SPLASH 2011 Superhuman Integration Techniques

Andre Kessler
November 19th, 2011

1 Series Expansions
Before we get to our integration techniques, we need to review the notion of a series expansion. The
techniques and methods developed here are fundamental to the derivation of most of the results we will
employ, so it is necesary to have a very solid foundation in this area. To begin, all of the ideas about series
expansions start with the finite geometric series
n
X 1 x n+1
1 + x + x2 + + xn = xk = .
k=0
1 x

We can verify the closed form on the right hand side of the above equation by multiplying each side by
the quantity 1 x. This clearly telescopes the sum and gives us the desired result, as long as x 6= 1. In
the case that x = 1, the sum is easily computable by just realizing its now the sum of n + 1 ones, but it
is important to note that our formula actually still holds we just need to be more careful and take the
following limit:
1 x n+1 (n + 1)x n
lim = lim = n + 1.
x1 1 x x1 1
We can also consider infinite geometric series of the form 1 + x + x 2 + x 3 + . This clearly converges as
long as |x| < 1, because in those cases the term x n+1 in our closed-form expression for the series goes to 0
as n goes to infinity, and the result we are left with is well-defined. Specifically, we have the result that

X 1
2
1 + x + x + = xk = .
k=0
1 x

Now, we can turn these ideas on their head and say instead that 1 + x + x 2 + is the power series
representation of the function f (x) = 1/(1 x) for any x with magnitude less than one. In other words,
we are expanding a function which is not a polynomial as an infinite-degree polynomial. We can do this
simply because the two sides are equal when the series converges. If the series diverges for a particular
value of x, the two sides are no longer equal. However, the function f (x) may still be well defined even
when the power series does not converge. For example, consider x = 1 in the standard geometric series.
In this case, we have the seqeuence 1 1 + 1 1 + . Does this converge? If we group the terms one
way and add them, we get a sum of 0: (1 1) + (1 1) + = 0. On the other hand, we can group the
terms differently and get a value of 1 as follows: 1 + (1 + 1) + (1 + 1) + = 1. Clearly this is not well
defined. Our function, on the other hand, is perfectly reasonable at x = 1 : now f (x) = 1/2. This is
a very important notion to remember when we define a function on some set of numbers using a power
series. For example, later on we will define the zeta function by
1
X
(s ) = .
n=1 ns

The series on the right hand side above converges for all s with |s| > 1. However, it turns out that we can
consider (s) as a function defined on all complex numbers s 6= 1. This is called analytic continuation, and
while we will not discuss this in depth, it is important to realize that power series can be limited in their
description of a function, and understanding where the series converges is necessary to determine where
a power series expansion will be useful.

1
From our humble little expansion for 1/(1 x) we can pull an infinite variety of different expansions;
some examples are shown below.
1
= 1 + x2 + x4 + x6 +
1 x2
!
1 11 1 x3 x6
= = 1 + +
8 + x3 8 1 x
 3

8 8 64
8
ex 1
x
= x
= 1 + e x + e 2x + e 3x +
e 1 1e
Now, the question logically arises: how can we find these series expansions in general? Suppose we want
to find a polynomial T (x) that approximates a function f (x) around x = x0 . We can start off by simply
letting T (x) = f (x0 ). This is a good approximation directly at x = x0 , but its probably not very good
anywhere else. To make this a better approximation, we should make sure that the functions have the
same slope at the point, which is equivalently the statement T 0 (x0 ) = f 0 (x0 ). We can do this by saying
T 0 (x) = f 0 (x0 ) T (x) = f (x0 ) + f 0 (x0 )x. Now lets make T (x) an even better approximation by making
the second derivatives agree:
1
T 00 (x) = f 00 (x0 ) T 0 (x) = f 0 (x0 ) + f 00 (x0 )x T (x) = f (x0 ) + f 0 (x0 )x + f 00 (x0 )x 2 .
2!
This process can clearly be continued indefinitely to obtain an arbitrarily accurate expansion, so if we
continue it forever we can write out the Taylor expansion of f (x). It is important to notice that this only
works if f (x) is infinitely differentiable at x = x0 .
f (k) (x )
0
X
f (x) = (x x0 )k
k=0
k!

