1 Linearisation & Differentials
1 Linearisation & Differentials
Since
1
f ′ (x) = √ ,
2 1+x
√ √
we have f (0.96) = 1.96 = 1.4 = 7/5, and f ′ (0.96) = 1/(2 1.96) = 10/28 = 5/14.
7 5 24 5x 37
L(x) = + x− = + .
5 14 25 14 35
0.5
0.0
x
√
Figure 1. The solid line is f (x) = 1 + x, and its linearisation L(x) at x = 0.96 is the broken line.
Clearly, as we move away from the neighbourhood of x = 0.96, the error becomes larger; but for a small
neighbourhood around 0.96, L(x) will be reasonably accurate.
1
It is straightforward to generalise the preceding result to obtain linearisations at some
point x = a of f (x).
L(x) = 1 + kx.
Verify this result on your own. This result implies that for small x, we have (1 + x)k ≈
1 + kx. ♣
The correct answer (to 3 decimal places) is 1.105, so the error is only about 0.5%. ♣
L(x) = 0 + (1)(x − 0) = x.
This means that sin x ≈ x for small values of x, leading to the result
sin x
≈ 1.
x
Hence intuitively, the limit of sin x/x as x → 0 should be 1. Note that L(x) has helped
us to study the behaviour of sin x/x for small values of x. ♣
2
We know from our geometrical study of the meaning of a derivative that
dy δy f (x + h) − f (x)
= lim = lim .
dx h→0 δx h→0 h
Accordingly, it does not make sense to view dy/dx as a ratio. For example,
d(sin x)
= cos x,
dx
and it seems ridiculous that we can treat the derivative as a ratio and write
because it is not clear what is meant from such operations. Fortunately, the concept of a
differential justifies such operations, and vastly extends the usefulness of the derivative.
dy = f ′ (x)dx.
It is not hard to see why Definition 2 works. From our earlier understanding of the
derivative, the tangent line at any point on f (x) has slope f ′ (x). If we complete a right-
angled triangle with small base dx, then the height dy must be f ′ (x)dx. This alternative
view of the derivative allows us to use the derivative as a means of estimating change.
According to Definition 2, if the ratio dy/dx exists, then it will be equal to the derivative.
For example, if dx 6= 0, then
dy f ′ (x)dx
= = f ′ (x).
dx dx
However, it is useful to see that we can obtain the same result in the following way
x (x + 1)dx − xd(x + 1) (x + 1)dx − xdx dx
d = = = .
x+1 (x + 1) 2 (x + 1) 2 (x + 1)2
3
We can use differentials to estimate change in the value of f (x), as we move a small
distance dx away from some particular point of the function.
√
EXAMPLE 5. Let f (x) = 1 + x. What is the estimated relative change in f (x) as we
move away from x = 0.96 to x = 1 (that is, dx = 0.04)?
√
SOLUTION. The true relative change is given by (f (1) − f (0.96))/f (0.96) = ( 2 −
1.4)/1.4 ≈ 0.01015. Let’s see how good is our differential estimation. The estimated
relative change is given by
df f ′ (0.96)(0.04) 0.04
= = ≈ 0.01020,
f (0.96) f (0.96) 2(1.96)
4kr 3 df
dV dr
= =4 .
V kr 4 r
A relative increase of 10% in r means dr/r = 0.1, so dV /V = 0.4. In words, this means
a 10% increase in r will result in 40% increase in flow. Doctors can use this principle to
calculate how large should the balloon at the end of a cathether be inflated in order to
restore a certain percentage of flow in a clogged artery. ♣
4
2 Newton’s Method for Finding Roots
A numerical method known as Newton’s Method (or Newton-Raphson’s Method) is used
for solving equations of the form
f (x) = 0.
This particular method is an example of numerical methods that make use of gradient
information and linearisation. It works in the following way. From a starting point x0 ,
the method finds the linearisation of f (x) at x0 , and then solves for
f (x0 )
x1 = x0 − .
f ′ (x0 )
Subsequently, the process is iterated using x1 as the input value, until convergence is
achieved, or a specified number of iterations has been reached. Usually, when the differ-
ence between two consecutive solutions xn and xn−1 is less than some threshold value, say
10−5 , we say that convergence has been attained. Figure 2 illustrates Newton’s method
in action for finding the positive root of (x − 1)(x + 1) = 0.
8
6
(x − 1) * (x + 1)
4
2
0
x2 x1 x0
Figure 2. The linearisation of f (x) at the initial guess value x0 = 2.5 intersects the x-axis at x = 1.45,
which is an improved guess of the root. By successively applying linearisations of f (x) at such improved
guesses, Newton’s Method leads to the solution x = 1.
5
EXAMPLE 7. Find the solution cos x = x.
In general, Newton’s Method will work for “well-behaved” functions, but it is im-
portant to be cautious in interpreting the solutions. When multiple roots exist, different
starting points will lead to different roots, so it is important to try to find out from theory
(if possible), how many solutions exist. In addition, iterations may occasionally “hang”
if f ′ (xn ) = 0, since there will not be an intersection point to define xn+1 . In this case, we
should restart the process using a different starting point. Finally, for some functions,
Newton’s Method may fail to converge due to a “cycling effect”.
KTF 2009/2010(2)
6