A Calculus Exam Misprint Ten Years Later

In Problem 99-005 Michael Renardy (Virginia Tech) and Thomas Hagen (Univer-
sity of Memphis) raised the following question: does the series

n=1
(2 + sin n)
n
3
n
n
converge? This question had been given on a calculus exam by mistake.
Solution by Albert Stadler
1
(Herrliberg, Switzerland).
A heuristic argument runs as follows. Note that
(2 + sin n)
n
n3
n
=
1
n
_
1
1 sin n
3
_
n
.
So the main contributions to the sum

n=1
(2 + sin n)
n
n3
n
stem from those integers n for which sin n is close to 1, which occurs if and only if n is close
to (2q +
1
2
), where q is a nonnegative integer. However, the sequence {(2q +
1
2
)}
q0
is
equidistributed mod 1, because is irrational. So sin n is close to 1 only for a small portion
of ns, too small actually to make the sum diverge. Now lets turn this heuristic argument
into a proof!
Let 0 < <
1
4
and let (x) be an even function that is 10 times continuously dierentiable
on (, ) and satises (x) = 1 for |x| 1, (x) = 0 for |x| 2, and (x) 0 for all
real x. An instance of such a function is
(x) =

1, |x| 1,
21!
(10!)
2
_
2
|x|
(t 1)
10
(2 t)
10
dt, 1 < |x| < 2,
0, |x| 2.
It is readily veried that
(i) (x) = (x) for all x,
1
E-mail: albert.stadler@ubs.com
(ii) (x) 0 for all x,
(iii)
(k)
= 0, |x| < 1 or |x| > 2, k 1,
(iv) lim
x1
(x) =
21!
(10!)
2
_
2
1
(t 1)
10
(2 t)
10
dt = 1,
lim
x2
(x) = 0,
(v)
(k)
(t) =
21!
(10!)
2
d
k1
dt
k1
[(t 1)
10
(2 t)
10
]
=
21!
(10!)
2

k1
j=0
_
k1
j
_
10!
(10j)!
10!
(10k+1+j)!
(t 1)
10j
(t 2)
10k+1+j
, 1 < t < 2,
so lim
t1

(k)
(t) = lim
t2

(k)
(t) = 0, 1 k 10.
Put (x) = (x/) for
1
2
x
1
2
and extend (x) to a 1-periodic function on the whole
real line. Then (x) can be expanded into a Fourier series
(x) =

n=
a
n
e
2inx
,
whose coecients a
n
are given by
a
n
=
_ 1
2

1
2
(x) e
2inx
dx.
The condition on dierentiability implies that if n = 0,
a
n
=
1
(2in)
10
_ 1
2

1
2

(10)
(x) e
2inx
dx,
which follows from the original representation of a
n
by 10-fold partial integration. Clearly,

(10)
(x) =
10

(10)
(x/). So, if n = 0,
|a
n
|
1
(2n)
10
_ 1
2

1
2

(10)
(x)

dx =
1
(2n)
10
_ 1
2

1
2

(10)
(x/)

dx (1)
=
1
(2n)
10

9
_ 1
2

1
2

(10)
(x)

dx =
1
(2n)
10

9
_
2
2

(10)
(x)

dx =
c
1

9
n
10
for some absolute constant c
1
. For n = 0 we have
(2) |a
0
|
_ 1
2

1
2
|(x)|dx =
_ 1
2

1
2
|(x/)|dx =
_ 1
2

1
2
|(x)|dx =
_
2
2
|(x)|dx = c
2
.
Put
=
1 cos
3
=
2 sin
2
(/2)
3
.
Then sin n (1 3, 1] if and only if there is an integer q such that ((4q + 1)/2) = 1.
This is because sin n = cos(n/2) (1 3, 1] implies that there is an integer q such that
|n (4q + 1)/2| < and therefore ((4q + 1)/2) = 1, and vice versa. Given a positive
integer N, put

N
:=
2N

n=N+1
(2 + sin n)
n
n3
n
.
Let
A = {n|N < n 2N and 1 sin n < 3},
A

= {n|N < n 2N and 1 sin 3}.

Then
0 <
N
<
1
N
2N

n=N+1
_
1
1 sin n
3
_
n
(3)
=
1
N

nA
_
1
1 sin n
3
_
n
+
1
N

nA

_
1
1 sin n
3
_
n
<
1
N

nA
1 + (1 )
N

1
N

nA
1 + exp(N).
Let x = min
nZ
|x n| denote the distance from x to the nearest integer. Let a, b be
integers, a < b. If x is not an integer, then

n=a
e
2inx

e
2iax
ba

n=0
e
2inx

e
2i(ba+1)x
1
e
2ix
1

2
|e
ix
e
ix
|
=
1
| sinx|

1
2x
.
Let
B =
_
n

N < n 2N and

n

2
(4q + 1)

< for some q Z

_
,
C =
_
q

N + 1
1
4


2
(4q + 1) 2N +
1
4
_
=
_
q

N + 1
1
4


2
2q 2N +
1
4


2
_
.
Then
1
N

nA
1 =
1
N

nB
1
1
N

qC

2
(4q + 1)
_
(4)
=
1
N

k=
a
k
e
i
2
k

qC
e
4i
2
qk

1
N
|a
0
|
_
N
2
+ 1
_
+
1
2N

k=0
|a
k
|
2k
c
3
+
c
1
2N

k=0
1

9
k
10
2k
,
where we have used the estimates (1) and (2).
Now we refer to a theorem of Hata [1] on rational approximations to that implies
2n n
8
for all suciently large n. Note. In fact, Hata proves that
|q p|
1
q
7.0161
for integers p, q with q > q
0
, so 2n n
8
holds with ample room to spare. This allows
us to bound the sum in (4) and obtain
(5)
1
N

A
1 c
3
+
c
1
2N

k=0
1

9
k
2
c
3
+
c
4
N
9
.
To sum up, by (3) and (5),
0 <
N
< c
3
+
c
4
N
9
+ exp(N) c
3
+
c
4
N
9
+ exp(c
5

2
N).
Now we choose = N
1/10
and obtain 0 <
N
< c
6
N
1/10
. This concludes the proof because

n=2
K
+1
(2 + sin n)
n
n3
n
=

j=K

2
j c
6

j=K
2
j/10
< .
We thank the anonymous referee for some useful comments that made the text more readable.
REFERENCE
[1] M. Hata, Rational approximations to and some other numbers, Acta Arith., 63 (1993),
pp. 335349.