CHAPTER 4

Derivatives (2)

“Leibniz never thought of the derivative as a limit”

http://www.gap-system.org/~history/Biographies/Leibniz.html

In chapter 2 we saw two mathematical problems which led to expressions of the form 00 . Now that we know
how to handle limits, we can state the definition of the derivative of a function. After computing a few
derivatives using the definition we will spend most of this section developing the differential calculus, which is
a collection of rules that allow you to compute derivatives without always having to use basic definition.

1. Derivatives Defined

1.1. Definition. Let f be a function which is defined on some interval (c, d) and let a be some number
in this interval.
The derivative of the function f at a is the value of the limit
f (x) − f (a)
(17) f 0 (a) = lim .
x→a x−a
f is said to be differentiable at a if this limit exists.
f is called differentiable on the interval (c, d) if it is differentiable at every point a in (c, d).

1.2. Other notations. One can substitute x = a + h in the limit (17) and let h → 0 instead of x → a.
This gives the formula
f (a + h) − f (a)
(18) f 0 (a) = lim ,
h→0 h
Often you will find this equation written with x instead of a and ∆x instead of h, which makes it look like
this:
f (x + ∆x) − f (x)
f 0 (x) = lim .
∆x→0 ∆x
The interpretation is the same as in equation (6) from §4. The numerator f (x + ∆x) − f (x) represents the
amount by which the function value of f changes if one increases its argument x by a (small) amount ∆x. If
you write y = f (x) then we can call the increase in f
∆y = f (x + ∆x) − f (x),
0
so that the derivative f (x) is
∆y
f 0 (x) = lim .
∆x
∆x→0
Gottfried Wilhelm von Leibniz, one of the inventors of calculus, came up with the idea that one should
write this limit as
dy ∆y
= lim ,
dx ∆x→0 ∆x
the idea being that after letting ∆x go to zero it didn’t vanish, but instead became an infinitely small quantity
which Leibniz called “dx.” The result of increasing x by this infinitely small quantity dx is that y = f (x)
increased by another infinitely small quantity dy. The ratio of these two infinitely small quantities is what we
call the derivative of y = f (x).
41
 There are no “infinitely small real numbers,” and this makes Leibniz’ notation difficult to justify. In the
20th century mathematicians have managed to create a consistent theory of “infinitesimals” which allows you
to compute with “dx and dy” as Leibniz and his contemporaries would have done. This theory is called “non
standard analysis.” We won’t mention it any further1. Nonetheless, even though we won’t use infinitely small
numbers, Leibniz’ notation is very useful and we will use it.

2. Direct computation of derivatives

2.1. Example – The derivative of f (x) = x2 is f 0 (x) = 2x . We have done this computation before
in §2. The result was
f (x + h) − f (x) (x + h)2 − x2
f 0 (x) = lim = lim = lim (2x + h) = 2x.
h→0 h h→0 h h→0

Leibniz would have written

dx2
= 2x.
dx

2.2. The derivative of g(x) = x is g 0 (x) = 1 . Indeed, one has

g(x + h) − g(x) (x + h) − x h
g 0 (x) = lim = lim = lim = 1.
h→0 h h→0 h h→0 h

In Leibniz’ notation:
dx
= 1.
dx
This is an example where Leibniz’ notation is most misleading, because if you divide dx by dx then you
should of course get 1. Nonetheless, this is not what is going on. The expression dx
dx is not really a fraction
since there are no two “infinitely small” quantities dx which we are dividing.

2.3. The derivative of any constant function is zero . Let k(x) = c be a constant function. Then
we have
k(x + h) − k(x) c−c
k 0 (x) = lim = lim = lim 0 = 0.
h→0 h h→0 h h→0
Leibniz would have said that if c is a constant, then
dc
= 0.
dx

2.4. Derivative of xn for n = 1, 2, 3, . . . . To differentiate f (x) = xn one proceeds as follows:

f (x) − f (a) xn − a n
f 0 (a) = lim = lim .
x→a x−a x→a x − a

We need to simplify the fraction (xn − an )/(x − a). For n = 2 we have

x2 − a 2
= x + a.
x−a
For n = 1, 2, 3, . . . the geometric sum formula tells us that
xn − an
(19) = xn−1 + xn−2 a + xn−3 a2 + · · · + xan−2 + an−1 .
x−a

1But if you want to read more on this you should see Keisler’s calculus text at
http://www.math.wisc.edu/~keisler/calc.html
I would not recommend using Keisler’s text and this text at the same time, but if you like math you should remember that it
exists, and look at it (later, say, after you pass 221.)

