Linear Algebra, Numerical and Complex Analysis

LINEAR ALGEBRA, NUMERICAL AND COMPLEX ANALYSIS
MA11004
SECTIONS 1 and 2
Dr. Jitendra Kumar
Professor
Department of Mathematics
Indian Institute of Technology Kharagpur
West Bengal 721302, India
Webpage: http://www.facweb.iitkgp.ac.in/~jkumar/
Polynomial Interpolation
 Polynomial Interpolation
 Existence and Uniqueness
 Error in Interpolating Polynomials
Dr. Jitendra Kumar – IIT KHARAGPUR 2

Interpolation
Interpolation is a process of estimating values between known data points or approximating complicated
functions by simple polynomials or determining a polynomial that fits a set of given points.
Applications :
1. Constructing the function when it is not given explicitly and only the values of function are given at
some points.
2. Replacing complicated function by an interpolating simpler function (usually polynomials)

so that many operations such as determination of roots, differentiation, integrations or
other such operations may be performed.

Fundamental principle behind polynomial interpolation
Weirstrass Approximation Theorem
Suppose that 𝑓 is defined and continuous on [𝑎, 𝑏]. For each 𝜖 > 0, there exists a
polynomial 𝑃(𝑥) with the property that
𝑓 𝑥 −𝑃 𝑥 < 𝜖; ∀ 𝑥 ∈ 𝑎, 𝑏 .

Why not Taylor’s Polynomial ? Consider Taylor’s Polynomial of 𝑒 𝑥 around 𝑥 = 0.
25
𝑥2 𝑥3 𝑥4 𝑥5
𝑃5 𝑥 = 1 + 𝑥 + + + +
20
20 𝑥 2 6 24 120
𝑥𝑒𝑥
𝑒
𝑒𝑥 𝑥2 𝑥3 𝑥4
15
15 𝑒 𝑃5 𝑃4 𝑥 = 1 + 𝑥 + + +
2 6 24
10
10
𝑃𝑃33
5 𝑃1 𝑥2 𝑥2 𝑥3
𝑃
𝑃2 𝑥 = 1 + 𝑥 + 𝑃3 𝑥 = 1 + 𝑥 + +
5 𝑃𝑃
000 𝑃1 2 2 6
0 𝑃0
0 𝑃 1 𝑥
-2 -1.5 = 1 +0 𝑥0.5 1
-1 -0.5 1.5 2 2.5 3
𝑃0 𝑥 = 1; 𝑃1 𝑥 = 1 + 𝑥
-5 -2 -1.5𝑃1-1 𝑥-0.5= 0 1 0.5
+ 𝑥1 1.5 2 2.5 3
Taylor’s polynomials agree as closely as possible with a given function at a specific point,
so they concentrate their accuracy near that point.
For ordinary computation purposes it is more efficient to use methods that include
information at various points.
Existence and uniqueness for polynomial interpolation
For (𝑛 + 1) data points there is one and only one polynomial of order ≤ 𝑛 that passes through all the points.
For example, there is only one straight line (a first order polynomial) that passes through two points.
Consider for simplicity, a second order polynomial
𝑓 𝑥 = 𝑎0 + 𝑎1 𝑥 + 𝑎2 𝑥 2
A straight forward method for computing the coefficients of a polynomial of degree 𝑛 is

based on the fact that (𝑛 + 1) data points are required (3 data points in this example)
to determine (𝑛 + 1) unknowns (3 unknowns 𝑎0 , 𝑎1 , 𝑎2 in this example)

Polynomial to be fitted with the given data 𝑓 𝑥 = 𝑎0 + 𝑎1 𝑥 + 𝑎2 𝑥 2
Suppose that there are 3 given data points 𝑥0 , 𝑓 𝑥0 , 𝑥1 , 𝑓 𝑥1 , 𝑥2 , 𝑓 𝑥2 .
If the given polynomial passes through the given data points then it must satisfy them, i.e.,
𝑓 𝑥0 = 𝑎0 + 𝑎1 𝑥0 + 𝑎2 𝑥02
1 𝑥0 𝑥02 𝑎0 𝑓(𝑥0 )
𝑓 𝑥1 = 𝑎0 + 𝑎1 𝑥1 + 𝑎2 𝑥12 ⟺ 1 𝑥1 𝑥12 𝑎1 = 𝑓(𝑥1 )
1 𝑥2 𝑥22 𝑎2 𝑓(𝑥2 )
𝑓 𝑥2 = 𝑎0 + 𝑎1 𝑥2 + 𝑎2 𝑥22

1 𝑥0 𝑥02 𝑎0 𝑓(𝑥0 )
1 𝑥1 𝑥12 𝑎1 = 𝑓(𝑥1 )
1 𝑥2 𝑥22 𝑎2 𝑓(𝑥2 )
This system of equations has a unique solution as
1 𝑥0 𝑥02
Det 1 𝑥1 𝑥12 = (𝑥0 − 𝑥1 )(𝑥1 − 𝑥2 )(𝑥2 − 𝑥0 ) ≠ 0 if 𝑥0 , 𝑥1 , 𝑥2 are distinct.
1 𝑥2 𝑥22
In practice, it is observed that the above system of equations is ill-conditioned.
Whether they are solved with an elimination method or with a more efficient algorithm,
the resulting coefficient can be highly inaccurate, in particular for, large 𝑛.
Therefore, we have some mathematical formats (interpolating formats) in which

such calculation can be avoided.

Error in Interpolating Polynomials
Let 𝑥0 , 𝑥1 , … , 𝑥𝑛 be (𝑛 + 1) points and let 𝑥 be a point belonging to the domain of a given function 𝑓.
Assume that 𝑓 ∈ 𝐶 𝑛+1 𝐼𝑥 , where 𝐼𝑥 is the smallest interval containing the nodes 𝑥0 , 𝑥1 , … , 𝑥𝑛 and 𝑥.
Then the interpolation error at the point 𝑥 is given by
𝜉 𝑓 𝑛+1
𝐸𝑛 𝑥 = 𝑓 𝑥 − 𝑃𝑛 𝑥 = 𝑥 − 𝑥0 𝑥 − 𝑥1 ⋯ (𝑥 − 𝑥𝑛 ) where 𝜉 ∈ 𝐼𝑥 .
𝑛+1 !

