Cubic Spline Interpolation

MATH 375, Numerical Analysis

J. Robert Buchanan

Department of Mathematics

Spring 2019
History
Given nodes and data {(x0 , f (x0 )), (x1 , f (x1 )), . . . , (xn , f (xn ))}
we have interpolated using:
I Lagrange and Hermite interpolating polynomials of degree
n, with n + 1 coefficients,
unfortunately,
History
Given nodes and data {(x0 , f (x0 )), (x1 , f (x1 )), . . . , (xn , f (xn ))}
we have interpolated using:
I Lagrange and Hermite interpolating polynomials of degree
n, with n + 1 coefficients,
unfortunately,

I such polynomials can possess large oscillations, and

I the error term can be difficult to construct and estimate.
History
Given nodes and data {(x0 , f (x0 )), (x1 , f (x1 )), . . . , (xn , f (xn ))}
we have interpolated using:
I Lagrange and Hermite interpolating polynomials of degree
n, with n + 1 coefficients,
unfortunately,

I such polynomials can possess large oscillations, and

I the error term can be difficult to construct and estimate.

An alternative is piecewise polynomial approximation, but of

what degree polynomial?
History
Given nodes and data {(x0 , f (x0 )), (x1 , f (x1 )), . . . , (xn , f (xn ))}
we have interpolated using:
I Lagrange and Hermite interpolating polynomials of degree
n, with n + 1 coefficients,
unfortunately,

I such polynomials can possess large oscillations, and

I the error term can be difficult to construct and estimate.

An alternative is piecewise polynomial approximation, but of

what degree polynomial?
I Piecewise linear results are not differentiable at xi ,
i = 0, 1, . . . , n.
I Piecewise quadratic results are not twice differentiable at
xi , i = 0, 1, . . . , n.
I Piecewise cubic!
Cubic Splines

I A cubic polynomial p(x) = a + bx + cx 2 + dx 3 is specified

by 4 coefficients.
I The cubic spline is twice continuously differentiable.
I The cubic spline has the flexibility to satisfy general types
of boundary conditions.
I While the spline may agree with f (x) at the nodes, we
cannot guarantee the derivatives of the spline agree with
the derivatives of f .
Cubic Spline Interpolant (1 of 2)

Given a function f (x) defined on [a, b] and a set of nodes

a = x0 < x1 < x2 < · · · < xn = b,

a cubic spline interpolant, S(x), for f (x) is a piecewise cubic

polynomial with components Sj (x) defined on [xj , xj+1 ] for
j = 0, 1, . . . , n − 1.
a0 + b0 (x − x0 ) + c0 (x − x0 )2 + d0 (x − x0 )3


 if x0 ≤ x ≤ x1
a1 + b1 (x − x1 ) + c1 (x − x1 )2 + d1 (x − x1 )3 if x1 ≤ x ≤ x2





 .. ..
. .

S(x) =

 ai + bi (x − xi ) + ci (x − xi )2 + di (x − xi )3 if xi ≤ x ≤ xi+1

 .. ..
. .




an−1 + bn−1 (x − xn−1 ) + cn−1 (x − xn−1 )2 + dn−1 (x − xn−1 )3 if xn−1 ≤ x ≤ xn

Cubic Spline Interpolant (2 of 2)

The cubic spline interpolant will have the following properties.

I S(xj ) = f (xj ) for j = 0, 1, . . . , n.
I Sj (xj+1 ) = Sj+1 (xj+1 ) for j = 0, 1, . . . , n − 2.
I Sj0 (xj+1 ) = Sj+1
0 (x
j+1 ) for j = 0, 1, . . . , n − 2.
00 00
I Sj (xj+1 ) = Sj+1 (xj+1 ) for j = 0, 1, . . . , n − 2.
I One of the following boundary conditions (BCs) is satisfied:

I S 00 (x0 ) = S 00 (xn ) = 0 (free or natural BCs).

