Chapter 12 Arithmetic of Power Series

Introduction. In Chapter 7, we have dealt with the representability of a function by a power series. A natural question arises as to the representability of the sums, products and quotients of functions by power series in terms of the power series of the functions and whether we can formally add, multiply or divide the power series representing the functions to give a power series for the sum, product and quotient.

12.1 Sums of Power Series

The situation of representing sum of functions by power series is quite simple. Theorem 1. Suppose the function f is represented by

anxn n= 0

with radius of
2

convergence r1 and g is represented by the sum f + g is represented by convergence r min(r1 , r2).

b n x n with radius of convergence r . Then n=0 the power series (a n + b n )x n with radius of n=0

Proof. If min(r1 , r2) = 0, then we have nothing to prove. Suppose now 0 < min(r1 , r2). Let c be a real number such that 0 < c < min(r1 , r2). Then the n-th partial sum s n (x) = a k x k converges absolutely and uniformly to f on [c, c] by Theorem 4 and
k=0 n n

Remark 16 of Chapter 7 since c < r1. Similarly we deduce that the n-th partial sum t n (x) = b k x k converges absolutely and uniformly to g on [c, c], since c < r2. This means given any > 0, there exists a positive integer N1 such that for all integer n N1 and for all x in [c, c], |s n (x) f (x)| < --------------------------------- (1) 2 and there exists a positive integer N2 such that for all integer n N2 and for all x in [c, c], |t n (x) g(x)| < --------------------------------- (2). 2 Let N = max (N1, N2). It follows by (1) and (2) that for all integer n N and for all x in [c, c], |s n (x) + t n (x) ( f (x) + g(x))| [ |s n (x) f (x)| + |t n (x) gf (x)| < + = . 2 2 n k This means that s n (x) + t n (x) = (a k + b k )x converges uniformly to f + g on [c, c].
k=0 k=0

Take a real number c' such that 0 < c < c' < min(r1 , r2).
n k k=0 n

Then by Theorem 4 of

Chapter 7, s n (c ) + t n (c ) = (a k + b k )(c ) is convergent. It follows by Proposition 5 Chapter 7 that s n (x) + t n (x) = (a k + b k )x k converges absolutely to f + g on [c, c].
k=0

(We may also deduce this directly by noting that for |x| < c,

convergent and since for each integer n 0, (a n + b n )x n [ a n x n + b n x n , by the Comparison Test, converges for any x such that |x| c and for any c such that 0 < c < min(r1 , r2).
Ng Tze Beng

anx n=0

and

bnxn n=0

are

(a n + b n )x n n=0

is convergent for |x| c. ) Therefore,

(a n + b n )x n n=0

Chapter 12 Arithmetic of Power Series

Therefore, if r is the radius of convergence of the series (a n + b n )x n , r > c for any 0 < c < min(r1 , r2). Hence, r min(r1 , r2).
n=0

Remark. Theorem 1 is valid when any one of r1 or r2 is infinite and when both r1 and r2 are infinite and in this case r is infinity.

Example 2. We know from Example 21 Chapter 8 that n ln(1 + x) = (1) n+1 x n for 1 < x 1. n=1 and that by Theorem 19 Chapter 9, 1 $ 3 $ 5(2n 1) n 1 =1+ x for 1 < x < 1. 2 $ 4 $ 6(2n) n=1 1x 1 $ 3 $ 5(2n 1) (1) n+1 n Therefore, the power series 1 + + x converges to n 2 $ 4 $ 6(2n) n =1 1 $ 3 $ 5(2n 1) 1 is divergent (by + ln(1 + x) for |x| < 1. Now since 1 + 2 $ 4 $ 6(2n) n =1 1x Comparison Test) and (1) n+1 1 n is convergent (by the Leibnitz's Alternating Series n =1 1 $ 3 $ 5(2n 1) (1) n+1 Test), the sum 1 + is divergent. Similarly, because + n 2 $ 4 $ 6(2n) n =1 1 $ 3 $ 5(2n 1) (1) n 1+ (1) n is convergent and (1) n+1 n is divergent, the sum 2 $ 4 $ 6(2n) n =1 n =1 1 $ 3 $ 5(2n 1) (1) n+1 1+ + (1) n is divergent. Therefore, the radius of n 2 $ 4 $ 6(2n) n =1 1 $ 3 $ 5(2n 1) (1) n+1 n convergence of the series 1 + + x is less than or equal n 2 $ 4 $ 6(2n) n =1 to 1. But by Theorem 1, its radius of convergence is exactly 1. In particular, it converges for all x such that 1 < x < 1 and diverges elsewhere. Given two series, we can form two kinds of products, the first is the termwise product and the second is the Cauchy product which is a natural product for consideration of product of two functions represented by power series.