Error in Interpolating Polynomials
𝑓 𝑛+1𝜉
𝐸𝑛 𝑥 = 𝑓 𝑥 − 𝑃𝑛 𝑥 = 𝑥 − 𝑥0 𝑥 − 𝑥1 ⋯ (𝑥 − 𝑥𝑛 ) where 𝜉 ∈ 𝐼𝑥 .
𝑛+1 !
Proof : Note that the result is obviously true if 𝑥 coincides with any of the interpolating nodes.
For simplicity, let us assume 𝑤𝑛+1 𝑥 = 𝑥 − 𝑥0 𝑥 − 𝑥1 ⋯ (𝑥 − 𝑥𝑛 )
Now, define for any 𝑥 ∈ 𝐼𝑥 , the function
𝑤𝑛+1 𝑡 𝐸𝑛 𝑥
𝐺 𝑡 = 𝐸𝑛 𝑡 − ; 𝑡 ∈ 𝐼𝑥
𝑤𝑛+1 𝑥
Since 𝑓 ∈ 𝐶 𝑛+1 (𝐼𝑥 ) and 𝑤𝑛+1 is a polynomial, then 𝐺 ∈ 𝐶 𝑛+1 (𝐼𝑥 ).

𝑤𝑛+1 𝑡 𝐸𝑛 𝑥 𝐸𝑛 𝑥 = 𝑓 𝑥 − 𝑃𝑛 𝑥
𝐺 𝑡 = 𝐸𝑛 𝑡 − ; 𝑡 ∈ 𝐼𝑥
𝑤𝑛+1 𝑥
𝑤𝑛+1 𝑥 = 𝑥 − 𝑥0 𝑥 − 𝑥1 ⋯ (𝑥 − 𝑥𝑛 )
Note that 𝐺(𝑡) has (𝑛 + 2) distinct zeros in 𝐼𝑥 since
𝑤𝑛+1 𝑥𝑖 𝐸𝑛 𝑥
𝐺 𝑥𝑖 = 𝐸𝑛 𝑥𝑖 − = 0; 𝑖 = 0, 1, 2, … , 𝑛.
𝑤𝑛+1 𝑥
𝑤𝑛+1 𝑥 𝐸𝑛 𝑥
𝐺 𝑥 = 𝐸𝑛 𝑥 − =0
𝑤𝑛+1 𝑥
Then using Rolle’s theorem, 𝐺 ′ has atleast (𝑛 + 1) distinct zeros.
By recursion it follows that 𝐺 (𝑗) admits at least 𝑛 + 2 − 𝑗 distinct zeros.
𝑛+1 𝑛+1
⇒𝐺 has atleast one zero, which we denote by 𝜉, i.e., 𝐺 𝜉 =0

𝑤𝑛+1 𝑡 𝐸𝑛 𝑥 𝑤𝑛+1 𝑥 = 𝑥 − 𝑥0 𝑥 − 𝑥1 ⋯ (𝑥 − 𝑥𝑛 )
𝐺 𝑡 = 𝐸𝑛 𝑡 − ; 𝑡 ∈ 𝐼𝑥
𝑤𝑛+1 𝑥 𝑛+1
𝐺 𝜉 =0
𝑛+1
𝑛+1 (𝑛+1) 𝑤𝑛+1 𝑡 𝐸𝑛 𝑥
⟹𝐺 𝑡 = 𝐸𝑛 𝑡 −
𝑤𝑛+1 𝑥
Note that 𝐸𝑛 𝑡 = 𝑓 𝑡 − 𝑃𝑛 𝑡 ⟹ 𝐸𝑛𝑛+1 𝑡 = 𝑓 𝑛+1

𝑡 as 𝑃𝑛
𝑛+1
𝑡 =0
𝑛+1 𝑛+1
𝑤𝑛+1 𝑡 = 𝑛+1 ! & 𝐺 𝜉 =0
𝑛+1
𝑛 + 1 ! 𝐸𝑛 (𝑥) 𝑓 𝑛+1 𝜉
⟹ 0=𝑓 𝜉 − ⟹ 𝐸𝑛 𝑥 = 𝑤 (𝑥)
𝑤𝑛+1 (𝑥) 𝑛 + 1 ! 𝑛+1

Polynomial interpolation is the method of determining a polynomial that fits a set of given points

For (𝑛 + 1) data points there is one and only one polynomial of order ≤ 𝑛 that passes through
all the points.

𝜉 𝑓 𝑛+1
𝐸𝑛 𝑥 = 𝑓 𝑥 − 𝑃𝑛 𝑥 = 𝑥 − 𝑥0 𝑥 − 𝑥1 … (𝑥 − 𝑥𝑛 )
𝑛+1 !

 Forward and Backward Difference Table
 Newton’s Forward Interpolation Formula
 Newton’s Backward Interpolation Formula

RECALL: Previous Lecture
Polynomial interpolation is the method of determining a polynomial that fits a set of given points

For (𝑛 + 1) data points there is one and only one polynomial of order ≤ 𝑛 that passes through
all the points.

𝑛+1
𝜉 𝑓
𝐸𝑛 𝑥 = 𝑓 𝑥 − 𝑃𝑛 𝑥 = 𝑥 − 𝑥0 𝑥 − 𝑥1 ⋯ (𝑥 − 𝑥𝑛 )
𝑛+1 !

Different Methods of Determining Interpolating Polynomials
Although there is a unique 𝑛th order polynomial that fits (𝑛 + 1) data points, there are a variety of
mathematical formats in which this polynomial can be expressed.
 Newton’s forward and backward interpolating polynomial
 Newton’s Divided Difference Formula
 Lagrange Interpolation Formula

Newton’s Forward and Backward Interpolation Formula
Let the tabular points 𝑥0 , 𝑥1 , … , 𝑥𝑛 be equally spaced, i.e., 𝑥𝑖 = 𝑥0 + 𝑖ℎ, 𝑖 = 0, 1, … , 𝑛
Finite Difference Operator
 The Shift operator: 𝐸𝑓 𝑥𝑖 = 𝑓 𝑥𝑖 + ℎ
 The Forward difference operator: ∆𝑓 𝑥𝑖 = 𝑓 𝑥𝑖 + ℎ − 𝑓 𝑥𝑖
 The Backward difference operator: 𝛻𝑓 𝑥𝑖 = 𝑓 𝑥𝑖 − 𝑓 𝑥𝑖 − ℎ
ℎ ℎ
 The Central difference operator: 𝛿𝑓 𝑥𝑖 = 𝑓 𝑥𝑖 + − 𝑓 𝑥𝑖 −
2 2