Armed with the general Taylor expansion, we can arrive at expressions for some functions. When are
these expressions valid?
xn
X
ex =
k=0
n!
X x 2k+1
sin x = (1)k
k=0
(2k + 1)!
X x 2k
cos x = (1)k
k=0
(2k)!
X n 
(1 + x)n = xk
k=0
k
1 2k 
X
p = xk
1 4x k=0 k

X x 2k+1
arctan x = (1)k
k=0
2k + 1
X x k+1
ln (1 x) =
k=0
k +1

2
 1
In particular, expanding gives us 1 + x + x 2 + as expected. Another slick method for finding
1 x
the series for sin x and cos x involves separating the real and imaginary parts from the two sides of the
expansion for the expression e i x = cos x + i sin x. Importantly, we can differentiate series or integrate
series along with their corresponding functions as much as we please; this is how we obtain the expansion
for arctan x (integrating the expansion for 1/(1 + x 2 )). As an example of this process, consider trying to
1
find the expansion for ln (1 x). We observe that this is an antiderivative of , so
1 x
Zx
dt
ln (1 x) =
0 1 t
Z xX
= tk dt
0 k=0
Z
X x
= tk dt
k=0 0
X x k+1
=
k=0
k +1

In the second-to-last step, we interchanged the order of summation and integration. I will not spend a lot
of time justifying this step, as it is rather tedious; suffice it to say that as long as the sequence or integral
does not have some strange convergence issues, you are allowed to do this pretty much whenever the
results make sense.
Suppose we want to determine the first three terms in the series expansion for tan x about x = 0. The
tangent function has derivatives of all orders at this point, so our series clearly exists. However, it would
be quite annoying to compute term by term from the general Taylor series formula. Instead of doing so,
we will divide the two series for sine and cosine as follows, and use a geometric series expansion.

sin x
tan x =
cos x
3 x5
x x3! +
!
5!
x3 x5 1
= 2 4
= x +  2

x x 3! 5! x4
1 2!
+ 4!
1 x
2!
4!
+
! ! !2
x3 x5 x2 x4 x2
= x + 1 + + + +

3! 5! 2! 4! 2!
! !
x3 x5 x2 5x 4
= x + 1+ + +
6 120 2 24
x3 x5 x3 x5 5x 5 x3 2x 5
=x + + + + = x + + +
6 120 2 12 24 3 15
Notice several things about this computation. First, we only kept the terms that could contribute some-
thing to the first three terms anything with a power of x larger than 5 cant possibly contribute to the
first three terms. Next, notice that we only have odd powers in our Taylor expansion; this is because tan x
is an odd function. Finally, similar methods can be applied for the division of series in any number of
cases like this one.

3
2 Generating Functions
Z 1
Okay, now its finally time to compute some integrals. Lets consider x 5 ln2 x d x. Were going to use
0
the fact that a power series expansion for a function about the same point is unique. Look closely at the
following manipulation.
Z1 Z1 Z1
n
x n+
dx = x x dx = x n e ln x d x
0 0 0
Z 1 k
X
= xn lnk x d x
0 k=0
k!
k
X Z 1
= x n lnk x d x
k=0
k! 0

Z 1
Thus, if we can find the terms of the power series expansion for x n+ d x in some other way, we can
0 Z1
then compare coefficients to determine the value of any one of the integrals x n lnk x d x. But we already
0
know how to evaluate the first integral and how to expand the resulting function, so we proceed to do so
as below.
1+n+ 1
Z1 
x
x n+ d x =

1 + n + 

0 0
1 1 1
= =
1+n + 1 + n 1 + n+1

X (1)k
= k
k+1
k=0 (n + 1)

Comparing coefficients of k , we immediately obtain that

1 k!(1)k
Z
n k
x ln x d x =
0 (n + 1)k+1

and therefore the specific integral we wanted to compute at the beginning is simply
1 2!(1)2 1
Z
5 2
x ln x d x = = .
0 63 108

2.1 Exercises
Z 1
1. Compute x ln x d x.
0
Z 2
2. Determine the value of ln4 x d x.
0

2011
ln x
Z
3. (HMMT 2011) Determine the value of d x.
1 x

4
 Z 1 Z
n
4. Make the substitution x = ln u in the integral ln u d u to show that x n e x d x = n! and
0 0
verify this result in some other way (for example, via integration by parts). This integral will be the
basis of our attempt to extend the factorial function to non-integer arguments.

