Erum Dost

Assignment 6

1. If f :[a,b] →  is a bounded function such that f (x) = 0 except for

x ∈{c1 ,c2 ,…,cn } ⊂ [a,b], show that f is Riemann integrable on [a,b] and determine the
Riemann integral value.

Proof: Suppose f (x) = 0 except for x ∈[a,b] \ {c1 ,…,cn } and let f (ci ) = ki for
i = 1,…,n, where ci ∈[a,b] for each i such that c1 < … < cn . Let ε > 0.
Suppose that c1 ≠ a and cn ≠ b.
Fix
{
K = max ki :i = 1,…,n . }
Let
δ = min {ε 2nK, c1 − a 3, c2 − c1 3,…, cn − cn−1 3, b − cn 3} .
Define a partition
P = {a, c1 − δ , c1 + δ ,…,cn − δ , cn + δ , b} .
Then
n
U(P, f ) − L(P, f ) = ∑ ki ( 2δ ) ≤ 2nKδ < ε ,
i=1
therefore f is Riemann integrable on [a,b]. If c1 = a, then remove the point c1 − δ from
the partition. Similarly, if cn = b, then remove the point cn + δ from the partition. Then
n
U(P, f ) − L(P, f ) ≤ ∑ ki ( 2δ ) ≤ 2nKδ < ε .
i=1
Therefore f is Riemann integrable in all cases. Since U(P, f ) ≥ 0 and L(P, f ) ≤ 0 such
that for ε > 0, U(P, f ) − L(P, f ) < ε therefore U(P, f ) < ε . Thus for any ε > 0, we
define a partition for which 0 ≤ U(P, f ) < ε . Then we must have that
b
∫ a
f dx = 0.
Therefore
b
∫a
f dx = 0. ■
 Erum Dost

2. Suppose that f is any bounded function on [a,b] and that for any number c ∈( a,b ) ,
the restriction of f to [c,b] is Riemann integrable. Show that f is integrable on [a,b]
and that
b b

∫ f = lim ∫ f .
a
c→a +
c

Proof: Let ε > 0. Define M = sup f and m = inf f for x ∈[a,b]. Choose δ > 0 such that
(M − m)δ ≤ ε 2. Since f is integrable on [ a + δ , b ], define a partition P1 of [ a + δ , b ]
such that Uδ ( f , P1 ) − Lδ ( f , P1 ) < ε 2. Consider the partition P2 of [ a,b ] defined by
adding the interval [ a, a + δ ] to P1 .
Then U( f , P2 ) − L( f , P2 ) ≤ Uδ ( f , P1 ) − Lδ ( f , P1 ) + (M − m)δ < ε 2 + ε 2 = ε .
Hence f is integrable on [a,b]. Define K = max { M , m } . Then for any ε > 0, there
exists δ > 0 such that Kδ < ε . Then for all c such that a < c < a + δ ,
b b c b b

∫ f dx − ∫ f dx = ∫ f dx ≤ Kδ < ε . Therefore ∫ f dx = lim ∫ f dx. ■

a c a a
c→a +
c

⎧ g1 , a < x < d,
3. Suppose that −∞ < a < d < b < ∞ and g(x) = ⎨ (g1 , g2 = constants), and
⎩ g2 , d < x < b,
let g(a), g(b), and g(d) be arbitrary. Suppose that f is defined on [a,b], continuous
b
from the right at a and from the left at b, and continuous at d. Show that ∫a
f (x)dg(x)
exists, and find its value.

Proof: Define a partition P = { x0 , x1 ,…, xn } of [a,b] such that P < min ( d − a, b − d ) .

Let c1 ,c2 ,…,cn be points in the Riemann-Stieltjes sum of f with respect to g over P.
Then xk−1 ≤ ck ≤ xk , 1 ≤ k ≤ n. Consider the following two cases.
i) If d ∉[ xk−1 , xk ], then g(xk ) − g(xk−1 ) = 0. Thus, if x j−1 < d < x j for some j in
{1,2,…,n}, then σ = f (c1 )[ g1 − g(a)] + f (c j )(g1 − g2 ) + f (cn )[ g(b) − g2 ].
ii) If c = x j for some j in {1,2,…,n} , then σ = f (c1 )[ g1 − g(a)] + f (c j−1 )[ g(d) − g1 ]
+ f (c j )[ g2 − g(d)] + f (cn )[ g(b) − g2 ].
Since f defined on [a,b] is continuous from the right at a and from the left at b, and
continuous at d, therefore in both cases σ → f (a)[ g1 − g(a)] + f (d)(g2 − g1 ) + f (b)[ g(b) − g2 ]
b
= ∫ f (x)dg(x) as P → 0. ■
a
 Erum Dost

4.
b
a) Give an example where ∫a
f (x)dg(x) exists even though f is unbounded on
[a,b]. (Thus, the analog of Theorem 3.1.2 does not hold for the Riemann-Stieltjes
integral.)
b) State and prove an analog of Theorem 3.1.2 for the case where g is increasing.

a) Example: Let f be an arbitrary function that is unbounded on [a,b]. Suppose

b
g(x) = 1. Then dg(x) = 0. Therefore ∫a
f (x)dg(x) = 0. ■

b) Analog of Theorem 3.1.2: If g is increasing and f is unbounded on [a,b], then

b
∫a
f (x)dg(x) does not exist.