I S 0 (x0 ) = f 0 (x0 ) and S 0 (xn ) = f 0 (xn ) (clamped BCs).
Example (1 of 7)
Construct a piecewise cubic spline interpolant for the curve
passing through
{(5, 5), (7, 2), (9, 4)},
with natural boundary conditions.
y
6

x
4 5 6 7 8 9 10
Example (2 of 7)
This will require two cubics:

S0 (x) = a0 + b0 (x − 5) + c0 (x − 5)2 + d0 (x − 5)3

S1 (x) = a1 + b1 (x − 7) + c1 (x − 7)2 + d1 (x − 7)3

Since there are 8 coefficients, we must derive 8 equations to

solve.
Example (2 of 7)
This will require two cubics:

S0 (x) = a0 + b0 (x − 5) + c0 (x − 5)2 + d0 (x − 5)3

S1 (x) = a1 + b1 (x − 7) + c1 (x − 7)2 + d1 (x − 7)3

Since there are 8 coefficients, we must derive 8 equations to

solve.
The splines must agree with the function (the y -coordinates) at
the nodes (the x-coordinates).

5 = S0 (5) = a0
2 = S0 (7) = a0 + 2b0 + 4c0 + 8d0
2 = S1 (7) = a1
4 = S1 (9) = a1 + 2b1 + 4c1 + 8d1
Example (3 of 7)

The first and second derivatives of the cubics must agree at

their shared node x = 7.

S00 (7) = b0 + 4c0 + 12d0 = b1 = S10 (7)

S000 (7) = 2c0 + 12d0 = 2c1 = S100 (7)
Example (3 of 7)

The first and second derivatives of the cubics must agree at

their shared node x = 7.

S00 (7) = b0 + 4c0 + 12d0 = b1 = S10 (7)

S000 (7) = 2c0 + 12d0 = 2c1 = S100 (7)

The final two equations come from the natural boundary

conditions.

S000 (5) = 0 = 2c0

S100 (9) = 0 = 2c1 + 12d1
Example (4 of 7)

All eight linear equations together form the system:

5 = a0
2 = a0 + 2b0 + 4c0 + 8d0
2 = a1
4 = a1 + 2b1 + 4c1 + 8d1
0 = b0 + 4c0 + 12d0 − b1
0 = 2c0 + 12d0 − 2c1
0 = 2c0
0 = 2c1 + 12d1
Example (5 of 7)

The solution is:

i ai bi ci di

17 5
0 5 − 0
8 32
1 15 5
1 2 − −
4 16 32
Example (6 of 7)
The natural cubic spline can be expressed as:

17 5 3
 5 − 8 (x − 5) + 32 (x − 5) if 5 ≤ x ≤ 7



S(x) =
 2 − 1 (x − 7) + 15 (x − 7)2 − 5 (x − 7)3 if 7 ≤ x ≤ 9



4 16 32
y
6

x
4 5 6 7 8 9 10
Example (7 of 7)

We can verify the continuity of the first and second derivatives

from the following graphs.
S''HxL
S'HxL

1.5
1.5
1.0

0.5
1.0
x
6 7 8 9
-0.5

-1.0 0.5

-1.5

-2.0 x
6 7 8 9
General Construction Process

Given n + 1 nodal/data values:

{(x0 , f (x0 )), (x1 , f (x1 )), . . . , (xn , f (xn ))} we will create n cubic
polynomials.
General Construction Process

Given n + 1 nodal/data values:

{(x0 , f (x0 )), (x1 , f (x1 )), . . . , (xn , f (xn ))} we will create n cubic
polynomials.
For j = 0, 1, . . . , n − 1 assume

Sj (x) = aj + bj (x − xj ) + cj (x − xj )2 + dj (x − xj )3 .

We must find aj , bj , cj and dj (a total of 4n unknowns) subject to

the conditions specified in the definition.
First Set of Equations
Let hj = xj+1 − xj then

Sj (xj ) = aj = f (xj )
Sj+1 (xj+1 ) = aj+1 = Sj (xj+1 ) = aj + bj hj + cj hj2 + dj hj3 .

So far we know aj for j = 0, 1, . . . , n − 1 and have n equations

and 3n unknowns.

a1 = a0 + b0 h0 + c0 h02 + d0 h03
..
.
aj+1 = aj + bj hj + cj hj2 + dj hj3
..
.
2 3
an = an−1 + bn−1 hn−1 + cn−1 hn−1 + dn−1 hn−1
First Derivative

The continuity of the first derivative at the nodal points

produces n more equations.
For j = 0, 1, . . . , n − 1 we have

Sj0 (x) = bj + 2cj (x − xj ) + 3dj (x − xj )2 .

Thus

Sj0 (xj ) = bj
0
Sj+1 (xj+1 ) = bj+1 = Sj0 (xj+1 ) = bj + 2cj hj + 3dj hj2

Now we have 2n equations and 3n unknowns.

