LOGARITHM, EXPONENTIAL, DERIVATIVE, AND INTEGRAL

MATH 152, SECTION 55 (VIPUL NAIK)

Corresponding material in the book: Section 7.2, 7.3, 7.4.

What students should definitely get: The definition of logarithm as an integral, its key properties.
The differentiation and integration formulas for logarithm and exponential, the key ideas behind combining
these with the chain rule and u-substitution to carry out other integrals.

Executive summary
0.1. Logarithm and exponential: basics.
R x logarithm is a one-to-one function with domain (0, ) and range R, and is defined as
(1) The natural
ln(x) := 1 (dt/t).
(2) The natural logarithm is an increasing function that is concave down. It satisfies the identities
ln(1) = 0, ln(ab) = ln(a) + ln(b), ln(ar ) = r ln a, and ln(1/a) = ln a.
(3) The limit limx0 ln(x) is and the limit limx ln(x) is +. Note that ln goes off to + at
even though its derivative goes to zero as x +.
(4) The derivative of ln(x) is 1/x and the derivative of ln(kx) is also 1/x. The derivative of ln(xr ) is
r/x.
(5) The antiderivative of 1/x is ln |x|+C. What this really means is that the antiderivative is ln(x)+C
when x is negative and ln(x)+C when x is positive. If we consider 1/x on both positive and negative
reals, the constant on the negative side is unrelated to the constant on the positive side.
(6) e is defined as the unique number x such that ln(x) = 1. e is approximately 2.718. In particular, it
is between 2 and 3.
(7) The inverse of the natural logarithm function is denoted exp, and exp(x) is also written as ex . When
x is a rational number, ex = ex (i.e., the two definitions of exponentiation coincide). In particular,
e1 = e, e0 = 1, etc.
(8) The function exp equals its own derivative and hence also its own antiderivative. Further, the
derivative of x 7 emx is memx . Similarly, the integral of emx is (1/m)emx + C.
(9) We have exp(x + y) = exp(x) exp(y), exp(rx) = (exp(x))r , exp(0) = 1, and exp(x) = 1/ exp(x).
All of these follow from the corresponding identities for ln.
Actions...
(1) We can calculate ln(x) for given x by using the usual methods of estimating the values of integrals,
applied to the function 1/x. We can also use the known properties of logarithms, as well as approx-
imate ln values for some specific x values, to estimate ln x to a reasonable approximation. For this,
it helps to remember ln 2, ln 3, and ln 5 or ln 10.
(2) Since both ln and exp are one-to-one, we can cancel ln from both sides of an equation and similarly
cancel exp. Technically, we cancel ln by applying exp to both sides, and we cancel exp by applying
ln to both sides.

0.2. Integrations involving logarithms and exponents. Words/actions ...

(1) If the numerator is the derivative of theRdenominator, the integral is the logarithm of the (absolute
value of) the denominator. In symbols, g 0 (x)/g(x) dx = ln |g(x)| + C.
0
(2) More generally, whenever we see an expression of the R form g (x)/g(x) inside theR integrand, we
should consider the substitution u = ln |g(x)|. Thus, f (ln |g(x)|)g 0 (x)/g(x) dx = f (u) du where
R = ln |g(x)|. R
u
(3) R f (ex )ex dx = f (u) du where u = ex .
(4) ex [f (x) + f 0 (x)] dx = ex f (x) + C.
1
 (5) ef (x) f 0 (x) dx = ef (x) + RC.
R

(6) Trigonometric
R integrals: tan x dx = R ln | cos x| + C, and similar integration
R formulas for cot, sec
and csc: cot x dx = ln | sin x| + C, sec x = ln | sec x + tan x| + C, and csc x dx = ln | csc x
cot x| + C.