42
If you don’t remember the geometric sum formula, then you could also just verify (19) by carefully multiplying
both sides with x − a. For instance, when n = 3 you would get
x × (x2 + xa + a2 ) = x3 +ax2 +a2 x
−a × (x2 + xa + a2 ) = −ax2 −a2 x −a3
(x − a) × (x2 + ax + a2 ) = x3 −a3

With formula (19) in hand we can now easily find the derivative of xn :
xn − an
f 0 (a) = lim
x→a x − a

= lim xn−1 + xn−2 a + xn−3 a2 + · · · + xan−2 + an−1


x→a
= an−1 + an−2 a + an−3 a2 + · · · + a an−2 + an−1 .
Here there are n terms, and they all are equal to an−1 , so the final result is
f 0 (a) = nan−1 .
One could also write this as f 0 (x) = nxn−1 , or, in Leibniz’ notation
dxn
= nxn−1 .
dx
This formula turns out to be true in general, but here we have only proved it for the case in which n is a
positive integer.

3. Differentiable implies Continuous

3.1. Theorem. If a function f is differentiable at some a in its domain, then f is also continuous at a.

Proof. We are given that

f (x) − f (a)
lim
x→a x−a
exists, and we must show that
lim f (x) = f (a).
x→a
This follows from the following computation


lim f (x) = lim f (x) − f (a) + f (a) (algebra)
x→a x→a
f (x) − f (a)
= lim · (x − a) + f (a) (more algebra)
x→a x−a
 
f (x) − f (a)
= lim · lim (x − a) + lim f (a) (Limit Properties)
x→a x−a x→a x→a
0
= f (a) · 0 + f (a) (f 0 (a) exists)
= f (a).


4. Some non-differentiable functions

4.1. A graph with a corner. Consider the function

(
x for x ≥ 0,
f (x) = |x| =
−x for x < 0.
This function is continuous at all x, but it is not differentiable at x = 0.
43
 To see this try to compute the derivative at 0,
|x| − |0| |x|
f 0 (0) = lim = lim = lim sign(x).
x→0 x − 0 x→0 x x→0

We know this limit does not exist (see §7.2)

If you look at the graph of f (x) = |x| then you see what is wrong: the graph has a corner at the origin
and it is not clear which line, if any, deserves to be called the tangent to the graph at the origin.

y = |x|

tangent?

tangent?

tangent?

tangent?

Figure 1. The graph of y = |x| has no tangent at the origin.

4.2. A graph with a cusp. Another example of a function without a derivative at x = 0 is

p
f (x) = |x|.
When you try to compute the derivative you get this limit
p
0 |x|
f (0) = lim =?
x→0 x
The limit from the right is p
|x| 1
lim = lim √ ,
x&0 x x&0 x
which does not exist (it is “+∞”). Likewise, the limit from the left also does not exist (’tis “−∞). Nonetheless,
a drawing for the graph of f suggests an obvious tangent to the graph at x = 0, namely, the y-axis. That
observation does not give us a derivative, because the y-axis is vertical and hence has no slope.

4.3. A graph with absolutely no tangents, anywhere. The previous two examples were about
functions which did not have a derivative at x = 0. In both examples the point x = 0 was the only point where
the function failed to have a derivative. It is easy to give examples of functions which are not differentiable
at more than one value of x, but here I would like to show you a function f which doesn’t have a derivative
anywhere in its domain.
To keep things short I won’t write a formula for the function, and merely show you a graph. In this
graph you see a typical path of a Brownian motion, i.e. t is time, and x(t) is the position of a particle which
undergoes a Brownian motion – come to lecture for further explanation (see also the article on wikipedia).
To see a similar graph check the Dow Jones or Nasdaq in the upper left hand corner of the web page at
http://finance.yahoo.com in the afternoon on any weekday.

