CE 341/441 - Lecture 6 - Fall 2004

LECTURE 6

LAGRANGE INTERPOLATION
th
• Fit N + 1 points with an N degree polynomial

g(x)
f2
f(x)
f3 f4
f1
f0 fN

x0 x1 x2 x3 x4 ... xN

• f ( x ) = exact function of which only N + 1 discrete values are known and used to estab-
lish an interpolating or approximating function g ( x )

• g ( x ) = approximating or interpolating function. This function will pass through all

specified N + 1 interpolation points (also referred to as data points or nodes).

p. 6.1
 CE 341/441 - Lecture 6 - Fall 2004

• The interpolation points or nodes are given as:

xo f ( xo ) ≡ f o
x1 f ( x1 ) ≡ f 1
x2 f ( x2 ) ≡ f 2
:
xN f ( xN ) ≡ f N

• There exists only one N th degree polynomial that passes through a given set of N + 1
points. It’s form is (expressed as a power series):
2 3 N
g ( x ) = ao + a1 x + a2 x + a3 x + … + a N x

where a i = unknown coefficients, i = 0, N ( N + 1 coefficients).

• No matter how we derive the N th degree polynomial,

• Fitting power series
• Lagrange interpolating functions
• Newton forward or backward interpolation
The resulting polynomial will always be the same!

p. 6.2
 CE 341/441 - Lecture 6 - Fall 2004

Power Series Fitting to Define Lagrange Interpolation

• g ( x ) must match f ( x ) at the selected data points ⇒

2 N
g ( xo ) = f o ⇒ ao + a1 xo + a2 xo + … + a N xo = f o
2 N
g ( x1 ) = f 1 ⇒ ao + a1 x1 + a2 x1 + … + a N x1 = f 1
: :
2 N
g( xN ) = f N ⇒ ao + a1 x N + a2 x N + … + a N x N = f N

• Solve set of simultaneous equations

2 N
1 xo xo … xo ao fo
2 N
1 x1 x1 … x1 a1
=
f1
… … … … … : :
2 N aN fN
1 xN xN … xN

• It is relatively computationally costly to solve the coefficients of the interpolating func-

tion g ( x ) (i.e. you need to program a solution to these equations).

p. 6.3
 CE 341/441 - Lecture 6 - Fall 2004

Lagrange Interpolation Using Basis Functions

• We note that in general g ( x i ) = f i

• Let
N
g( x) = ∑ f i V i( x)
i=0

where V i ( x ) = polynomial of degree N associated with each node i such that

0 i ≠ j
V i( x j) ≡ 
1 i = j

• For example if we have 5 interpolation points (or nodes)

g ( x3 ) = f o V o ( x3 ) + f 1 V 1 ( x3 ) + f 2 V 2 ( x3 ) + f 3 V 3 ( x3 ) + f 4 V 4 ( x3 )

Using the definition for V i ( x j ) : V 0 ( x 3 ) = 0 ; V 1 ( x 3 ) = 0 ; V 2 ( x 3 ) = 0 ; V 3 ( x 3 ) = 1 ;

V 4 ( x 3 ) = 0 ,we have:

g ( x3 ) = f 3

p. 6.4
 CE 341/441 - Lecture 6 - Fall 2004

• How do we construct V i ( x ) ?

• Degree N
• Roots at x o, x 1, x 2, …x i – 1, x i + 1, …, x N (at all nodes except x i )

• V i ( xi ) = 1

• Let W i ( x ) = ( x – x o ) ( x – x 1 ) ( x – x 2 )… ( x – x i – 1 ) ( x – x i + 1 )… ( x – x N )

• The function W i is such that we do have the required roots, i.e. it equals zero at
nodes x o, x 1, x 2, ... , x N except at node x i

• Degree of W i ( x ) is N

• However W i ( x ) in the form presented will not equal to unity at x i

• We normalize W i ( x ) and define the Lagrange basis functions V i ( x )

( x – x o ) ( x – x 1 ) ( x – x 2 )… ( x – x i – 1 ) ( x – x i + 1 )… ( x – x N )
V i ( x ) = -----------------------------------------------------------------------------------------------------------------------------------------------------------
( x i – x o ) ( x i – x 1 ) ( x i – x 2 )… ( x i – x i – 1 ) ( x i – x i + 1 )… ( x i – x N )

p. 6.5
 CE 341/441 - Lecture 6 - Fall 2004

• Now we have V i ( x ) such that V i ( x i ) equals:

( x i – x o ) ( x i – x 1 ) ( x i – x 2 )… ( x i – x i – 1 ) ( 1 ) ( x i – x i + 1 )… ( x i – x N )
V i ( x i ) = -------------------------------------------------------------------------------------------------------------------------------------------------------------
- ⇒
( x i – x o ) ( x i – x 1 ) ( x i – x 2 )… ( x i – x i – 1 ) ( x i – x i + 1 )… ( x i – x N )

V i ( xi ) = 1

• We also satisfy V i ( x j ) = 0 for i ≠ j

( x2 – xo ) ( 1 ) ( x2 – x2 ) ⋅ ( x2 – x3 ) … ( x2 – x N )
e.g. V 1 ( x 2 ) = ---------------------------------------------------------------------------------------------------------- = 0
( x 1 – x o ) ( 1 ) ( x 1 – x 2 ) ( x 1 – x 3 )… ( x 1 – x N )

• The general form of the interpolating function g ( x ) with the specified form of V i ( x ) is:
N
g( x) = ∑ f iV i( x)
i=0

• The sum of polynomials of degree N is also polynomial of degree N

• g ( x ) is equivalent to fitting the power series and computing coefficients a o, …, a N .

p. 6.6
 CE 341/441 - Lecture 6 - Fall 2004

Lagrange Linear Interpolation Using Basis Functions

• Linear Lagrange ( N = 1 ) is the simplest form of Lagrange Interpolation

1
g( x) = ∑ f iV i( x)
i=0
⇒
g( x) = f oV o( x) + f 1V 1( x)

where
( x – x1 ) ( x1 – x ) ( x – xo )
V o ( x ) = --------------------- = --------------------- and V 1 ( x ) = --------------------
-
( xo – x1 ) ( x1 – xo ) ( x1 – xo )

V0 (x) V1(x)
1.0

(x)
x0 x1

p. 6.7
 CE 341/441 - Lecture 6 - Fall 2004

Example

• Given the following data:

xo = 2 f o = 1.5

x1 = 5 f 1 = 4.0

Find the linear interpolating function g ( x )

• Lagrange basis functions are:

5–x x–2
V o ( x ) = ----------- and V 1 ( x ) = -----------
3 3

• Interpolating function g(x) is:

g ( x ) = 1.5V o ( x ) + 4.0V 1 ( x )

p. 6.8
 CE 341/441 - Lecture 6 - Fall 2004

4
2 1.5 V0 (x)
x
x0 = 2 x1 = 5

4
4.0 V1(x)
2
x
x0 = 2 x1 = 5

g(x) = 1.5 V0(x) + 4.0V1(x)

x0 = 2 x1 = 5

p. 6.9
 CE 341/441 - Lecture 6 - Fall 2004

Lagrange Quadratic Interpolation Using Basis Functions

• For quadratic Lagrange interpolation, N=2
2
g( x) = ∑ f i V i( x)
i=0

g( x) = f oV o( x) + f 1V 1( x) + f 2V 2( x)

where
( x – x1 ) ( x – x2 )
V o ( x ) = ------------------------------------------
-
( xo – x1 ) ( xo – x2 )

( x – xo ) ( x – x2 )
V 1 ( x ) = ------------------------------------------
-
( x1 – xo ) ( x1 – x2 )

( x – xo ) ( x – x1 )
V 2 ( x ) = ------------------------------------------
-
( x2 – xo ) ( x2 – x1 )

p. 6.10
 CE 341/441 - Lecture 6 - Fall 2004

V0 (x) V1(x) V2(x)

1.0

x
x0 x1 x2

• Note that the location of the roots of V 0 ( x ) , V 1 ( x ) and V 2 ( x ) are defined such that the
basic premise of interpolation is satisfied, namely that g ( x i ) = f i . Thus:

g ( xo ) = V o ( xo ) f o + V 1 ( xo ) f 1 + V 2 ( xo ) f 2 = f 0

g ( x1 ) = V o ( x1 ) f o + V 1 ( x1 ) f 1 + V 2 ( x1 ) f 2 = f 1

g ( x2 ) = V o ( x2 ) f o + V 1 ( x2 ) f 1 + V 2 ( x2 ) f 2 = f 2

p. 6.11
 CE 341/441 - Lecture 6 - Fall 2004

Example
• Given the following data:
xo = 3 fo = 1

x1 = 4 f1 = 2

x2 = 5 f2 = 4

Find the quadratic interpolating function g ( x )

• Lagrange basis functions are

( x – 4)( x – 5)
V o ( x ) = ----------------------------------
(3 – 4)(3 – 5)

( x – 3)( x – 5)
V 1 ( x ) = ----------------------------------
(4 – 3)(4 – 5)