Proof: If g is increasing and f is unbounded on [a,b], where P is any partition of [a,b],

and M > 0, then there are Riemann-Stieltjes sums σ and σ ′ of f with respect to g
over P such that
σ −σ ′ ≥ M.
Let
n
σ = ∑ f (c j ) ⎡⎣ g(x j ) − g(x j−1 ) ⎤⎦
j=1

be a Riemann-Stieltjes sum of f with respect to g over the partition P of [a,b]. There

must be an integer i in {1,2,…,n} such that
M
f (c) − f (ci ) ≥
g(xi ) − g(xi−1 )
for some c in [ g(xi−1 ), g(xi )], because if it were not so, we would have
M
f (x) − f (c j ) < , x j−1 ≤ x ≤ x j , 1 ≤ j ≤ n.
g(x j ) − g(x j−1 )
Then
f (x) = f (c j ) + f (x) − f (c j ) ≤ f (c j ) + f (x) − f (c j )
M
≤ f (c j ) + , x j−1 ≤ x ≤ x j , 1 ≤ j ≤ n,
g(x j ) − g(x j−1 )
which implies that
M
f (x) ≤ max f (c j ) + , a ≤ x ≤ b,
1≤ j≤n g(x j ) − g(x j−1 )
contradicting the assumption that f is unbounded on [ a,b ].
Now suppose that c satisfies
M
f (c) − f (ci ) ≥ .
g(xi ) − g(xi−1 )
 Erum Dost

Consider the Riemann-Stieltjes sum,

n
σ ′ = ∑ f (c′j ) ⎡⎣ g(x j ) − g(x j−1 ) ⎤⎦
j=1

over the same partition P, where

⎧c j , j ≠ i,
c′j = ⎨
⎩c, j = i.
Since
σ − σ ′ = f (c) − f (ci ) [ g(xi ) − g(xi−1 )],
therefore
M
f (c) − f (ci ) ≥ ⇒ σ −σ ′ ≥ M. ■
g(xi ) − g(xi−1 )

5. Prove that the function f : [ 0,1] →  defined by f (0) = 1, f (x) = 0 if x is irrational,

and f (m / n) = 1 / n, if m,n ∈ are relatively prime is Riemann-integrable.

Proof: Let ε > 0. We will construct a partition P of [0,1] for which

U( f , P) − L( f , P) < ε . Since the irrationals are dense in  and f (x) = 0 for every
irrational, it follows that L( f , P) = 0 for every partition P. Consider the set of points
D = { x : f (x) ≥ ε 2} . If x ∈D then x must be a rational number of the form x = m n
with n ≤ 2 ε . The number of such points in the interval [0,1] is finite. Let {c1 ,c2 ,…,cN }
be a finite set of points in D ∩[0,1]. Consider a partition P constructed from disjoint
intervals around each ck with length less than ε (2N ). Since f (x) < 1, these intervals
contribute to the upper sum at most N ⋅(ε (2N ))⋅1 = ε 2. For all the other intervals we
have that f (x) ≤ ε 2. It follows that U( f , P) = ε 2 + ε 2 = ε . Since L( f , P) = 0,
therefore U( f , P) − L( f , P) < ε . Hence f (x) is Riemann-integrable. Since
1 1
0 = L( f , P) ≤ ∫ f (x) ≤ U( f , P) < ε for all ε , it follows that ∫ f (x) = 0. ■
0 0

6. Prove that Dirichlet’s discontinuous function is not Riemann-integrable.

⎧1 for x rational,
Proof: Consider Dirichlet’s discontinuous function f (x) = ⎨
⎩0 for x irrational.
Suppose P is a partition of [0,1]. Then the density of the rationals in  implies that
every subinterval of P will contain a point where f (x) = 1. It follows that U( f , P) = 1.
Since the irrationals are also dense in , then L( f , P) = 0. Since this is true for every
partition P, therefore the upper integral U( f ) = 1 and the lower integral L( f ) = 0. Since
U( f ) ≠ L( f ), therefore Dirichlet’s discontinuous function is not Riemann-integrable. ■
 Erum Dost

7. State why each integral is improper, and determine whether or not it is convergent. If
the improper integral is convergent, find its value.

1
i) ∫0
ln(x)dx is an improper integral since the function ln(x) has a vertical
asymptote at x = 0 where lim+ ln(x) = −∞.
x→0
1 1
∫ ln(x)dx = lim ∫ ln(x)dx
0 t→0 + t

= lim+ [ x ln(x) − x ]t
1

t→0

= lim+ [ (1ln(1) − 1) − (t ln(t) − t)]

t→0

= lim+ ( t − 1− t ln(t))
t→0

= lim+ ( t − 1) − lim+ ( t ln(t))

t→0 t→0

⎛ ln(t) ⎞
= −1− lim+ ⎜
t→0 ⎝ 1 t ⎟ ⎠
⎛ 1t ⎞
= −1− lim+ ⎜ 2⎟
t→0 ⎝ −1 / t ⎠

= −1+ lim+ ( t )
t→0

= −1

1
Therefore the improper integral ∫
0
ln(x)dx converges to −1. ■

2 1 1
ii) ∫1 x ln(x)
dx is an improper integral since the function
x ln(x)
is not

1
continuous at x = 0 and lim+ = ∞.
x→1 x ln(x)

2 1 2 1
∫1 x ln(x) dx = lim
t→1+ ∫t
x ln(x)
dx

Substitution: Let u = ln(x) ⇒ du = (1 x)dx. Then x = t ⇒ u = ln(t).