Newton’s Forward and Backward Interpolation Formula
ℎ
Let 𝑓𝑖 = 𝑓 𝑥𝑖 𝑓𝑖+1 = 𝑓 𝑥𝑖 + ℎ 𝑓 1 = 𝑓 𝑥𝑖 +
𝑖+2 2
It can easily be verified that
∆𝑓𝑖 = 𝛻𝑓𝑖+1 = 𝛿𝑓 1
𝑖+2
Also, note that
∆≡𝐸−1 ∆𝑓𝑖 = 𝐸𝑓𝑖 − 𝑓𝑖
𝛻 ≡ 1 − 𝐸 −1 𝛻𝑓𝑖 = 𝑓𝑖 − 𝐸 −1 𝑓𝑖
1 1 1 1
− −2
𝛿 ≡𝐸 −𝐸2 2 𝛿𝑓𝑖 = 𝐸 2 𝑓𝑖 −𝐸 𝑓𝑖

Higher Order Differences
∆2 𝑓 𝑥𝑖 = ∆ ∆𝑓 𝑥𝑖
= ∆ 𝑓𝑖+1 − 𝑓𝑖
= 𝑓𝑖+2 − 𝑓𝑖+1 + 𝑓𝑖+1 − 𝑓𝑖
= 𝑓𝑖+2 − 2𝑓𝑖+1 + 𝑓𝑖
Similarly, 𝛻 2 𝑓𝑖 = 𝛻 𝛻𝑓𝑖 = 𝛻 𝑓𝑖 − 𝑓𝑖−1
= 𝑓𝑖 − 𝑓𝑖−1 − 𝑓𝑖−1 − 𝑓𝑖−2
= 𝑓𝑖 − 2𝑓𝑖−1 + 𝑓𝑖−2

Forward Difference Table Backward Difference Table
𝑥 𝑓(𝑥) ∆𝑓 ∆2 𝑓 ∆3 𝑓 𝑥 𝑓(𝑥) 𝛻𝑓 𝛻2𝑓 𝛻3𝑓
𝑥0 𝑓0 𝑥0 𝑓0 𝛻𝑓1
∆𝑓0 2
∆ 𝑓0 𝛻 2 𝑓2
𝑥1 𝑓1 ∆3 𝑓0 𝑥1 𝑓1 𝛻 3 𝑓3
∆𝑓1 𝛻𝑓2
∆2 𝑓1 𝛻 2 𝑓3
𝑥2 𝑓2 𝑥2 𝑓2 𝛻𝑓3
∆𝑓2
𝑥3 𝑓3 𝑥3 𝑓3
∆𝑓𝑖 = 𝑓𝑖+1 − 𝑓𝑖 𝛻𝑓𝑖 = 𝑓𝑖 − 𝑓𝑖−1
∆2 𝑓𝑖 = ∆𝑓𝑖+1 − ∆𝑓𝑖 𝛻 2 𝑓𝑖 = 𝛻𝑓𝑖 − 𝛻𝑓𝑖−1
∆3 𝑓𝑖 = ∆2 𝑓𝑖+1 − ∆2 𝑓𝑖 𝛻 3 𝑓𝑖 = 𝛻 2 𝑓𝑖 − 𝛻 2 𝑓𝑖−1

Difference Table – Numerical Example
2
𝑥 𝑓(𝑥) ∆Τ𝛻 ∆2 Τ𝛻
Δ𝑓0 = 1 Δ2 𝑓0 = −2
0 1
1
1 2 −2
−1
2 1 𝛻𝑓2 = −1 𝛻 2 𝑓2 = −2

Newton’s Forward Difference Formula
1. Linear Interpolation : The simplest way to connect two data points with a straight line.
Given that : 𝑥0 𝑥1
approximating
𝑓 𝑥0 𝑓 𝑥1 polynomial
Consider a general equation of straight line

𝑃1 𝑥 = 𝑏0 + 𝑏1 𝑥 − 𝑥0 𝑓(𝑥1 ) 𝑃1 (𝑥)
At the point 𝑥 = 𝑥0 : 𝑓(𝑥0 )
𝑃1 𝑥0 = 𝑓 𝑥0 = 𝑏0
actual
At the point 𝑥 = 𝑥1 : function
𝑃1 𝑥1 = 𝑓 𝑥1 = 𝑏0 + 𝑏1 𝑥1 − 𝑥0
𝑥0 𝑥1
𝑓 𝑥1 − 𝑓(𝑥0 )
⇒ 𝑏1 =
𝑥1 − 𝑥0

𝑃1 𝑥 = 𝑏0 + 𝑏1 𝑥 − 𝑥0
𝑏0 = 𝑓 𝑥0
𝑓 𝑥1 − 𝑓(𝑥0 )
𝑏1 =
𝑥1 − 𝑥0
Let us consider the equidistant data points, then
𝑓 𝑥0 + ℎ − 𝑓(𝑥0 ) ∆𝑓0
𝑏1 = =
ℎ ℎ
∆𝑓0
Interpolating Polynomial 𝑃1 𝑥 = 𝑓 𝑥0 + (𝑥 − 𝑥0 )
ℎ

2. Quadratic Interpolation :
𝑥0 𝑥1 𝑥2
Suppose 3 data points are given:
𝑓 𝑥0 𝑓 𝑥1 𝑓 𝑥2
Consider a second order polynomial
𝑃2 𝑥 = 𝑏0 + 𝑏1 𝑥 − 𝑥0 + 𝑏2 𝑥 − 𝑥0 𝑥 − 𝑥1
At the point 𝑥 = 𝑥0 : 𝑏0 = 𝑓(𝑥0 )
𝑓 𝑥1 − 𝑓(𝑥0 )
At the point 𝑥 = 𝑥1 : 𝑏1 =
𝑥1 − 𝑥0
∆𝑓0
In the case of equidistant data points : 𝑏1 =
ℎ

∆𝑓0
𝑃2 𝑥 = 𝑏0 + 𝑏1 𝑥 − 𝑥0 + 𝑏2 𝑥 − 𝑥0 𝑥 − 𝑥1 𝑏0 = 𝑓(𝑥0 ) 𝑏1 =
ℎ
Δ𝑓0
At the point 𝑥 = 𝑥2 : 𝑓 𝑥2 = 𝑓 𝑥0 + 𝑥2 − 𝑥0 + 𝑏2 𝑥2 − 𝑥0 𝑥2 − 𝑥1
ℎ
⇒ 𝑓 𝑥2 = 𝑓 𝑥0 + 2𝑓 𝑥1 − 2𝑓 𝑥0 + 2! ℎ2 𝑏2
𝑓 𝑥2 − 2𝑓 𝑥1 + 𝑓(𝑥0 ) ∆2 𝑓0
⇒ = 𝑏2 ⇒ 𝑏2 =
2! ℎ2 2! ℎ2
Interpolating polynomial :
∆𝑓0 ∆2 𝑓0
𝑃2 𝑥 = 𝑓 𝑥0 + 𝑥 − 𝑥0 + (𝑥 − 𝑥0 )(𝑥 − 𝑥1 )
ℎ 2! ℎ2

