find
I a function ψ : R → R such that

ψ(xi ) = fi for i = 0, 1, . . . , n.

♣ Such a function ψ is called an interpolant

♣ There can be many interpolants depending on the function-type.

Harsha Hutridurga (IIT Bombay) MA 214 3 / 24

 classes of iterpolants

♣ polynomial interpolation:

ψ(x) = a0 + a1 x + a2 x2 + · · · + an xn

with a0 , a1 , . . . , an ∈ R

♣ trigonometric interpolation:
n
X n
X
ψ(x) = a0 + ak cos(kx) + bk sin(kx)
k=1 k=1

with a0 , a1 , . . . , an , b1 , . . . , bn ∈ R

Harsha Hutridurga (IIT Bombay) MA 214 4 / 24

 classes of iterpolants (contd.)

♣ rational interpolation:

a0 + a1 x + a2 x2 + · · · + an xn
ψ(x) =
b0 + b1 x + b2 x2 + · · · + bm xm
with a0 , a1 , . . . , an , b0 , b1 , . . . , bm ∈ R

♣ exponential interpolation:
n
X
ψ(x) = ak eλk x
k=0

with a0 , a1 , . . . , an , λ0 , λ1 , . . . , λn ∈ R

Harsha Hutridurga (IIT Bombay) MA 214 5 / 24

 polynomial interpolation

Q: Given a set of data points,

is it always possible to find a polynomial interpolant?

sets of polynomials

♣ A polynomial is a function of the form

p(x) = a0 + a1 x + a2 x2 + · · · + an xn

Here a0 , a1 , . . . , an ∈ R are called coefficients

and the integer n ≥ 0 is called the degree of p(x) (if an 6= 0)
♣ Let Pn be the set of all polynomials of degree n and smaller
I P0 has all constant functions
I P1 has all constant functions and all linear functions
I P2 has all constant, linear and quadratic functions

Harsha Hutridurga (IIT Bombay) MA 214 6 / 24

 some warm-up

Q: Given a set of data points,

is it always possible to find a polynomial interpolant?
I If only one data point was given, say (x0 , f0 ) ∈ R2 , then take
p(x) = f0 for all x ∈ R
Note that p(x) ∈ P0 and is an interpolant
I The case of two data points: (x0 , f0 ) and (x1 , f1 ).
Take the unique line connecting the two:
 
f1 − f0
y(x) = f0 + (x − x0 )
x1 − x0
Note that y(x) ∈ P1 and is an interpolant
Q: In the case of three data points,
do we get a polynomial interpolant in P2 ?
Q: If yes, will that polynomial be unique?
Harsha Hutridurga (IIT Bombay) MA 214 7 / 24
 main result

Theorem (Joseph-Louis Lagrange)

Take n + 1 distinct points x0 , x1 , . . . , xn ∈ R
and take real numbers f0 , f1 , . . . , fn .
Then there exists a unique polynomial p ∈ Pn such that

p(xi ) = fi for i = 0, 1, . . . , n.

♣ The proof involves

I a construction to build an interpolant p ∈ Pn
I showing uniqueness by fundamental theorem of algebra

♣ Terminology: The points xi are called interpolation points

Harsha Hutridurga (IIT Bombay) MA 214 8 / 24

 fundamental theorem of algebra

♣ Fundamental theorem of algebra is about roots of polynomials

♣ i.e. all those x such that p(x) = 0
♣ The theorem essentially says that
I a polynomial of degree n can have at most n real zeroes
I if complex zeroes are counted (counted with multiplicity),
it also says that degree n polynomial has exactly n zeroes
√
♣ Take f (x) = 1 + x2 with only complex roots i = −1 and −i

Harsha Hutridurga (IIT Bombay) MA 214 9 / 24

 uniqueness of the polynomial interpolant

♣ Suppose there are two polynomial interpolants: p(x), q(x) ∈ Pn

I Observe that the new function w(x) := p(x) − q(x) ∈ Pn
I Observe also that

w(xi ) = p(xi ) − q(xi ) = 0 for i = 0, 1, . . . , n

i.e. w(x) has n+1 distinct real zeros

I This contradicts the fundamental theorem of algebra
I Hence w(x) = 0 for all x, i.e. p(x) = q(x)
I We have, thus, proved uniqueness
♣ Some comments:
I We have not shown the existence of an interpolant in Pn yet
I We have essentially demonstrated the following:
F If there is an interpolant in Pn , then it has to be unique

Harsha Hutridurga (IIT Bombay) MA 214 10 / 24

 existence of the polynomial interpolant

♣ Recall that we have n + 1 pairs: (xi , fi ) for i = 0, 1, . . . , n

♣ Idea due to Lagrange – lagrange’s interpolation formula
I Construct nth -degree polynomials Ln0 (x), Ln1 (x), . . . , Lnn (x) such that
(
n 1 if i = k
Lk (xi ) =
0 if i 6= k

I Construct a polynomial p ∈ Pn as
n
X
p(x) := fk Lnk (x)
k=0

♣ Note that p(xi ) = fi for i = 0, 1, . . . , n

♣ We have, thus, found the polynomial interpolant p in Pn

Harsha Hutridurga (IIT Bombay) MA 214 11 / 24

 lagrange polynomials Lnk (x)

♣ We need nth -degree polynomials Lnk (x) such that

(
n 1 if i = k
Lk (xi ) =
0 if i 6= k

♣ Take
(x − x0 ) . . . (x − xk−1 )(x − xk+1 ) . . . (x − xn )
Lnk (x) :=
(xk − x0 ) . . . (xk − xk−1 )(xk − xk+1 ) . . . (xk − xn )