1 ln(2) 1
lim+ = lim+ ∫ du
x→1 x ln(x) t→1 ln(t ) u

= lim+ [ ln(u)]ln(t )
ln(2)

t→1

= lim+ ( ln(ln(2)) − ln(ln(c)))

t→1

=∞
2 1
Therefore the improper integral ∫
1 x ln(x)
dx diverges. ■
 Erum Dost

2x
iii) ∫1 x −1
dx is an improper integral since the function has a vertical
asymptote at x = 1.
2 x 2 x
∫1 x − 1 dx = lim
t→1+ ∫t
x −1
dx
1 1
Substitution: Let u = x ⇒ du = 1, dv = ( x − 1) 2 ⇒ v = 2 ( x − 1) 2 .
−

x
Then ∫ dx = 2x x − 1 − 2 ∫ x − 1 dx
x −1
2 3
= 2x x − 1 − 2 ⋅ ( x − 1) 2
3
4 3
= 2x x − 1 − ( x − 1) 2 .
3
3 x=2
x ⎡ 4 ⎤
dx = lim+ ⎢ 2x x − 1 − ( x − 1) 2 ⎥
2
Thus lim+ ∫
t→1 t x −1 t→1 ⎣ 3 ⎦ x=t
⎡8 ⎛ 4 3
⎞⎤
= lim+ ⎢ − ⎜ 2t t − 1 − ( t − 1) 2 ⎟ ⎥
t→1
⎣3 ⎝ 3 ⎠⎦
⎡8 4 3
⎤
= lim+ ⎢ − 2t t − 1 + ( t − 1) 2 ⎥
t→1 ⎣ 3 3 ⎦
8
=
3
2 x 8
Therefore the improper integral ∫ dx converges to . ■
1 x −1 3
 Erum Dost

Assignment 5

1. Suppose f : D → , and c ∈D and c is a cluster point of D. Then f is differentiable at

c if and only if for every sequence { xn } of points D − {c} converging to c, the sequence
⎧ f (xn ) − f (c) ⎫
⎨ ⎬ → f '(c) as x → c.
⎩ x−c ⎭

f (x) − f (c)
Proof: Let f '(c) = lim = L.
x→cx−c
(⇒) Suppose that lim f (x) = L. Let ε > 0. Then ∃ δ > 0 such that f (x) − L < ε if
x→c

0 < x − c < δ . Let { xn } be a sequence in D − {c} which converges to c for all n. Then
∃ N ∈ such that 0 < xn − c < δ for all n ≥ N. It follows that f (xn ) − L < ε whenever
n ≥ N. Thus lim f (xn ) = L.
(⇐) Suppose that lim f (x) ≠ L. Then ∃ ε > 0 such that for each δ > 0 there is an x ∈D
x→c

with 0 < x − c < δ but f (x) − L ≥ ε . In particular, for every n ∈ there is an xn ∈D

such that xn − c < 1 / n and f (xn ) − L ≥ ε . Then xn → c but { f (xn )} does not converge
to L, a contradiction. ■

2. Prove Darboux’s Theorem: If f is differentiable on [a,b] and if k is a number between

f '(a) and f '(b), then there is at least one point c ∈(a,b) such that f '(c) = k.

Proof: Let I = [a,b] and suppose that f '(a) < k < f '(b). Define g on I by
g(x) := kx − f (x) for x ∈I. Since g is continuous, it attains its maximum value on I.
Since g'(a) = k − f '(a) > 0 , there exists a δ > 0 such that g(x) > g(a) for x ∈I where
a < x < a + δ . Hence, the maximum of g does not occur at x = a. Similarly, since
g'(b) = k − f '(b) < 0, there exists a δ > 0 such that g(x) > g(b) for x ∈I where
b − δ < x < b. Hence, the maximum does not occur at x = b. Therefore, g attains its
maximum at some c ∈(a,b). Then 0 = g'(c) = k − f '(c). Hence, f '(c) = k. ■

3. Prove: If f is defined on a neighborhood of x0 , then f is differentiable at x0 if and only

f (x) − f (x0 )
if the discontinuity of h(x) = at x0 is removable.
x − x0

Proof: We note that h(x) has a removable discontinuity at x0 if lim h(x) exists. Then the
x→x0