44
 p
y= |x|

The tangent is vertical

Figure 2. Tangent to the graph of y = |x|1/2 at the origin

5. Exercises

101. Compute the derivative of the following functions 103. For which value(s) is the function defined by
(
ax + b for x < 0
f (x) = x2 − 2x f (x) =
x − x2 for x ≥ 0
1
g(x) = differentiable at x = 0? Sketch the graph of the function
x
k(x) = x3 − 17x f for the values a and b you found.
2 104. For which value(s) is the function defined by
u(x) =
1+x
(
√ ax2 + b for x < 1
v(x) = x f (x) =
x − x2 for x ≥ 1
1
w(x) = √ differentiable at x = 0? Sketch the graph of the function
x
f for the values a and b you found.
using either (17) or (18). 105. For which value(s) is the function defined by
(
ax2 for x < 2
102. Which of the following functions is differentiable at f (x) =
x = 0? x + b for x ≥ 2
p differentiable at x = 0? Sketch the graph of the function
f (x) = x|x|, g(x) = x |x|, f for the values a and b you found.
π
h(x) = x + |x|, k(x) = x2 sin , 106. Group Problem.
x
π True or false: If a function f is continuous at some
`(x) = x sin .
x x = a then it must also be differentiable at x = a?

These formulas do not define k and ` at x = 0. We define 107. Group Problem.

k(0) = `(0) = 0. True or false: If a function f is differentiable at
some x = a then it must also be continuous at x = a?
x(t)

Figure 3. A Brownian motion. Note how the graph doesn’t seem to have a a tangent anywhere at all.

45
 6. The Differentiation Rules

You could go on and compute more derivatives from the definition. Each time you would have to compute
a new limit, and hope that there is some trick that allows you to find that limit. This is fortunately not
necessary. It turns out that if you know a few basic derivatives (such as dxn /dx = nxn−1 ) the you can find
derivatives of arbitrarily complicated functions by breaking them into smaller pieces. In this section we’ll
look at rules which tell you how to differentiate a function which is either the sum, difference, product or
quotient of two other functions.

dc
Constant rule: c0 = 0 =0
dx
du ± v du dv
Sum rule: (u ± v)0 = u0 ± v 0 = ±
dx dx dx
duv du dv
Product rule: (u · v)0 = u0 · v + u · v 0 = v+u
dx dx dx
u du dv
 u 0 u0 · v − u · v 0 d v −u
Quotient rule: = v = dx dx
v v2 dx v2

Table 1. The differentiation rules

The situation is analogous to that of the “limit-properties” (P1 ). . . (P6 ) from the previous chapter which
allowed us to compute limits without always having to go back to the epsilon-delta definition.

6.1. Sum, product and quotient rules. In the following c and n are constants, u and v are functions
of x, and 0 denotes differentiation. The Differentiation Rules in function notation, and Leibniz notation, are
listed in figure 1.
Note that we already proved the Constant Rule in example 2.2. We will now prove the sum, product and
quotient rules.

6.2. Proof of the Sum Rule. Suppose that f (x) = u(x)+v(x) for all x where u and v are differentiable.
Then
f (x) − f (a)
f 0 (a) = lim (definition of f 0 )
x→a x−a



u(x) + v(x) − u(a) + v(a)
= lim (use f = u + v)
x→a x−a
 
u(x) − u(a) v(x) − v(a)
= lim + (algebra)
x→a x−a x−a
u(x) − u(a) v(x) − v(a)
= lim + lim (limit property)
x→a x−a x→a x−a
0 0
= u (a) + v (a) (definition of u0 , v 0 )

6.3. Proof of the Product Rule. Let f (x) = u(x)v(x). To find the derivative we must express the
change of f in terms of the changes of u and v
f (x) − f (a) = u(x)v(x) − u(a)v(a)
= u(x)v(x) − u(x)v(a) + u(x)v(a) − u(a)v(a)



= u(x) v(x) − v(a) + u(x) − u(a) v(a)

