Taylor Polynomials — Approximating Functions Near a

Specified Point
Suppose that you are interested in the values of some function f (x) for x near some fixed
point x0 . The function is too complicated to work with directly. So you wish to work instead
with some other function F (x) that is both simple and a good approximation to f (x) for x
near x0 . We’ll consider a couple of examples of this scenario later. First, we develop several
different approximations.

1. Zeroth Approximation — the Constant Approximation

The simplest functions are those that are constants. The first approximation will be by a
constant function. That is, the approximating function will have the form F (x) = A, for
some constant A. To ensure that F (x) is a good approximation for x close to x0 , we choose
A so that f (x) and F (x) take exactly the same value when x = x0 .

F (x) = A so F (x0 ) = A = f (x0 ) =⇒ A = f (x0 )

Our first, and crudest, approximation rule is

f (x) ≈ f (x0 ) (1)

Here is a figure showing the graphs of a typical f (x) and approximating function F (x). At
y
y = f (x)

y = F (x) = f (x0 )

x0 x

x = x0 , f (x) and F (x) take the same value. For x very near x0 , the values of f (x) and F (x)
remain close together. But the quality of the approximation deteriorates fairly quickly as x
moves away from x0 .

2. First Approximation — the Tangent Line, or Linear, Approxi-

mation
We now develop a better approximation by allowing the approximating function to be a
linear function of x and not just a constant function. That is, we allow F (x) to be of the
form A + Bx, for some constants A and B. To ensure that F (x) is a good approximation

c Joel Feldman. 2014. All rights reserved. 1 January 29, 2013

for x close to x0 , we choose A and B so that f (x0 ) = F (x0 ) and f ′ (x0 ) = F ′ (x0 ). Then f (x)
and F (x) will have both the same value and the same slope at x = x0 .
F (x) = A + Bx =⇒ F (x0 ) = A + Bx0 = f (x0 )
F ′ (x) = B =⇒ F ′ (x0 ) = B = f ′ (x0 )
Substituting B = f ′ (x0 ) into A + Bx0 = f (x0 ) gives A = f (x0 ) − x0 f ′ (x0 ) and consequently
F (x) = A + Bx = f (x0 ) − x0 f ′ (x0 ) + xf ′ (x0 ) = f (x0 ) + f ′ (x0 )(x − x0 ). So, our second
approximation is
f (x) ≈ f (x0 ) + f ′ (x0 )(x − x0 ) (2)
You may recall that y = f (x0 ) + f ′ (x0 )(x − x0 ) is exactly the equation of the tangent line
to the curve y = f (x) at x0 . Here is a figure showing the graphs of a typical f (x) and
approximating function F (x). Observe that the graph of f (x0 ) + f ′ (x0 )(x − x0 ) remains

y y = F (x) = f (x0 ) + f ′ (x0 )(x − x0 )

y = f (x)

x0 x

close to the graph of f (x) for a much larger range of x than did the graph of f (x0 ).

3. Second Approximation — the Quadratic Approximation

We next develop a still better approximation by allowing the approximating function be to
a quadratic function of x. That is, we allow F (x) to be of the form A + Bx + Cx2 , for some
constants A, B and C. To ensure that F (x) is a good approximation for x close to x0 , we
choose A, B and C so that f (x0 ) = F (x0 ) and f ′ (x0 ) = F ′ (x0 ) and f ′′ (x0 ) = F ′′ (x0 ).
F (x) = A + Bx + Cx2 =⇒ F (x0 ) = A + Bx0 + Cx20 = f (x0 )
F ′ (x) = B + 2Cx =⇒ F ′ (x0 ) = B + 2Cx0 = f ′ (x0 )
F ′′ (x) = 2C =⇒ F ′′ (x0 ) = 2C = f ′′ (x0 )
Solve for C first, then B and finally A.
C = 21 f ′′ (x0 ) =⇒ B = f ′ (x0 ) − 2Cx0 = f ′ (x0 ) − x0 f ′′ (x0 )
=⇒ A = f (x0 ) − Bx0 − Cx20 = f (x0 ) − x0 [f ′ (x0 ) − x0 f ′′ (x0 )] − 21 f ′′ (x0 )x20
Then build up F (x).
F (x) = f (x0 ) − f ′ (x0 )x0 + 12 f ′′ (x0 )x20 (this line is A)
′ ′′
+ f (x0 ) x − f (x0 )x0 x (this line is Bx)
+ 12 f ′′ (x0 )x2 (this line is Cx2 )
= f (x0 ) + f ′ (x0 )(x − x0 ) + 21 f ′′ (x0 )(x − x0 )2

c Joel Feldman. 2014. All rights reserved. 2 January 29, 2013

Our third approximation is

f (x) ≈ f (x0 ) + f ′ (x0 )(x − x0 ) + 21 f ′′ (x0 )(x − x0 )2 (3)

It is called the quadratic approximation. Here is a figure showing the graphs of a typical
f (x) and approximating function F (x). This third approximation looks better than both

y
y = f (x)
y = F (x) = f (x0 ) + f ′ (x0 )(x − x0 ) + 21 f ′′ (x0 )(x − x0 )2

x0 x

the first and second.

4. Still Better Approximations – Taylor Polynomials

We can use the same strategy to generate still better approximations by polynomials of any
degree we like. Let’s approximate by a polynomial of degree n. The algebra will be simpler
if we make the approximating polynomial F (x) of the form

a0 + a1 (x − x0 ) + a2 (x − x0 )2 + · · · + an (x − x0 )n

Because x0 is itself a constant, this is really just a rewriting of A0 + A1 x + A2 x2 + · · · + An xn .

For example,

a0 + a1 (x − x0 ) + a2 (x − x0 )2 = a0 + a1 x − a1 x0 + a2 x2 − 2a2 xx0 + a2 x20

= (a0 − a1 x0 + a2 x20 ) + (a1 − 2a2 x0 )x + a2 x2
= A0 + A1 x + A2 x2

with A0 = a0 − a1 x0 + a2 x20 , A1 = a1 − 2a2 x0 and A2 = a2 . The advantage of the form

a0 + a1 (x − x0 ) + · · · is that x − x0 is zero when x = x0 , so lots of terms in the computation
drop out. We determine the coefficients ai by the requirements that f (x) and its approximator
F (x) have the same value and the same first n derivatives at x = x0 .