3 Product Expansions
Why did we even look at infinite sums in the first place? It was because we were interested in representing
some function f (x) which is not a polynomial as a sort of infinite-degree polynomial. So far we have
been looking at the coefficients of power series in other words, the coefficients of these infinite-degree
polynomials. But coefficient form is not the only way to think of a polynomial we can also consider
its factored form. An ordinary polynomial has finitely many zeros. However, functions (for example,
sin x and cos x) can have infinitely many zeros. This begs the question as to whether it is possible to
represent such functions in factored form in other words, as a product of terms that describe the zeros
of the function. Weierstrasss Factorization Theorem answers with a resounding YES for a particular class
of functions in complex analysis known as entire functions1 , and we will shortly see how to work with
these infinite product representations. First, however, we will discuss conditions for convergence of such
infinite products.

3.1 Convergence of Infinite Products

By infinite product we mean something in the form

Y
un
n=1

where un is some non-zero complex number. We will say that the above equation converges if the sequence
of partial products u1 u2 un converges to a nonzero limit. Clearly, a necessary condition for an infinite
product to converge is that lim un = 1 (why?). Thus, it makes sense to write un = 1 + an for some
n Y Y
sequence an with lim an = 0 in our infinite product representations instead: un = (1 + an ). How
n
can we now connect this to series? Consider writing a convergent infinite product as the exponential of a
sum, which we are certainly allowed since we restricted convergence of infinite products to exclude zero.
( )
Y X
1 + an = exp ln (1 + an )


n=1 n=1

Thus, we can answer

X questions about the convergence of the infinite product by simply considering the
convergence of ln (1 + an ). If an consists entirely of positive terms, we can make an even stronger
Y
statement: if all of the terms an have the same sign, then the convegence of the infinite product (1+an )
X
is implied by the convergence of the series an (to see this, simply consider expanding the logarithm
X Y  1

in ln (1 + an ) using our techniques from before). For example, 1+ diverges to infinity, but
n=1 n
 1 e e sinh
Y 
1 + 2 converges to = .
n=1 n 2
1
Basically, a function that is infinitely differentiable everywhere in the complex plane. Alternately, an entire function is a
function whose power series expansion (say, about z = 0) converges everywhere in the complex plane.

5
3.2 Representation of Functions
Now, if we want to represent a polynomial p(x) that has roots r1 , r2 , . . . , rn , then we can write it in
factored form as
n

p(0) Y x
p(x) = (x r1 )(x r2 ) (x rn ) = p(0) 1
(1)n r1 r2 rn k=1
rk

If some entire function f (x) has infinitely many roots, then by the Weierstrass Factorization Theorem,
sin x
we can write it as an infinite product of its zeros. Take, for example, f (x) = , with f (0) = 1. Then
x
we can write
sin x x x x x
=1 1 1+ 1 1+
x ! ! 2 ! 2
x2 x2 x2
= 1 2 1 2 2 1 2 2
2 3
!
Y x2
= 1 2 2
n=1 n

Notice that the order in which we multiplied out the terms was important. Multiplying only the terms
for positive roots first would have resulted in a divergent infinite product, so we need to be careful while
doing these manipulations that we do not separate our product in ways that are not allowed.
We are now going to be able to use our infinite product expansion to obtain a very interesting result.
Consider expanding the infinite product by multiplying out the terms. This will be an infinite sum; in
fact, it will be a power series for f (x). Just from reading off the terms in the infinite sum, we can see that
the expansion will be
sin x 1 1 1
 
=1 + + + x 2 + O(x 4 ).
x 2 22 2 32 2
However, the power series expansion for any function f (x) is unique, and we know the expansion for
sin x
is 1 x 2 /6 + O(x 4 ). The two coefficients must be the same, so we know that
x
1 1 1 1 1 1 1 1
X 2
+ + + = = 1 + + + + = = .
2 22 2 32 2 6 22 32 42 n=1 n
2 6

This important result was first discovered by Euler and brought him a great deal of fame.