46
Now divide by x − a and let x → a:
f (x) − f (a) v(x) − v(a) u(x) − u(a)
lim = lim u(x) + v(a)
x→a x−a x→a x−a x−a
(use the limit properties)
  v(x) − v(a)   u(x) − u(a) 
= lim u(x) lim + lim v(a)
x→a x→a x−a x→a x−a
= u(a)v 0 (a) + u0 (a)v(a),
as claimed. In this last step we have used that
u(x) − u(a) v(x) − v(a)
lim = u0 (a) and lim = v 0 (a)
x→a x−a x→a x−a
and also that
lim u(x) = u(a)
x→a
This last limit follows from the fact that u is continuous, which in turn follows from the fact that u is
differentiable.

6.4. Proof of the Quotient Rule . We can break the proof into two parts. First we do the special
case where f (x) = 1/v(x), and then we use the product rule to differentiate
u(x) 1
f (x) = = u(x) · .
v(x) v(x)
So let f (x) = 1/v(x). We can express the change in f in terms of the change in v
1 1 v(x) − v(a)
f (x) − f (a) = − = .
v(x) v(a) v(x)v(a)
Dividing by x − a we get
f (x) − f (a) 1 v(x) − v(a)
= .
x−a v(x)v(a) x−a
Now we want to take the limit x → a. We are given the v is differentiable, so it must also be continuous and
hence
lim v(x) = v(a).
x→a
Therefore we find
f (x) − f (a) 1 v(x) − v(a) v 0 (a)
lim = lim lim = .
x→a x−a x→a v(x)v(a) x→a x−a v(a)2
That completes the first step of the proof. In the second step we use the product rule to differentiate f = u/v
 u 0  1 0 1
 0
1 u0 v0 u0 v − uv 0
f0 = = u· = u0 · + u · = −u 2 = .
v v v v v v v2

6.5. A shorter, but not quite perfect derivation of the Quotient Rule . The Quotient Rule
can be derived from the Product Rule as follows: if w = u/v then
(20) w·v =u
By the product rule we have
w 0 · v + w · v 0 = u0 ,
so that
u0 − w · v 0 u0 − (u/v) · v 0 u0 · v − u · v 0
w0 = = = .
v v v2
Unlike the proof in §6.4 above, this argument does not prove that w is differentiable if u and v are. It only
says that if the derivative exists then it must be what the Quotient Rule says it is.
The trick which is used here, is a special case of a method called “implicit differentiation.” We have an
equation (20) which the quotient w satisfies, and from by differentiating this equation we find w0 .
47
 6.6. Differentiating a constant multiple of a function . Note that the rule
(cu)0 = cu0
follows from the Constant Rule and the Product Rule.

6.7. Picture of the Product Rule. If u and v are quantities which depend on x, and if increasing x
by ∆x causes u and v to change by ∆u and ∆v, then the product of u and v will change by
(21) ∆(uv) = (u + ∆u)(v + ∆v) − uv = u∆v + v∆u + ∆u∆v.
If u and v are differentiable functions of x, then the changes ∆u and ∆v will be of the same order of magnitude
as ∆x, and thus one expects ∆u∆v to be much smaller. One therefore ignores the last term in (21), and thus
arrives at
∆(uv) = u∆v + v∆u.
Leibniz would now divide by ∆x and replace ∆’s by d’s to get the product rule:
∆(uv) ∆v ∆u
=u +v .
∆x ∆x ∆x

u∆v ∆u∆v
∆v

v uv v∆u

u ∆u

Figure 4. The Product Rule. How much does the area of a rectangle change if its sides u and v are
increased by ∆u and ∆v? Most of the increase is accounted for by the two thin rectangles whose areas
are u∆v and v∆u. So the increase in area is approximately u∆v + v∆u, which explains why the product
rule says (uv)0 = uv 0 + vu0 .

7. Differentiating powers of functions

7.1. Product rule with more than one factor. If a function is given as the product of n functions,
i.e.
f (x) = u1 (x) × u2 (x) × · · · × un (x),
then you can differentiate it by applying the product rule n − 1 times (there are n factors, so there are n − 1
multiplications.)
After the first step you would get

0
f 0 = u01 u2 · · · un + u1 u2 · · · un .