F (x) = a0 + a1 (x − x0 ) + a2 (x − x0 )2 + · · · + an (x − x0 )n
=⇒ F (x0 ) = a0 = f (x0 )
F ′ (x) = a1 + 2a2 (x − x0 ) + 3a3 (x − x0 )2 + · · · + nan (x − x0 )n−1
=⇒ F ′ (x0 ) = a1 = f ′ (x0 )
F ′′ (x) = 2a2 + 3 × 2a3 (x − x0 ) + · · · + n(n − 1)an (x − x0 )n−2
=⇒ F ′′ (x0 ) = 2a2 = f ′′ (x0 )

c Joel Feldman. 2014. All rights reserved. 3 January 29, 2013

 F (3) (x) = 3 × 2a3 + · · · + n(n − 1)(n − 2)an (x − x0 )n−3
=⇒ F (3) (x0 ) = 3 × 2a3 = f (3) (x0 )
..
.
F (n) (x) = n!an =⇒ F (n) (x0 ) = n!an = f (n) (x0 )

Here n! = n(n − 1)(n − 2) · · · 1 is called n factorial. Hence

a0 = f (x0 ) a1 = f ′ (x0 ) a2 = 1 ′′
2!
f (x0 ) a3 = 1 (3)
3!
f (x0 ) ··· an = 1 (n)
n!
f (x0 )

and the approximator, which is called the Taylor polynomial of degree n for f (x) at x = x0 ,
is

f (x) ≈ f (x0 ) + f ′ (x0 ) (x−x0 ) + 2!1 f ′′ (x0 ) (x−x0 )2 + 3!1 f (3) (x0 ) (x−x0 )3 + · · ·+ n!1 f (n) (x0 ) (x−x0 )n

or, in summation notation,

n
X
1 (ℓ)
f (x) ≈ ℓ!
f (x0 ) (x − x0 )ℓ (4)
ℓ=0

where we are using the standard convention that 0! = 1.

5. The ∆x, ∆y Notation

Suppose that we have two variables x and y that are related by y = f (x), for some function
f . For example, x might be the number of cars manufactured per week in some factory and
y the cost of manufacturing those x cars. Let x0 be some fixed value of x and let y0 = f (x0 )
be the corresponding value of y. Now suppose that x changes by an amount ∆x, from x0 to
x0 + ∆x. As x undergoes this change, y changes from y0 = f (x0 ) to f (x0 + ∆x). The change
in y that results from the change ∆x in x is

∆y = f (x0 + ∆x) − f (x0 )

Substituting x = x0 + ∆x into the linear approximation (2) yields the approximation

f (x0 + ∆x) ≈ f (x0 ) + f ′ (x0 )(x0 + ∆x − x0 ) = f (x0 ) + f ′ (x0 )∆x

for f (x0 + ∆x) and consequently the approximation

∆y = f (x0 + ∆x) − f (x0 ) ≈ f (x0 ) + f ′ (x0 )∆x − f (x0 ) =⇒ ∆y ≈ f ′ (x0 )∆x (5)

for ∆y. In the automobile manufacturing example, when the production level is x0 cars
per week, increasing the production level by ∆x will cost approximately f ′ (x0 )∆x. The
additional cost per additional car, f ′ (x0 ), is called the “marginal cost” of a car.
If we use the quadratic approximation (3) in place of the linear approximation (2)

f (x0 + ∆x) ≈ f (x0 ) + f ′ (x0 )∆x + 21 f ′′ (x0 )∆x2

c Joel Feldman. 2014. All rights reserved. 4 January 29, 2013

we arrive at the quadratic approximation

∆y = f (x0 + ∆x) − f (x0 )

≈ f (x0 ) + f ′ (x0 )∆x + 21 f ′′ (x0 )∆x2 − f (x0 )
=⇒ ∆y ≈ f ′ (x0 )∆x + 21 f ′′ (x0 )∆x2 (6)

for ∆y.

6. Examples

Example 1
As an initial example, we compute, approximately, tan 46◦ , using the constant approximation
(1), the linear approximation (2) and the quadratic approximation (3). To do so, we choose
π π
f (x) = tan x, x = 46 180 radians and x0 = 45 180 = π4 radians. This is a good choice for x0
because
• x0 = 45◦ is close to x = 46◦ . Generally, the closer x is to x0 , the better the quality of
our various approximations.

• We know the values of all trig functions at 45◦ .

The first step in applying our approximations is to compute f and its first two derivatives
at x = x0 .

f (x) = tan x =⇒ f (x0 ) = tan π4 = 1 √

′ −2 ′ 1 2
f (x) = (cos x) =⇒ f (x0 ) = cos2 (π/4)
= (1/√1 2)2 = 2 1
√
sin(π/4) 1/ 2 45◦
f ′′ (x) = −2 −cossin3 xx =⇒ f ′′ (x0 ) = 2 cos = 2 √ 1
= 2 1/2 =4
3 (π/4)
(1/ 2)3 1
π π π
As x − x0 = 46 180 − 45 180 = 180
radians, the three approximations are

f (x) ≈ f (x0 ) =1
′ π
f (x) ≈ f (x0 ) + f (x0 )(x − x0 ) = 1 + 2 180 = 1.034907

2
f (x) ≈ f (x0 ) + f ′ (x0 )(x − x0 ) + 21 f ′′ (x0 )(x − x0 )2 = 1 + 2 180
π
+ 12 4 π
180
= 1.035516

For comparison purposes, tan 46◦ really is 1.035530 to 6 decimal places.

Example 1

Warning 2.