4 The Gamma Function

Z
Earlier, in one of the exercises, we derived that x n e x d x = n!. This is all right and good for nonneg-
Z 0
p
ative integers n, but what if we want to know x e x d x? Even if we assumed that our expression is
0
true for all values of n that cause the integral to converge, we still wouldnt know how to determine the
value of (1/2)!. Thus, wed like to determine an extension of the factorial function to complex arguments.
Well, what are the defining properties of the factorial function in the first place, anyways? The most
fundamental requirements are the recursion relation n! = n (n 1)! and 0! = 1. Thus well define the
function (z) which satisfies the following relations: (z + 1) = z(z), (1) = 1, and is logarithmically

6
convex. It turns out that this function is unique; additionally, (z) = (z 1)!. Now, well take n to be a
very large positive integer and write a limit definition of the gamma function:
z terms
z }| {
(n + z)! n! (n + 1)(n + 2) (n + z) n!n z
(z) = =
(n + z)(n + z 1) (z + 1)z (n + z)(n + z 1) (z + 1)z (n + z)(n + z 1) (z + 1)z

n!n z
(z) := lim
n (n + z)(n + z 1) (z + 1)z
This will be our way of defining the Gamma function. It is easy to check that it satisfies all of our desired
properties in this form; moreover, this form is valid for all complex z, excluding the non-positive integers
(why?). Okay, hold on to your chairs, because were about to do something pretty crazy.

1 (n + z)(n + z 1) (z + 1)z n + z n 1+ z z +1
 
z ln n
= lim = z lim e
(z) n n!n z n n n 1 1
 z 
z
  Y n z
= z lim e z ln n 1 + 1+ (1 + z) = z lim e z ln n 1+
n n n 1 n
k=1
k
n z n z
z ln n+z nj=1 1j
Y P Y
z ln n z/k z/k z/k
= z lim e e e 1+ = z lim e e 1+
n
k=1
k n
k=1
k
z
= ze z
Y
e z/k 1 +
k=1
k

This infinite product reminds us in several ways of the infinite product for sin z; in fact, we can pull
apart the expansion for that exact function to obtain the Gamma function reflection identity.
!
Y z2 Y
z Y z
sin z = z 1 2 = z e z/k 1 + e z/k 1
k=1 k k=1
k k=1 k
1 1
= z =
z(z) (z)(z) (z)(1 z)

The above manipulations give us the famous result that (z)(1 z) = .
sin z

4.1 Exercises
Z
2
1. The integral e x d x frequently appears in the study of probability distributions. Compute it.
0

2. Evaluate the following integrals in terms of the Gamma function, and simplify as much as possible.
Z Z Z
2p
(a) x 3 3x d x (c) x 3 e x cos x d x (e) x2x x d x
0 0 0
Z p
Z Z p
3
x 3
(b) x 2 e 2 x
dx (d) x3e dx (f) x 2 x e x sin x d x
0 0 0
Z p p
x
3. Determine the value of f () = e cos x d x for > 0. Is your simple expression f ()
0
defined at = 0? What about the integral?

7
4.2 Expansions of (1 + )
Z
So far we know how to integrate things such as x 2 e x d x, but we dont immediately have a way to
Z 0
2 x
determine x ln x e d x. From the methods we used earlier, we know that a good way to proceed
0 Z
would be to consider the integral x 2+ e x d x = (3 + ) = (2 + )(1 + )(1 + ). It turns out that
0
evaluating these integrals will come down to requiring an expansion for (1+), so we will use the infinite
product developed earlier to find one. We know that
1
(1 + ) = () = e e /k
Y
1+
k=1
k

so we can write an expansion for the natural logarithm of (1 + ) as follows.

