MATA266
4
4.3 Smoothing: Low-order
Polynomial Models
Smoothing
Use a low order polynomial instead of a high(er) order polynomial.
This polynomial will then not pass through all the points.
2 decisions:
• The order/degree of the polynomial to use
• Criterion to use for the best-fitting polynomial
Divided differences
– To answer the question about which order
polynomial to use, we first take a look at the
concept of divided differences
X Y d dd ddd dddd
0 1
y2 y1 2 1
1
x2 x1 1 0
d 2 d1 2 1
1 2 0.5
x3 x1 2 0
dd 2 dd1 1 0.5
y3 y2 4 2 0.1667
2 x4 x1 3 0
x3 x2 2 1
d3 d 2 4 2 ddd 2 ddd1 0.33 0.167
2 4 1 0.04167
x4 x2 3 1 x5 x1 4 0
y4 y3 8 4 dd3 dd 2 2 1
4 0.3333
x4 x3 3 2 x5 x2 4 1
d 4 d3 8 4
3 8 2
x5 x3 4 2
y5 y4 16 8
8
x5 x4 4 3
4 16
A different formulation
f xi f x j
f xi , x j
xi x j
f xi , x j f x j , xk
f xi , x j , xk
xi x k
f xn , xn 1 ,, x2 f xn 1 , xn 2 ,, x1
f xn , xn 1,, x2 , x1
xn x1
Pertinent properties of polynomials
in terms of their order.
Look for a pattern:
-- the degree is 1 and ,
-- the degree is 2 and ,
-- the degree is 3 and ,
For an degree polynomial , we will have and
We consider the divided differences as approximations to the
derivative.
Therefore, to decide which order polynomial to use on the
data, we set up a divided differences table and look for the
column where the values are almost equal(constant) and if
the column after contains values that are close to zero, it
further confirms our suspicion of which order polynomial can
be considered appropriate.
So, if the column has values close to equal, we suspect we
should use a polynomial of order .
For this
data we will
use a cubic
polynomial
Note how the 3rd order differences are
constant and the 4th order differences are 0.
Stopping
distance
revisited
– Without an explicative model
(theory-based) available, what
can the data tell us about what
type of model to use?
– As the 2nd order differences are
somewhat similar and the 3rd
order differences are small, we
can conclude that a 2nd degree
polynomial might be a good
starting point for a model.
– We see that a 3rd degree model
can also work well, so we have
to decide if the extra complexity
will be worth it or not.
Explicative vs Empirical
– In chapter 2 we found the
explicative stopping distance
model:
– With the data available we
can use least squares fitting
to determine the best 2nd
degree polynomial of the
form . We obtain:
Growth of a yeast culture