All of our derivative formulae for trig functions were developed under the assumption
that angles are measured in radians. Those derivatives appeared in the approxima-
tion formulae that we used in Example 1, so we were obliged to express x − x0 in
radians.

c Joel Feldman. 2014. All rights reserved. 5 January 29, 2013

 Example 3
Let’s find all Taylor polyomials for sin x and cos x at x0 = 0. To do so we merely need com-
pute all derivatives of sin x and cos x at x0 = 0. First, compute all derivatives at general
x.

f (x) = sin x f ′ (x) = cos x f ′′ (x) = − sin x f (3) (x) = − cos x f (4) (x) = sin x · · ·
(7)
g(x) = cos x g ′ (x) = − sin x g ′′ (x) = − cos x g (3) (x) = sin x g (4) (x) = cos x · · ·

The pattern starts over again with the fourth derivative being the same as the original
function. Now set x = x0 = 0.

f (x) = sin x f (0) = 0 f ′ (0) = 1 f ′′ (0) = 0 f (3) (0) = −1 f (4) (0) = 0 · · ·

(8)
g(x) = cos x g(0) = 1 g ′(0) = 0 g ′′(0) = −1 g (3) (0) = 0 g (4) (0) = 1 · · ·

For sin x, all even numbered derivatives are zero. The odd numbered derivatives alternate
between 1 and −1. For cos x, all odd numbered derivatives are zero. The even numbered
derivatives alternate between 1 and −1. So, the Taylor polynomials that best approximate
sin x and cos x near x = x0 = 0 are

sin x ≈ x − 3!1 x3 + 5!1 x5 − · · ·

cos x ≈ 1 − 2!1 x2 + 4!1 x4 − · · ·

Here are graphs of sin x and its Taylor poynomials (about x0 = 0) up to degree seven.

sin x ≈ x sin x ≈ x − 3!1 x3

sin x ≈ x − 3!1 x3 + 5!1 x5 sin x ≈ x − 3!1 x3 + 5!1 x5 − 7!1 x7

c Joel Feldman. 2014. All rights reserved. 6 January 29, 2013

To get an idea of how good these Taylor polynomials are at approximating sin and cos, let’s
concentrate on sin x and consider x’s whose magnitude |x| ≤ 1. (If you’re writing software
to evaluate sin x, you can always use the trig identity sin(x) = sin(x − 2nπ), to easily restrict
to |x| ≤ π, and then use the trig identity sin(x) = − sin(x ± π) to reduce to |x| ≤ π2 and then
use the trig identity sin(x) = ∓ cos( π2 ± x)) to reduce to |x| ≤ π4 .) If |x| ≤ 1 radians (recall
that the derivative formulae that we used to derive the Taylor polynomials are valid only
when x is in radians), or equivalently if |x| is no larger than 180
π
≈ 57◦ , then the magnitudes
of the successive terms in the Taylor polynomials for sin x are bounded by

|x| ≤ 1 1
3!
|x|3 ≤ 1
6
1
5!
|x|3 ≤ 1
120
≈ 0.0083
1
7!
|x|7 ≤ 1
7!
≈ 0.0002 1
9!
|x|9 ≤ 1
9!
≈ 0.000003 1
11!
|x|11 ≤ 1
11!
≈ 0.000000025

From these inequalities, and the graphs on the previous page, it certainly looks like, for x
not too large, even relatively low degree Taylor polynomials give very good approximations.
We’ll see later how to get rigorous error bounds on our Taylor polynomial approximations.
Example 3

Example 4
Suppose that you are ten meters from a vertical pole. You were contracted to measure the
height of the pole. You can’t take it down or climb it. So you measure the angle subtended
by the top of the pole. You measure θ = 30◦ , which gives

h = 10 tan 30◦ = 10
√
3
≈ 5.77m

But there’s a catch. Angles are hard to measure accurately. Your contract specifies that the
height must be measured to within an accuracy of 10 cm. How accurate did your measurement
of θ have to be?
Solution. For simplicity, we are going to assume that the pole is perfectly straight and
perfectly vertical and that your distance from the pole was exactly 10 m. Write h = h0 + ∆h,
where h is the exact height and h0 = √103 is the computed height. Their difference, ∆h, is the
error. Similarly, write θ = θ0 + ∆θ where θ is the exact angle, θ0 is the measured angle and
∆θ is the error. Then

h0 = 10 tan θ0 h0 + ∆h = 10 tan(θ0 + ∆θ)

We apply ∆y ≈ f ′ (x0 )∆x, with y replaced by h and x replaced by θ. That is, we apply
∆h ≈ f ′ (θ0 )∆θ. Choosing f (θ) = 10 tan θ and θ0 = 30◦ and substituting in

2
f ′ (θ0 ) = 10 sec2 θ0 = 10 sec2 30◦ = 10 √2
3
= 40
3

we see that the error in the computed value of h and the error in the measured value of θ are
related by
∆h ≈ 40
3
∆θ or ∆θ ≈ 403
∆h

c Joel Feldman. 2014. All rights reserved. 7 January 29, 2013

 3 3 180
To achieve |∆h| ≤ 0.1m, we better have |∆θ| smaller than 0.1 40 radians or 0.1 40 π
= 0.43◦ .

Example 4

Example 5
Suppose that the radius of a sphere has been measured with a percentage error of at most
ε%. Find the corresponding approximate percentage error in the surface area and volume of
the sphere.
Solution. Suppose that the exact radius is r0 and that the measured radius is r0 + ∆r.
Then the absolute error in the measurement is |∆r| and, by definition, the percentage error
is 100 |∆r|
r0
. We are told that 100 |∆r|
r0
≤ ε. The surface area of a sphere of radius r is
2
A(r) = 4πr . The error in the surface area computed with the measured radius is

∆A = A(r0 + ∆r) − A(r0 ) ≈ A′ (r0 )∆r

The corresponding percentage error is

|∆A| ′
100 A(r 0)
≈ 100 |A A(r
(r0 )∆r|
0)
= 100 8πr4πr
0 |∆r|
2 = 2 × 100 |∆r|
r0
≤ 2ε
0

The volume of a sphere of radius r is V (r) = 43 πr 3 . The error in the volume computed with
the measured radius is

∆V = V (r0 + ∆r) − V (r0 ) ≈ V ′ (r0 )∆r

The corresponding percentage error is

′ 4πr02 |∆r|
100 V|∆V |
(r0 )
≈ 100 |V V(r(r0 )∆r|
0)
= 100 4πr03 /3
= 3 × 100 |∆r|
r0
≤ 3ε

We have just computed an approximation to ∆V . In this problem, we can compute the

exact error
V (r0 + ∆r) − V (r0 ) = 43 π(r0 + ∆r)3 − 34 πr03
Applying (a + b)3 = a3 + 3a2 b + 3ab2 + b3 with a = r0 and b = ∆r, gives

V (r0 + ∆r) − V (r0 ) = 34 π[r03 + 3r02 ∆r + 3r0 ∆r 2 + ∆r 3 − r03 ]

= 43 π[3r02 ∆r + 3r0 ∆r 2 + ∆r 3 ]

