Linearization, Differentials, and Taylor Polynomials

by Arlan Rodrigo

A. LINEARIZATION

Generally speaking, given a function, it is impossible to compute for all the function values at each number
in its domain.

For example, let , and f x = x we want to compute for 4.001 . It would be impossible to get the exact
value but we can give a good approximation to it by finding the equation of the tangent line T x to the
curve at x = 4 and then evaluate T 4.001 , which is easily computed. The idea for linearization is that the
tangent line to the curve at x = 4 gives a good approximation to f x for values of x sufficiently "close" to
4 since both and the tangent line T x at x = 4 have the same rate of change there. In other words, when
x is very near 4, the tangent line becomes almost indistinguishable to the curve. This means that for values
of x near 4 on f x , it can be given a good approximation by its function value at the tangent line instead.

50
O func d 1 Cx
50
func := 1 Cx (1)
O ad0
a := 0 (2)
d
O plot eval func, x = a $ x Ka Ceval func, x = a , func , x = 0 ..9 ;
dx

1
O plot $ x Ka Ceval func, x = a , func , x = 0 ..9 ;
3

d
O eval func, x = a $ x Ka Ceval func, x = a
dx
50 x C1 (3)
O eval (3), x = .0002
1.0100 (4)
O
O
O 1.0002^50;
1.010049157 (5)
Note that this approximation works since 4.001 is "close enough" to 4. Here, we say that the tangent line T
is a form of linearization of f at x = 4

Note that there are other linearizations of f at x = 4. However, as mentioned earlier, the tangent line T
( whenever it exists ) gives us the best linearization of a function at a number "a" since both f and T have
the same rate of change at "a". So from now on, we will refer to the tangent line of f at "a" as the linear -
ization of f at that point. More precisely, the linearization at the number "a" will be given by :

T x = f a Cf ' a x Ka , and we have T x z f x when x is "near" "a"

So in the previous example, f 4.001 z T 4.001 = 2.00025

Note : The challenge for approximating a certain number "f a " by this method is finding an appropriate
function f where "f a " is a function value for some number "a" in its domain, and then finding another
appropriate number in its domain sufficiently close to "a" where the derivative is easily calculated.

Also note that differentiability at some open interval containing " a" is necessary to find tangent lines.

Question : What is the utility of linearization ?

Some Examples :
50 50
1.) Find a linearization of f x = 1 Cx where k is a constant at x = 0. Hence approximate 1.0002
1
2.) Find a linearization of (a) f x = x C at x = 1 (b) f x = sin x at x = 0
x
3.) Estimate the value of the following by means of an appropriate linearization
1
(a) tan 44 + (b)
1002
4.) Let f x = 2 x K3 x3 C5 x2 K3 x C2. Use a linear approximation to estimate f 0.01 . Then, show
4

f 0.01 z T x at x = 0 evaluated at 0.01

that the linear approximation is an underestimate of the actual value.
K1
5.) The following facts are known about the function f x : (a) f 2 = 4 (b) f ' x = x4 C1 for
all x. Use a linear approximation to estimate f 2.05
6.) Given f x , approximate the function values at x = 0.99 by using a linearization at x = 1
2
(a) f x = 2 (b) 2 x
x

B. DIFFERENTIALS

The concept of linearlization can be translated in terms of differentials. Let y = f x . If f is a differentiable

function, we let dx be any real number. And at any pt. x on the domain of f, the slope of the tangent line is
given by f ' x . We let the variable dy = f ' x dx so that dy is a function of x. The geometric interpre-
tation is as follows :

Notice that by letting 6x = dx, 6y = f x Cdx Kf x , and this can be approximated by dy if dx is

sufficiently "small". In other words, 6y = f x Cdx Kf x z dy, f x Cdx z f x Cdy. Let us
re-do the previous example of approximating 4.001 but now by using differentials.
 1
We have y = f x = x . Then dy = dx. Since we want to estimate 4.001, we let x = 4, dx = .001
2 x
1
Then with this, dy = $.001 = .00025.
4
So 4.001 = f 4.001 = f 4 C.001 z 2 C.00025 = 2.00025
dy
Relative Error in approximating 6y : , Percentage error = Relative Error x 100
y

Some Examples :

1.) Re-do the previous example 3 using differentials.