12.2 Termwise Product

Definition 3. If
n=0

a n and

n=0

b n , then the termwise product of these two series is

defined to be

anbn. n=0 an n=0

and

Theorem 4. If

bn n=0

are absolutely convergent, then

anbn n=0

is also

absolutely convergent. Proof. Since

positive integer N such that n N | bn | < 1. Therefore, for all integer n such that n N, we have.
2 Ng Tze Beng

n=0

b n is convergent,

|bn| 0 as n .

Therefore, there exists a

Chapter 12 Arithmetic of Power Series

Since

n=0

an

|an bn | = |an| |bn| | an | is convergent,

n=0

a N+n is also convergent.

------------------------ (1) Therefore, by inequality (1)

and the Comparison Test,

is convergent. convergent.

a N+n b N+n Therefore, a n b n n=0

n=0

is convergent and so

anbn n=0

is absolutely

An easy consequence of Theorem 4 is the following: If

n=0

a n is absolutely convergent, then

n=0

a2 n is absolutely convergent.
n=0

The converse of this statement is evidently false as divergent. It is of course not true that if

1 is convergent but n2
2

n=0

1 n

is

a2 a n is convergent, then n n is absolutely n=0 =0 is divergent.

convergent. For instance

(1) n 1 is convergent but (1) n 1 n n n =0 n =0

b n x n are power series then the termwise product of a n x n and n n=0 =0 these two power series is defined to be a n b n x n . n=0 Definition 5. If Theorem 6. If

n=0

a n x n has radius of convergence R1 > 0 and

b n x n has radius of n=0

convergence R2 > 0, then the termwise product R1R2 .

a n b n x n has radius of convergence n=0

Proof. Take any real number c such that 0 < c < R1R2 . We shall show that c < R and c < R . Therefore, 2 1 a n b n c n is convergent. Now c < R 1 R 2 u R R2 1 n=0 c c 1 1 n ) < R 1 and x 2 = 2 (R 2 + ) < R 2 . Hence a n x 1 and b n x n x 1 = 2 (R 1 + 2 are R2 R1 n=0 n=0 absolutely convergent. Therefore, by Theorem 4, the termwise product
n=0

n anxn 1bnx2

n=0

a n b n (x 1 x 2 ) n

is absolutely convergent.

Observe that

c2 c 1 c2 c 1 x1x2 = 1 4 (R 1 R 2 + 2c + R R ) = 2 + 4 (R 1 R 2 + R R ) > 2 + 2 ( R 1 R 2 $ 1 2 1 2
Hence by Proposition 5 Chapter 7,
n=0

anbncn

c2 ) = c R1R2

is absolutely convergent. Thus, by

a n b n x n is absolutely convergent for all x such that |x| c. n=0 Since c is any real number such that 0 < c < R R , this means that a n b n x n is n=0
Proposition 5 Chapter 7
1 2

3 Ng Tze Beng

Chapter 12 Arithmetic of Power Series

absolutely convergent for all x such that |x| < R1R2. convergence of Example 7. If
n=0

a n b n x n is greater than or equal to R R .

1 2

Therefore the radius of

n=0

a n x n has radius of convergence R1 > 0 and

n=0

b n x n has radius of

a n b n x n may have radius of n=0 convergence exactly equal to R R . Take a n x n = 1n x n and b n x n = 1n x n . n=0 n=0 2 n=0 n=0 3 1 Then R = 2 and R = 3. The termwise product a n b n x n = n x n . By the Ratio 6 n=0
convergence R2 > 0, then the termwise product
1 2 1 2

Test (Theorem 18 Chapter 7 ), since lim nd which is precisely equal to R1R2 . Example 8. If

a n +1 b n +1 1 = , the radius of convergence is 6, 6 anbn

anxn n=0
x 3
n 2n+1

has radius of convergence R1 > 0 and

b n x n has radius of n=0

2n

convergence R2 > 0, a termwise product may have its radius of convergence strictly greater than R1R2 . Let

a n x n = n n=0 =0
an = 1 3 1 5 n 1 3 n

+ x 5

2n

and

b n x n = n n=0 =0

x 5

2n+1

+ x 3

. That is,

if n is even and n m 0

if n is odd positive integer

n

and b n =

if n is even and n m 0 .