The linear approximation, ∆V ≈ 4πr02 × ∆r, is recovered by retaining only the first of the
three terms in the square brackets. Thus the difference between the exact error and the linear
approximation to the error is obtained by retaining only the last two terms in the square
brackets. This has magnitude
4
3
π 3r0 ∆r 2 + ∆r 3 = 43 π 3r0 + ∆r ∆r 2

or in percentage terms
1 4 2 ∆r 3
π 3r0 ∆r 2 + ∆r 3 = 100 3 ∆r = 100 3∆r ∆r ∆r ε ε





100 4 33 r02
+ r03 r0 r0
1+ 3r0
≤ 3ε 100
1+ 300
3
πr0

c Joel Feldman. 2014. All rights reserved. 8 January 29, 2013

Thus the difference between the exact error and the linear approximation is roughly a factor
ε
of 100 smaller than the linear approximation 3ε.
Example 5

Example 6
When an aircraft crosses the Atlantic ocean at a speed of u mph, the flight costs the company
u 240,000
C(u) = 100 + 3
+ u

dollars per passenger. When there is no wind, the aircraft flies at an airspeed of 550mph.
Find the approximate savings, per passenger, when there is a 35 mph tail wind and estimate
the cost when there is a 50 mph head wind.
Solution. Let u0 = 550. When the aircraft flies at speed u0 , the cost per passenger is C(u0).
By (5), a change of ∆u in the airspeed results in an change of

∆C ≈ C ′ (u0 )∆u = 31 − 240,000

   1 240,000 
u 2 ∆u = 3
− 5502 ∆u ≈ −.460∆u
0

in the cost per passenger. With the tail wind ∆u = 35 and the resulting

∆C ≈ −.460 × 35 = −16.10

so there is a savings of $16.10. With the head wind ∆u = −50 and the resulting

∆C ≈ −.4601 × (−50) = 23.01

so there is an additional cost of about $23.00.

Example 6

Example 7
To compute the height h of a lamp post, the length s of the shadow of a two meter pole is
measured. The pole is 6 m from the lamp post. If the length of the shadow was measured
to be 4 m, with an error of at most one cm, find the height of the lamp post and estimate
the percentage error in the height.
Solution. By similar triangles,
h
s 6+s 2 12 2
= =⇒ h = (6 + s) = +2
2 h s s 6 s
The length of the shadow was measured to be s0 = 4 m. The corresponding height of the
lamp post is h0 = 12s0
+ 2 = 12
4
+ 2 = 5 m. If the error in the measurement of the length of
the shadow was ∆s, then the exact shadow length was s = s0 + ∆s and the exact lamp post
height is h = f (s0 + ∆s), where f (s) = 12
s
+ 2. The error in the computed lamp post height
is ∆h = h − h0 = f (s0 + ∆s) − f (s0 ). By (5),

∆h ≈ f ′ (s0 )∆s = − 12
s2
∆s = − 12
42
∆s
0

c Joel Feldman. 2014. All rights reserved. 9 January 29, 2013

 1 12 1 3
We are told that |∆s| ≤ 10 m. Consequently |∆h| ≤ 42 10
= 40
(approximately). The
percentage error is then approximately

100 |∆h|
h0
3
≤ 100 40×5 = 1.5%

Example 7

7. The Error in the Taylor Polynomial Approximations

Any time you make an approximation, it is desirable to have some idea of the size of the error
you introduced. We will now develop a formula for the error introduced by the approximation
f (x) ≈ f (x0 ). This formula can be used to get an upper bound on the size of the error, even
when you cannot determine f (x) exactly.
By simple algebra
f (x) − f (x0 )
f (x) = f (x0 ) + (x − x0 ) (9)
x − x0
The coefficient f (x)−f
x−x0
(x0 )
of (x − x0 ) is the average slope of f (t) as t moves from t = x0 to
t = x. In the figure below, it is the slope of the secant joining the points (x0 , f (x0 )) and

y


x, f (x)

y = f (t)
x0 , f (x0 )

x0 c x t

(x, f (x)). As t moves x0 to x, the instantaneous slope f ′ (t) keeps changing. Sometimes it is
larger than the average slope f (x)−f
x−x0
(x0 )
and sometimes it is smaller than the average slope.
But there is a theorem, called the Mean–Value Theorem, which says that there must be
some number c, strictly between x0 and x, for which f ′ (c) = f (x)−f
x−x0
(x0 )
. Substituting this into
formula (9) gives

f (x) = f (x0 ) + f ′ (c)(x − x0 ) for some c strictly between x0 and x (10)

Thus the error in the approximation f (x) ≈ f (x0 ) is exactly f ′ (c)(x − x0 ) for some c strictly
between x0 and x. There are formulae similar to (10), that can be used to bound the error
in our other approximations. One is

f (x) = f (x0 )+f ′ (x0 )(x−x0 )+ 12 f ′′ (c)(x−x0 )2 for some c strictly between x0 and x (11)

c Joel Feldman. 2014. All rights reserved. 10 January 29, 2013

It implies that the error in the approximation f (x) ≈ f (x0 ) + f ′ (x0 ) (x − x0 ) is exactly
1 ′′
2
f (c) (x − x0 )2 for some c strictly between x0 and x. In general
f (x) =f (x0 ) + f ′ (x0 ) (x − x0 ) + · · · + 1 (n)
n!
f (x0 ) (x − x0 )n
+ 1
(n+1)!
f (n+1) (c) (x − x0 )n+1 for some c strictly between x0 and x (12)
That is, the error introduced when f (x) is approximated by its Taylor polynomial of degree
n, is precisely the last term of the Taylor polynomial of degree n + 1, but with the derivative
evaluated at some point between x0 and x, rather than exactly at x0 . These error formulae
are proven in the next (optional) section.
Example 8
Suppose we wish to approximate sin 46◦ using Taylor polynomials about x0 = 45◦ . Then, we
would define
f (x) = sin x x0 = 45◦ = 45 180
π
radians x = 46◦ = 46 180
π
radians x − x0 = π
180
radians
The first few derivatives of f at x0 are
f (x) = sin x f (x0 ) = √1
2

f ′ (x) = cos x f ′ (x0 ) = √1

2

f ′′ (x) = − sin x f ′′ (x0 ) = − √12

f (3) (x) = − cos x
The constant, linear and quadratic approximations for sin 46◦ are
sin 46◦ ≈ f (x0 ) = √1
2
= 0.70710678
sin 46◦ ≈ f (x0 ) + f ′ (x0 )(x−x0 ) √1 + √1 π