1. Logarithms: the adventure begins

1.1. Finding an antiderivative of the reciprocal function. Recall that the process of differentiation
never gave us fundamentally new functions, because the derivatives of all the basic functions that we knew
were expressible in terms of other basic functions, and using the operations of pointwise combination and
composition did not allow us to break ground into new functions. The situation differs somewhat for
integration. We have seen that we often come across functions for which we have no clue as to how to find
an antiderivative. We now discuss how to handle one such function.
This function is the function 1/x, which, for now, we will assume to be a function on (0, ). We want to
find an antiderivative for this function.
The basic results of integration tell us that one way of defining an antiderivative is by using a definite
integral from a fixed value to x, as long as that fixed value is in the domain. For reasons that are not
immediately obvious, we choose the fixed value (the reference point) as 1. We thus define the following
function:
Z x
dt
L(x) :=
1 t
Note that this is the unique antiderivative which has the property that its value at 1 is 0. By definition,
L0 (x) = 1/x for all x. What further information can we derive about L?
1.2. Using the multiplicative transform. By the u-substitution method, we can readily verify that, for
a, b > 0:
Z aZ ab
dt dt
=
1 t b t
The key thing that is special about 1/x is that the multiplicative factor on the dt part cancels the
multiplicative factor on the t part.
This gives us that:

L(a) L(1) = L(ab) L(b)

Since L(1) = 0, we obtain that L is a function satisfying the property:

L(ab) = L(a) + L(b) a, b > 0

Thus, even though we do not have an explicit description of L, we know that L converts products to sums.
In particular, we also see, for instance, that:

L(an ) = nL(a) a > 0, n Z

In particular, L(1/a) = L(a).
We can further see that for any rational number r, we have:

L(ar ) = rL(a) a > 0, r Q

In other words, the function L converts products to sums and pulls the exponent into a multiple. We also
know that since L0 (x) > 0 for all x > 0, L is continuous and increasing. In particular, we see that L is a
one-to-one map on (0, ).
What is the range of L? Consider a = 2. Then, L(a) = L(2) > 0. As n , L(an ) = nL(a) , and
as n , L(an ) = nL(a) . Since L is increasing, we can use this to see that limx L(x) =
and limx0 L(x) = . Further, by the intermediate value theorem, we see that the range of L is R.
The upshot: L is a continuous increasing one-to-one function from (0, ) to R that sends 1 to 0 and
converts products to sums.
2
1.3. L for (natural) logarithm. The function L that is described above is termed the natural logarithm
function. It is ubiquitous in mathematics, and is denoted ln. Thus, we have the definition:
Z x
dt
ln x := x>0
1 t
It turns out that this natural logarithm behaves in ways very similar to logarithms to base 10. A quick
primer for those who didnt live in prehistoric times: in the olden days, when people had to do multiplications
by hand, they used a tool called logarithm tables to do these multiplications. The logarithm tables basically
converted the multiplication problem to an addition problem.
Here is the principle on which the logarithm tables worked. These tables allowed you to, for a given
number x, find the approximate value of r such that 10r = x. This value of r is called log10 x. Then, if you
had to multiply x and y, you first found log10 x and log10 y. It turns out that log10 (xy) = log10 (x) + log10 y,
because if 10r = x and 10s = y, then 10r+s = xy by properties of exponents. Thus, to find xy, we find log10 x
and log10 (y) and add them. Then, there are antilogarithm tables, that allow us to find the antilogarithm of
this sum that we have computed (or basically, raise 10 to the power of that number).
The principle of logarithm tables was later converted to a mechanical device called the slide rule. How
many people have used slide rules? What a slide rule does is use a logarithmic scale, i.e., it places numbers
on a scale in such a way that the distance between the positions of two numbers is determined by their
quotient. So, on a logarithmic scale, the distance between 1 and 10 is the same as the distance between 10
and 100, and also the same as the distance between 0.01 and 0.1. The distance between 3 and 7 is the same
as the distance between 30 and 70. (If youre interested in pictures of slide rules, do a Google image search.
I havent included any picture here because of potential copyright considerations).
A slide rule comprises two logarithmic scales (using the same calibration) but one of them can slide against
each other. We can use the sliding scale to add lengths along the scale, but since the scale is logarithmic,
this ends up multiplying the numbers. You may have heard about how people with an abacus can often
do simple calculations faster than people with a calculator. It turns out that people with a slide rule can
usually do multiplications faster than people with a calculator.1
Logarithmic scales are used in many measurements. Here are some examples:
(1) The Richter magnitude scale measures the intensity of earthquakes. It is calibrated logarithmi-
cally to base 10. An earthquake one point higher on the Richter scale is ten times as intense.
(2) The pH scale in chemistry is a logarithmic scale to measure the concentration of the H + (more
precisely, H3 O+ ) ions. It is a negative logarithmic scale to base 10. An increase in the pH value by
1 corresponds to a decrease in the hydronium ion concentration to 1/10 of its original value.
(3) The decibel scale, used for sound levels and other level measurements, is a logarithmic scale where
an increase in 10 points along the scale corresponds to a ten-fold increase in amplitude. Thus, 20dB
is ten times as loud as 10dB.
The natural logarithm function can be thought of as creating a logarithmic scale on the positive reals.
But the question we are concerned with is: what precisely is this scale? How does this compare with the
usual logarithm to base 10? Our hunch is that there should be a number e such that ln(x) is the value r
such that x = er . What must this number e be?