( x – 3)( x – 4)
V 2 ( x ) = ----------------------------------
(5 – 3)(5 – 4)

• Interpolating function g(x) is:

g ( x ) = 1.0V o ( x ) + 2.0V 1 ( x ) + 4.0V 2 ( x )

p. 6.12
 CE 341/441 - Lecture 6 - Fall 2004

1.0 V0 (x)
1.0
x
x0 = 3 x2 = 5
x1 = 4

2.0 2.0 V1(x)

x
x0 = 3 x1 = 4 x2 = 5

4.0 4.0 V2(x)

x
x0 = 3 x1 = 4 x2 = 5

g(x) = 1.0 V0(x) + 2.0V1(x) + 4.0V2(x)

4.0

x0 = 3 x1 = 4 x2 = 5

p. 6.13
 CE 341/441 - Lecture 6 - Fall 2004

Lagrange Cubic Interpolation Using Basis Functions

• For Cubic Lagrange interpolation, N=3

Example
• Consider the following table of functional values (generated with f ( x ) = ln x )

i xi fi
0 0.40 -0.916291
1 0.50 -0.693147
2 0.70 -0.356675
3 0.80 -0.223144

• Find g ( 0.60 ) as:

( x – x1 ) ( x – x2 ) ( x – x3 ) ( x – xo ) ( x – x2 ) ( x – x3 )
g ( x ) = f o ----------------------------------------------------------------- + f 1 -----------------------------------------------------------------
( xo – x1 ) ( xo – x2 ) ( xo – x3 ) ( x1 – xo ) ( x1 – x2 ) ( x1 – x3 )

( x – xo ) ( x – x1 ) ( x – x3 ) ( x – xo ) ( x – x1 ) ( x – x2 )
+ f 2 ----------------------------------------------------------------- + f 3 -----------------------------------------------------------------
( x2 – xo ) ( x2 – x1 ) ( x2 – x3 ) ( x3 – xo ) ( x3 – x1 ) ( x3 – x2 )

p. 6.14
 CE 341/441 - Lecture 6 - Fall 2004

( 0.60 – 0.50 ) ( 0.60 – 0.70 ) ( 0.60 – 0.80 )

g ( 0.60 ) = – 0.916291 ⋅ ------------------------------------------------------------------------------------------------
( 0.40 – 0.50 ) ( 0.40 – 0.70 ) ( 0.40 – 0.80 )

( 0.60 – 0.40 ) ( 0.60 – 0.70 ) ( 0.60 – 0.80 )

– 0.693147 ⋅ ------------------------------------------------------------------------------------------------
( 0.50 – 0.40 ) ( 0.50 – 0.70 ) ( 0.50 – 0.80 )

( 0.60 – 0.40 ) ( 0.60 – 0.50 ) ( 0.60 – 0.80 )

– 0.356675 ⋅ ------------------------------------------------------------------------------------------------
( 0.70 – 0.40 ) ( 0.70 – 0.50 ) ( 0.70 – 0.80 )

( 0.60 – 0.40 ) ( 0.60 – 0.50 ) ( 0.60 – 0.70 )

– 0.223144 ⋅ ------------------------------------------------------------------------------------------------
( 0.80 – 0.40 ) ( 0.80 – 0.50 ) ( 0.80 – 0.70 )

g ( 0.60 ) = – 0.509976

p. 6.15
 CE 341/441 - Lecture 6 - Fall 2004

Errors Associated with Lagrange Interpolation

• Using Taylor series analysis, the error can be shown to be given by:

e( x) = f ( x) – g( x)

( N + 1)
e( x) = L( x) f (ξ) xo ≤ ξ ≤ x N

where
N+1 th
f (ξ) = N + 1 derivative of f w.r.t. x evaluated at ξ

( x – x o ) ( x – x 1 )… ( x – x N ) th
L ( x ) = ----------------------------------------------------------------
- = an N + 1 degree polynomial
( N + 1 )!