= 2 2 180
= 0.71944812

2
sin 46◦ ≈ f (x0 ) + f ′ (x0 )(x−x0 ) + 21 f ′′ (x0 )(x−x0 )2 = √1 + √1 π
− √12 π


2 2 180 180
= 0.71934042
The errors in those approximations are
error in 0.70710678 = f ′ (c)(x − x0 ) π


= (cos c) 180
π 2
error in 0.71944812 = 21 f ′′ (c)(x − x0 )2 = − 12 (sin c)


180
1 (3) π 3
− x0 )3 = − 3!1 (cos c)


error in 0.71923272 = 3!
f (c)(x 180

In each of these three cases c must lie somewhere between 45◦ and 46◦ . No matter what c is,
we know that | sin c| ≤ 1 and cos c| ≤ 1. Hence
π


error in 0.70710678 ≤ 180 < 0.018
2
error in 0.71944812 ≤ 21 180
π


< 0.00015
π 3
error in 0.71934042 ≤ 3!1 180


< 0.0000009

Example 8

c Joel Feldman. 2014. All rights reserved. 11 January 29, 2013

 Example 9 (ex and e)
Let f (x) = ex . Then

f (x) = ex ⇒ f ′ (x) = ex ⇒ f ′′ (x) = ex ···

f (0) = e0 = 1 ⇒ f ′ (0) = e0 = 1 ⇒ f ′′ (0) = e0 = 1 ···

Applying (12) with f (x) = ex and x0 = 0, and using that f (m) (x0 ) = ex0 = e0 = 1 for all m,
x2
ex = f (x) = 1 + x + 2!
+···+ xn
n!
+ 1
(n+1)!
ec xn+1 (13)

for some c between 0 and x. We can use this to find approximate values for the number e,
with any desired degree of accuracy. Just setting x = 1 in (13) gives

e=1+1+ 1
2!
+···+ 1
n!
+ 1
(n+1)!
ec (14)

for some c between 0 and 1. Since ec increases as c increases, this says that 1 + 1 + 2!1 + · · ·+ n!1
e
is an approximate value for e with error at most (n+1)! . The only problem with this error
bound is that it contains the number e, which we do not know. Fortunately, we can again
use (14) to get a simple upper bound on how big e can be. Just setting n = 2 in (14), and
again using that ec ≤ e, gives

e ≤ 1 + 1 + 2!1 + 3!e =⇒ 1 − 16 e ≤ 1 + 1 + 2!1 = 52 =⇒ e ≤ 52 × 56 = 3

So we now know that 1 + 1 + 2!1 + · · · + n!1 is an approximate value for e with error at most
3 3 3
(n+1)!
. For example, when n = 9, (n+1)! = 10! < 10−6 so that

1+1+ 1
2!
+···+ 1
9!
≤e≤1+1+ 1
2!
+···+ 1
9!
+ 10−6

with
1 1 1 1 1 1 1 1
1+1+ 2!
+ 3!
+ 4!
+ 5!
+ 6!
+ 7!
+ 8!
+ 9!
=1 + 1 + 0.5 + 0.16̇ + 0.0416̇ + 0.0083̇ + 0.00138̇ + 0.0001984 + 0.0000248 + 0.0000028
=2.718282

to six decimal places.

Example 9

Example 10 (Example 4 Revisited)

In Example 4 (measuring the height of the pole), we used the linear approximation

f (θ0 + ∆θ) ≈ f (θ0 ) + f ′ (θ0 )∆θ (15)

π
with f (θ) = 10 tan θ and θ0 = 30 180 to get

∆h
∆h = f (θ0 + ∆θ) − f (θ0 ) ≈ f ′ (θ0 )∆θ =⇒ ∆θ ≈
f ′ (θ0 )

c Joel Feldman. 2014. All rights reserved. 12 January 29, 2013

While this procedure is fairly reliable, it did involve an approximation. So that you could
not 100% guarantee to your client’s lawyer that an accuracy of 10 cm was achieved. If we
use the exact formula (10), with the replacements x → θ0 + ∆θ and x0 → θ0

f (θ0 + ∆θ) = f (θ0 ) + f ′ (c)∆θ for some c between θ0 and θ0 + ∆θ

in place of the approximate formula (2), this legality is taken care of.

∆h
∆h = f (θ0 + ∆θ) − f (θ0 ) = f ′ (c)∆θ =⇒ ∆θ = for some c between θ0 and θ0 + ∆θ
f ′ (c)

Of course we do not know exactly what c is. But suppose that we know that the angle was
somewhere between 25◦ and 35◦ . In other words suppose that, even though we don’t know
precisely what our measurement error was, it was certainly no more than 5◦ . Since sec(c)
increases with c (for c between 0 and 90◦ ), f ′ (c) = 10 sec2 (c) must certainly be smaller than
10 sec2 35◦ < 14.91, which means that f∆h .1 .1 180
′ (c) must be at least 14.91 radians or 14.91 π = .38◦ .
A measurement error of 0.38◦ or less is certainly acceptable.
Example 10

8. Derivation of the Error Formulae (Optional)

Fix any real number x0 and natural number n. Define

En (x) = f (x) − f (x0 ) − f ′ (x0 )(x − x0 ) − · · · − 1 (n)

n!
f (x0 )(x − x0 )n

This is the error introduced when one approximates f (x) by its Taylor polynomial of degree
n (about x0 ). We shall now prove that

En (x) = 1
(n+1)!
f (n+1) (c) (x − x0 )n+1 (16n )

for some c strictly between x0 and x. In fact, we have already used the Mean–Value Theorem
to prove that E0 (x) = f ′ (c) (x − x0 ), for some c strictly between x0 and x. This was the
content of (10). To deal with n ≥ 1, we need the following generalization of the Mean–Value
Theorem. (Choosing G(x) = x reduces Theorem 11 to the Mean–Value Theorem.)

Theorem 11 (Generalized Mean–Value Theorem).

Let the functions F (x) and G(x) both be defined and continuous on a ≤ x ≤ b and
both be differentiable on a < x < b. Furthermore, suppose that G′ (x) 6= 0 for all
a < x < b. Then, there is a number c obeying a < c < b such that
F (b)−F (a) F ′ (c)
G(b)−G(a)
= G′ (c)

c Joel Feldman. 2014. All rights reserved. 13 January 29, 2013

Proof. Define
     
h(x) = F (b) − F (a) G(x) − G(a) − F (x) − F (a) G(b) − G(a)

Observe that h(a) = h(b) = 0. So, by the Mean–Value Theorem, there is a number c obeying
a < c < b such that
h(b)−h(a)
= h′ (c) = F (b) − F (a) G′ (c) − F ′ (c) G(b) − G(a)
   
0= b−a

As G(a) 6= G(b) (otherwise the Mean–Value Theorem would imply the existence of an a <
x < b obeying G′ (x) = 0), we may divide by G′ (c) G(b) − G(a) which gives the desired
 

result.