∆𝑓0 ∆𝑓0 ∆2 𝑓0
𝑃1 𝑥 = 𝑓 𝑥0 + (𝑥 − 𝑥0 ) 𝑃2 𝑥 = 𝑓 𝑥0 + 𝑥 − 𝑥0 + (𝑥 − 𝑥0 )(𝑥 − 𝑥1 )
ℎ ℎ 2! ℎ2
Generalized Formula :
We can now write the Newton’s forward difference formula based on (𝑛 + 1) nodal points 𝑥0 , 𝑥1 , … , 𝑥𝑛 as:
∆𝑓0 ∆2 𝑓0 ∆𝑛 𝑓0
𝑃𝑛 𝑥 = 𝑓 𝑥0 + 𝑥 − 𝑥0 + 𝑥 − 𝑥0 𝑥 − 𝑥1 2
+ ⋯ + 𝑥 − 𝑥0 𝑥 − 𝑥1 𝑥 − 𝑥𝑛−1
ℎ 2! ℎ 𝑛! ℎ𝑛
𝑥 − 𝑥0
If we put = 𝑢 then it takes the following form:
ℎ
𝑢(𝑢 − 1) 2 𝑢 𝑢−1 ⋯ 𝑢−𝑛+1 𝑛

𝑃𝑛 𝑥0 + ℎ𝑢 = 𝑓0 + 𝑢∆𝑓0 + ∆ 𝑓0 + ⋯ + ∆ 𝑓0
2! 𝑛!

Alternative Derivation :
𝑥 − 𝑥0
𝑓 𝑥 = 𝑓 𝑥0 + ℎ = 𝑓 𝑥0 + 𝑢ℎ
ℎ
= 𝐸 𝑢 𝑓 𝑥0
= 1 + ∆ 𝑢 𝑓 𝑥0
𝑢(𝑢 − 1) 2 𝑢 𝑢−1 ⋯ 𝑢−𝑛+1 𝑛

= 𝑓0 + 𝑢 ∆𝑓0 + ∆ 𝑓0 + ⋯ + ∆ 𝑓0 + ⋯
2! 𝑛!
Neglecting the difference ∆𝑛+1 𝑓0 and higher order differences , we get the above generalized
formula.

Newton’s Backward Difference Formula :
𝛻𝑓𝑛 (𝑥 − 𝑥𝑛 )(𝑥 − 𝑥𝑛−1 ) 2 𝑥 − 𝑥𝑛 𝑥 − 𝑥𝑛−1 … (𝑥 − 𝑥1 ) 𝑛

𝑃𝑛 𝑥 = 𝑓𝑛 + 𝑥 − 𝑥𝑛 + 𝛻 𝑓𝑛 + ⋯ + 𝛻 𝑓𝑛
ℎ 2! ℎ2 𝑛! ℎ𝑛
𝑥 − 𝑥𝑛
𝑓 𝑥 = 𝑓 𝑥𝑛 + ℎ = 𝑓 𝑥𝑛 + ℎ𝑢 = 𝐸 𝑢 𝑓 𝑥𝑛 = 1 − 𝛻 −𝑢 𝑓 𝑥𝑛 𝛻 ≡ 1 − 𝐸 −1
ℎ
𝑢 𝑢+1 2 𝑢 𝑢+1 … 𝑢+𝑛−1 𝑛

= 𝑓 𝑥𝑛 + 𝑢𝛻𝑓 𝑥𝑛 + 𝛻 𝑓 𝑥𝑛 + ⋯ + 𝛻 𝑓 𝑥𝑛 + ⋯
2! 𝑛!
Neglecting the difference 𝛻 𝑛+1 𝑓(𝑥𝑛 ) and higher order differences, we get:
𝑢(𝑢 + 1) 2 𝑢 𝑢+1 … 𝑢+𝑛−1 𝑛

𝑃𝑛 𝑥𝑛 + ℎ𝑢 = 𝑓𝑛 + 𝑢𝛻𝑓𝑛 + 𝛻 𝑓𝑛 + ⋯ + 𝛻 𝑓𝑛
2! 𝑛!

∆𝑓0 ∆2 𝑓0 ∆𝑛 𝑓0
𝑃𝑛 𝑥 = 𝑓 𝑥0 + 𝑥 − 𝑥0 + 𝑥 − 𝑥0 𝑥 − 𝑥1 + ⋯ + 𝑥 − 𝑥0 𝑥 − 𝑥1 𝑥 − 𝑥𝑛−1
ℎ 2! ℎ2 𝑛! ℎ𝑛
𝛻𝑓𝑛 𝛻 2 𝑓𝑛
𝑃𝑛 𝑥 = 𝑓𝑛 + 𝑥 − 𝑥𝑛 + 𝑥 − 𝑥𝑛 𝑥 − 𝑥𝑛−1 2
+⋯
ℎ 2! ℎ
𝛻 𝑛 𝑓𝑛
+ 𝑥 − 𝑥𝑛 𝑥 − 𝑥𝑛−1 … 𝑥 − 𝑥1
𝑛! ℎ𝑛

Example : Using Newton forward and backward interpolation formula, find the quadratic polynomial
which takes the following values
𝑥 0 1 2
𝑓(𝑥) 1 2 1
Solution : The difference table
𝑥 𝑓(𝑥) 𝛻𝑓/∆𝑓 𝛻 2 𝑓/∆2 𝑓
0 1
1 −2
1 2
−1
2 1

Newton’s Forward Formula: 𝑥 𝑓(𝑥) 𝛻𝑓/∆𝑓 𝛻 2 𝑓/∆2 𝑓
0 1
(𝑥 − 𝑥0 ) (𝑥 − 𝑥0 )(𝑥 − 𝑥1 ) 2 1 −2
1 2
𝑃2 𝑥 = 𝑓0 + ∆𝑓0 + ∆ 𝑓0 −1
ℎ 2! ℎ2 2 1
𝑥−0 𝑥−0 𝑥−1

=1+ .1 + ⋅ (−2)
1 2! ⋅ 1
= 1 + 𝑥 − 𝑥2 + 𝑥
= 1 + 2𝑥 − 𝑥 2

Newton’s Backward Formula: 𝑥 𝑓(𝑥) 𝛻𝑓/∆𝑓 𝛻 2 𝑓/∆2 𝑓
0 1
1 −2
(𝑥 − 𝑥2 ) (𝑥 − 𝑥2 )(𝑥 − 𝑥1 ) 2 1 2
𝑃2 𝑥 = 𝑓2 + 𝛻𝑓2 + 𝛻 𝑓2 −1
ℎ 2! ℎ2 2 1
𝑥−2 𝑥−2 𝑥−1