1 5

if n is odd positive integer

1 1 Then lim sup |a n | n = 1 and lim sup |b n | n = 1 . Therefore, by the Cauchy-Hadamard 3 3 nd nd Formula (Theorem 19 Chapter 7), the radius of convergence R1 of a n x n and the

radius of convergence R2 of

n=0

bnxn

n=0

are the same and equals 3. But the termwise

product

n=0

1 x n whose radius of convergence by the Cauchy Hadamard a n b n x n = (15 )n

Formula is 15 and is strictly greater than R1R2 =9. Remark. Note that if f (x) =
n=0

a n x n and n=0

g(x) =

b n x n , then the termwise product n=0

a n b n x n is not equal to f (x) g(x) in general. Take the two series in Example 7. f (x) = a n x n = 1n x n and g(x) = b n x n = 1n x n . Then f (1) = 2 and g(1) = 3 2 n=0 n=0 2 n=0 n=0 3
. Therefore, f (1) g(1) = 3. The termwise product of the two power series is a n b n x n = 1n x n and converges to the value 6 ! f (1)g(1) when x = 1. 5 6 n=0

4 Ng Tze Beng

Chapter 12 Arithmetic of Power Series

12. 3 Cauchy Product

b n x n . We now come to a product of power a n x n and g(x) = n n=0 =0 series of a n x n and b n x n which when it converges gives the value of the product n=0 n=0 Suppose f (x) = f (x) g(x). First we give the definition for series of constant terms. b n are two series. a n and n n=0 =0 product of these two series to be the series c n , where n=0 n c n = a k b nk . k=0 Definition 9. Suppose

Then we define the Cauchy

b n are absolutely convergent, then the Cauchy product a n and n n=0 =0 c is also absolutely convergent. Moreover, if a = A , b = B and n n n c n = C, n=0 n=0 n=0 n=0 Theorem 10. If then C = A B. Proof. We shall prove Theorem 10 when both an and bn are non-negative for all
n=0

integer n 0. Suppose

a n and

n=0

b n are convergent.
n n k=0 j=0

Then the product,

n n

bn a n n n=0 =0

bk a n k n=0 =0

=lim a k lim b = lim a lim b -- (1) nd nd k nd j nd j

k=0

j=0

a n b k is convergent for b k is convergent means k =0 n each integer n 0, or equivalently that a n lim exists. Since b a n is convergent nd j n=0 j= 0 Now for each integer n 0, a n
k=0

b a k lim nd j k=0 j=0

and
k=0

is convergent
n n n n

a k lim b =lim a k lim b = lim a lim b . nd nd k nd j nd j nd j

j=0 k=0 j=0 k=0 j=0
k=0

This explains the identity (1) above. For each integer n 0, let l n = Thus by (1), ln is convergent and converges to Now l n+1 =
n+1 k=0 n =0

ak

k=0

bk .

an
n

ak

n+1 k=0 n

bk =
k =0

k=0

ak
n k=0

k =0

bk +

n =0

bn .
n +1 k=0

k=0

a k b n+1 + a n+1

bk

= l n + a n+1 Consider the array

b k + b n +1

ak

---------------------------- (2)

5 Ng Tze Beng

Chapter 12 Arithmetic of Power Series

a0b0| a0b1| a0b2| a1b0 a2b0 a3b0 a4b0 a1b1| a1b2| a2b1 a3b1 a4b1 a2b2| a3b2 a4b2

a0b3| a1b3| a2b3| a3b3| a4b3

a 0 b 4 a 1 b 4

a0bk a1bk

a 0 b k+ 1 a 1 b k+ 1 a 2 b k+ 1 a 3 b k+ 1 a 4 b k+1 a k b k+ 1

a 2 b 4 a 2 b k a 3 b 4 a 3 b k a 4 b 4 a 4 b k a k b 4

akb0 akb1 akb2 akb3 akbk a k+1 b 0 a k+1 b 1 a k+1 b 2 a k+1 b 3 a k+1 b 4 a k+1 b k a k+1 b k+1