Proof of (16n ). To prove (161 ), that is (16n ) for n = 1, simply apply the Generalized Mean–
Value Theorem with F (x) = E1 (x) = f (x) − f (x0 ) − f ′ (x0 )(x − x0 ), G(x) = (x − x0 )2 , a = x0
and b = x. Then F (a) = G(a) = 0, so that
F (b) F ′ (c̃) f (x)−f (x0 )−f ′ (x0 )(x−x0 ) f ′ (c̃)−f ′ (x0 )
G(b)
= G′ (c̃)
=⇒ (x−x0 )2
= 2(c̃−x0 )

for some c̃ strictly between x0 and x. By the Mean–Value Theorem (the standard one, but
′ ′ (x )
with f (x) replaced by f ′ (x)), f (c̃)−f
c̃−x0
0
= f ′′ (c), for some c strictly between x0 and c̃ (which
forces c to also be strictly between x0 and x). Hence
f (x)−f (x0 )−f ′ (x0 )(x−x0 )
(x−x0 )2
= 12 f ′′ (c)

which is exactly (161 ).

At this stage, we know that (16n ) applies to all (sufficiently differentiable) functions for
n = 0 and n = 1. To prove it for general n, we proceed by induction. That is, we assume
that we already know that (16n ) applies to n = k − 1 for some k (as is the case for k = 1, 2)
and we wish to prove that it also applies to n = k. We apply the Generalized Mean–Value
Theorem with F (x) = Ek (x), G(x) = (x − x0 )k+1 , a = x0 and b = x. Then F (a) = G(a) = 0,
so that
F (b) F ′ (c̃) Ek (x) Ek′ (c̃)
= ′ =⇒ = (17)
G(b) G (c̃) (x − x0 )k+1 (k + 1)(c̃ − x0 )k
for some c̃ between x0 and x. But
d h i
Ek′ (c̃) = f (x) − f (x0 ) − f ′ (x0 ) (x − x0 ) − · · · − k!1 f (k) (x0 ) (x − x0 )k
hdx i x=c̃
′ ′ 1 (k) k−1
= f (x) − f (x0 ) − · · · − (k−1)! f (x0 )(x − x0 )
x=c̃
′ ′
= f (c̃) − f (x0 ) − · · · − 1
(k−1)!
f (k) (x0 )(c̃ − x0 ) k−1
(18)

The last expression is exactly the definition of Ek−1 (c̃), but for the function f ′ (x), instead of
the function f (x). But we already know that (16k−1 ) is true. So, substituting n → k − 1,
f → f ′ and x → c̃ into (16n ), we already know that (18), i.e. Ek′ (c̃), equals

(k−1+1)
1
(k−1+1)!
f′ (c)(c̃ − x0 )k−1+1 = 1 (k+1)
k!
f (c) (c̃ − x0 )k

c Joel Feldman. 2014. All rights reserved. 14 January 29, 2013

for some c strictly between x0 and c̃, and hence also strictly between x0 and x. Substituting
this into (17) gives

Ek (x) Ek′ (c̃) f (k+1) (c) (c̃ − x0 )k 1

k+1
= k
= k
= f (k+1) (c)
(x − x0 ) (k + 1)(c̃ − x0 ) (k + 1) k! (c̃ − x0 ) (k + 1)!

which is exactly (16k ).

So we now know that

• if, for some k, (16k−1) is true for all k times differentiable functions,

• then (16k ) is true for all k + 1 times differentiable functions.

Repeatedly applying this for k = 2, 3, 4, · · · (and recalling that (161 ) is true) gives (16k ) for
all k.

9. Taylor Series
Fix a real number x0 and suppose that all derivatives of the function f (x) exist. We have
seen in (12) that, for any natural number n,

f (x) = Pn (x) + En (x) (19)

where
Pn (x) = f (x0 ) + f ′ (x0 ) (x − x0 ) + · · · + 1 (n)
n!
f (x0 ) (x − x0 )n (19a)
is the Taylor polynomial of degree n for the function f (x) and expansion point x0 and

En (x) = f (x) − Pn (x) = 1

(n+1)!
f (n+1) (c) (x − x0 )n+1 (19b)

is the error introduced when we approximate f (x) by the polynomial Pn (x). If it happens
that En (x) tends to zero as n → ∞, then we have the exact formula

f (x) = lim Pn (x)

n→∞

for f (x). This is usually written

∞
X
1 (n)
f (x) = n!
f (x0 ) (x − x0 )n (20)
n=0

and is called the Taylor series of f (x) with expansion point x0 .

Example 12 (Exponential Series)
This happens with the exponential function f (x) = ex . Recall from (13) that, for all natural
numbers n and all real numbers x,

ex = 1 + x + 21 x2 + 3!1 x3 + · · · + 1 n
n!
x + ec
(n+1)!
xn+1

c Joel Feldman. 2014. All rights reserved. 15 January 29, 2013

for some c strictly between 0 and x. Now consider any fixed real number x. As c runs from 0
to x, ec runs from e0 = 1 to ex . In particular, ec is always between 1 and ex and so is smaller
than 1 + ex . Thus the error term
ec |x|n+1
|En (x)| = xn+1 ≤ [ex + 1]
(n + 1)! (n + 1)!
|x| n+1
Let’s call en (x) = (n+1)! . We claim that as n increases towards infinity, en (x) decreases
(quickly) towards zero. To see this, let’s compare en (x) and en+1 (x).

|x|n+2
en+1 (x) (n+2)! |x|
= =
en (x) |x|n+1 n+2
(n+1)!