X X X (1) j +1 j
ln (1 + ) = + ln 1 + = +

k k k j

k=1 k=1 j =1 jk
X (1) j j
X (1) j j X
X 1
= + = +
k=1 j =2 jkj j =2 j k=1 kj
X (1)n (n)
= + n .
n=2 n

4.2.1 Exercises
1. Evaluate the following integrals in terms of the Gamma function, and simplify as much as possible.
Z Z Z
2
(a) ln x e x d x (c) x ln x e 2x d x (e) ln2 x e x d x
0 0 0
x 2 ln x p
Z Z Z
(b) ln2 x e x d x (d) x
dx (f) x ln x e 3x d x
0 0 0

2. Take = 1 in our expansion for ln (1 + ) in order to obtain an infinite series representation for .
Does this sum converge quickly or slowly? What can you do to make it converge faster?

3. Take = 1/2 in our expansion for ln (1 + ) in order to obtain an infinite series representation
for . Does this sum converge quickly or slowly? What can you do to make it converge faster?
Z
p 1
 
x
4. Evaluate the integral 3
x ln x e d x. Your expression should contain terms of the form + .
0 3
How might you deal with these terms? Consider the infinite product expansion of the gamma func-
tion, and see if you can derive the relation
p
1 2 2 3
   
(1 + ) + + = 3+1 (1 + 3).
3 3 3

8
4.3 The Beta Function
A particularly lucrative result can be obtained by considering the product ()(). Making the substi-
tutions u = x + y and t = x/u will allow us to obtain the celebrated Beta function integral.
Z Z
()() = x 1 e x d x y 1 e y d y
Z0 Z 0

d x x 1 y 1 e xy

= dy
Z0 Z0 u
d x x 1 (u x)1 e u

= du
0 0
Z Zu
x 1 1 x 1 u

= du dx u u 1 e
0 0 u u
Z Z1
u d t u 1 t 1 u 1 (1 t )1 e u

= du
0 0
Z Z 1
+1 u
= u e du t 1 (1 t )1 d t
0 0
Z 1
= ( + ) t 1 (1 t )1 d t
0

Rearranging our expression, we can write

1 ()()
Z
t 1 (1 t )1 d t = = B(, ).
0 ( + )

For simple integer values of and , this expression is nothing all that special, but its real power comes
Z 1 Z1
p
when we consider integrals such as t (1 t ) d t or t 1 t . Previously we would have needed to
0 0
make somewhat annoying trig substitutions, but now we can simply plug and chug to get the answer
quickly. From the beta  we can also quickly derive the Legendre Duplication Formula, which
 function
1
allows us to express + z in terms of known quantities. Consider B(z, z), or
2

(z)(z)
Z1
= t z1 (1 t ) z1 d t
(2z) 0
u + 1 z1 u + 1 z1 d u
Z1   
= [1 ]
1 2 2 2
Z 1
1 z1
= 2z2 1 u2 du
2 0
1 (1/2)(z)
Z1
1
= 2z1 x 1/2 (1 x) z1 d x = 2z1
2 0 2 12 + z

1
 
Solving for + z , we see that
2
1 p (2z)
 
+ z = 212z .
2 (z)

9
Finally, we can obtain an extremely useful expression involving the Beta function and powers of trigono-
metric functions. Consider substituting x = u 2 and y = v 2 in the product ()() as follows.
Z Z
()() = x 1 e x d x y 1 e y d y
0 Z 0
 Z 
21 u 2 21 v 2
= 2 u e du 2 v e dv
0 0
Z Z /2 h i
2
=4 r dr d r 2+22 e r cos21 sin21
0 0
Z Z /2
2+21 r 2
=4 r e dr cos21 sin21 d
0 0
Z Z /2
=2 u +1 e u d u cos21 sin21 d
0 0
Z /2
= 2 ( + ) cos21 sin21 d
0

Solving for the integral, we obtain the result that

Z /2 ()()
cos21 sin21 d = .
0 2 ( + )

This allows us to obtain previously known results much more easily such as
Z /2 5
sin6 cos6 d =
0 2048