⎧⎪h(x) if x ∈Dh and x ≠ x0 ,

function g(x) = ⎨ lim h(x) if x = x0 ,
⎪⎩ x→x0
 Erum Dost

is continuous at x0 . If the discontinuity of h(x) at x0 is removable then lim h(x) exists

x→x0

if and only if f '(x0 ) exists. ■

4. Suppose that f '(0) exists and f (x + y) = f (x) f (y) for all x and y. Prove that f ' exists
for all x.

Proof: Since f (x) = f (x) f (0), either f = 0 or f (0) = 1. If f = 0, then f ' exists for all x.
f (x0 + h) − f (x0 ) f (x0 ) f (h) − f (x0 )
If f (0) = 1 and x0 is arbitrary, then lim = lim =
h→0 h h→0 h
f (h) − 1
f (x0 )lim = f (x0 ) f '(0). Therefore f ' exists for all x. ■
h→0 h

5. Prove by induction: If n ≥ 1 and f (n) (x0 ) and g (n) (x0 ) exist, then so does ( fg)(n) (x0 ),
n
⎛ n⎞
and ( fg) (x0 ) = ∑ ⎜ ⎟ f (m ) (x0 )g (n−m ) (x0 ). Hint: See Exercise 1.2.19. This is Leibniz’s
(n)

⎝ m⎠
m=0

rule for differentiating a product.

Proof: By Theorem 2.3.4, P1 is true. Suppose that n ≥ 1 and Pn is true. If f (n+1) (x0 ) and
g (n+1) (x0 ) exist, then f (n) and g (n) exist on some neighborhood N of x0 . Then Pn implies
⎛ n⎞
( )
n
d (k )
that ( fg)(n) (x) = ∑ ⎜ ⎟ f (k ) (x)g (n−m ) (x) if x ∈N. [1] Then, f (x)g (n−k ) (x) =
m=0 ⎝ m⎠ dx
f (k+1) (x)g (n−k ) (x) + f (k ) (x)g (n−k+1) (x) for x ∈N. Substituting this into equation [1] yields
n
⎛ n⎞
( fg)(n+1) (x) = ∑ ⎜ ⎟ ⎡⎣ f (m+1) (x)g (n−m ) (x) + f (m ) (x)g (n−m+1) (x) ⎤⎦
m=0 ⎝ m ⎠
n+1
⎛ n ⎞ (m ) n
⎛ n ⎞ (m )
= ∑⎜ ⎟ f (x)g (n−m+1)
(x) + ∑ ⎜⎝ m ⎟⎠ f (x)g
(n−m )+1
(x)
m=1 ⎝ m − 1⎠ m=0

⎡⎛ n ⎞ ⎛ n ⎞ ⎤ (m )
n
= f (x)g (x) + ∑ ⎢⎜
(n+1)
⎟ + ⎜ ⎟ ⎥ f (x)g (n−m+1) (x) + f (n+1) (x)g(x)
m=1 ⎣ ⎝ m − 1⎠ ⎝ k⎠ ⎦
n+1
⎛ n + 1⎞ (m )
= ∑⎜ ⎟ f (x)g (n−m+1) (x), x ∈N by Exercise 1.2.19.
m=0 ⎝ m ⎠
n+1
⎛ n + 1⎞ (m )
Let x = x0 . Then ( fg) (n+1)
(x0 ) = ∑ ⎜ ⎟ f (x0 )g (n−m+1) (x0 ) yields Pn+1 . Therefore
m=0 ⎝ m ⎠
n
⎛ n⎞
( fg)(n) (x0 ) exists such that ( fg)(n) (x0 ) = ∑ ⎜ ⎟ f (m ) (x0 )g (n−m ) (x0 ). ■
m=0 ⎝ m ⎠
 Erum Dost

6. Prove or give a counterexample: If f is differentiable on a neighborhood of x0 , then f

satisfies a Lipschitz condition on some neighborhood of x0 .

⎧⎪ x 2 sin (1 / x 2 ) if x ≠ 0,
Counterexample: Let f (x) = ⎨
⎪⎩0 if x = 0.
⎛ 1⎞ 2 ⎛ 1⎞ x 2 sin(1 / x 2 )
If x ≠ 0, then f '(x) = 2x sin ⎜ 2 ⎟ − cos ⎜ 2 ⎟ . Since f '(0) = lim = 0, f is
⎝x ⎠ x ⎝x ⎠ x→0 x
differentiable on (−∞,∞). However, f does not satisfy a Lipschitz condition at x0 = 0.
Consider the set of points { xn }n=1 and { yn }n=1 both with zero as a limit point such that
∞ ∞

f (xn ) − f (yn )
= ∞. Let xn = ( 2nπ + π / 2 ) and yn = ( 2nπ )
−1/2 −1/2
lim for n = 1,2,…
n→∞ xn − y n
Suppose f (x) satisfies a Lipschitz condition. Then there exists M such that
f (xn ) − f (yn )
M≥
xn − yn