So, when n is bigger than, for example 2|x|, we have en+1 (x)
en (x)
< 12 . That is, increasing the
index on en (x) by one decreases the size of en (x) by a factor of at least two. As a result en (x)
must tend to zero as n → ∞. Consequently lim En (x) = 0 and
n→∞

h i ∞
X
ex = lim 1 + x + 21 x2 + 3!1 x3 + · · · + 1 n
n!
x = 1 n
n!
x (21)
n→∞
n=0

Example 12

Example 13 (Sine and Cosine Series)

The trigonometric functions sin x and cos x also have widely used Taylor series expansions
about x0 = 0. Reviewing (7) we see that every derivative of sin x and cos x is one of ± sin x
and ± cos x. Consequently, when we apply (19b) we always have f (n+1) (c) ≤ 1 and hence
|x|n+1 |x|n+1
|En (x)| ≤ (n+1)! . We have already seen in Example 12, that (n+1)! (which we called en (x)
in Example 12) converges to zero as n → ∞. Consequently, for both f (x) = sin x and
f (x) = cos x, we have lim En (x) = 0 and
n→∞
h i
f (x) = lim f (0) + f ′ (0) x + · · · + 1 (n)
n!
f (0) xn
n→∞

Reviewing (8), we conclude that

∞
X
1 3 1 5
sin x = x − 3!
x + 5!
x −··· = (−1)n (2n+1)!
1
x2n+1
n=0
∞ (22)
X
1 2 1 4 n
cos x = 1 − 2!
x + 4!
x −··· = (−1) 1
(2n)!
x2n
n=0

Example 13

c Joel Feldman. 2014. All rights reserved. 16 January 29, 2013

10. Evaluating Limits Using Taylor Expansions
Taylor polynomials provide a good way to understand the behaviour of a function near a
specified point and so are useful for evaluating complicated limits. We’ll see examples of this
shortly.
We’ll just start by recalling, from (12), that if, for some natural number n, the function
f (x) has n + 1 derivatives near the point x0 , then

f (x) = Pn (x) + En (x)

where
Pn (x) = f (x0 ) + f ′ (x0 ) (x − x0 ) + · · · + 1 (n)
n!
f (x0 ) (x − x0 )n
is the Taylor polynomial of degree n for the function f (x) and expansion point x0 and

En (x) = f (x) − Pn (x) = 1

(n+1)!
f (n+1) (c) (x − x0 )n+1

is the error introduced when we approximate f (x) by the polynomial Pn (x). Here c is some
unknown number between x0 and x. As c is not known, we do not know exactly what the
error En (x) is. But that is usually not a problem. In taking the limit x → x0 , we are only
interested in x’s that are very close to x0 , and when x is very close x0 , c must also be very
close to x0 . As long as f (n+1) (x) is continuous at x0 , f (n+1) (c) must approach f (n) (x0 ) as
x → x0 . In particular there must be constants M, D > 0 such that f (n+1) (c) ≤ M for all
M
c’s within a distance D of x0 . If so, there is another constant C (namely (n+1)! ) such that

En (x) ≤ C|x − x0 |n+1 whenever |x − x0 | ≤ D

There is some notation for this behavour.

11. The Big O Notation

Definition 14 (Big O).

Let x0 and m be real numbers. We say “F (x) is of order |x − x0 |m near x0 ” and we

write F (x) = O |x − x0 |m if there exist constants C, D > 0 such that

F (x) ≤ C|x − x0 |m whenever |x − x0 | ≤ D (23)

Whenever O |x − x0 |m appears in an algebraic expression, it just stands for some

(unknown) function F (x) that obeys (23). This is called “big O” notation.

Example 15
Let f (x) = sin x and x0 = 0. Then

f (x) = sin x f ′ (x) = cos x f ′′ (x) = − sin x f (3) (x) = − cos x f (4) (x) = sin x ···
f (0) = 0 f ′ (0) = 1 f ′′ (0) = 0 f (3) (0) = −1 f (4) (0) = 0 ···

c Joel Feldman. 2014. All rights reserved. 17 January 29, 2013

and the pattern repeats. Thus f (n+1) (c) ≤ 1 for all real numbers c and all natural numbers
n. So the Taylor polynomial of, for example, degree 3 and its error term are

sin x = x − 3!1 x3 + cos c 5

5!
x
= x − 3!1 x3 + O(|x| ) 5

1
under Definition 14, with C = and any D > 0. Similarly, for any natural number n,
5!

sin x = x − 3!1 x3 + 5!1 x5 − · · · + (−1)n (2n+1)!

1
x2n+1 + O |x|2n+3

cos x = 1 − 2!1 x2 + 4!1 x4 − · · · + (−1)n (2n)!

1
x2n + O |x|2n+2

Example 15

Example 16
dm x
Let n be any natural number. Since dxm
e = ex for every integer m ≥ 0,
x2 x3
ex = 1 + x + 2!
+ 3!
+···+ xn
n!
+ ec
(n+1)!
xn+1

for some c between 0 and x. If, for example, |x| ≤ 1, then |ec | ≤ e, so that the error term
ec
(n+1)!
xn+1 ≤ C|x|n+1 with C = e
(n+1)!
whenever |x| ≤ 1
e
So, under Definition 14, with C = (n+1)!
and D = 1,

x2 x3
ex = 1 + x + xn
+ O |x|n+1


2!
+ 3!
+···+ n!

Example 16

Example 17
Let f (x) = ln(1 + x) and x0 = 0. Then

f ′ (x) = 1
1+x
f ′′ (x) = − (1+x)
1
2 f (3) (x) = 2
(1+x)3
f (4) (x) = − (1+x)
2×3
4 f (5) (x) = 2×3×4
(1+x)5

f ′ (0) = 1 f ′′ (0) = −1 f (3) (0) = 2 f (4) (0) = −3! f (5) (0) = 4!

We can see a pattern for f (n) (x) forming here — f (n) (x) is a sign times a ratio with

• the sign being + when n is odd and being − when n is even. So the sign is (−1)n−1 .
• The denominator is a power of (1 + x). The power is just n.
• The numerator is a product 2 × 3 × 4 × · · · . The last integer in the power is n − 1, at
least for n ≥ 2. So the product, for n ≥ 2, is 2 × 3 × 4 × · · · × (n − 1). The notation
n!, read “n factorial”, means 1 × 2 × 3 × · · · × n, so the numerator is (n − 1)!, at least
for n ≥ 2. By convention, 0! = 1, so the numerator is (n − 1)! for n = 1 too.

c Joel Feldman. 2014. All rights reserved. 18 January 29, 2013

Thus, for any natural number n,
(n−1)!
f (n) (x) = (−1)n−1 (1+x) 1 (n)
f (0) xn = (−1)n−1 (n−1)! xn = (−1)n−1 xn
n
n n! n!

so
x2 x3
− · · · + (−1)n−1 xn + En (x)
n
ln(1 + x) = x − 2
+ 3

with
En (x) = 1
(n+1)!
f (n+1) (c) (x − x0 )n+1 = (−1)n (n+1)(1+c)
1
n+1 x
n+1

If we choose, for example D = 21 , then for any x obeying |x| ≤ 1

2
, we have |c| ≤ 1
2
and
|1 + c| ≥ 12 so that
n+1
1
= O |x|n+1


|En (x)| ≤ (n+1)(1/2)n+1 |x|

2n+1
under Definition 14, with C = n+1
and D = 1. Thus we may write

x2 x3
− · · · + (−1)n−1 xn + O |x|n+1
n

ln(1 + x) = x − 2
+ 3
(24)

Example 17

Remark 18.