/2 p

2
Z
in addition to entirely new, previously unknown results such as the fact that sin d = 2 (3/4).
0

4.3.1 Logarithms and the Beta Function

We can apply regularization techniques similar to those developed earlier in order to compute integrals in-
volving logarithms of x or 1x over the interval 0 to 1 (or, similarly, integrals involving logarithms of sines
Z1
and cosines over the interval 0 to /2). What if we want to determine the value of ln x ln (1 x) d x?
0
Similar to our earlier expansion methods, we will consider the integral
1 (1 + )(1 + )
Z
x (1 x) d x = .
0 (2 + + )

10
The integral we are looking for corresponds to the term in the double Taylor expansion. Hence, we
can manipulate this expression of Gamma functions to yield the desired term.
(1 + )(1 + ) 1 (1 + )(1 + )
=
(2 + + ) 1++ (1 + + )
(2) 2


2
= 1 ( + ) + ( + ) exp ( + ) + + 2 +
2
(2)

+ ( + ) ( + )2
2
= 1 ( + ) + ( + )2 e (2)

2 2
2
= 1 + + 2 + + (1 (2) + ) our term is 2 .
6

This technique takes some getting used to, but once you have it down you can obtain some very important
results all that is required is algebraic fortitude.

4.3.2 Exercises
1. Evaluate the following integrals in terms of the Gamma function, and simplify as much as possible.
Z1 Z /2 Z1
ln x
(a) p dx (e) sin x ln sin x d x (i) x ln x ln (1 x) d x
0 1 x 0 0
Z1 Z1 Z /2 
1 1 1

(b) ln dx (f) ln2 x ln (1 x) d x (j) csc x dx
0 x 1 x 0 0 x
Z1 Z 1 ln x Z
2 3 (g) d x
(c) x (1 x) ln x d x p (k) sin2 x cos8 x d x
0 0 x(1 x) 0
Z1 Z /2 Z p
p
(d) x3 1 x d x (h) sec x ln sin x d x (l) sin4 x sin x d x
0 0 0

2. Make the substitution u = 1 + 1/t in the Beta function integral to obtain

u 1
Z
= B(, ).
0 (1 + u)+
x m1 m
Z
From this obtain n
dx = csc and then compute the following integrals.
0 x +1 n n
Z p p
x ln2 x
ln x Z
Z 3
x x
(a) 3
dx (b) (c) dx
0 x +1
2 2
0 x4 + 1 0 (x + 1)

4.4 n-Dimensional Hyperspheres

Consider an n-dimensional hypersphere. When we think of the word sphere, we are usually thinking of a
2-sphere (the surface itself is 2-dimensional and it is embedded in three-dimensional space). Thus, it makes
sense to set d := n +1 as the dimension of the space the sphere is embedded in. Additionally, to be more
specific, we will define an n-sphere as the surface that satisfies the equation

x12 + x22 + + xn+1

2
= r2 (1)

11
where the radius r is some positive number. Clearly, we can parametrize this surface using n angular vari-
ables 1 , 2 , . . . , n1 , (simply think of the case of the 2-sphere, where we get away with parametrizing
the surface with just the variables , ) with 0 i and 0 2. The specific parametrization of
x1 , x2 , . . . , xn+1 is shown below.

x1 = r cos 1
x2 = r sin 1 cos 2
x3 = r sin 1 sin 2 cos 3
..
.
xn = r sin 1 sin 2 sin n1 cos
xn+1 = r sin 1 sin 2 sin n1 sin

To easily obtain the surface area, we need to determine what the area differential d x1 d x2 d xn1 is in
terms of these angular variables. Clearly, for 1 k n 1, we have d xk = r sin 1 sin 2 sin k , so we
can write the surface area Sn as
Z Z Z Z 2
Sn = d 1 d 2 d n1 d r sin 1 sin n1 r sin 1 sin n2 r sin 1 r
 
0 0 0 0
Z Z Z Z 2
= d 1 d 2 d n1 d r n sinn1 1 sinn2 2 sin n1
0 0 0 0
 Z  Z 2
Z  Z
n n1 n2
=r sin 1 d 1 sin 2 d 2 sin n1 d n1 d
0 0 0 0

n2 12 n1
2
12 (1) 12
n
=r 2
n+1
2
n2 32
n+1
2 2 2 d /2 d 1
= rn = r .
n+1 (d /2)
2