Equations Derived So Far

aj+1 = aj + bj hj + cj hj2 + dj hj3 (for j = 0, 1, . . . , n − 1)

b1 = b0 + 2c0 h0 + 3d0 h02
..
.
bj+1 = bj + 2cj hj + 3dj hj2
..
.
2
bn = bn−1 + 2cn−1 hn−1 + 3dn−1 hn−1

The unknowns are bj , cj , and dj for j = 0, 1, . . . , n − 1.

Second Derivative

The continuity of the second derivative at the nodal points

produces n more equations.
For j = 0, 1, . . . , n − 1 we have

Sj00 (x) = 2cj + 6dj (x − xj ).

Thus

Sj00 (xj ) = 2cj

00
Sj+1 (xj+1 ) = 2cj+1 = Sj00 (xj+1 ) = 2cj + 6dj hj

Now we have 3n equations and 3n unknowns.

Equations Derived So Far

aj+1 = aj + bj hj + cj hj2 + dj hj3 (for j = 0, 1, . . . , n − 1)

bj+1 = bj + 2cj hj + 3dj hj2 (for j = 0, 1, . . . , n − 1)
2c1 = 2c0 + 6d0 h0
..
.
2cj+1 = 2cj + 6dj hj
..
.
2cn = 2cn−1 + 6dn−1 hn−1

The unknowns are bj , cj , and dj for j = 0, 1, . . . , n − 1.

Summary of Equations

For j = 0, 1, . . . , n − 1 we have

aj+1 = aj + bj hj + cj hj2 + dj hj3

bj+1 = bj + 2cj hj + 3dj hj2
cj+1 = cj + 3dj hj .

Note: The quantities aj and hj are known.

Summary of Equations

For j = 0, 1, . . . , n − 1 we have

aj+1 = aj + bj hj + cj hj2 + dj hj3

bj+1 = bj + 2cj hj + 3dj hj2
cj+1 = cj + 3dj hj .

Note: The quantities aj and hj are known.

Solve the third equation for dj and substitute into the other two
equations.
cj+1 − cj
dj =
3hj
This eliminates n equations of the third type.
Solving the Equations (1 of 3)

aj+1 = aj + bj hj + cj hj2 + dj hj3

bj+1 = bj + 2cj hj + 3dj hj2

Solving the Equations (1 of 3)

aj+1 = aj + bj hj + cj hj2 + dj hj3

cj+1 − cj
 
2
= aj + bj hj + cj hj + hj3
3hj
hj2
= aj + bj hj + (2cj + cj+1 )
3
bj+1 = bj + 2cj hj + 3dj hj2
Solving the Equations (1 of 3)

aj+1 = aj + bj hj + cj hj2 + dj hj3

cj+1 − cj
 
2
= aj + bj hj + cj hj + hj3
3hj
hj2
= aj + bj hj + (2cj + cj+1 )
3
bj+1 = bj + 2cj hj + 3dj hj2
cj+1 − cj
 
= bj + 2cj hj + 3 hj2
3hj
= bj + hj (cj + cj+1 )
Solving the Equations (1 of 3)

aj+1 = aj + bj hj + cj hj2 + dj hj3

cj+1 − cj
 
2
= aj + bj hj + cj hj + hj3
3hj
hj2
= aj + bj hj + (2cj + cj+1 )
3
bj+1 = bj + 2cj hj + 3dj hj2
cj+1 − cj
 
= bj + 2cj hj + 3 hj2
3hj
= bj + hj (cj + cj+1 )

Solve the first equation for bj .

1 hj
bj = (aj+1 − aj ) − (2cj + cj+1 )
hj 3
Solving the Equations (2 of 3)

Replace index j by j − 1 to obtain

1 hj−1
bj−1 = (aj − aj−1 ) − (2cj−1 + cj ).
hj−1 3
Solving the Equations (2 of 3)

Replace index j by j − 1 to obtain

1 hj−1
bj−1 = (aj − aj−1 ) − (2cj−1 + cj ).
hj−1 3

We can also re-index the earlier equation

bj+1 = bj + hj (cj + cj+1 )

to obtain
bj = bj−1 + hj−1 (cj−1 + cj ).
Substitute the equations for bj−1 and bj into the remaining
equation. This step eliminates n equations of the first type.
Solving the Equations (3 of 3)

1 hj
(aj+1 − aj ) − (2cj + cj+1 )
hj 3
1 hj−1
= (aj − aj−1 ) − (2cj−1 + cj ) + hj−1 (cj−1 + cj )
hj−1 3

Collect all terms involving c to one side.