The big O notation has a few properties that are useful in computations and taking
limits. All follow immediately from Definition 14.

1. If p > 0, then lim O(|x|p ) = 0.

x→0

2. For any real numbers p and q, O(|x|p) O(|x|q ) = O(|x|p+q ).

(This is just because C|x|p × C ′ |x|q = (CC ′ )|x|p+q .)
In particular, axm O(|x|p) = O(|x|p+m), for any constant a and any integer
m.

3. For any real numbers p and q, O(|x|p) + O(|x|q ) = O(|x|min{p,q} ).

(For example, if p = 2 and q = 5, then C|x|2 + C ′ |x|5 = C + C ′ |x|3 |x|2 ≤

(C + C ′ )|x|2 whenever |x| ≤ 1.)

4. For any real numbers p and q with p > q, any function which is O(|x|p) is also
O(|x|q ) because C|x|p = C|x|p−q |x|q ≤ C|x|q whenever |x| ≤ 1.

c Joel Feldman. 2014. All rights reserved. 19 January 29, 2013

12. Evaluating Limits Using Taylor Expansion — Examples

Example 19
ln(1+x)
In this example we’ll use the Taylor polynomial of Example 17 to evaluate lim x
and
x→0
lim (1 + x)a/x . The Taylor expansion (24) with n = 1 tells us that
x→0

ln(1 + x) = x + O(|x|2 )
That is, for small x, ln(1 + x) is the same as x, up to an error that is bounded by some
constant times x2 . So, dividing by x, x1 ln(1 + x) is the same as 1, up to an error that is
bounded by some constant times |x|. That is
1
x
ln(1 + x) = 1 + O(|x|)
But any function that is bounded by some constant times |x|, for all x smaller than some
constant D > 0, necessarily tends to 0 as x → 0. Thus
2)
lim ln(1+x) = lim x+O(|x|
 
x x
= lim 1 + O(|x|) = 1
x→0 x→0 x→0

and
2 )]
lim (1 + x)a/x = lim e /x ln(1+x) = lim e /x [x+O(|x| = lim ea+O(|x|) = ea
a a

x→0 x→0 x→0 x→0

Here we have used that if F (x) = O(|x| ), that is if |F (x)| ≤ C|x|2 for some constant C,
2

then xa F (x) ≤ C ′ |x| for the new constant C ′ = |a|C, so that F (x) = O(|x|). We have also
used that the exponential is continuous — as x tends to zero, the exponent of ea+O(|x|) tends
to a so that ea+O(|x|) tends to ea .
Example 19

Example 20
In this example we’ll evaluate the harder limit
cos x − 1 + 12 x sin x
lim
x→0 [ln(1 + x)]4
The first thing to notice about this limit is that, as x tends to zero, the numerator, which
is cos x − 1 + 21 x sin x, tends to cos 0 − 1 + 12 · 0 · sin 0 = 0 and the denominator [ln(1 + x)]4
tends to [ln(1 + 0)]4 = 0 too. So both the numerator and denominator tend to zero and we
may not simply evaluate the limit of the ratio by taking the limits of the numerator and
denominator and dividing. To find the limit, or show that it does not exist, we are going to
have to exhibit a cancellation between the numerator and the denominator. To develop a
strategy for evaluating this limit, let’s do a “little scratch work”, starting by taking a closer
look at the denominator. By Example 17,
ln(1 + x) = x + O(x2 )
This tells us that ln(1 + x) looks a lot like x for very small x. So the denominator [x + O(x2 )]4
looks a lot like x4 for very small x. Now, what about the numerator?

c Joel Feldman. 2014. All rights reserved. 20 January 29, 2013

• If the numerator looks like some constant times xp with p > 4, for very small x, then the
ratio will look like the constant times xx4 = xp−4 and will tend to 0 as x tends to zero.
p

• If the numerator looks like some constant times xp with p < 4, for very small x, then
the ratio will look like the constant times xx4 = xp−4 and will tend to plus or minus ∞
p

(depending on the sign of the constant) as x tends to zero.

4
• If the numerator looks like Cx4 , for very small x, then the ratio will look like Cx x4
=C
and will tend to C as x tends to zero.

The moral of the above “scratch work” is that we need to know the behaviour of the numer-
ator, for small x, up to order x4 . Any contributions of order xp with p > 4 may be put into
error terms O(|x|p). Now we are ready to evaluate the limit. Using Examples 15 and 17,

cos x − 1 + 21 x sin x 1 − 21 x2 + 4!1 x4 + O(x6 ) − 1 + 21 x x − 3!1 x3 + O(|x|5)

   
lim = lim
x→0 [ln(1 + x)]4 x→0 [x + O(x2 )]4
1
( 4!1 − 2×3! )x4 + O(x6 ) + x2 O(|x|5)
= lim
x→0 [x + O(x2 )]4
( 1 − 1 )x4 + O(x6 ) + O(x6 )
= lim 4! 2×3! by Remark 18, part 2.
x→0 [x + O(x2 )]4
( 1 − 1 )x4 + O(x6 )
= lim 4! 2×3! by Remark 18, parts 2, 3.
x→0 [x + x O(|x|)]4
1
( 4!1 − 2×3! )x4 + x4 O(x2 )
= lim by Remark 18, part 2.
x→0 x4 [1 + O(|x|)]4
( 1 − 1 ) + O(x2 )
= lim 4! 2×3!
x→0 [1 + O(|x|)]4
1 1
= 4!
− 2×3! by Remark 18 part 1.
1 1
− 12 = − 4!1


= 3! 4

Example 20

c Joel Feldman. 2014. All rights reserved. 21 January 29, 2013