=
( 2nπ + π / 2 )
−1

( 2nπ )−1/2 − ( 2nπ + π / 2 )−1/2

= 4n ( (2nπ )−1/2 + (2nπ + π / 2)−1/2 ) → ∞ as n → ∞.
Thus no such M can exist. Hence f (x) does not satisfy a Lipschitz condition at x0 = 0. ■

e−1/x
2

7. Find lim n = 0 , n ∈.

x→0 x

Proof by induction: First suppose that n = 0. Then we can prove directly that
lim e−1/x = 0. Let ε > 0 be arbitrary and choose δ = 1 / ln(1 / ε ). Then 0 < x < δ
2

x→0

e−1/x
( ) ( )
2

−1/x 2
( )
< ε . For n < 0, lim − n = lim x n e−1/x = lim x n ⋅ lim e−1/x = 0 ⋅ 0 = 0.
2 2
implies that e
x→0 x x→0 x→0 x→0

Therefore Pk follows immediately if k ≤ 0. Suppose that n ≥ 0 and Pk is true for k ≤ n.

e−1/x x − n−1 −(n + 1)x − n−2 n + 1 e−1/x
2 2

Then lim n+1 = lim 1/x 2 = lim 2 = lim = 0. Hence, Pn+1 is true. ■
x→0 x x→0 e x→0 −2e1/x / x 3 2 x→0 x n−1
 Erum Dost

8. Suppose f : [ a,b ] → , c is an interior point of [ a,b ], and there is an n ∈ for which

f , and all of its n th derivatives are continuous on a neighborhood of c with
f '(c) = f ''(c) =  = f (n−1) (c) = 0 but f (n) (c) ≠ 0.
(i) If n is even and f (n) (c) > 0, then f (c) is a relative minimum.
(ii) If n is even and f (n) (c) < 0, then f (c) is a relative maximum.
(iii) If n is odd, then f (c) is neither a relative maximum or relative minimum.

Proof: Applying Taylor’s Theorem at c, we find that for x ∈[ a,b ] we have

f (n) (k)
f (x) = Pn−1 (x) + Rn−1 (x) = f (c) + (x − c)n ,
n!
where k is some point between c and x. Since f (n)
is continuous, if f (n) (c) ≠ 0, then
there exists an interval I containing c such that f (n) (x) will have the same sign as f (n) (c)
for x ∈I. If x ∈I, then k ∈I thus f (n) (k) and f (n) (c) have the same sign.
(i) If n is even and f (n) (c) > 0, then for x ∈I, f (n) (k) > 0 and (x − c)n ≥ 0. Then
Rn−1 (x) ≥ 0, hence f (x) ≥ f (c) for x ∈I. Therefore, f has a relative
minimum at c.
(ii) If n is even and f (n) (c) < 0, then it follows that for x ∈I, Rn−1 (x) ≤ 0. Hence
f (x) ≤ f (c) for x ∈I. Therefore, f has a relative maximum at c.
(iii) If n is odd, then (x − c)n is positive if x > c and negative if x < c. If x ∈I,
then Rn−1 (x) will have opposite signs to the left and right of c. Therefore, f
has neither a relative minimum nor a relative maximum at c. ■
 Erum Dost

Assignment 8

1. Prove the Weierstrass M-test: For each n ∈, let fn be a function defined on a set
∞
S ⊆ , and let M n > 0 be a real number satisfying fn (x) ≤ M n for all x ∈S. If ∑M n
n=1
∞
converges, then ∑f n converges uniformly on S.
n=1

∞
Proof: Let ε > 0 be arbitrary. Since ∑M n converges, there exists an N such that
n=1
n ≥ m ≥ N implies
M m+1 + M m+2 ++ M n < ε .
Since
fm+1 (x) + fm+2 (x) ++ fn (x) ≤ M m+1 + M m+2 ++ M n ,
∞
then by the Cauchy criterion for uniform convergence of series, ∑f n converges
n=1
uniformly. ■

2. Prove by induction that the function f (x) = e−1 x for x ≠ 0 and f (0) = 0 has
2

derivatives of all orders at every point in  and that all of these derivatives vanish at
x = 0. Hence this function is not given by its Taylor expansion about x = 0.

Proof: For any n ∈, it is known that

lim y n e− y = 0
y→∞

since the exponential function grows faster than any power. Substituting y = 1 x 2 , we
obtain
e−1 x
2

lim =0
x→0 p(x)

for any polynomial p. By the definition of the derivative,

e−1 x
2
f (x) − f (0)
f '(0) = lim = lim = 0.
x→0 x x→0 x
We will prove by induction that
p (x)
f (n) (x) = n e−1 x , x ≠ 0,
2

qn (x)
where pn and qn are polynomials. It is clear that (2) is true for n = 0, with
p0 (x) = q0 (x) = 1. Suppose that (2) holds for some n ≥ 0, and let x ≠ 0.
 Erum Dost

