Linear and quadratic approximation

November 11, 2013

Definition: Suppose f is a function that is differentiable on an interval I containing the

point a. The linear approximation to f at a is the linear function

L(x) = f (a) + f 0 (a)(x − a), for x in I.

Now consider the graph of the function and pick a point P not he graph and look at
the tangent line at that point. As you zoom in on the tangent line, notice that in a small
neighbourhood of the point, the graph is more and more like the tangent line. In other
words, the values of the function f in a neighbourhood of the point are being approximated
by the tangent line.
Suppose f is differentiable on an interval I containing the point a. The change in the
value of f between two points a and a + ∆x is approximately ∆y = f 0 (a).∆x, where a + ∆x
is in I.
Uses of Linear approximation
• To approximate f near x = a, use

f (x) ≈ L(x) = f (a) + f 0 (a)(x − a).

Alternately,
f (x) − f (a) ≈ f 0 (a)(x − a).

• To approximate the change ∆y in the dependent variable given a change ∆x in the

independent variable, use
∆y ≈ f 0 (a)∆x.

Differentials
Let f be differentiable on an interval containing x. A small change in x is denoted by
the differential dx. The corresponding change in y = f (x) is approximate by the differential
dy = f 0 (x)dx; that is

∆y = f (x + dx) − f (x) ≈ dy = f 0 (x)dx.

1
Quadratic approximation
Recall that if a function f is differentiable at a point a, then it can be approximated
near a by its tangent line. We say that the tangent line provides a linear approximation to
f at the point a. Recall also that the tangent line at the point (a, f (a)) is given by

y − f (a) = f 0 (a)(x − a) a or y = f (a) + f 0 (a)(x − a).

The linear approximation function p1 (x) is the polynomial

p1 (x) = f (a) + f 0 (a)(x − a)

of degree one; this is just the equation to the tangent line at that point. This polynomial
has the property that it matches f in value and slope at the point a. In other words,

p1 (a) = f (a) and p01 (a) = f 0 (a).

We would like to do better, namely we would also like to get a better approximation to
the function in that we match concavity of f as well at this point. Recall that the concavity
information is coded in the second derivative. We thus create a quadratic approximating
polynomial p2 (x) of degree two defined by:

p2 (x) = f (a) + f 0 (a)(x − a) + c2 (x − a)2 ; p2 (x) = p1 (x) + c2 (x − a)2 ,

where c2 = 1/2f 00 (a). So we have

1 1
p2 (x) = f (a) + f 0 (a)(x − a) + f 00 (a)(x − a)2 = p1 (x) + f 00 (a)(x − a)2 .
2 2
Note that
p2 (a) = f (a), p02 (a) = f 0 (a), p002 (a) = f 00 (a),
where we assume that f and its first and second derivatives exist at a. The polynomial
p2 (x) is the quadratic approximating polynomial for f at the point a. The quadratic
approximation gives a better approximation to the function near a than the linear approx-
imation.
In solving linear approximation problems, you should first look for the function f (x)
as well as the point a, so that you can approximate f at a point close to a. The advantage
of linear approximation is the following; the function f that one is considering might be
very complicated, but *near* a point a in its domain, we are able to estimate the value of
the function using a polynomial of degree one (i.e. a linear function) which is far simpler
than the original function. For example, let us consider the following
√
PROBLEM: Approximate the number 4 1.1.

2
 We use the fact that a curve lies very close to its tangent (linear approximation) near
√
the point of tangency. So let f (x) = 4 x. First we look for a point a close to 1.1, such that
f (a) is easy to compute, but f (x) is difficult to compute for x close to a. So we use the
tangent line at (a, f (a)) and use it√as an approximation to the curve y = f (x).
Clearly, 1 is close to 1.1 and 4 1 is easy to compute! So we take a = 1. Then the
tangent line is
y = f (a) + f 0 (a)(x − a).
As f (a) = 1 and f 0 (a) = 1
4a3/4
. The equation to the tangent line is

f (a) + f 0 (a)(x − a),

and
f (x) ≈ f (a) + f 0 (a)(x − a).
Putting a = 1 and x = 1.1, we have
√
4
√
4 1 1
1.1 ≈ 1 + (1.1 − 1) = 1 + (0.1) = 1.025.
4.13/4 4