We note from (2) that ln+1 is obtained from ln by adding the term,
n +1 k=0

a n+1

b k + b n+1

Note also that ln is obtained by adding the terms in the n n square in the above array. Hence ( ln ) is convergent means the series obtained by summing up in succession the terms in the k k square by adding the next (2k+1) terms ak

k =0

ak .

bi + bk
i= 0

k1

(a 1 b 0 + a 1 b 1 + a 0 b 1 ), then add (a 2 b 0 + a 2 b 1 + a 2 b 2 + b 2 a 0 + b 2 a 1 ) and so on, is convergent. Since all the terms are non-negative, the series obtained from the above by adding only one term at a time is convergent and converges to the same limit. We define this series as below by defining its terms. We let d1 = a0 b0 , d2 = a1b0, d3 = a1b1 , d4 = a0b1, d5 = a2 b0, d6 = a2b1, d7 = a2b2, d8 = b2a0, d9 = b2a1 and in general, d k2+i = a k b i1 , for i = 1, , k + 1, d k2+i = b k a i(k+1) , for i = k + 1, , 2k. Hence, for integer n 1,
ln =
k=0

ai i=0

is convergent. That is the series, starting with a0 b0, then add

bk a k k =0

(n+1) 2 k=1

d k --------------------------- (3)

Since ( ln ) is convergent and since it is non-negative and monotone increasing, it is bounded above. The n-th partial sum for each integer n 1 sn = dk
k=1 n

is also bounded above. This can be deduced as follows. s n = d k [ d k = l n1 [lim l nd n

k=1 k=1 n n2

Since ( sn ) is monotone increasing and bounded above, by the Monotone Convergence Theorem (Theorem 15 Chapter 2), ( sn ) is convergent. Then any subsequence of ( sn ) is convergent and converges to the same limit (see Proposition 19 Chapter 3). By (3), for each integer n 1, s n2 = l n1 . Therefore,

6 Ng Tze Beng

Chapter 12 Arithmetic of Power Series

each integer n 0. Then, t0 = d0 =a0 b0 and

b k = AB a k k k=0 =0 It follows that lim s = AB. This means d n = AB. Now we examine the series nd n n =1 n n c . Recall that c = a b . Let t = n n n n k n k c k be the n-th partial sum for n=0 n =0 =0 k =0 k=0 lim s 2 =lim l =lim nd n nd n nd
n n n+1 k=0

t n+1 = c k + c n+1 = c k + a k b n+1k ------------------- (4)

k=0

Let

n=0

g n be the series obtained from

n=0

n =0

cn =

k=0

n =0 k =0

a k b n k

without the bracketing.

Since each ai bj corresponds uniquely to some dk and conversely each dk corresponds uniquely to some ai bj , by (4),

g n is a rearrangement of

n =1

d n.

Therefore, since

g n = d n = AB. d n is (absolutely) convergent , n =0 n =1 n =1 solution). Now c n =

n =0 n =0 k =0

(see Tut 10 Question 1 and

n=0

a k b nk is obtained from
the order of the

g n by group the terms in

Therefore, This

brackets
n =0

proves the theorem for the case when an ,bn are non-negative for all integer n 0. In general, if b n are absolutely convergent, then a n and b n are a n and n n=0 =0 n=0 n=0 convergent. It follows that d n , where d is defined as before, is absolutely n =1 convergent, because by the above argument d n is convergent and converges to n =1 n b n . Let g n be the series obtained from c n = a k b nk a n n n =0 =0 n=0 n =0 n =0 k =0 without the bracketing. Then as noted and observed before, g n is a rearrangement n=0 of d n . Since d n is absolutely convergent, n =1 n =1 gn = dn. n=0 n=1 We have observed above that c n = g n . Hence c n = d n . Now from (3) n=0 n=0 n=0 n=1
n

c n = n a k b n k =0 k =0

without
n

altering =
n =0

terms.

g n = AB.