Then
f (n+1) (x) = ⎡⎣ f (n) (x) ⎤⎦′

⎡ p (x)e−1 x ⎤′
2

=⎢ n ⎥
⎣ qn (x) ⎦
pn (x) ⎛ 2 −1 x 2 ⎞ ⎛ qn (x) pn′ (x) − qn′ (x)pn (x) ⎞ −1 x 2
= ⎜ e ⎟⎠ + ⎜ ⎟⎠ e
qn (x) ⎝ x 3 ⎝ qn2 (x)
⎛ 2 p (x) qn (x) pn′ (x) − qn′ (x)pn (x) ⎞ −1 x 2
=⎜ 3 n + ⎟⎠ e
⎝ x qn (x) qn2 (x)
⎛ 2 p (x)qn (x) + x 3qn (x) pn′ (x) − x 3qn′ (x)pn (x) ⎞ −1 x 2
=⎜ n ⎟⎠ e
⎝ x 3qn2 (x)
pn+1 (x) −1 x 2
= e
qn+1 (x)
for some polynomials pn+1 and qn+1 . Suppose that f (n) (0) = 0 for some n. Then by
induction,
f (n) (x) − f (n) (0)
f (n+1) (0) = lim
x→0 x
pn (x) −1 x 2
e −0
qn (x)
= lim
x→0 x
e−1 x
2

= pn (0)lim
x→0 x ⋅ q (x)
n

=0
by (1). ■

∞
3. Prove that if ∑a x n
n
converges for x = x0 and diverges for x = x1 , then
n=0
∞
(a) ∑a x n
n
converges absolutely for x < x0 , and
n=0
∞
(b) ∑a x n
n
diverges for x > x1 .
n=0

Proof:
∞ ∞
(a) Suppose x0 ≠ 0, ∑a x n
n 0 converges and x < x0 . Since ∑a x n
n 0 converges, there
n=0 n=0

exists M ≥ 0 such that an x0n ≤ M for all n ∈. Therefore,

 Erum Dost

n n
x x
an x = a x
n
≤M n
n 0 for all n ∈.
x0 x0
n
∞
x x
Since
x0
< 1, ∑M x is a convergent geometric series. By the Comparison Test, the
n=0 0
∞
series ∑a x n
n
converges absolutely.
n=0
∞ ∞
(b) Let x ∈ and x > x1 . Suppose ∑ an x n converges. Then by (a), the series ∑a x n
n 1
n=0 n=0
∞
converges absolutely, which is a contradiction. Thus ∑a x n
n
diverges for x > x1 . ■
n=0

4. How could one use the Ratio Test to establish criteria for the radius of convergence?

∞
an+1
Proof: Consider the power series ∑ a (x − x ) . Suppose that the limit lim
n
an
0
n
exists or
n→∞
n=0

diverges to infinity. By Theorem 4.5.3 of the text, the radius of convergence R of the
1 a a
power series is given by = lim n+1 . Now suppose that lim n+1 = L exists in the
R n→∞ an n→∞ a
n
∞
extended reals. We apply the Ratio Test to the series ∑ a (x − x )
n 0
n
to obtain
n=0

an+1 (x − x0 )n+1 a
lim = lim n+1 ⋅ x − x0 = L x − x0 .
n→∞ an (x − x0 ) n n→∞ an
By the Ratio Test, the series converges absolutely if L x − x0 < 1, or if x − x0 < 1 L; and
diverges if L x − x0 > 1, or if x − x0 > 1 L . It follows that R = 1 L by the definition of R
in Theorem 4.5.3. ■

5. Suppose that ∑a n diverges and that {an } is bounded. Prove the radius of
convergence of the power series ∑a x n
n
is equal to 1.

Proof: Since {an } is bounded, there exists M ≥ 0 such that an ≤ M . It follows that for
any x ∈, an x n = an x ≤ M x . If x < 1, then ∑M x
n n n
is a convergent geometric
series. By the Comparison Test, ∑a x n
n
converges where the definition of absolute
convergence implies that ∑ an x n converges for x < 1. Therefore, the radius of
convergence R of the power series satisfies R ≥ 1. However, the series diverges when
x = 1, implying that R ≤ 1. It follows that the radius of convergence of the power series
∑a xn
n
is equal to 1. ■
 Erum Dost

6. Write the Taylor series for 1− x centered at 0. Prove that the series converges to
1− x for x ∈(−1,0].

Proof: Let f (x) = 1− x and observe that f '(x) = − (1− x )

−1 2
2 and for n ≥ 2,

f (n) (x) = ( −1)

( 2n − 3)!! 1− x − n+1 2
2n+1
n ( )
2
The Taylor series for f centered at 0 is then
x ∞ 2n−1 ( 2n − 3)!! n
1− + ∑ ( −1) x
2 n=2 n!2 n
Note that for x ∈(−1,0] and n ≥ 2,
( 2n − 1)!! 1− x − n−1 2 ≤ ( 2n − 1)!!
f (n+1) (x) = ( )
2 n+1 2 n+1
Let sn (x) be the nth partial sum of the Taylor series for f. Then for x ∈(−1,0] and n ≥ 2,

sn (x) − f (x) ≤
( 2n − 1)!!x n+1
( n + 1)!2 n+1
≤
( 2n − 1)!!
( n + 1)!2 2n+1
≤
( 2n )!!
n!2 2n+1
1
= n+1
2
Taking limits as n → ∞ gives lim sn (x) = f (x) for x ∈(−1,0]. Therefore the Taylor series
n→∞

for 1− x centered at 0 converges to 1− x for x ∈(−1,0]. ■

 Erum Dost

Assignment 7

1. Prove that absolute convergence implies convergence.

Proof: Let X := ( xn ) be a sequence in . Then ∑x n is absolutely convergent if the

series ∑ xn is convergent in . Since ∑ xn is convergent, the Cauchy Criterion
implies that, given any ε > 0 there exists N(ε ) ∈ such that if m ≥ n ≥ N(ε ), then
xn+1 + xn+2 ++ xm < ε .
By the Triangle Inequality,
xn+1 + xn+2 ++ xm ≤ xn+1 + xn+2 ++ xm < ε .
Since ε > 0 is arbitrary, the Cauchy Criterion implies that ∑x n converges. ■