2.) One side of a right triangle is known to be 20 cm. long and the opposite angle is measured as 30 + , with
a
possible error of G1 +
(a) Use differentials to estimate the error in computing the length of the hypotenuse.
(b) What is the percentage error ?
3.) The radius of a circular disk is given as 24 cm. with a maximum error in measurement of 0.2 cm.
(a) Use differentials to estimate the maximum error in the calculated area of the disk.
(b) What is the relative error ? What is the percentage error ?
4.) Use differentials to approximate the volume of material needed to make a rubber ball if the radius of the
1
hollow inner core is 2 in. and the thickness of the rubber is in.
8
5.) Find dy and 6y for the values of x and 6x
1
(a) y = x2, x = 2, 6x = 0.5 (b) y = , x = 3, 6x =K0.2
x

C. TAYLOR POLYNOMIALS

** we will assume here that we have a function f that is "many" times differentiable for some open interval

containing the number "a". **

As we have mentioned in the previous discussion on linearization, the tangent line T x is the best linear
approximation to the graph of the function f near "a". Even though linearization has been a good tool for
us thus far, we can try to be more ambitious and try to give a better approximation of f near "a" by using
higher degree polynomials. For example, we can try to make a quadratic approximation T2 x to
f x = x at x = 4 by stipulating the following conditions for T2 x :

(a) f 4 = T2 4 (b) f ' 4 = T2 ' 4 (c) f '' 4 = T2 '' 4

And we know that any quadratic polynomial can be expressed as A CB$ x K4 CC$ x K4 2, for some
2
constants A, B, C. Hence, we can let T2 x = A CB$ x K4 CC$ x K4 . So what are the constants
A, B, and C ? Answer : __________________

This polynomial is the quadratic approximation of f at x = 4. For higher degree polynomial

approximations,
given that a function f is n times differentiable, we require that each nth derivative of the function f
evaluated at 4 to be equal to the nth derivative of the polynomial Tn x evaluated at 4. In general, the
number 4 can be made aritrary here and set as a variable "a". Such polynomials are called Taylor
polynomials of degree n centered at the number "a". ( named after English Mathematician Brook
Taylor ). A special case wherein a = 0 is called a Maclaurin polynomial named after Scottish
Mathematician
Colin Maclaurin.

Note : Higher degree Taylor polynomials tend to be better approximations for a function near the number
"a",
and actually for some Taylor polynomials, higher degrees would also mean a better approximation evn for
numbers that are very far from "a". But this is not always the case, however. We illustrate by two compa-
risons: (a) f x = sin x b f x = 1 Cx

O restart : with plots :

a d 0 : b d 25 : c d 0 : d d 6 : p d 4 : n d 25 :
fdx/ x :

for k from 1 to n do
m k d unapply convert taylor f x , x = p, k , polynom , x :
end do:

an1 d plot f x , x = a ..b, view = c ..d, color = RED :

for k from 1 to n do
an2 k d plot m k x , x = a ..b, color = BLUE :
end do:

q d plots display seq an2 k , k = 1 ..n , insequence = true :

display an1, q ;

O
** something to think about : what is the nth degree Taylor polynomial for f x = sin x evaluated at
x=0

Exercises :
1.) Find the quadratic approximation T2 x of f x = x at x = 1, and by this, estimate 1.02
2.) Find the quadratic approximation T2 x of f x = cos x at x = 0, and by this, estimate cos .05
1
3.) Find the quadratic approximation T2 x of f x = at x = 0
x K2