3 3
hj−1 cj−1 +2cj (hj−1 +hj )+hj cj+1 = (aj+1 −aj )− (aj −aj−1 )
hj hj−1

for j = 1, 2, . . . , n − 1.
Solving the Equations (3 of 3)

1 hj
(aj+1 − aj ) − (2cj + cj+1 )
hj 3
1 hj−1
= (aj − aj−1 ) − (2cj−1 + cj ) + hj−1 (cj−1 + cj )
hj−1 3

Collect all terms involving c to one side.

3 3
hj−1 cj−1 +2cj (hj−1 +hj )+hj cj+1 = (aj+1 −aj )− (aj −aj−1 )
hj hj−1

for j = 1, 2, . . . , n − 1.
Remark: we have n − 1 equations and n + 1 unknowns.
Natural Boundary Conditions

If S 00 (x0 ) = S000 (x0 ) = 2c0 = 0 then c0 = 0 and if

S 00 (xn ) = Sn−1
00 (x ) = 2c = 0 then c = 0.
n n n
Natural Boundary Conditions

If S 00 (x0 ) = S000 (x0 ) = 2c0 = 0 then c0 = 0 and if

S 00 (xn ) = Sn−1
00 (x ) = 2c = 0 then c = 0.
n n n

Theorem
If f is defined at a = x0 < x1 < · · · < xn = b then f has a unique
natural cubic spline interpolant.
Natural BC Linear System (1 of 3)

In matrix form the system of n + 1 equations has the form

Ac = y where
 
1 0 0 0 ··· 0
 h0 2(h0 + h1 )
 h 1 0 · · · 0 

 0 h1 2(h1 + h2 ) h2 ··· 0 
A= .
 
. . . ..
 .. .. .. .. 
 . 

 0 0 0 hn−2 2(hn−2 + hn−1 ) hn−1 
0 0 0 0 ··· 1

Note: A is a tridiagonal matrix.

Natural BC Linear System (2 of 3)

The vector y on the right-hand side is formed as

0
 
3 3

 h1 (a2 − a1 ) − h0 (a1 − a0 )


3 3
h2 (a3 − a2 ) − h1 (a2 − a1 )
 
 
 .. 
y=
 . 

3 3
hn−2 (an−1 − an−2 ) − hn−3 (an−2 − an−3 )
 
 
3 3
 
hn−1 (an − an−1 ) − hn−2 (an−1 − an−2 )
 
0

Note: A is a tridiagonal matrix.

Natural BC Linear System (3 of 3)

 

c0
 0
3 3
 c1   h1 (a2 − a1 ) − h0 (a1 − a0 )

   3 3

c2 h2 (a3 − a2 ) − h1 (a2 − a1 )
   
  
A =

.. .. 
.   .
 
   
 cn−1 3 3
hn−1 (an − an−1 ) − hn−2 (an−1 − an−2 )
 



cn 0

We solve this linear system of equations using a common

algorithm for handling tridiagonal systems.
Natural Cubic Spline Algorithm
INPUT {(x0 , f (x0 )), (x1 , f (x1 )), . . . , (xn , f (xn ))}
STEP 1 For i = 0, 1, . . . , n − 1 set ai = f (xi ); set
hi = xi+1 − xi .
STEP 2 For i = 1, 2, . . . , n − 1 set
3 3
αi = (ai+1 − ai ) − (ai − ai−1 ).
hi hi−1
STEP 3 Set l0 = 1; set µ0 = 0; set z0 = 0.
STEP 4 For i = 1, 2, . . . , n − 1 set
hi
li = 2(xi+1 − xi−1 ) − hi−1 µi−1 ; set µi = ; set
li
αi − hi−1 zi−1
zi = .
li
STEP 5 Set ln = 1; set cn = 0; set zn = 0.
STEP 6 For j = n − 1, n − 2, . . . , 0 set cj = zj − µj cj+1 ; set
aj+1 − aj hj (cj+1 + 2cj ) cj+1 − cj
bj = − ; set dj = .
hj 3 3hj
STEP 7 For j = 0, 1, . . . , n − 1 OUTPUT aj , bj , cj , dj .
Example (1 of 4)

Construct the natural cubic spline interpolant for

f (x) = ln(ex + 2) with nodal values:

x f (x)
−1.0 0.86199480
−0.5 0.95802009
0.0 1.0986123
0.5 1.2943767

Calculate the absolute error in using the interpolant to

approximate f (0.25) and f 0 (0.25).
Example (2 of 4)

In this case n = 3 and

h0 = h1 = h2 = 0.5

with

a0 = 0.86199480, a1 = 0.95802009,
a2 = 1.0986123, a3 = 1.2943767.