2. Use the comparison test to prove the limit comparison test (stated below).
⎧a ⎫
Suppose {an } and {bn } are positive real sequences such that ⎨ n ⎬ converges to L < ∞.
⎩ bn ⎭
i) If L ≠ 0, then ∑ a and ∑ b either both diverge or both converge together.
n n

ii) If L = 0 and ∑ b converges, then ∑ a converges.

n n

Proof:
i) (→ ) Suppose that an bn converges to L > 0. Let ε = L 2. By the definition
an L
of convergence, there exists N such that if n ≥ N, −L < or,
bn 2
L an 3L
equivalently, < <
2 bn 2
for n ≥ N. Assume the series ∑a n converges.

2
Since bn <
L
an for n ≥ N, the series ∑b n converges by the comparison test.
2
Now assume that the series ∑a n diverges. Since bn >
3L
an for n ≥ N, the

series ∑b n diverges by the comparison test.

(←) The backward argument implies that if the series ∑b n converges, so
does the series ∑a , n and if the series ∑b n diverges, so does the series
∑a .n Instead, note that bn an converges to 1 L , which is also strictly
positive. Therefore an and bn play symmetric roles. Hence if L ≠ 0, then
∑a n and ∑b n either both diverge or both converge together.
 Erum Dost

ii) If L = 0, then for every ε > 0 there exists N such that if n ≥ N, 0 < an < ε bn .
Thus, if ε = 1 then 0 < an < bn . Therefore if ∑b n converges, then ∑a n

converges. ■

∞
1 π
3. Assume the series ∑n 2
converges to
6
. Use the integral test to “confirm” this
n=1
resulting using S10 and S20 .

Proof: Applying the integral test, we obtain

∞ ∞ ∞ ∞
∞
1 π2 1
∫n+1 ∑
f ≤
n=1
an − Sn ≤ ∫
n
f ⇒ ∫
n+1
x 2
dx ≤
6
− Sn ≤ ∫
n
x2
dx
∞ ∞
1 π2 1
⇒− ≤ − Sn ≤ −
x n+1 6 xn
1 π2 1
⇒ ≤ − Sn ≤
n +1 6 n
1 π 2
1
Evaluating at n = 10 we obtain ≤ − S10 ≤ .
11 6 10
1 π 2
1
Evaluating at n = 20 we obtain ≤ − S20 ≤ .
21 6 20
1
Observe that both sides of the inequalities approach zero since lim = 0 and
n→∞ n + 1

1 ⎛π2 ⎞ ∞
1 π
lim = 0 . By the Squeeze Theorem, lim ⎜ − Sn ⎟ = 0. Therefore ∑ 2 = . ■
n→∞ ⎝ 6 ⎠ n=1 n 6
n→∞ n

1 1 1 1
4. Show that series + + + + is convergent, but that both the Ratio and the
12 2 3 32 4 3
Root Tests fail to apply.

∞
1
1 1 1
Proof: Note that ∑n p
p
1 2
+ p + p + converges for p > 1. The series
=
3
n=1
1 1 1 1 1 1 1 1
2
+ 3 + 2 + 3 + is bounded above by 2 + 2 + 2 + 2 +, which is a convergent
1 2 3 4 1 2 3 4
1 1 1 1
p − series. By the Comparison Test, the series 2 + 3 + 2 + 3 + converges. By the
1 2 3 4
a 1 (n − 1) 2
a
Ratio Test, lim sup n+1 = lim 3
= ∞, but lim inf n+1 = 0. Therefore the Ratio
an n→∞ 1n an
1n
1n ⎛ 1⎞
Test does not apply. By the Root Test, lim sup an = lim ⎜ 2 ⎟ = 1. Therefore the Root
n→∞ ⎝ n ⎠

Test does not apply. ■

 Erum Dost

∞
5. Let {an } be a sequence of real numbers. Prove that ∑(a n − an+1 ) converges if and
n=1

only if {an } converges. If the series ∑(a n − an+1 ) converges, what is its sum?

n
Proof: Let sn = ∑ ( ak − ak+1 ). Observe that sn = ( a1 − a2 ) + ( a2 − a3 ) ++ ( an − an+1 ) = a1 − an+1 .
k=1

By the algebraic limit theorem, ( sn ) converges if and only if ( an+1 ) converges. Hence,
∞

∑(a n − an+1 ) converges if and only if ( an ) converges. Let lim an = a. Then for the series
n→∞
n=1

∑(a n − an+1 ) , the partial sums are sn = a1 − an+1 , so ∑(a n − an+1 ) converges to a1 − a. ■

6. Write an infinite series for the repeating decimal for the rational number 5 9 and
prove your series converges to 5 9.