5 The Zeta Function

The Riemann Zeta function is defined for complex numbers s with real part greater than 1 by the follow-
ing series.
1 1 X 1
(s) = 1 + s + s + =
2 3 k=1
ks
The sum on the right converges absolutely whenever Re[s] > 1. To see that we encounter problems for
1 1 1 1 1
s = 1, suppose the sum converges to a particular limit L = 1 + + + . Then L/2 = + + + , so
2 3 2 4 6
1 1
1+ + + = L/2 as well, but every corresponding term of the sequence of odd terms is larger than the
3 5
sequence with even terms, contradiction. To see the Zeta functions connection to the prime numbers,
observe that we can factor the series as an infinite product over all primes:
1
X Y ps
(s) = = .
k=1
ks p ps 1

12
Additionally, there are some specific integrals in which the Zeta function makes a very important appear-
ance. Consider the following manipulations.
x n1 x n1 e x
Z Z Z X
dx = = x n1 e x e k x d x
0 ex 1 0 1 e x 0 k=0

X Z
X 1
Z
n1 (k+1)x
= x e dx = n
u n1 e u d u
k=0 0 k=0
(k + 1) 0

= (n) (n)
x n1
Z
A virtually identical sequence of manipulations can be used to show that = (n)(n), where
ex + 1 0
the Dirichlet eta function (n) is defined as a sort of alternating zeta function:

1 1 (1)k+1
X 
1

(n) = 1 + + = = 1 n1 (n).
2n 3n k=1
kn 2

As usual, considering a more general form of the integral allows us to differentiate through the integral
sign and produce many more highly useful results:
x n1
Z
ax
= a n (n) (n)
0 e 1
xnex
Z
= (n + 1) (n)
0 (e x 1)2
xn
Z
= (n + 1) [ (n) (n + 1)]
0 (e x 1)2
Although it is beyond the scope of this course to derive (it required a good background in complex analy-
sis), we will find the reflection identity for the zeta function useful. It states that

(2s) = 22s 2s1 sin s(1 2s) (1 2s).

This identity allows us to compute the value of the zeta function for various values that were previously
inaccessible. For example, it is clear that (2n) = 0 for positive integers n. The negative even integers
are thus known as the trivial zeroes of the zeta function, since they trivially exist due to the sine function
apparent in this reflection identity. Additionally, we can pull an expression for the values of the Zeta
function at the negative odd integers, as well as an expression for the derivative of the function at the
negative evens.
2(2n + 1)!
(2n 1) = (1)n+1 (2n + 2)
(2)2n+2
(2n)!
0 (2n) = (1)n (2n + 1)
2(2)2n
From these results, we can assign value of 1/12 to the divergent sum
1
1 + 2 + 3 + 4 + = (1) =
12
It turns out that this can actually be a meaningful result in various physical applications.

13
6 Problems
This selection of problems is intended to both be a source of practice for material from the lecture, as well
as an invitation to results that I havent had time to discuss. Some of these problems will require you to
develop your own results to solve; some of these problems are very hard. Still, if you persevere, you will
undoubtedly gain a much greater understanding of the material and be able to apply it in a wider variety
of situations.

1. Evaluate the following integrals. Make sure that all your answers are simplified as much as possi-
ble by applying the Gamma reflection identity if necessary and by reducing the argument of any
Gamma functions to something less than one.
Z Z Z
2 x x ln x
(a) x ln x e d x (d) x cos 2x sin 3x e 5x d x (g) dx
3
0 0 0 x +1
ln x 3
Z 1 Z1
Z
sin x p
(b) p dx
(e) d x (h) x 2 x ln (1 x)2 d x
0 x 1
0 x 0
Z 2
Z x 1 2
cos x (f) 4x dx

(c) p dx 0 2
cosh 2x
1 e 2e 2x
cos x + 1
0 x cos x

X 1 1
2. It is fairly simple to convince yourself that = e 1 + e 1 = cosh 1. With the logic behind
(2k)! 2 k=0
this sum in mind, see if you can figure out how to sum the following series.