In the second step you apply the product rule to (u2 u3 · · · un )0 . This yields
f 0 = u01 u2 · · · un + u1 u02 u3 · · · un + u2 (u3 · · · un )0
 

0
= u01 u2 · · · un + u1 u02 u3 · · · un + u1 u2 u3 · · · un .

48
Continuing this way one finds after n − 1 applications of the product rule that

0
(22) u1 · · · un = u01 u2 · · · un + u1 u02 u3 · · · un + · · · + u1 u2 u3 · · · u0n .

7.2. The Power rule . If all n factors in the previous paragraph are the same, so that the function f
is the nth power of some other function,

n
f (x) = u(x) ,
then all terms in the right hand side of (22) are the same, and, since there are n of them, one gets
f 0 (x) = nun−1 (x)u0 (x),
or, in Leibniz’ notation,
dun du
(23) = nun−1 .
dx dx

7.3. The Power Rule for Negative Integer Exponents . We have just proved the power rule (23)
assuming n is a positive integer. The rule actually holds for all real exponents n, but the proof is harder.
Here we prove the Power Rule for negative exponents using the Quotient Rule. Suppose n = −m where
m is a positive integer. Then the Quotient Rule tells us that
0
(um )0


0 1 Q.R.
(un )0 = u−m = m
= − m 2.
u (u )
Since m is a positive integer, we can use (23), so (um )0 = mum−1 , and hence
mum−1 · u0
(un )0 = − = −mu−m−1 · u0 = nun−1 u0 .
u2m

7.4. The Power Rule for Rational Exponents . So far we have proved that the power law holds if
the exponent n is an integer.
We will now see how you can show that the power law holds even if the exponent n is any fraction,
n = p/q. The following derivation contains the trick called implicit differentiation which we will study in
more detail in Section 15.
So let n = p/q where p and q are integers and consider the function
w(x) = u(x)p/q .
Assuming that both u and w are differentiable functions, we will show that
p p
(24) w0 (x) = u(x) q −1 u0 (x)
q

Raising both sides to the qth power gives

w(x)q = u(x)p .
Here the exponents p and q are integers, so we may apply the Power Rule to both sides. We get
qwq−1 · w0 = pup−1 · u0 .
Dividing both sides by qwq−1 and substituting up/q for w gives
pup−1 · u0 pup−1 · u0 pup−1 · u0 p
w0 = q−1
= p(q−1)/q
= p−(p/q)
= · u(p/q)−1 · u0
qw qu qu q
which is the Power Rule for n = p/q.
This proof is flawed because we did not show that w(x) = u(x)p/q is differentiable: we only showed what
the derivative should be, if it exists.
49
 7.5. Derivative of xn for integer n . If you choose the function u(x) in the Power Rule to be u(x) = x,
then u0 (x) = 1, and hence the derivative of f (x) = u(x)n = xn is
f 0 (x) = nu(x)n−1 u0 (x) = nxn−1 · 1 = nxn−1 .
We already knew this of course.

7.6. Example – differentiate a polynomial . Using the Differentiation Rules you can easily differ-
entiate any polynomial and hence any rational function. For example, using the Sum Rule, the Power Rule
with u(x) = x, the rule (cu)0 = cu0 , the derivative of the polynomial
f (x) = 2x4 − x3 + 7
is
f 0 (x) = 8x3 − 3x2 .

7.7. Example – differentiate a rational function. By the Quotient Rule the derivative of the
function
2x4 − x3 + 7
g(x) =
1 + x2
is
(8x3 − 3x2 )(1 + x2 ) − (2x4 − x3 + 7)2x
g 0 (x) =
(1 + x2 )2
6x − x + 8x − 3x2 − 14x
5 4 3
= .
(1 + x2 )2
If you compare this example with the previous then you see that polynomials simplify when you differentiate
them while rational functions become more complicated.
√
7.8. Derivative of the square root . The derivative of f (x) = x = x1/2 is
1 1 1 1
f 0 (x) = x1/2−1 = x−1/2 = 1/2 = √
2 2 2x 2 x
where we used the power rule with n = 1/2 and u(x) = x.