=1+ ⋅ −1 + ⋅ (−2)
1 2! ⋅ 1
= 1 − 𝑥 + 2 − (𝑥 2 − 3𝑥 + 2)
= 1 + 2𝑥 − 𝑥 2

Note : If we wish to add another data point, say 𝑥3 , 𝑓 𝑥3 = 3, 10 we need to add only another row to
the difference table in order to apply forward difference formula.
𝑥 𝑓(𝑥) 𝛻𝑓/∆𝑓 𝛻 2 𝑓/∆2 𝑓 ∆3 𝑓/𝛻 3 𝑓
0 1
1 −2
1 2 −1 12
2 1 10
9
3 10
𝑥−0 𝑥−1 𝑥−2

𝑃2 𝑥 = 1 + 2𝑥 − 𝑥2 + (12)
3!
= 1 + 2𝑥 − 𝑥 2 + 2𝑥 𝑥 2 − 3𝑥 + 2
= 2𝑥 3 − 7𝑥 2 + 6𝑥 + 1

Example : Construct Newton’s forward interpolation polynomial for the following table :
𝑥 4 6 8 10
𝑦 1 3 8 16
Hence evaluate the interpolating polynomial for 𝑥 = 5.
Solution : Difference table:
𝑥 𝑦 ∆𝑦 ∆2 𝑦 ∆3 𝑦
4 1
2 3 0
6 3
8 8 5 3
10 16 8

(𝑥 − 4) 𝑥−4 𝑥−6 𝑥 𝑦 ∆𝑦 ∆2 𝑦 ∆3 𝑦
𝑃 𝑥 =1+ ⋅2+ 2
⋅3+0
2 2! 2
4 1
2 3 0
6 3
3 2 8 8 5 3
= 1 + 𝑥 − 4 + 𝑥 − 10𝑥 + 24 10 16 8
8
11 3 2
=6− 𝑥+ 𝑥
4 8
11 3
𝑃 5 =6− × 5 + × 25 = 1.625
4 8
Note : In this example, given 4 data points we get only a second degree polynomial.

Example: Apply Newton’s backward difference formula to the data below:
𝑥 1 2 3 4 5
𝑦 1 −1 1 −1 1
Difference Table:
𝑥 𝑓(𝑥) 𝛻𝑓 𝛻2𝑓 𝛻 3 𝑓3 𝛻 4 𝑓3
1 1 −2
2 −1 4 −8
2 −4 16
3 1 −2 8
4
4 −1 2
5 1

Backward Difference Formula 𝑥 𝑓(𝑥) 𝛻𝑓 𝛻2𝑓 𝛻 3 𝑓3 𝛻 4 𝑓3
1 1 −2
𝛻𝑓4 (𝑥 − 𝑥4 )(𝑥 − 𝑥3 ) 2 2 −1 4 −8
𝑃4 𝑥 = 𝑓4 + 𝑥 − 𝑥4 + 2
𝛻 𝑓4 2 −4 16
ℎ 2! ℎ 3 1 −2 8
4
(𝑥 − 𝑥4 )(𝑥 − 𝑥3 )(𝑥 − 𝑥2 ) 3 4 −1 2
+ 3
𝛻 𝑓4 5 1
3! ℎ
(𝑥 − 𝑥4 )(𝑥 − 𝑥3 )(𝑥 − 𝑥2 )(𝑥 − 𝑥1 ) 4

+ 𝛻 𝑓4
4! ℎ4
(𝑥 − 5)(𝑥 − 4) (𝑥 − 5)(𝑥 − 4)(𝑥 − 3)
𝑃4 𝑥 = 1 + 𝑥 − 5 2 + 4+ 8
2 6
(𝑥 − 5)(𝑥 − 4)(𝑥 − 3)(𝑥 − 2)
+ 16
2 100 2 24
= 𝑥 4 − 8𝑥 3 + 𝑥 − 56𝑥 + 31
3 3

Example : Approximate the function 𝑒 𝑥 on the interval [0, 1] by using polynomial interpolation with
𝑥0 = 0, 𝑥1 = 0.5, 𝑥2 = 1. Let 𝑃2 (𝑥) denote the interpolation polynomial.
𝑎) Find the interpolating polynomial using any method.
𝑏) Find an upper bound for the error magnitude max |𝑒 𝑥 − 𝑃2 (𝑥)| .

0≤𝑥≤1
𝑐) Compare the actual error at different points of your choice with the error bound.

(𝑥 − 𝑥0 ) (𝑥 − 𝑥0 )(𝑥 − 𝑥1 ) 2
𝑎) Difference table : 𝑃2 𝑥 = 𝑓0 + ∆𝑓0 + ∆ 𝑓0
ℎ 2! ℎ2
𝑥 𝑓(𝑥) ∆𝑓 ∆2 𝑓
0 1
0.6487 0.4208
0.5 1.6487
1 2.7183 1.0696
0.6487 0.4208
𝑃2 𝑥 = 1 + 𝑥 − 0 + 𝑥−0 𝑥 − 0.5
0.5 2 × 0.52
1
= 1 + 1.2974 𝑥 + 𝑥 2 − 𝑥 × 0.8416
2
= 0.8416𝑥 2 + 0.8766𝑥 + 1

𝑛+1
𝜉 𝑓
𝑏) 𝑓 𝑥 − 𝑃𝑛 𝑥 = 𝑥 − 𝑥0 𝑥 − 𝑥1 ⋯ (𝑥 − 𝑥𝑛 )
𝑛+1 !
𝑥
1 1
max 𝑒 − 𝑃2 𝑥 ≤ max 𝑒 𝑡 max 𝑥−0 𝑥− 𝑥−1
𝑥∈[0,1] 6 𝑡∈ 0,1 𝑥∈ 0,1 2
1 1
= ×𝑒× 𝑥−0 𝑥− 𝑥−1 = 0.0218
6 2 𝑥=
3± 3
6
1 1 1
Let 𝑔 = 𝑥 𝑥 − 𝑥−1 ⟹ 𝑔′ = 𝑥− 𝑥−1 +𝑥 𝑥−1 +𝑥 𝑥− =0
2 2 2
1
1 3± 9−4×3× 3± 3
2
⟹ 3𝑥 2 − 3𝑥 + = 0 ⟹ 𝑥 = =
2 2×3 6

𝑏) Error bound
max 𝑒 𝑥 − 𝑃2 𝑥 ≤ 0.0218
𝑥∈[0,1]
𝑐) Comparison with actual error
𝑥 0.1 0.3 0.6 0.9

|𝑒 𝑥 − 𝑃2 (𝑥)| 0.0091 0.0111 0.0068 0.0110

 Newton’s Divided - Difference Interpolating Polynomial
 Lagrange's Interpolation Formula