I These polynomials do satisfy the above condition

I Written in compact form
n
Y (x − xj )
Lnk (x) =
(xk − xj )
j=0, j6=k

♣ Observe that Lnk (x) are all nth -degree polynomials

Harsha Hutridurga (IIT Bombay) MA 214 12 / 24
 polynomial interpolation – recipe of lagrange

♣ Start with the given n + 1 pairs

(x0 , f0 ), (x1 , f1 ), . . . , (xn , fn )

♣ Build the n + 1 Lagrange polynomials, each of degree n:

n
Y (x − xj )
Lnk (x) = for k = 0, 1, . . . , n
(xk − xj )
j=0, j6=k

♣ Construct the polynomial interpolant p ∈ Pn as a weighted sum:

n
X
p(x) = fk Lnk (x)
k=0

Harsha Hutridurga (IIT Bombay) MA 214 13 / 24

 an example with three data points

♣ Given three data points

 
1 1
(−1, 1), (0, 0), ,
2 4
♣ Lagrange polynomials
(x − 0)(x − 12 )
 
2 1
L20 (x) = = x x−
(−1 − 0)(−1 − 12 ) 3 2
(x + 1)(x − 12 )
 
2 1
L1 (x) = = −2(x − 1) x −
(0 + 1)(0 − 12 ) 2
(x + 1)(x − 0) 4
L22 (x) = 1 1 = x (x + 1)
( 2 + 1)( 2 − 0) 3
♣ Polynomial interpolant:
       
2 1 1 1 4
p(x) = 1 x x − + 0 −2(x − 1) x − + x (x + 1)
3 2 2 4 3
= x2
Harsha Hutridurga (IIT Bombay) MA 214 14 / 24
 an example with three data points (contd.)

♣ Plotting the Lagrange polynomials and the interpolant

I Interpolant p(x) = x2

Harsha Hutridurga (IIT Bombay) MA 214 15 / 24

 an example with four data points

♣ We add another point to previous data set:

 
1 1
(−1, 1), (0, 0), , , (1, 1)
2 4

♣ Observe that all four points lie on the same curve y = x2

Q: Will the recipe of Lagrange yield p(x) = x2 ?
Q: Will the recipe yield a polynomial of degree 3?

Harsha Hutridurga (IIT Bombay) MA 214 16 / 24

 an example with four data points (contd.)

xi −1 0 0.5 1
fi 1 0 0.25 1

♣ Lagrange polynomials

(x − 0)(x − 12 )(x − 1) x3 x2 x
L30 (x) = = − + −
(−1 − 0)(−1 − 12 )(−1 − 1) 3 2 6
(x + 1)(x − 12 )(x − 1)
L31 (x) = 1 = 2x3 − x2 − 2x + 1
(0 + 1)(0 − 2 )(0 − 1)
(x + 1)(x − 0)(x − 1) 8x3 8x
L32 (x) = = − +
( 21 + 1)( 12 − 0)( 12 − 1) 3 3
(x + 1)(x − 0)(x − 12 ) x2 x
L33 (x) = = x 3
+ −
(1 + 1)(1 − 0)(1 − 12 ) 2 2

Harsha Hutridurga (IIT Bombay) MA 214 17 / 24

 an example with four data points (contd.)

♣ Observe that all the Lagrange polynomials are of degree 3

♣ Constructing the interpolant:
 3
x2 x

x
+ 0 2x3 − x2 − 2x + 1
 
p(x) = 1 − + −
3 2 6
8x3 8x x2 x
   
1 3
+ − + +1 x + −
4 3 3 2 2

= x2

♣ Note that the fourth data point turns out to be redundant

Harsha Hutridurga (IIT Bombay) MA 214 18 / 24

 an example with four data points (contd.)

♣ Plotting the Lagrange polynomials and the interpolant

I Interpolant p(x) = x2

Harsha Hutridurga (IIT Bombay) MA 214 19 / 24

 another example – perturbed data
♣ We are going to slightly perturb the data in previous example
I Data from previous example
xi −1 0 0.5 1
fi 1 0 0.25 1
I We will slightly perturb the data
xi −1 0 0.5 1
fi 1 0 0.25 1.1
♣ Observe that interpolation points are same for both data sets
♣ Hence Lagrange polynomials L3k (x) are same for both data sets
3
Y (x − xj )
L3k (x) =
(xk − xj )
j=0, j6=k

♣ Polynomial interpolant for perturbed data set

3
X x3 21x2 x
p(x) = fk L3k (x) = + −
10 20 20
k=0

Harsha Hutridurga (IIT Bombay) MA 214 20 / 24

 another example – perturbed data (contd.)

♣ Plotting the Lagrange polynomials and the interpolant

x3 21x2 x
I Interpolant p(x) = 10 + 20 − 20

Harsha Hutridurga (IIT Bombay) MA 214 21 / 24

 story so far

♣ Given n + 1 pairs (x0 , f0 ), (x1 , f1 ), . . . , (xn , fn )

I there exists a unique p ∈ Pn such that

p(xi ) = fi

♣ Recipe of Lagrange
n n
X Y (x − xj )
p(x) = fk Lnk (x) where Lnk (x) =
(xk − xj )
k=0 j=0, j6=k

Harsha Hutridurga (IIT Bombay) MA 214 22 / 24

 try it out

♣ Find a polynomial interpolant for the data points

xi −2 −1 0 1 2
fi 28 18 10 4 0
♣ Suppose a nth -degree polynomial passes through n + 1 data points.

I Does there exist a (n + 1)th -degree polynomial passing through

same data points?
I Lagrange’s Zero One Blocks
♣ Does the ordering of x0 , x1 , . . . , xn matter in Lagrange’s recipe?

Harsha Hutridurga (IIT Bombay) MA 214 23 / 24

 thank you for your attention

dedicated email address for this course

[email protected]

Harsha Hutridurga (IIT Bombay) MA 214 24 / 24