8. Exercises

108. Let f (x) = (x2 + 1)(x3 + 3). Find f 0 (x) in two ways: ax + b
115. f (x) =
(a) by multiplying and then differentiating, cx + d
(b) by using the product rule. 1
Are your answers the same? 116. f (x) =
(1 + x2 )2
109. Let f (x) = (1 + x2 )4 . Find f 0 (x) in two ways, first by x
expanding to get an expression for f (x) as a polynomial 117. f (x) = √
1+ x
in x and then differentiating, and then by using the power
rule. Are the answers the same?
r
1−x
118. f (x) =
110. Prove the statement in §6.6, i.e. show that (cu)0 = 1+x
c(u0 ) follows from the product rule. q
√
3
Compute the derivatives of the following functions. 119. f (x) = x + x
(try to simplify your answers)
t
120. ϕ(t) = √
1+ t
111. f (x) = x + 1 + (x + 1)2 r
x−2 1−s
112. f (x) = 121. g(s) =
x4 + 1 1+s
1
−1 q
√
113. f (x) = 122. h(ρ) = 3 ρ + ρ
1+x
p
114. f (x) = 1 − x2 123. Group Problem.

50
 Using derivatives to approximate numbers. (b) Show that for any pair of functions u and v one
(a) Find the derivative of f (x) = x4/3 . has
(b) Use (a) to estimate the number (uv)0 u0 v0
= +
1274/3
− 125 4/3 uv u v
2 (u/v)0 u0 v0
= −
approximately without a calculator. Your answer should u/v u v
0
have the form p/q where p and q are integers. [Hint: un


u0
Note that 53 = 125 and take a good look at equation =n
un u
(17).] √
(c)
√ Approximate in the same way the numbers 143 and 125. (a) Find f 0 (x) and g 0 (x) if
145 (Hint: 12 × 12 = 144). 1 + x2 2x4 + 7
f (x) = 4
, g(x) = .
124. Group Problem. 2x + 7 1 + x2
(Making the product and quotient rules look nicer.) Note that f (x) = 1/g(x).
Instead of looking at the derivative of a function you can (b) Is it true that f 0 (x) = 1/g 0 (x)?
look at the ratio of its derivative to the function itself, i.e. (c) Is it true that f (x) = g −1 (x)?
you can compute f 0 /f . This quantity is called the log- (d) Is it true that f (x) = g(x)−1 ?
arithmic derivative of the function f for reasons that
will become clear later this semester.
126. Group Problem. (a) Let x(t) = (1 − t2 )/(1 + t2 ),
(a) Compute the logarithmic derivative of these func- y(t) = 2t/(1 + t2 ) and u(t) = y(t)/x(t). Find dx/dt,
tions (i.e. find f 0 (x)/f (x)) dy/dt.
F (x) = x, g(x) = 3x, h(x) = x2 (b) Now that you’ve done (a) there are two different ways
of finding du/dt. What are they, and use one of both to
k(x) = −x2 , `(x) = 2007x2 , m(x) = x2007 find du/dt.

9. Higher Derivatives

9.1. The derivative is a function. If the derivative f 0 (a) of some function f exists for all a in the
domain of f , then we have a new function: namely, for each number in the domain of f we compute the
derivative of f at that number. This function is called the derivative function of f , and it is denoted by f 0 .
Now that we have agreed that the derivative of a function is a function, we can repeat the process and try to
differentiate the derivative. The result, if it exists, is called the second derivative of f . It is denoted f 00 .
The derivative of the second derivative is called the third derivative, written f 000 , and so on.
The nth derivative of f is denoted f (n) . Thus
f (0) = f, f (1) = f 0 , f (2) = f 00 , f (3) = f 000 , . . . .
Leibniz’ notation for the nth derivative of y = f (x) is
dn y
= f (n) (x).
dxn

9.2. Example. If f (x) = x2 − 2x + 3 then

f (x) = x2 − 2x + 3
f 0 (x) = 2x − 2
f 00 (x) = 2
f (3) (x) = 0
f (4) (x) = 0
..
.
All further derivatives of f are zero.
51