Proof: Note that 5 9 = 0.5555… Written as an infinite series, the repeating decimal for
∞
55 55 55 55
the rational number 5 9 is equivalent to + 2
100 100 100
+ 3
+ = ∑ n
. This series is
n=1 100
∞
of the form ∑ ar n
where a and r are constants. For r ≠ 1, the partial sums are given by
n=0
n
sn = ∑ ar k . Multiplying the partial sum by (1− r), we obtain
k=0
n n n
(1− r)∑ ar k = ∑ ar k − ∑ ar k+1
k=0 k=0 k=0

= a + ar + ar 2 ++ ar n − ( ar + ar 2 ++ ar n + ar n+1 )

= a − ar n+1
n a (1− r n+1 )
∑ ar k
=
1− r
.
k=0
∞
a a
For r < 1, we have lim r n+1 = 0. Hence lim s n =
n→∞ n→∞ 1− r
. Therefore ∑ ar n
=
1− r
if r < 1.
n=0
∞ ∞
55 55 55
Then for the infinite series ∑ 100 n
= ∑
100 n=0 100 n
for the rational number 5 9 we have
n=1
∞
55 1 55 55 100 55 100 5
a=
100
and r =
100
< 1. It follows that ∑ 100 n
= = = .■
1− (1 100 ) 99 100 9
n=1
 Erum Dost

7. Suppose ∑ an converges absolutely and {bn } is bounded. Prove that ∑a bn n

converges absolutely.

Proof: Since {bn } is bounded, there exists M ≥ 0 such that bn ≤ M . Since ∑a n

converges then by the Algebraic Limit Theorem, ∑M a n also converges. Since

anbn ≤ M an , we may use the Comparison Test to assert that ∑ab n n converges.
Finally, the definition of absolute convergence implies ∑a bn n converges absolutely. ■

8. If ∑a n converges absolutely, prove that ∑a 2

n converges. Is the converse true? Is the
statement true if ∑a n converges conditionally?

Proof: By definition of absolute convergence, ∑a n must converge. By Corollary 4.3.6,

if ∑a n converges, then lim an = 0. Then there exists an N such that n ≥ N implies
n→∞
∞
an < 1. Hence an2 < an for n ≥ N. By the Comparison Test, ∑a 2
n converges and
n=N
∞
therefore so does ∑a 2
n as there are only a finite number of terms before N. Since
n=1

an2 ≥ 0, the convergence is absolute. This result does not hold without absolute
( −1)n .
convergence. Consider ∑ n
This series is convergent by Cauchy’s convergence

1
criterion; however, the series ∑ diverges. Now consider
( −1)n+1
n
∑ n
which converges

1
conditionally; however, ∑ diverges. ■
n

9. Determine the values of r for which ∑ nr n

converges.

Proof: Applying the Ratio Test, we have

R = lim
( n + 1) r n+1
n→∞ nr n
⎛ n + 1⎞
= r lim ⎜ ⎟
n→∞ ⎝ n ⎠

⎛ 1⎞
= r lim ⎜ 1+ ⎟
n→∞ ⎝ n⎠
= r.
 Erum Dost

Hence the series converges when R = r < 1 and diverges when r > 1.
∞
At r = 1, we have ∑ n, which diverges by the nth − term test for divergence.
n=1
∞

∑ n ( −1) , which also diverges by the nth − term test for divergence.
n
At r = −1, we have
n=1

Therefore the series ∑ nr n

converges if and only if r < 1. ■

10. Use Raabe’s Test to show that if {an } is a sequence of non-zero real numbers and if
⎡ ⎛ a ⎞⎤
the limit a := lim ⎢ n ⎜ 1− n+1 ⎟ ⎥ exists, then ∑a n is absolutely convergent when a > 1
n→∞
⎢⎣ ⎝ an ⎠ ⎥⎦
and is not absolutely convergent when a < 1.

Raabe’s Test: Let A := ( an ) be a sequence of non-zero real numbers.

an+1 a
i) If there exists numbers a > 1 and K ∈ such that ≤ 1− for n ≥ K,
an n
then ∑a n is absolutely convergent.
an+1 a
ii) If there exist real numbers a ≤ 1 and K ∈ such that ≥ 1− for
an n
n ≥ K, then ∑a n is not absolutely convergent.

Proof: Suppose the limit exists and that a > 1. If a1 is any number with a > a1 > 1, then
⎛ a ⎞ a a
there exists K ∈ such that a1 < n ⎜ 1− n+1 ⎟ for n > K. Therefore n+1 < 1− 1 for
⎝ an ⎠ an n
n ≥ K, where Raabe’s Test implies that ∑ an is absolutely convergent for a > 1. Now
suppose that the limit exists and that a < 1. If a1 is any number with a < a1 < 1, then there
⎛ a ⎞ a a
exists K ∈ such that a1 > n ⎜ 1− n+1 ⎟ for n > K. Therefore n+1 > 1− 1 for n ≥ K,
⎝ an ⎠ an n
where Raabe’s Test implies that ∑a n is not absolutely convergent for a < 1. ■