X 1
X 1
X 1
(a) (c) (e)
k=0
(2k + 1)! k=0
(3k + 1)! k=0
(4k)!
X 1 X 1 X 1
(b) (d) (f)
k=0
(3k)! k=0
(3k + 2)! k=0
(4k + 1)!

X
From this you should be able to figure out how you would represent the series aq k+ p x q k+ p in
k=0

X
terms of the function f (x) = an x n .
k=0

d 0 (z)
3. The Digamma function (z) is defined by (z) = ln (z) = . Write a power series expan-
Z dz (z)
p
3
sion for (1 + z) and evaluate the integral ln x e 2x d x in terms of the Gamma and Digamma
0
functions.

4. It turns out that the digamma function can be evaluated at all positive rational arguments. Deter-
mine (1/3).

5. For each of the following integrals, determine if the integral converges and if so, evaluate it.
Z dx Z /2 Z /2  1

(a) 2 2
p (b) sec x ln sin x d x (c) csc x dx
0 x(e x 1) 0 0 x2

14
 /2 
e x
Z p
1
Z  Z
2
(d) 4
csc x dx (g) ln x e x dx (j) p dx
0 x4 0 0 x x
/2  sin 8x 3 1
Z Z
1 1
Z 
(e) csc3 x d x (h) 3
dx (k) x ln2 x ln2 (1 x 2 ) d x
3 2x x
0 x 0 0
1 ln (1 x) 3 x 3 ln x
Z
cos 8x
Z Z
(f) dx (i) dx (l) dx
0 x 0 x3 0 (x 4 + 1)3

6. What dimension d has the hypersphere with maximum volume? And what is that volume?

X 1
7. Consider f (s; ) = for some complex number that is not an integer. Re-express
k=
(k + ) s
s
X x
the sum f (s; ) as a logarithm of a infinite product, and use your results to evaluate the
s=1 s
following sums.
 1 1

1 1
X  X
k
(a) (d) (1)
k=0
3k + 1 3k + 2 k=0 (4k + 1)3 (4k + 3)3
X  1 1
 X

1 1

(b) (e) +
k=0
6k + 1 6k + 5 k=0 (12k + 1)
2
(12k + 11)2


1 1 1 1
X   X
k
(c) (1) (f) +
k=0
6k + 1 6k + 5 k=0 (6k + 1)4 (6k + 5)4

7 References
If you have found the concepts of series expansions and the Gamma function interesting, there are a
number of books that you can check out to learn more about these fascinating subjects - and to get a
more rigorous background in the material, since given the time constraints I had to skim through parts
of the proofs. First and foremost, this lecture is very strongly based on the material taught by Jonathan
A. Osborne in his Advanced Mathematical Techniques course at TJHSST, the material of which can also
be found in the textbook he wrote for the course, Advanced Mathematical Techniques: for Scientists and
Engineers, Second Edition. Without that course, I could not be teaching this to you today as I would simply
not know these techniques in the first place! For a strong basis in complex analysis, there is no better place
to turn than A. I. Markushevichs work Theory of Functions of a Complex Variable. For a (significantly!)
cheaper, condensed version of Markushevichs text, I suggest Richard A. Silvermans text Introductory
Complex Analysis. And finally, for a much more theoretical take on the subject, I suggest William A.
Veechs A Second Course in Complex Analysis. In other areas, I suggest Julian Havils book Gamma:
Exploring Eulers Constant as a very readable exposition of the many different facets of Eulers . If the
Zeta function and the Riemann hypothesis interest you, John Derbyshires book Prime Obsession is a very
well-written, fascinating discussion of the Riemann hypothesis it reads like a gripping mystery novel but
the mathematics remains present and clearly exposited. Finally, I suggest H. M. Edwards Riemanns Zeta
Function as a good advanced text for learning about the developments in theory caused by Riemanns
landmark 1859 paper, ranging from the basic reflection identities to the prime number theorem, Fourier
analysis, and other topics.

15