Recall: Previous Lecture
∆𝑓0 ∆2 𝑓0 ∆𝑛 𝑓0
𝑃𝑛 𝑥 = 𝑓 𝑥0 + 𝑥 − 𝑥0 + 𝑥 − 𝑥0 𝑥 − 𝑥1 2
+ ⋯ + 𝑥 − 𝑥0 𝑥 − 𝑥1 𝑥 − 𝑥𝑛−1
ℎ 2! ℎ 𝑛! ℎ𝑛
𝛻𝑓𝑛 𝛻 2 𝑓𝑛
𝑃𝑛 𝑥 = 𝑓𝑛 + 𝑥 − 𝑥𝑛 + 𝑥 − 𝑥𝑛 𝑥 − 𝑥𝑛−1 +⋯
ℎ 2! ℎ2
𝛻 𝑛 𝑓𝑛
+ 𝑥 − 𝑥𝑛 𝑥 − 𝑥𝑛−1 … 𝑥 − 𝑥1
𝑛! ℎ𝑛
Note that both formulas were derived taking equidistant nodal points!
Newton’s Divided - Difference Interpolating Polynomial
Linear Polynomial : Given that : 𝑥0 𝑥1

𝑓 𝑥0 𝑓 𝑥1
𝑃1 𝑥 = 𝑏0 + 𝑏1 𝑥 − 𝑥0
At the point 𝑥 = 𝑥0 : 𝑓 𝑥0 = 𝑏0
At the point 𝑥 = 𝑥1 : 𝑓 𝑥1 = 𝑏0 + 𝑏1 𝑥1 − 𝑥0
𝑓 𝑥1 − 𝑓(𝑥0 )
⇒ 𝑏1 = = 𝑓 𝑥1 , 𝑥0 divided difference
𝑥1 − 𝑥0
𝑃1 𝑥 = 𝑓 𝑥0 + 𝑓[𝑥1 , 𝑥0 ](𝑥 − 𝑥0 )

Quadratic polynomial: Given data points: (𝑥0 , 𝑓(𝑥0 )), (𝑥1 , 𝑓(𝑥1 )), (𝑥2 , 𝑓(𝑥2 ))
In this case, we can fit a polynomial of degree 2: 𝑃2 𝑥 = 𝑏0 + 𝑏1 𝑥 − 𝑥0 + 𝑏2 𝑥 − 𝑥0 𝑥 − 𝑥1
At the point 𝑥 = 𝑥0 : 𝑏0 = 𝑓 𝑥0
At the point 𝑥 = 𝑥1 : 𝑏1 = 𝑓[𝑥1 , 𝑥0 ]
𝑓 𝑥2 − 𝑓(𝑥1 ) 𝑓 𝑥1 − 𝑓(𝑥0 )
− 𝑓 𝑥2 , 𝑥1 − 𝑓[𝑥1 , 𝑥0 ]
𝑥2 − 𝑥1 𝑥1 − 𝑥0
At the point 𝑥 = 𝑥2 : 𝑏2 = =
𝑥2 − 𝑥0 𝑥2 − 𝑥0
= 𝑓[𝑥2 , 𝑥1 , 𝑥0 ]
𝑃2 𝑥 = 𝑓 𝑥0 + 𝑓 𝑥1 , 𝑥0 𝑥 − 𝑥0 + 𝑓[𝑥2 , 𝑥1 , 𝑥0 ](𝑥 − 𝑥0 )(𝑥 − 𝑥1 )

Generalized formula for Newton’s Divided-Difference Interpolating Polynomial
𝑃𝑛 𝑥 = 𝑓 𝑥0 + 𝑥 − 𝑥0 𝑓 𝑥1 , 𝑥0 + 𝑥 − 𝑥0 𝑥 − 𝑥1 𝑓 𝑥2 , 𝑥1 , 𝑥0 + ⋯
+ 𝑥 − 𝑥0 𝑥 − 𝑥1 … 𝑥 − 𝑥𝑛−1 𝑓 𝑥𝑛 , 𝑥𝑛−1 , … , 𝑥0
This is called Newton’s divided difference interpolating polynomial.

Lagrange Interpolating Polynomial
Linear Polynomial : Given data points : 𝑥0 , 𝑓 𝑥0 , 𝑥1 , 𝑓 𝑥1 1
𝑃1 𝑥 = ෍ 𝐿𝑖 𝑥 𝑓(𝑥𝑖 )
𝑃1 𝑥 = 𝑓 𝑥0 + 𝑓[𝑥1 , 𝑥0 ](𝑥 − 𝑥0 ) 𝑖=0
𝑓 𝑥1 − 𝑓 𝑥0
𝑃1 𝑥 = 𝑓 𝑥0 + 𝑥 − 𝑥0
𝑥1 − 𝑥0
(𝑥 − 𝑥1 ) (𝑥 − 𝑥0 )
= 𝑓 𝑥0 + 𝑓(𝑥1 )
(𝑥0 −𝑥1 ) (𝑥1 −𝑥0 )
𝐿0 (𝑥) 𝐿1 (𝑥)
𝐿0 & 𝐿1 are called Lagrange polynomials of degree 1.

1
(𝑥 − 𝑥1 ) (𝑥 − 𝑥0 )
Linear Polynomial : 𝑃1 𝑥 = ෍ 𝐿𝑖 𝑥 𝑓(𝑥𝑖 ) 𝐿0 𝑥 = 𝐿1 𝑥 =
(𝑥0 −𝑥1 ) (𝑥1 −𝑥0 )
𝑖=0
1
𝑥 − 𝑥𝑗
𝐿𝑖 𝑥 = ෑ ; 𝑖 = 0, 1
Generalized Lagrange Interpolating Polynomial 𝑥𝑖 − 𝑥𝑗
𝑗=0
𝑗≠𝑖
Given data points : 𝑥0 , 𝑓 𝑥0 , 𝑥1 , 𝑓 𝑥1 , … , 𝑥𝑛 , 𝑓 𝑥𝑛 .
𝑃𝑛 𝑥 = ෍ 𝐿𝑖 𝑥 𝑓(𝑥𝑖 )
𝑖=0
𝑛
𝑥 − 𝑥𝑗 𝑥 − 𝑥0 𝑥 − 𝑥1 … 𝑥 − 𝑥𝑖−1 𝑥 − 𝑥𝑖+1 … (𝑥 − 𝑥𝑛 )
𝐿𝑖 𝑥 = ෑ =
𝑥𝑖 − 𝑥𝑗 𝑥𝑖 − 𝑥0 𝑥𝑖 − 𝑥1 … 𝑥𝑖 − 𝑥𝑖−1 𝑥𝑖 − 𝑥𝑖+1 … (𝑥𝑖 − 𝑥𝑛 )
𝑗=0
𝑗≠𝑖
Lagrange’s polynomial of degree 𝑛