(See Tut 10 Question 2 and solution.)

s (n+1) 2 =

(n+1) 2

k=1

dk = ln =
n

bk . a k k k=0 =0
n

2 =lim l n =lim Therefore, lim s nd (n+1) nd nd

( sn ) is convergent,

n =1

2 = AB. Hence, s =lim s d n =lim nd n nd (n+1)

bk a k k k=0 =0
n

= AB . Since we already knew

n=0

cn =

n=1

d n = AB. cn

Note that for each integer n 0, c n [ a k b nk . We have already shown that

n=0 k=0

ak

k=0

b nk is convergent. It follows then by the Comparison test that

n=0

convergent and so

c n is absolutely convergent.
7

n=0

is

Ng Tze Beng

Chapter 12 Arithmetic of Power Series

If

n=0

a n and

n=0

b n are conditionally convergent, it need not follow that the Cauchy

See the following example:

product

c n is convergent. n=0

Example 11. Let

a0 = a1 = b0 = b1 = 0 and a n = b n = (1) n 1 for integer n 2. ln(n) a n and b n are convergent. Since 1 > 1 for ln(n) n n=0 n=0 integer n 2 and 1 is divergent, by the Comparison Test 1 is divergent. n n =1 n=2 ln(n) Hence a n and b n are conditionally convergent. n=0 n=0 By the Alternating Series Test, c0 = c1 = c2 = c3 = 0 and for integer n > 3, n 1 1 1 c n = a k b nk = (1) n + ++ . ln(n 2) ln(2) ln(2) ln(n 2) ln(3) ln(n 3) k=0 n 3 . By application of L'Hpital's Rule (see Thus, for n even and > 3, c n m (ln(n 2)) 2 n3 remark after Example 39 Chapter 4), d , Therefore, by the (ln(n 2)) 2 Comparison test, c2n . Similarly for n odd and > 3, c n [ n 3 2 d . Hence the series c n cannot (ln(n 2)) n=0 converge.

b n be two series where a n and n n=0 =0

12.4 Multiplication of Power Series

Theorem 12. f 2 (x) =

Suppose f 1 (x) =

n=0

anxn

has radius of convergence R1 and

2

b n x n has radius of convergence R n=0 cnxn, n=0

. Then the product f (x) = f 1 (x) f 2 (x)

is represented by a power series

where c n = a k b nk ,
k=0

in the interval (R, R), where R = min(R1 , R2)

Proof. Let x be in the interval (R, R). Then x (R1 , R1 ) and x (R2, R2 ). Therefore,
n=0

a n x n and
n=0

n=0

b n x n are absolutely convergent (see Theorem 4 Chapter 7).

n k =0 n

Then by Theorem 10, the Cauchy product of these two series,

d n , where d n = (a k x k )(b nk x nk ) =
k=0

a k b nk k=0

x n,

is absolutely convergent. That is, if c n = a k b nk for each integer 0, then is absolutely convergent. Moreover, the Cauchy product,
n =0

cnxn n=0

cnxn =

n =0

anxn

n=0

bnxn
8

= f 1 ( x) f 2 ( x) .

Ng Tze Beng

Chapter 12 Arithmetic of Power Series

Hence

c n x n is absolutely convergent for |x| < R. n=0

ln(1 + x) Example 13. We can express as a power series for |x| < 1. 1+x n First we note that 1 = (1) n x n and ln(1 + x) = (1) n x for |x| < 1. That is, n 1 + x n=0 n=1 the radius of convergence for both series is 1. Therefore, by Theorem 12, for |x| < 1,
ln(1 + x) = xn 1+x n=1 n 1 where H n = . k=1 k

k=1

(1) nk

(1) k = (1) n x n k n=1

k=1

1 = (1) n+1 H x n , n k n=1

12. 5. Quotient of Power Series.

We now consider division of power series. Plainly it is sufficient to study the quotient of a power series 1/ a n x n since
n =0 n =0

n=0

bn

xn

anxn

bnxn n=0

for if we can represent product with Let f (x) =

n=0

bnx

n =0 n

1 as a power series we can then obtain the Cauchy anxn

n=0

1 , anxn

by Theorem 12.

first the case that f (x) 0 at least in a neighbourhood containing 0. Hence the leading coefficient a0 0. We may assume that a0 = 1. This is seen as follows. We can write n n n anx = a0 a a0 x . The series
n=0

n=0

a n x n . Suppose

n=0

a n x n has radius of convergence R > 0.