The linear system resembles,

    
1.0 0.0 0.0 0.0 c0 0.0
 0.5 2.0 0.5 0.0   c1   0.267402 
Ac =  0.0 0.5 2.0 0.5   c2  =  0.331034  = y
   

0.0 0.0 0.0 1.0 c3 0.0

Example (3 of 4)
The coefficients of the piecewise cubics:
i ai bi ci di
0 0.861995 0.175638 0.0 0.0656509
1 0.95802 0.224876 0.0984763 0.028281
2 1.09861 0.344563 0.140898 −0.093918
Example (3 of 4)
The coefficients of the piecewise cubics:
i ai bi ci di
0 0.861995 0.175638 0.0 0.0656509
1 0.95802 0.224876 0.0984763 0.028281
2 1.09861 0.344563 0.140898 −0.093918
The cubic spline:

0.861995 + 0.175638(x + 1) if −1 ≤ x ≤ −0.5



+ 0.0656509(x + 1)3





 0.95802 + 0.224876(x + 0.5)


S(x) = + 0.0984763(x + 0.5)2 if −0.5 ≤ x ≤ 0
 + 0.028281(x + 0.5)

 3

1.09861 + 0.344563x if 0 ≤ x ≤ 0.5




+ 0.140898x 2 − 0.093918x 3

Example (4 of 4)

y
1.3

1.2

1.1

1.0

x
-1.0 -0.8 -0.6 -0.4 -0.2 0.2 0.4

f (0.25) S(0.25) Abs. Err. f 0 (0.25) S 0 (0.25) Abs. Err.

1.18907 1.19209 3.02154 × 10−3 0.390991 0.3974 6.40839 × 10−3
Clamped Boundary Conditions (1 of 2)

If S 0 (a) = S00 (a) = f 0 (a) = b0 then

1 h0
f 0 (a) = (a1 − a0 ) − (2c0 + c1 )
h0 3
which is equivalent to

3
h0 (2c0 + c1 ) = (a1 − a0 ) − 3f 0 (a).
h0

This replaces the first equation in our system of n equations.

Clamped Boundary Conditions (2 of 2)

Likewise if S 0 (b) = Sn0 (b) = f 0 (b) = bn then

bn = bn−1 + hn−1 (cn−1 + cn )

1 hn−1
= (an − an−1 ) − (2cn−1 + cn ) + hn−1 (cn−1 + cn )
hn−1 3
1 hn−1
= (an − an−1 ) + (cn−1 + 2cn )
hn−1 3

which is equivalent to

3
hn−1 (cn−1 + 2cn ) = 3f 0 (b) − (an − an−1 ).
hn−1

This replaces the last equation in our system of n equations.

Clamped BC Linear System (1 of 2)

Theorem
If f is defined at a = x0 < x1 < · · · < xn = b and differentiable at
x = a and at x = b, then f has a unique clamped cubic spline
interpolant.
In matrix form the system of n equations has the form Ac = y
where
 
2h0 h0 0 0 ··· 0
 h0 2(h0 + h1 )
 h1 0 ··· 0 

 0 h 1 2(h 1 + h2 ) h 2 · · · 0 
A= .
 
. . . ..
 .. .. .. .. 
 . 

 0 0 0 hn−2 2(hn−2 + hn−1 ) hn−1 
0 0 0 ··· hn−1 2hn−1

Note: A is a tridiagonal matrix.

Clamped BC Linear System (2 of 2)

3
− a0 ) − 3f 0 (a)
 
  h0 (a1
c0 3 3
h1 (a2 − a1 ) − h0 (a1 − a0 )
 
 
 c1 3 3
h2 (a3 − a2 ) − h1 (a2 − a1 )
  
   
 c2
..
  