2. The back and forth of things: logarithm and exponential

2.1. In search of e. If such a number e exists, then it must be the unique number x satisfying ln(x) =
1. Further, since ln is an increasing function, we can try locating e between two consecutive integers by
determining ln 2, ln 3, and so on. Actually, we can be more clever.
We can begin by trying toR compute ln 2. We could do this using upper and lower sums. We could also
x
do it by noting that ln x = 1 dt/t, and must be located between the antiderivatives of x1/2 and x3/2 .
We did this approximation a few weeks ago and found that ln 2 is located between 0.58 and 0.83. Further
approximations using either this method or upper and lower sums for partitions yields than ln 2 is between
0.69 and 0.70. We will assume ln 2 0.7 for calculations.
1On the other hand, the calculator is a lot more versatile than the slide rule, and is probably faster for computing roots,
multiplying long sequences of numbers, or combinations of multiplication and addition.
3
 Now that we know ln 2, we do not need to do any more messy work with upper and lower sums. We know
that 2 < 2 2 < 3. We also have
that ln(2 2) = (3/2) ln 2 1.05, which is bigger than 1. Thus, ln 3 > 1, so
2 < e < 3. In fact, 2 < e < 2 2 2.83. We can also calculate, for instance, that the cuberoot of 2 is about
1.26. Thus, ln(2.52) is approximately (4/3)(ln 2) which is approximately 0.93. Thus, we get that e should be
bigger than 2.52. A similar process of successive approximations yields that the value of e is approximately
2.718281828. You should know e to at least three decimal places: 2.718.
Clearly, since ln(e) = 1, we have that ln(ep/q ) = p/q for integers p and q, with q 6= 0. It is not clear
a priori what we would mean by the notation er for irrational numbers r, but whatever we may mean, it
should be the case that ln(er ) = r. In other words, the function x 7 ex must be the inverse function of
the one-to-one function ln. The function x 7 ex is also called the exponentiation function, and sometimes
denoted exp. By the way, the letter e could be thought of as standing for exponentiation, but historically it
is believed to be named after Leonhard Euler, a prolific mathematician who studied the properties both of
the number e and the exponentiation function.

2.2. Logarithms and exponents: some rules. Here are some of the rules:

(1) ln is a one-to-one function from (0, ) to R and exp is a one-to-one function from R to (0, ). The
two functions are inverse functions of each other.
(2) ln converts products to sums and exp converts sums to products. In other words, ln(xy) = ln(x) +
ln(y) and exp(x + y) = exp(x) exp(y).
(3) ln(1) = 0 and exp(0) = 1.
(4) ln(1/x) = ln x and exp(x) = 1/ exp(x).
(5) ln(xr ) = r ln x and exp(rx) = (exp(x))r .
(6) Both exp and ln are continuous and increasing functions.
(7) exp has the x-axis as a horizontal asymptote as x , while ln has the y-axis as a vertical
asymptote as x 0.
(8) exp is concave up and ln is concave down.

Here are the graphs:

The logarithm graph, zoomed in for small inputs:

The logarithm graph, zoomed out:

4
 The exponential graph:

Here are the logarithm and exponential graphs together, so that we can see that the graphs are reflections
of each other about the y = x line:

5
2.3. Numerical shennanigans. In his autobiographical book, Richard Feynmann discusses how he im-
pressed a bunch of mathematicians by being able to calculate natural logarithms of many numbers to one or
two decimal places. However, what he did was hardly impressive. It turns out that remembering the natural
logarithms of a few numbers allows us to compute them approximately for many numbers.
For instance, it is useful to remember that ln(2) 0.6931, ln(3) 1.0986, ln(10) 2.3026, and ln(7)
1.9459. We can now calculate the logarithm values for most integers. How? Using the fact that logarithms
translate multiplication to addition. Thus, ln(4) = 2 ln(2) 1.3862, while ln(5) = ln(10) ln(2) 1.6095.
In fact, we can readily calculate the natural logarithm of any positive integer all of whose prime factors
are among 2, 3, 5, and 7. What about ln(11)? While we cannot calculate this precisely, we can calculate
ln(10) and ln(12) and thus obtain reasonable upper and lower bounds for ln(11). Even better, we know that
ln(120) < 2 ln(11) < ln(125), and since we can calculate both ln(120) and ln(125), we get a pretty small
range for ln(11).
In fact, Feynman was able to impress physicists by doing calculations that essentially relied on only two
facts: ln(2) 0.7 and ln(10) 2.3.
If we are able to quickly calculate natural logarithms, a happy corollary of that is that we can quickly
integrate dx/x on intervals.

2.4. Domain and range issues. For domain computations in the past, we used the following basic guide-
lines:
(1) Things in the denominator must be nonzero.
(2) Things with squareroots or even roots must be nonnegative.
(3) Things with squareroots or even roots in the denominator must be positive.
We now add two more criteria:
6
 (4) Things under logarithm must be positive.
(5) Things under logarithm of the absolute value must be nonzero.

2.5. Logarithm of the absolute value. The natural logarithm function is defined only for positive reals.
However, we can extend it to a function on all reals by taking the absolute value first, i.e., we look at the
function x 7 ln(|x|). This is an even function and its graph is obtained by takingthe graph of the logarithm
function and adding its mirror image about the y-axis
It turns out that the derivative of ln(|x|) is 1/x. In particular, we see that ln(|x|) serves as an antiderivative
of 1/x for all nonzero x. This is an improvement on ln(x), which worked only for positive x. However, we
should be careful because the domain of 1/x as well as of ln |x| excludes zero. Hence, the behavior on the
positive and negative side are totally independent of each other. We shall return to this point in a later
lecure.

3. Formulas for derivatives and integrals

3.1. Derivative and integral formulas for logarithms. The main formula that we have, which follows
from our definition of natural logarithm, is the following:

d 1
(ln x) =
dx x
This is the formula for x > 0. A more general version, for x 6= 0, is:

d 1
(ln |x|) =
dx x
The corresponding antiderivative formula for x > 0 is:
Z
dx
= ln(x) + C
x
In general, the antiderivative formula is:
Z
dx
= ln(|x|) + C
x
However, it should be remembered that this formula is valid only when we are working with x either in
(0, ) or in (, 0), i.e., we cannot use the formula to cross between the interval (0, ) and (, 0). In
fact, if we have a function f : R \ {0} R such that f 0 (x) = 1/x, then we can guarantee that f (x) ln(|x|) is
constant on x > 0 and is constant on x < 0. However, these constants may differ. The behavior on the (0, )
connected component does not in any way constrain the behavior on the (, 0) connected component.

3.2. The exponential and its derivative and integral formulas. Recall that exp is the inverse of the
ln function. Thus, we can use the rule for differentiating the inverse function to find exp0 . We have:

1 1
exp0 (x) = = = exp(x)
ln0 (exp x) 1
exp(x)
Thus, we have the remarkable property that the exponential function, i.e., the function x 7 ex , is its
own derivative. Another way of thinking about this is that the rate of growth of the exponential function is
equal to its value. Note that this also implies that the exponential function is infinitely differentiable and all
higher derivatives equal the same function.
We rewrite the above in Liebniz notation:

d x
(e ) = ex
dx
We also note the corresponding statement for indefinite integration:
Z
ex dx = ex + C

7
3.3. Some corollaries. Using the above, we obtain the following identities for the logarithm and exponent:

d 1
(ln(kx)) =
dx x
d r
(ln(xr )) =
dx x
d mx
(e ) = memx
Zdx
1 mx
emx dx = e +C
m

Each of these identities can be derived in two ways: either by using the properties of logarithms and
exponents on the inside and then differentiating, or by first differentiating and then simplifying. For instance,
for the second identity, we can either simplify ln(xr ) as r ln(x) first and then pull the constant r out before
differentiating, or we can use the chain rule to obtain (1/xr ) rxr1 . It is gratifying to know that the answers
we obtain both ways are the same.