We consider

anx

n=0

and

n=0

n=0

an n a 0 x have the same radius of convergence R while the Then 1 $ = a 0 1

leading coefficient of

an n a x is 1. n=0 0
n =0

1 anxn

a n a n=0 0

. xn

We seek a power series solution, we have

bnxn n=0
n =0

such that

n =0

1 = b n x n. n=0 n anx

Hence for such a

anxn $

bnxn n =0

= 1.

Hence by the definition of Cauchy product, cnxn = 1 n=0

9 Ng Tze Beng

Chapter 12 Arithmetic of Power Series

and so we have the equations c0 = 1 and cn = 0 for all integer n 1. That means, n 1, n = 0, , c n = a k b n k = k=0 0, n m 1 That is, b0 = 1 b 1 = a 1 b 0 b 2 = (a 1 b 1 + a 2 b 0 ) --------------------------------------- (A) n b n = a k b n k k=1 In this way we define a formal inverse

anxn. b n x n to n n=0 =0
0

Theorem 14. Suppose

a n x n is a power series with a n=0

= 1 and with a non-zero

radius of convergence R. Then the formal inverse of f (x) = a n x n as given by (A) n=0 is a power series expansion for 1 with a non-zero radius of convergence. That is, f ( x) 1 = b xn, n f (x) n =0 where the bn's are determined recursively by (A). Proof. It is sufficient to show that

Note that by assumption, f (0) = a0 = 1 0. Since the radius of convergence of a n x n is positive, by continuity f (x) 0 in a neighbourhood of 0. Thus f (1x) is at n=0 least defined in a neighbourhood of 0. We shall give some estimate of bk's to use the Cauchy Hadamard formula to show that

b n x n has a non-zero radius of convergence. n=0

b n x n has a positive radius of convergence. n=0 Let r be any real number such that 0 < r < R. Then a n r n is convergent. n=0
the formal inverse

Therefore,

an rn 0. (See Proposition 10 Chapter 6). Therefore, the sequence ( anrn ) is bounded and so there exists a constant M 1 such that | an rn | M for integer n 0. Hence, for integer n 0, M ------------------------------------- (1) |a n | [ r n Then using (1), we get, | b0 | = 1 M , |b1 | = |a1 b0| = |a1| M/r , M + M M = 2 M2 |b 2 | = |(a 1 b 1 + a 2 b 0 )| [ |a 1 b 1 | + |a 2 b 0 | [ M r r r2 r2 2 3 M M M M 2M |b 3 | [ |a 1 b 2 | + |a 2 b 1 | + |a 3 b 0 | [ M r $ 2 r2 + r2 $ r + r3 $ M [ 2 r3
10 Ng Tze Beng

Chapter 12 Arithmetic of Power Series

n We claim that |b n | [ 2 n M for any integer n 0. Evidently from the above it is true n r for n = 0,1,2 and 3. Assuming that the claim is true for k = 0, 1,2, , n-1, we shall show that it is true for k = n. By (A), n n 2 nk M nk by (1) and the induction hypothesis b n [ a k b n k [ M $ k r n k k=1 k=1 r

Hence, by mathematical induction |b n | [ 2 n M r n for any integer n 0. 1 1 M Therefore, for all integer n 1, |b n | n [ 2 r . It follows that sup{|b n | n : n m k} [ 2 M r 1 for integer k 1. Therefore, lim sup b n | n [ 2 M . By Theorem 19 Chapter 7 (Cauchy r
nd

1 1 nk n+1k [ r M [ r n 2 n k =1 n n n nk nM [ M 2 [ 2 n n r k=1 r

2 n k M n k=1
n

Hadamard Formula), the radius of convergence of b n x n is given by n=0 1 r > 0. 1 m 2M lim sup b n | n
nd

In practice we can obtain the coefficients bn's by long division and for region of convergence estimate of M for a particular r need to be sought.

We now come to the case when the leading coefficient a0 =0 so that f (0) = 0. We can find the reciprocal 1/ f (x) as a power series in a deleted neighbourhood of 0. Suppose f (x) = x p

radius of convergence R > 0. Then for x 0 and x in (R, R), f (x) anxn. xp = n =0

anxn n=0

and p is an integer 1 and a0 0. Suppose f (x) has

a n x n has the same radius of convergence R > 0. n=0 formal quotient of a n x n , n=0 1 = bnxn n=0 n
Note that
n =0

By Theorem 14, the

anx

has radius of convergence R' > 0.