Underestimates and overestimates

If f (x) is concave up in a neighbourhood of (a, f (a)), then the tangent line lies below
the graph of f , and the approximated value is an underestimate. In this case, the value
obtained by linear approximation is less than the actual value of f (x) in a neighbourhood
of a. If the tangent line lies above the graph of f , then the approximated value is an
overestimate. We remake that linear approximation gives good estimates when x is close
to a but the accuracy of the approximation gets worse when √ the points are farther away
from 1. Also, a calculator would give an approximation for 4 1.1, but linear approximation
gives an approximation over a small interval around 1.1.

Percentage Error
Suppose you have used linear or quadratic approximation centred around a to approx-
imate f (c), for a point c close to a.The percentage error is calculated using the formula
(T rue value) − (approximated value)
x100.
(T rue V alue)
Thus if you have used linear approximation using the linear polynomial, the approximated
value at a point c close to a is given by p1 (c) and the percentage error is given by
f (c) − p1 (c)
× 100.
f (c)
If you have used quadratic approximation using the formula p2 (x), we would use p2 (c)
instead of p1 in the above formula.

3
Taylor series
Recall that
p1 (x) = f (a) + f 0 (a)(x − a)
is the linear approximating polynomial for f (x) centred around a and
p2 (x) = f (a) + f 0 (a)(x − a) + f 00 (a)/2(x − a)2
is the quadratic polynomial. The polynomial p1 (x) is the equation of the tangent line at
the centre point (a, f (a), and these are used to estimate values of f (x) for points x that
are close to a. For example, you should check that for f (x) = Sin x, with centre a = 0, we
have p1 (x) = x. The quadratic approximation gives a better estimate for f (x) for x near a
when compared with the linear polynomial f1 (x).
Suppose a function f has derivatives f (k) (a) of all orders at the point a. The Taylor
polynomial of order (or degree) n is defined by
pn (x) = c0 + c1 (x − a) + c2 (x − a)2 + · · · + cn (x − a)n
(k)
where ck = f k!(a) .
Of course, p1 (x) and p2 (x) are as above. If we let n → ∞, we obtain a power series,
called the Taylor series for f centred at a. The special case of a Taylor series centred at 0
is called a Maclaurin series. The higher order Taylor polynomials give better and better
approximations for f (x) in a neighbourhood of the centre a.

Errors

Definition: The error in the linear approximation of a function f (x) is M where |R(x)| =
|f (x) − p1 (x)| ≤ M.
To find M , we should find an upper bound for the difference between the actual value
and the approximated value p1 (x). Notice that the difference between p1 and f increases
as we mover farther away from the centre a. Notice also that the difference increases if
the function bends away from the tangent. In other words, the larger the absolute value
of the second derivative at a, | f 00 (a) |, the greater the deviation of f (x) from the tangent
line. The linear approximation at a is more accurate for f , when the rate of change of f 0 ,
which is nothing else but f 00 is smaller.

Remainder
In general, write f (x) = pn (x) + Rn (x), where n = 1 or 2 so that p1 (x) is the linear
approximating polynomial and p2 (x) is the quadratic approximation polynomial. To esti-
mate the remainder term Rn (x), we find a number M such that | f (n+1 )(c) |≤ M , for all
c between a and x inclusive, Then the remainder satisfies
| x − a |n+1
| Rn (x) |=| f (x) − pn (x) |≤ M ,
(n + 1)!

4
where p1 (x) (respectively p2 (x)) is the linear (respectively quadratic) approximating poly-
nomial centred around a. A similar formula holds for the remainder term for the n-th order
Taylor polynomial.

Derivatives of inverse trigonometric functions

d −1 √ 1 for − 1 < x < 1.
dx (sin x) = 1−x2
d −1 −1
dx (cos x) = √
1−x2
for − 1 < x < 1.
d −1 1
for − ∞ < x < ∞.
dx (tan x) = √
1+x2
d −1 1
for − ∞ < x < ∞.
dx (cot x) = √
1+x2
d −1 1
for |x| > 1.
dx (sec x) = √
|x| x2 +1
d −1 1
for |x| > 1.
dx (cosec x) = √
|x| x2 −1