4. Application to indefinite and definite integration

4.1. The u-substitution: a textbook example. We begin with an easy example:
Z n
1
x ln x
dx
n
Here, n is an integer greater than 1.
Believe it or not, this integral actually came up in some asymptotic approximations I was doing some
time ago to figure out whether some numbers have large prime divisors! Let us first look at the indefinite
integral. The substitution u = ln x gives us:
Z Z
1 du
= = ln(u) = ln(ln x) + C
x ln x u
Note that we do not need to put absolute values here because on the interval of integration, ln is positive.
Now, we can evaluate between limits:

   
n ln n ln n
[ln(ln x)]n = ln(ln n) ln(ln n) = ln = ln = ln 2
ln n (1/2) ln n
So, the answer is ln 2, which, as we computed earlier, is approximate 0.693. Apparently, this is the rough
heuristic argument for why about 69.3% of the numbers have a prime divisor greater than their squareroot.
4.2. Numerator as derivative of denominator. The gist of this logarithmic substitution can be captured
by the formula:
Z 0
g (x)
dx = ln |g(x)| + C
g(x)
The proof of this proceeds via setting u = g(x). Thus, the general idea when using logarithmic substi-
tutions is to try to obtain the numerator as the derivative of the denominator. For instance, consider the
integral:
Z
x
dx
x2 + 1
Here, the derivative of the denominator is 2x, so we adjust by a factor of 2 to obtain:
Z
1 2x 1
dx = ln(|x2 + 1|) + C
2 x2 + 1 2
2
Note that in this case, since x + 1 is always positive, the absolute value can be dropped and we get
1 2
2 )(x + 1) + C. Further, this antiderivative is valid over all reals.
8
4.3. Trigonometric integrals involving logarithms. Recall so far that we have seen the antiderivatives
of sin, cos, sec2 , sec tan, csc2 , and csc cot. We also used these, along with trigonometric identities, to
compute antiderivatives for sin2 , cos2 , tan2 , and cot2 . All these results were obtained as corollaries of the
differentiation formulas.
We now try to obtain a formula to integrate cot. The key idea is to note that:
cos x
cot x =
sin x
Since cos is the derivative of sin, this matches up with the general pattern that we just discussed, and we
obtain that the antiderivative of cot is ln | sin |. In other words:
Z
cot x dx = ln | sin x| + C

Note that cot is undefined at multiples of , and so any integration of this sort is valid only if the entire
interval of integration lies strictly between two consecutive multiples of . It is also instructive to graph the
antiderivative of cot.
Here is the graph of cot and its antiderivative on the interval (0, ), where both are defined:

Similarly, we obtain:
Z
tan x dx = ln | cos x| + C = ln | sec x| + C

Here is the picture of tan and ln | cos | on the interval (/2, /2), where both are defined:

Note that those two expressions are the same because cos and sec are reciprocals of each other.
Let us look at a somewhat harder integral: the integral of the secant function:

sec2 x + sec x tan x

Z Z Z
sec x(sec x + tan x)
sec x dx = dx = dx
sec x + tan x tan x + sec x
The numerator is the derivative of the denominator, and we obtain:
Z
sec x dx = ln | sec x + tan x| + C
9
 In a similar vein, we obtain that:
Z
csc x dx = ln | csc x cot x| + C

4.4. Domain and range issues. When doing indefinite integration, it is often best to forget about issues
of domain and range and just let the algebraic manipulations flow. However, to interpret the results at the
end, it is important to look at the domain and range issues. Ideally, the antiderivative should be defined and
should make sense on all intervals where the function itself is continuous. Further, if there are points where
the function is continuous but the expression obtained for the antiderivative is not defined, we should try to
obtain the limit at that point.
4.5. An application: integrating the cube of the tangent function. Let us look at an application of
the above:
Z
tan3 x dx
Before we proceed, it is worth remarking how different integration is from differentiation. For differen-
tiation, there was just the formula for differentiating sin and cos, and everything else followed using the
product rule and quotient rule. We still memorized more, but that was mainly to speed things up, not out
of necessity. With integration, on the other hand, we need to have a whole bag of ad hoc tricks that we try
one after the other.
Let us look at this integral. The key thing to do here is to break down tan3 x = tan x tan2 x. Next, we
use tan2 x = sec2 x 1, and we have:
Z Z
2
tan x sec x dx tan x dx
The first integral can be quickly calculated using the chain rule or u-substitution, since sec2 x is the
derivative of tan x. The second integral comes from our formula, and we get:

tan2 x
+ ln | cos x| + C
2
4.6. A fancier formula. Here is a formula that uses the chain rule twice:
Z 0
g (x)f (ln |g(x)|)
Z
dx = f (u) du
g(x)
where u = ln(|g(x)|). For instance:

2x(ln(x2 + 1))3
Z Z
dx = u3 du
x2 + 1
where u = ln(x2 + 1). This further simplifies to:
1
[ln(x2 + 1)]4 + C
4
5. More tricks and techniques
5.1. Logarithmic differentiation. Logarithmic derivatives are both a conceptual and a computational
tool. Currently, we focus on the computational aspects. The idea is to use the same formula that we
obtained earlier, but in reverse:

d g 0 (x)
ln(|g(x)|) =
dx g(x)
Rearranging the terms yields:

d
g 0 (x) = g(x) ln(|g(x)|)
dx
10
 If g is a product of functions g1 , g2 , . . . , gn , then ln |g(x)| = ln |g1 (x)| + + ln |gn (x)|, and we get:
 0
g (x) g20 (x) g 0 (x)

g 0 (x) = g(x) 1 + + + n
g1 (x) g2 (x) gn (x)
Note that this expression is not really new and did not really require logarithms. You can in fact convince
yourself that it is just a reformulation of the product rule. When g = g1 g2 , for instance, this says that:
 0
g0

g
g0 = g 1 + 2
g1 g2
Substituting g = g1 g2 , this simplifies to the usual product rule. However, the logarithmic formulation has
some conceptual advantages.
For instance, suppose g(x) := x(x 1)(x 2). Then, we immediately obtain that:

g 0 (x) 1 1 1
= + +
g(x) x x1 x2
Further, of we have g(x) = g1 (x)a1 g2 (x)a2 . . . gn (x)an , we obtain that:

g 0 (x) a1 g10 (x) a2 g20 (x) an gn0 (x)

= + + +
g(x) g1 (x) g2 (x) gn (x)
3 4 5
So, if g(x) = x (x 1) (x 2) , we obtain that:

g 0 (x) 3 4 5
= + +
g(x) x x1 x2
5.2. Exponentiation tricks. We have already seen basic integration and differentiation identities for the
exponentiation function. There are some ways of combining these identities with the chain rule. I note some
special cases here.
(1) For any function f , the derivative of f (x)ex is (f (x) + f 0 (x))ex . Thus, the integral of g(x)ex is
f (x)ex + C where f + f 0 = g.
(2) (1) is particularly useful when integrating polynomial function times ex . This is because we can use
linear algebraR to find, for a given polynomial g, the unique polynomial f such that f + f 0 = g.
(3) The integral f (e )e dx is f (u) du where u = ex .
x x
R

(4) The integral ef (x) f 0 (x) dx is ef (x) + C.

Let us consider an example to illustrate this. Consider the function:

F (x) := (x2 + 5x + 1)ex

The derivative of this, by the product rule, turn out to be ex times the sum of x2 +5x+1 and its derivative,
giving:

F 0 (x) = (x2 + 7x + 6)ex

Note that this is a new polynomial times ex . An intersting question would be how we could reverse this
procedure, i.e., given g(x)ex where g is a polynomial, how do we find a polynomial f such that the derivaitve
of f (x)ex is g(x)ex ? By the product rule, we obtain that:

g(x) = f (x) + f 0 (x)

Thus, we need to find the coefficients of f . Let us do this in our concrete case where g(x) = x2 + 7x + 6.
We know that the degree of f 0 is strictly smaller than the degree of f , so f + f 0 has the same degree and
same leading coefficient as f . In this case, this forces f to be a quadratic polynomial of the form x2 + mx + n.
We then get f 0 (x) = 2x + m, and we obtain that:

f (x) + f 0 (x) = x2 + (m + 2)x + (m + n)

Since we are given that g(x) = x2 + 7x + 6, we can match coefficients and obtain:
11
 m + 2 = 7, m+n=6
Solving, we get m = 5, and n = 1, and we get f (x) = x2 + 5x + 1, recovering our original polynomial.
Although the specific procedure involving comparing coefficients does require that we are dealing with
polynomials, the general idea remains true in a broader sense: integrating g(x)ex is equivalent to finding
R

a function f such that f + f 0 = g. In some cases, it is easier to think of it as an integration problem, and
in others, it is easier to think of it in terms of the differental equation f + f 0 = g. The idea that these two
apparently different computations measure the same thing is extremely important.

12