A . =
   
 .. . 
3 3
  
hn−2 (an−1 − an−2 ) − hn−3 (an−2 − an−3 )
   
 cn−1   
3 3
hn−1 (an − an−1 ) − hn−2 (an−1 − an−2 )
 
cn  
3
3f 0 (b) − hn−1 (an − an−1 )
Coefficients of the Cubic Splines

Since aj for j = 0, 1, . . . , n is known, once we solve the linear

system for cj (again for j = 0, 1, . . . , n) then

1 hj
bj = (aj+1 − aj ) − (cj+1 + 2cj )
hj 3
1
dj = (cj+1 − cj )
3hj

for j = 0, 1, . . . , n − 1.
Clamped Cubic Spline Algorithm (1 of 2)

INPUT {(x0 , f (x0 )), (x1 , f (x1 )), . . . , (xn , f (xn ))}, f 0 (x0 ), and
f 0 (xn ).
STEP 1 For i = 0, 1, . . . , n − 1 set ai = f (xi ); set
hi = xi+1 − xi .
3(a1 − a0 )
STEP 2 Set α0 = − 3f 0 (x0 );
h0
3(an − an−1 )
αn = 3f 0 (xn ) − .
hn−1
STEP 3 For i = 1, 2, . . . , n − 1 set
3 3
αi = (ai+1 − ai ) − (ai − ai−1 ).
hi hi−1
α0
STEP 4 Set l0 = 2h0 ; µ0 = 0.5; z0 = .
l0
Clamped Cubic Spline Algorithm (2 of 2)

STEP 5 For i = 1, 2, . . . , n − 1 set

hi
li = 2(xi+1 − xi−1 ) − hi−1 µi−1 ; µi = ;
li
αi − hi−1 zi−1
zi = .
li
αn − hn−1 zn−1
STEP 6 Set ln = hn−1 (2 − µn−1 ); zn = ;
ln
cn = zn .
STEP 7 For j = n − 1, n − 2, . . . , 0 set cj = zj − µj cj+1 ;
aj+1 − aj hj (cj+1 + 2cj ) cj+1 − cj
bj = − ; dj = .
hj 3 3hj
STEP 8 For j = 0, 1, . . . , n − 1 OUTPUT aj , bj , cj , dj .
Example (1 of 4)

Construct the clamped cubic spline interpolant for

f (x) = ln(ex + 2) with nodal values:

x f (x)
−1.0 0.86199480
−0.5 0.95802009
0.0 1.0986123
0.5 1.2943767

Calculate the absolute error in using the interpolant to

approximate f (0.25) and f 0 (0.25).
Example (2 of 4)
In this case n = 3 and

h0 = h1 = h2 = 0.5

with

a0 = 0.86199480, a1 = 0.95802009,
a2 = 1.0986123, a3 = 1.2943767.

Note that f 0 (−1) ≈ 0.155362 and f 0 (0.5) ≈ 0.451863.

The linear system resembles,
    
1.0 0.5 0.0 0.0 c0 0.110064
 0.5 2.0 0.5 0.0   c1   0.267402 
Ac =   =  = y.
 0.0 0.5 2.0 0.5   c2   0.331034 
0.0 0.0 0.5 1.0 c3 0.181001
Example (3 of 4)
The coefficients of the piecewise cubics:
i ai bi ci di
0 0.861995 0.155362 0.0653748 0.0160031
1 0.95802 0.23274 0.0893795 0.0150207
2 1.09861 0.333384 0.11191 0.00875717
Example (3 of 4)
The coefficients of the piecewise cubics:
i ai bi ci di
0 0.861995 0.155362 0.0653748 0.0160031
1 0.95802 0.23274 0.0893795 0.0150207
2 1.09861 0.333384 0.11191 0.00875717
The cubic spline:

0.861995 + 0.155362(x + 1)


+ 0.0653748(x + 1)2



 if −1 ≤ x ≤ −0.5
3




 + 0.0160031(x + 1)
0.95802 + 0.23274(x + 0.5)

S(x) =

 + 0.0893795(x + 0.5)2 if −0.5 ≤ x ≤ 0
 + 0.0150207(x + 0.5)

 3

2
 1.09861 + 0.333384x + 0.11191x if 0 ≤ x ≤ 0.5



3

+ 0.00875717x
Example (4 of 4)

y
1.3

1.2

1.1

1.0

x
-1.0 -0.8 -0.6 -0.4 -0.2 0.2 0.4

f (0.25) S(0.25) Abs. Err. f 0 (0.25) S 0 (0.25) Abs. Err.