• Notes
• If f ( x ) = polynomial of degree M where M ≤ N , then
(N + 1)
f ( x) = 0 ⇒ e ( x ) = 0 for all x
Therefore g ( x ) will be an exact representation of f ( x )

p. 6.16
 CE 341/441 - Lecture 6 - Fall 2004

• Since in general ξ is not known, if the interval [ x o, x N ] is small and if f ( N + 1 ) ( x )

does not change rapidly in the interval
( N + 1) xo + x N
e( x) ≈ L( x) f ( x m ) where x m = ------------------- .
2

• f ( N + 1 ) can be estimated by using Finite Difference (F.D.) formulae

• L ( x ) will significantly effect the distribution of the error

• L ( x ) is a minimum at the center of [ x o, x N ] and a maximum near the edges

• e.g. using 6 point interpolation L ( x ) looks like:

0 1 2 3 4 5

• L ( x ) = 0 at all data points

• L ( x ) largest 0 ≤ x ≤ 1 4 ≤ x ≤ 5 . L ( x ) becomes very large outside of the interval.

p. 6.17
 CE 341/441 - Lecture 6 - Fall 2004

• As the size of the interpolating domain increases, so does the maximum error within
the interval

D = x N – x o ↑ ⇒ L max ↑ ⇒ e max ↑
x0 ≤ x ≤ x N x0 ≤ x ≤ x N

• As N increases from a small value, L max ↓ ⇒ e max ↓

x0 ≤ x ≤ x N x0 ≤ x ≤ x N

• However as N > N CRIT ⇒ L max ↑ for a given [ x o, x N ] and thus e max ↑

x0 ≤ x ≤ x N x0 ≤ x ≤ x N

• Therefore convergence as N ↑ does not necessarily occur!!

• Properties of f ( N + 1 ) ( ξ ) will also influence error as D and N vary

p. 6.18
 CE 341/441 - Lecture 6 - Fall 2004

Example

• Estimate the error made in the previous example knowing that f ( x ) = ln ( x ) (usually we
do not have this information).

(N + 1)
e( x) ≈ L( x) f ( xm )

( x – xo ) ( x – x1 ) ( x – x2 ) ( x – x3 ) (3 + 1)
e ( x ) ≈ ----------------------------------------------------------------------------- f ( xm )
( 3 + 1 )!

( 0.60 – 0.40 ) ( 0.60 – 0.50 ) ( 0.60 – 0.70 ) ( 0.60 – 0.80 ) ( 3 + 1 )

e ( 0.60 ) ≈ -------------------------------------------------------------------------------------------------------------------------------- f ( 0.6 )
( 3 + 1 )!

(4)
e ( 0.60 ) = 0.000017 f ( 0.6 )

p. 6.19
 CE 341/441 - Lecture 6 - Fall 2004

• We estimate the fourth derivative of f(x) using the analytical function itself
f ( x ) = ln x ⇒
(1) –1
f ( x) = x ⇒
(2) –2
f ( x) = –x ⇒
(3) –3
f ( x ) = 2x ⇒
(4) –4
f ( x ) = – 6x ⇒
(4)
f ( 0.6 ) = – 46.29
• Therefore
e ( 0.60 ) = – 0.00079

• Exact error is computed as:

E ( x ) = ln ( 0.60 ) – g ( 0.60 ) = – 0.00085

Therefore error estimate is excellent

(N + 1)
• Typically we would also have to estimate f ( x m ) using a Finite Difference (F.D.)
approximation (a discrete differentiation formula).

p. 6.20
 CE 341/441 - Lecture 6 - Fall 2004

SUMMARY OF LECTURES 5 AND 6

• Linear interpolation passes a straight line through 2 data points.

th
• Power series → N + 1 data points → N degree polynomial → find coefficients by
solving a matrix
th
• Lagrange Interpolation passes an N degree polynomial through N + 1 data points →
Use specialized nodal functions to make finding g ( x ) easier.
N
g( x) = ∑ f iV i( x)
i=0

where
g ( x ) = the interpolating function approximating f(x)

fi = the value of the function at the data (or interpolation) point i

V i ( x ) = the Lagrange basis function

• Each Lagrange polynomial or basis function is set up such that it equals unity at the
data point with which it is associated, zero at all other data points and nonzero in-
between.

p. 6.21
 CE 341/441 - Lecture 6 - Fall 2004

• For example when N = 2 → 3 data points

V0 V1 V2

0 1 2

g( x) = f oV o( x) + f 1V 1( x) + f 2V 2( x)

f0
g(x) f2

f1

p. 6.22
 CE 341/441 - Lecture 6 - Fall 2004

• Linear interpolation is the same as Lagrange Interpolation with N = 1

• Error estimates can be derived but depend on knowing f ( N + 1 ) ( x m ) (or at some point in
the interval).

( N + 1)
e( x) = L( x) f (ξ) xo ≤ ξ ≤ x N

where
N+1 th
f (ξ) = N + 1 derivative of f w.r.t. x evaluated at ξ
( x – x o ) ( x – x 1 )… ( x – x N ) th
L ( x ) = ----------------------------------------------------------------
- = an N + 1 degree polynomial
( N + 1 )!

p. 6.23