Therefore, by Theorem 12,

f (x) bnxn = 1 xp n =0 for x 0 and x in (R2, R2), where R2 = min (R, R'). Then in (R2, R2) {0}, b0 b1 1 = 1 n b x = + + + b + b p+k x k n p p p x f (x) x n=0 x p1 k=1 f ( x) In view of Theorem 14, division of power series g(x) can be performed as forming the product of f (x) and 1/g(x) and is equivalent to formal power series division.
11 Ng Tze Beng

Chapter 12 Arithmetic of Power Series

sin(x) Example 15. Find the first few terms of cos(x) by power series division. Recall from Chapter 11, 3 5 2n+1 sin(x) = x x + x + + (1) n x + 3! 5! (2n + 1)!
= (1) n x (2n + 1)! n =0
2n+1

2 4 n cos(x) = 1 x + x + + (1) n x + . 2! 4! (2n)! 2n x = (1) n (2n)! n=0 We perform long division as follows:

x + 1 x 2 + 2 x 5 + 17 x 7 + 62 x 9 2835 3 15 315
2 4 6 1 x + x x + | x 1 x3 + 6 2! 4! 6! x 1 x3 + 2

1 x5 1 x7 + 1 x9 + 5! 7! 9! 1 x5 1 x7 + 1 x9 + 4! 6! 8! 4 x5 + 6 x7 8 x9 + 5! 7! 7! 1 x5 + 6 x7 8 x9 + 6 4!3 6!3 2 x5 4 x7 + 1 x9 + 15 315 2268 2 x5 1 x7 + 1 x9 + 15 15 180 17 x 7 29 x 9 + 315 5670 17 x 7 17 x 9 + 315 630 62 x 9 + 2835

1 x3 3 1 x3 3

sin(x) 2 Therefore, cos(x) = x + 1 3x +

2 5 15 x

17 7 315 x

62 9 2835 x

Remark. Is the composition of two power series expandable as a power series function? The answer is "yes". But the proof will require analytic function theory. If the power series function f (x) = ) 0 for some neighbourhood of x0 and so by Darboux Theorem (Theorem 47 Chapter 4) in that neighbourhood either f ' (x ) > 0 or f ' (x ) < 0. Therefore, in a small neighbourhood of x0 , f (x) is either strictly increasing or strictly decreasing. Hence f (x) is injective in some interval containing x0 . Therefore, its inverse f 1 exists in that
12 Ng Tze Beng

a n (x x 0 ) n satisfies that f ' (x n=0

) 0, then by continuity f ' (x

Chapter 12 Arithmetic of Power Series

interval. The question we can ask is whether the inverse function is representable as powers of ( y f (x0)). The answer is yes but the proof would require an invariance of domain result.

12.6 Analytic Function

We have seen that not all infinitely differentiable functions can be expandable as a power series, (See Example 13 Chapter 8). For manipulation of function by power series, we need to ascertain if the function to be manipulated has a power series expansion, i.e., if it is analytic. The next theorem gives a necessary and sufficient condition for a function to be analytic at the origin. This gives another criterion for analyticity in addition to Theorem 15 of Chapter 8. Theorem 16. Suppose the function f is defined in a neighbourhood of the origin. Then f is analytic at the origin if and only if f is infinitely differentiable and that there exists a real number r > 0 and a real number K > 0 such that f (n) (x) [ r K n! (r |x|) n+1 for r < x < r. Proof. Suppose f is analytic at the origin. Then f has a power series expansion in some interval say (R, R) containing the origin. That is to say, f (x) =

anxn n=0

for all |x| < R. Then by Theorem 11 of Chapter 8, we can differentiate f (x) termwise any number of times in the interval (R, R) and n n f (n) (x) = d n a k x k = a k d n x k dx k=0 dx k=0 n = ak d n xk dx k=0 = a k k(k 1)(k (n 1))x kn
k=n

= ak
k=n

k! x kn (k n)!