1.18907 1.18991 1.97037 × 10−5 0.390991 0.390982 9.67677 × 10−6
Error Analysis

Theorem  
Let f ∈ C 4 [a, b] with max f (4) (x) = M. If S is the unique
 
a≤x≤b
clamped cubic spline interpolant to f with respect to nodes
a = x0 < x1 < · · · < xn = b, then for all x ∈ [a, b],

5M
|f (x) − S(x)| ≤ max (xj+1 − xj )4 .
384 0≤j≤n−1
Example

Earlier we found the clamped cubic spline interpolant for

f (x) = ln(ex + 2). In this example xj+1 − xj = 0.5 for all j.
Note that
2ex (4 − 8ex + e2x )
f (4) (x) =
(2 + ex )4
 
 (4) 
max f (x) ≈ 0.120398
−1≤x≤0.5

|f (0.25) − S(0.25)| = 1.97037 × 10−5

5(0.120398)
≤ (0.5)4
384
≈ 9.798 × 10−5 .
Natural Cubic Spline Example (1 of 3)

Consider the following data:

x f (x)
−0.5 −0.02475
−0.25 0.334938
0.0 1.101

The linear system takes the form

Ac = y
    
1.00 0.00 0.00 c0 0.00
 0.25 1.00 0.25   c1  =  4.8765 
0.00 0.00 1.00 c2 0.00
Natural Cubic Spline Example (2 of 3)

The coefficients of the natural cubic spline interpolant are

ai bi ci di
−0.02475 1.03238 0.0 6.502
0.334938 2.2515 4.8765 −6.502
and the piecewise cubic is
−0.02475 + 1.03238(x + 0.5) + 6.502(x + 0.05)3

if −0.5 ≤ x ≤ −0.25
S(x) =
0.334938 + 2.2515(x + 0.25) + 4.8765(x + 0.25)2 − 6.502(x + 0.25)3 if −0.25 ≤ x ≤ 0.
Natural Cubic Spline Example (3 of 3)

fHxL

0.8

0.6

0.4

0.2

x
-0.5 -0.4 -0.3 -0.2 -0.1
Clamped Cubic Spline Example (1 of 4)

Here we will find the√ clamped cubic spline interpolant to the

function f (x) = J0 ( x) at the nodes xi = 5i for i = 0, 1, . . . , 10.

x f (x)
0.0 1.0
5.0 0.0904053
10.0 −0.310045
.. ..
. .
50.0 0.299655

Note: f 0 (0) = −0.25 and f 0 (50) = −0.00117217.

Clamped Cubic Spline Example (2 of 4)

The tridiagonal linear system takes the following form

 
0.204243
 

 0.305487 

10 5 0 0 · · · 0 0 0 0 c0 
 0.184846 


 5 20 5 0 · · · 0 0 0 0   c1  
  
 0.100749 


 0 5 20 5 · · · 0 0 0 0    c2  
  
0.044242


.. .. ..   ..  =  0.008211 .
 
 . .  .  
.    
−0.012944
 
 0 0 0 0 ··· 5 20 5 0   c8   
−0.023582
    
 0 0 0 0 ··· 0 5 20 5   c9   
−0.027056
 
0 0 0 0 ··· 0 0 5 10 c10 



 −0.025905 
−0.011808
Clamped Cubic Spline Example (3 of 4)

The coefficients of the clamped cubic spline interpolant are

ai bi ci di
1 −0.25 0.0154655 −0.00036986
0.09040533 −0.1230843 0.009917643 −0.0002637577
−0.3100448 −0.0436897 0.005961278 −0.0001836499
−0.4024176 0.00214934 0.003206529 −0.0001229411
−0.3268753 0.02499404 0.001362412 −0.0000780158
−0.1775968 0.03276697 0.000192174 −0.0000454083
−0.0146336 0.03128308 −0.00048895 −0.0000224102
0.12675676 0.02471281 −0.00082510 −6.79522 × 10−6
0.22884382 0.01595213 −0.00092703 3.265389 × 10−6
0.28583684 0.00692671 −0.00087805 9.088463 × 10−6
Clamped Cubic Spline Example (4 of 4)

y
1.0

0.8

0.6

0.4

0.2

x
10 20 30 40 50
-0.2

-0.4
Homework

I Read Section 3.5

I Exercises: 1, 3d, 5d, 7d, 25