Example : Using Lagrange and the Newton-divided difference formulas, construct a polynomial of degree 2 or
less with the following data:
𝑥 1 2 4
𝑓(𝑥) 1 3 3
2
Solution : Lagrange interpolating polynomial: 𝑃2 𝑥 = ෍ 𝐿𝑖 𝑥 𝑓(𝑥𝑖 )
𝑖=0
(𝑥 − 𝑥1 )(𝑥 − 𝑥2 ) (𝑥 − 2)(𝑥 − 4) 1
𝐿0 𝑥 = = = 𝑥−2 𝑥−4
𝑥0 − 𝑥1 (𝑥0 − 𝑥2 ) (1 − 2)(1 − 4) 3
(𝑥 − 𝑥0 )(𝑥 − 𝑥2 ) (𝑥 − 1)(𝑥 − 4) 1
𝐿1 𝑥 = = =− 𝑥−1 𝑥−4
𝑥1 − 𝑥0 (𝑥1 − 𝑥2 ) (2 − 1)(2 − 4) 2
(𝑥 − 𝑥0 )(𝑥 − 𝑥1 ) (𝑥 − 1)(𝑥 − 2) 1
𝐿2 𝑥 = = = (𝑥 − 1)(𝑥 − 2)
𝑥2 − 𝑥0 (𝑥2 − 𝑥1 ) (4 − 1)(4 − 2) 6

1 1 1
𝐿0 𝑥 = 𝑥−2 𝑥−4 𝐿1 𝑥 = − 𝑥−1 𝑥−4 𝐿2 𝑥 = (𝑥 − 1)(𝑥 − 2)
3 2 6
2
Lagrange interpolating polynomial : 𝑃2 𝑥 = ෍ 𝐿𝑖 𝑥 𝑓(𝑥𝑖 )
𝑖=0
1 1 1
𝑃2 𝑥 = 𝑥 − 2 𝑥 − 4 × 1 − 𝑥 − 1 𝑥 − 4 × 3 + 𝑥 − 1 𝑥 − 2 × 3
3 2 6
1 2 3 1
= 𝑥 − 6𝑥 + 8 − 𝑥 2 − 5𝑥 + 4 + 𝑥 2 − 3𝑥 + 2
3 2 2
2 2 7
= − 𝑥 + 4𝑥 −
3 3

Newton’s Divided-Difference Formula:
Divided-Difference table
𝑥 𝑓(𝑥) 𝑓 𝑥𝑖 , 𝑥𝑖−1 𝑓 𝑥𝑖 , 𝑥𝑖−1 , 𝑥𝑖−2
1 1
2 2
2 3 −
0 3
4 3
𝑃2 𝑥 = 𝑓 𝑥0 + 𝑓 𝑥1 , 𝑥0 𝑥 − 𝑥0 + 𝑓[𝑥2 , 𝑥1 , 𝑥0 ](𝑥 − 𝑥0 )(𝑥 − 𝑥1 )
2
= 1 + 2 𝑥 − 1 − (𝑥 − 1)(𝑥 − 2)
3
2 2 7
= − 𝑥 + 4𝑥 −
3 3

Example: Use the Lagrange and the Newton-divided difference formulas to derive interpolating polynomial
from the following table: 𝑥 0 1 2 4 5 6
𝑓(𝑥) 1 14 15 5 6 19
Divided difference table:
𝑥 𝑓(𝑥) 𝑓 ⋅,⋅ 𝑓[⋅,⋅,⋅] 𝑓 ⋅,⋅,⋅,⋅ 𝑓 ⋅,⋅,⋅,⋅,⋅ 𝑓 ⋅,⋅,⋅,⋅,⋅,⋅
0 1
13
1 14 −6 1
1 0
2 15 −2 1 0
−5 0
4 5 2 1
5 6 1
6
6 19 13

𝑃5 𝑥 = 𝑓 𝑥0 + 𝑓 𝑥1 , 𝑥0 𝑥 − 𝑥0 + 𝑓 𝑥2 , 𝑥1 , 𝑥0 𝑥 − 𝑥0 𝑥 − 𝑥1 + 𝑓[𝑥3 , 𝑥2 , 𝑥1 , 𝑥0 ](𝑥 − 𝑥0 )(𝑥 − 𝑥1 )(𝑥 − 𝑥2 )
= 1 + 13 𝑥 − 6 𝑥 𝑥 − 1 + 1(𝑥)(𝑥 − 1)(𝑥 − 2)
= 𝑥 3 − 9𝑥 2 + 21𝑥 + 1
Divided difference table:
𝑥 𝑓(𝑥) 𝑓 ⋅,⋅ 𝑓[⋅,⋅,⋅] 𝑓 ⋅,⋅,⋅,⋅ 𝑓 ⋅,⋅,⋅,⋅,⋅ 𝑓 ⋅,⋅,⋅,⋅,⋅,⋅
0 1
13
1 14 −6 1
1 0
2 15 −2 1 0
−5 0
4 5 2 1
5 6 1
6
6 19 13

Lagrange’s Interpolation Formula 𝑥 0 1 2 4 5 6
5
𝑓(𝑥) 1 14 15 5 6 19
𝑃5 𝑥 = ෍ 𝐿𝑖 𝑥 𝑓(𝑥𝑖 )
𝑖=0
𝑥– 1 𝑥– 2 𝑥– 4 𝑥– 5 𝑥– 6 𝑥 𝑥– 2 𝑥– 4 𝑥– 5 𝑥– 6
= ×1 + × 14
(−1)(−2)(−4)(−5)(−6) 1 − 0 1 − 2 1 − 4 (1 − 5)(1 − 6)
𝑥 𝑥 –1 𝑥 – 4 𝑥 – 5 𝑥 – 6 𝑥 𝑥 –1 𝑥 –2 𝑥 – 5 𝑥 – 6
+ × 15 + ×5
2 − 0 2 − 1 2 − 4 (2 − 5)(2 − 6) 4 − 0 4 − 1 4 − 2 (4 − 5)(4 − 6)
𝑥 𝑥 –1 𝑥 –2 𝑥 –4 𝑥 – 6 𝑥 𝑥 –1 𝑥 –2 𝑥 –4 𝑥 –5
+ ×6+ × 19
5 − 0 5 − 1 5 − 2 (5 − 4)(5 − 6) 6 − 0 6 − 1 6 − 2 (6 − 4)(6 − 5)
= 𝑥 3 − 9𝑥 2 + 21𝑥 + 1