---------------------------------- (1) f (r) =

n =0

for all |x| < R. Let constant r > 0 be such that 0 < r < R. Then absolutely convergent. Therefore, 0,

anrn

is

anrn n=0

is convergent and so for each integer n --------------------------------------- (2)

anrn [ akrk .
k=0 k=0

Let K = a k r k . Then from (1) we obtain by the triangle inequality, f

(n)

(x) [ a k x kn
k =n

kn [ a k x kn r kn k! (k n)! r k=n k! $ 1 |x| [ akrk (k n)! r n r k=n

k! (k n)!

kn

13 Ng Tze Beng

Chapter 12 Arithmetic of Power Series

Therefore, for |x| < r,

(n)

k! $ 1 |x| f (x) [ a k r ( k n)! r n r k=n |x| kn k ! 1 [K $ by (2) (k n)! r n r k=n k n |x| |x| kn K n ! k ! k ! K [ rn = rn r r k=n (k n)! k=n (k n)!n! k n n! k |x| [K n r rn k =n (n+1) |x| n! ------------------------ (3) = K rn 1 r kn (n+1) |x| k |x| since x = 1 r . We deduce this series r < 1 so that k n r =n expansion as a special case of the following: 1 First note that = x k for |x| < 1. Then by Theorem 11 Chapter 8, we can 1x k =0 differentiate the function termwise any number of times in the interval (-1, 1). Therefore, for any integer n 0 and for |x| <1, dn 1 = dn dn xk k x = n n n dx 1 x dx k=0 k=0 dx
k

kn

= k(k 1)(k (n 1))x kn = Hence, It follows that n! ( 1 x ) n +1

k=n k=n

k! x kn (k n)! k! x kn. = ( k n)! k=n

k! 1 k x kn ------------------ (4) = x kn = n (1 x ) n+1 k=n (k n)!n! k=n for |x| < 1 and any integer n 0. It follows from (3) that (n+1) (n+1) n! 1 |x| K n! r f (n) (x) [ K = = r K n! r rn r n r | x| (r |x|) n+1 for |x| < r and any integer n 0. Now for the proof of the converse statement. Suppose f is infinitely differentiable and that there exists real numbers r > 0 and K > 0 such that for |x| < r and integer n 0, f (n) (x) [ r K n! . (r |x|) n+1 By Theorem 44 Chapter 4, the Lagrange form of the remainder for the Taylor expansion for f around the origin is given by 1 n+1 (n+1) R n (x) = (n+ f () 1)! x for some between 0 and x. Therefore for |x| < r/2, 1 x n+1 f (n+1) () for some between 0 and x. R n (x) = (n + 1)! r K (n + 1)! 1 rK [ x n +1 = x n+1 (r ||) n+2 (n + 1)! (r ||) n+2 rK [ x n+1 since || |x| (r |x|) n+2
14 Ng Tze Beng

Chapter 12 Arithmetic of Power Series

x rK ----------------------- (5) (r |x|) (r |x|) n+1 x x Now because |x| < r/2, < 1 and so d 0 as n . Hence it (r |x|) (r |x|) follows from (5) that Rn (x) d 0. Therefore, f is analytic on the interval ( r , r ). 2 2 This completes the proof. =

n +1

Exercises 17.
1. By squaring the left hand side of 1 = x n and using the Cauchy product of 1 x n =0 1 the right hand side with itself show that = (n + 1)x n for |x| < 1. ( 1 x) 2 n = 0

2. By using power series multiplication (Cauchy product), compute the first few terms in the expansion about x = 0 of the following: (a) cos2(x) (b) e x ln(1+ x) (c) (sin 1(x) )2 (d) sin(x) sin 1(x) (e) ex cos(x) (f) sin(x) cosh (x) 3. By using division of power series find the first few terms in the expansion about x = 0 of the following functions. x cos(x) cos(x) 1 (a) (b) x x (c) e 2 (d) (e) . e 1 1 + sin(x) cosh(x) 1x cosh(x) 4. The Bernoulli numbers Bn are defined by
Bn n x = x . Prove that e x 1 n = 0 n! B2k+1 = 0 for k = 1,2,3, and that B1 = 1/2 , B2 = 1/6, B4 = 1/30, 6 = 1/42 and B8 = 1/30 . Bn n 1 = (2 1)x n1 , where the Bn 's are defined in ex + 1 n = 0 n! question 4 above.

5. Show that

6. Show by using Cauchy product that 1 ln( 1 ) = n 1+ 1 + 1 ++ 1 n x 1x 1x 2 3 n=1 and state the range for which it is valid. 7. Show that for |x| < 1,

x n sin(n) = x n (n + 1) sin(n) . x n cos(n) n n=0 =0 n=0

15 Ng Tze Beng