CONCLUSIONS
Newton’s Divided-Difference Interpolating Polynomial
𝑃𝑛 𝑥 = 𝑓 𝑥0 + 𝑥 − 𝑥0 𝑓 𝑥1 , 𝑥0 + 𝑥 − 𝑥0 𝑥 − 𝑥1 𝑓 𝑥2 , 𝑥1 , 𝑥0 + ⋯
+ 𝑥 − 𝑥0 𝑥 − 𝑥1 … 𝑥 − 𝑥𝑛−1 𝑓 𝑥𝑛 , 𝑥𝑛−1 , … , 𝑥0
Lagrange Interpolating Polynomial
𝑛 𝑛
𝑥 − 𝑥𝑗
𝑃𝑛 𝑥 = ෍ 𝐿𝑖 𝑥 𝑓(𝑥𝑖 ) 𝐿𝑖 𝑥 = ෑ
𝑥𝑖 − 𝑥𝑗
𝑖=0 𝑗=0
𝑗≠𝑖

Linear Algebra, Numerical and Complex Analysis

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Linear Algebra, Numerical and Complex Analysis

Uploaded by

Copyright:

Available Formats

LINEAR ALGEBRA, NUMERICAL AND COMPLEX ANALYSIS

 Existence and Uniqueness

 Error in Interpolating Polynomials

Dr. Jitendra Kumar – IIT KHARAGPUR 2

2. Replacing complicated function by an interpolating simpler function (usually polynomials)

Dr. Jitendra Kumar – IIT KHARAGPUR 3

Weirstrass Approximation Theorem

polynomial 𝑃(𝑥) with the property that

Dr. Jitendra Kumar – IIT KHARAGPUR 4

Consider for simplicity, a second order polynomial

A straight forward method for computing the coefficients of a polynomial of degree 𝑛 is

Dr. Jitendra Kumar – IIT KHARAGPUR 6

Suppose that there are 3 given data points 𝑥0 , 𝑓 𝑥0 , 𝑥1 , 𝑓 𝑥1 , 𝑥2 , 𝑓 𝑥2 .

Dr. Jitendra Kumar – IIT KHARAGPUR 7

This system of equations has a unique solution as

In practice, it is observed that the above system of equations is ill-conditioned.

Therefore, we have some mathematical formats (interpolating formats) in which

Dr. Jitendra Kumar – IIT KHARAGPUR 8

Then the interpolation error at the point 𝑥 is given by

Dr. Jitendra Kumar – IIT KHARAGPUR 9

For simplicity, let us assume 𝑤𝑛+1 𝑥 = 𝑥 − 𝑥0 𝑥 − 𝑥1 ⋯ (𝑥 − 𝑥𝑛 )

Now, define for any 𝑥 ∈ 𝐼𝑥 , the function

Since 𝑓 ∈ 𝐶 𝑛+1 (𝐼𝑥 ) and 𝑤𝑛+1 is a polynomial, then 𝐺 ∈ 𝐶 𝑛+1 (𝐼𝑥 ).

Dr. Jitendra Kumar – IIT KHARAGPUR 10

Then using Rolle’s theorem, 𝐺 ′ has atleast (𝑛 + 1) distinct zeros.

By recursion it follows that 𝐺 (𝑗) admits at least 𝑛 + 2 − 𝑗 distinct zeros.

Dr. Jitendra Kumar – IIT KHARAGPUR 11

Note that 𝐸𝑛 𝑡 = 𝑓 𝑡 − 𝑃𝑛 𝑡 ⟹ 𝐸𝑛𝑛+1 𝑡 = 𝑓 𝑛+1

Dr. Jitendra Kumar – IIT KHARAGPUR 12

 Existence and Uniqueness

 Error in Interpolating Polynomials

Dr. Jitendra Kumar – IIT KHARAGPUR 13

 Newton’s Forward Interpolation Formula

 Newton’s Backward Interpolation Formula

Dr. Jitendra Kumar – IIT KHARAGPUR 14

 Existence and Uniqueness

 Error in Interpolating Polynomials

Dr. Jitendra Kumar – IIT KHARAGPUR 15

mathematical formats in which this polynomial can be expressed.

 Newton’s forward and backward interpolating polynomial

 Newton’s Divided Difference Formula

 Lagrange Interpolation Formula

Dr. Jitendra Kumar – IIT KHARAGPUR 16

Let the tabular points 𝑥0 , 𝑥1 , … , 𝑥𝑛 be equally spaced, i.e., 𝑥𝑖 = 𝑥0 + 𝑖ℎ, 𝑖 = 0, 1, … , 𝑛

Finite Difference Operator

 The Shift operator: 𝐸𝑓 𝑥𝑖 = 𝑓 𝑥𝑖 + ℎ

 The Forward difference operator: ∆𝑓 𝑥𝑖 = 𝑓 𝑥𝑖 + ℎ − 𝑓 𝑥𝑖

 The Backward difference operator: 𝛻𝑓 𝑥𝑖 = 𝑓 𝑥𝑖 − 𝑓 𝑥𝑖 − ℎ

Dr. Jitendra Kumar – IIT KHARAGPUR 17

It can easily be verified that

Also, note that

∆≡𝐸−1 ∆𝑓𝑖 = 𝐸𝑓𝑖 − 𝑓𝑖

Dr. Jitendra Kumar – IIT KHARAGPUR 18

= 𝑓𝑖+2 − 𝑓𝑖+1 + 𝑓𝑖+1 − 𝑓𝑖

Similarly, 𝛻 2 𝑓𝑖 = 𝛻 𝛻𝑓𝑖 = 𝛻 𝑓𝑖 − 𝑓𝑖−1

= 𝑓𝑖 − 𝑓𝑖−1 − 𝑓𝑖−1 − 𝑓𝑖−2

Dr. Jitendra Kumar – IIT KHARAGPUR 19

𝑥 𝑓(𝑥) ∆𝑓 ∆2 𝑓 ∆3 𝑓 𝑥 𝑓(𝑥) 𝛻𝑓 𝛻2𝑓 𝛻3𝑓

∆𝑓𝑖 = 𝑓𝑖+1 − 𝑓𝑖 𝛻𝑓𝑖 = 𝑓𝑖 − 𝑓𝑖−1

∆2 𝑓𝑖 = ∆𝑓𝑖+1 − ∆𝑓𝑖 𝛻 2 𝑓𝑖 = 𝛻𝑓𝑖 − 𝛻𝑓𝑖−1

Dr. Jitendra Kumar – IIT KHARAGPUR 20

Dr. Jitendra Kumar – IIT KHARAGPUR 21