Professor Oswaldo Rio Branco de Oliveira
http://www.ime.usp.br/~oliveira oliveira@ime.usp.br
This text shows and proves the long division method, a division table for two arbitrary power series, either real or complex. It also includes other ways of dividing, such as “the geometric series trick”. Several (23) examples are given. The efficiency of such division table is highlighted, here, with the rather illustratingBernoulli Numbers.
1. Introduction...2
2. Notations...4
3. The Algorithm (Long Division Algorithm)...5
4. Kid Examples...7
5. Examples...10
6. Polynomial Division: Euclidean X Long ...14
7. Avoiding the Long Division (the Geometric Series Trick)...17
8. Examples (using the Geometric Series Trick) ...19
9. Examples, using the Big-oh Notation “O”...20
10. Examples (Miscellaneous Methods)...25
11. Bernoulli Numbers...27
12. References...30
1. Introduction
Let us denote by z the variable in the complex plane.
Let us consider two complex power series
¢
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¤
az a0a1za2z2a3z3 and
bz b0b1zb2z2b3z3,
where a0, a1, a2, a3, . . . and b0, b1, b2, b3, . . . are all complex coefficients, with the power series bz satisfying the additional condition
b0 x0.
Let us assume that both series converge on a small neighborhood of the origin.
It’s known that the reciprocal function (or multiplicative inverse function) 1
bz
1
b0b1zb2z2b3z3
may, on a possibly smaller neighborhood of the origin, be written as a convergent power series with complex coefficients.
Since the multiplication of convergent power series is also a convergent power series, the division of the power series az for the power series bz is a power series that converges on a small neighborhood of the origin. Hence, we may write
az
bz c0c1zc2z2c3z3,for allSzS small enough.
Thus, for all z on a small enough neighborhood of the origin we have
ª
Q
n 0
anzn
ª
Q
j 0
bjzj
ª
Q
k 0
ckzk.
We recall that the multiplication of two power series satisfy the following rule (basically, the “same” rule that applies to the multiplication of two polynomials)
In short notation, we may write such infinite multiplication as
ª
Q
j 0
bjzj
ª
Q
k 0
ckzk
ª
Q
n 0
Q
jk n
bjck
zn.
These three last equations lead to the identity
ª
Q
n 0
anzn
ª
Q
n 0
Q
jk n
bjck
zn.
This identity, plus the Uniqueness Theorem for the Coefficients of a Power Se- ries (see Oliveira [6, p. 7; 5, p. 15], Beardon [2, pp. 112–113]), presents the relations from where we can finally determine all the coefficients that we are loo- king for: c0, c1, c2, c3, c4, . . . . In fact, we can recursively determine the coefficients c0, c1, c2, c3, c4, c5, . . . (in this exact order) by the well known relations
¢
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¤
a0 b0c0
a1 b0c1b1c0
a2 b0c2b1c1b2c0
a3 b0c3b1c2b2c2b3c0
a4 b0c4b1c3b2c3b3c1b4c0
a5 b0c5b1c4b2c3b3c2b4c1b5c0
a6 b0c6b1c5b2c4b3c3b4c2b5c1b6c0
a7 b0c7b1c6b2c5b3c4b4c3b5c2b6c1b7c0
a8 b0c8b1c7b2c6b3c5b4c4b5c3b6c2b7c1b8c0
a9 b0c9b1c8b2c7b3c6b4c5b5c4b6c3b7c2b8c1b9c0
a10 b0c10b1c9b2c8b3c7b4c6b5c5b6c4b7c3b8c2b9c1b10c0
.
Well, this really looks beautiful albeit quite cumbersome. Needless to say, the computations can be too tedious!
In view of this seemingly unpleasant strategy, we proceed by searching for a more suitable method for finding the much desired coefficients cns.
Remark. The very important “Euclidean division” employed for dividing an infi- nite power series by a polynomial (that is the case ifbzis, in fact, a polynomial) can be seen at Oliveira [7].
2. Notations
The notation that we are about to use for this so called long division process is the same one that we use for the Euclidean division for polynomials. In fact, this notation is the same one that we use since childhood for dividing natural numbers.
That is, by considering two natural numbers N and D, with Dx0, we write the Euclidean division as
N D
R Q meaning
¢
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¤
N DQR
or
N
D QDR,
where N stands for the numerator or dividend, Dx0 stands for the divisor, the letter Q stands for the unique quotient, and the letter R stands for the unique remainder.
Analogously, the Euclidean division for polynomials allow us to divide an arbitrary polynomialPzby a non null polynomialQz, with the sole condition that the degree of Pz is greater or equal to the degree ofQz.
Such Euclidean polynomial division gives us an unique quotient polynomial Dz, and anuniqueremainder polynomialRzwhose degree is strictly smaller than that of the divisor Qz, which is also called the denominator polynomial.
We then write
Pz Qz
Rz Dz meaning
¢
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¤
Pz QzDzRz or
Pz
Qz Dz RQz
z,
with degreeRdegreeQ. We recall that, as a convention, the zero polynomial has degree ª.
3. The Algorithm (Long Division Algorithm)
In this section we show an algorithm that gives us the complex coefficients c0, c1, c2, c3, c4, . . . related to the division
a0a1za2z2a3z3
b0b1zb2z2b3z3 c0c1zc2z2c3z3.
The division algorithm for power series, also called Long Division Algorithm, that we will see has the following three characteristics
Y First. It depends solely on the coefficients of the involved power series, and the order that these coefficients appear on the expressions of the re- lated power series. Hence, in the algorithm, we don’t need to explicit the monomialsz, z2, z3, z4, . . .(but we may write them down, if it is convenient).
Y Second. In the last section we saw a way of finding the numerical coefficients c0, c1, c2, c3, . . ., in this exact order. This exigence remains. That is, we will first find the independent term c0, then the coefficient c1 (of the termc1z), then the coefficient c2 (of the term c2z2), and so on.
Y Third. The algorithm is merely formal. This has two consequences.
X The fact that the algorithm gives to us the coefficients c0, c1, c2, . . ., does not guarantee at all the convergence of the power series cz c0 c1z c2z2 . Furthermore, the algorithm does not guarantee, by itself, the numerical identity az bzcz at a point z. What guarantee this identity are the results mentioned in the introduction.
X The algorithm can be applied to two arbitrary power series, either real or complex ones, either convergent or not. The computations are merely formal. Thus, given two completely arbitrary power series az a0a1za2z2andbz b0b1zb2z2, with the sole con- dition b0 x0, the algorithm gives the unique coefficients c0, c1, c2, . . . of a power seriescz c0c1zc2z2 that formally satisfy
a0a1za2z2 b0b1zb2z2c0c1zc2z2. In other words, we are merely saying that the coefficients c0, c1, c2, . . . satisfy the relationsa0 b0c0, a1 b0c1b1c0, a2 b0c2b1c1b2c0, . . ..
ALGORITHM DIVISION FOR a0a1za2z2a3z3
b0b1zb2z2b3z3 c0c1zc2z2c3z3.
The algorithm is given by
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
a0 a1 a2 a3 a4 . . . S b0 b1 b2 b3 b4 . . .
S
0 a1c0b1 a2c0b2 a3c0b3 a4c0b4 . . . S c0 c1 c2 c3 c4
0 0 a2c0b2c1b1 a3c0b3c1b2 a4c0b4c1b3 . . . S 0 0 0 a3c0b3c1b2c2b1 a4c0b4c1b3c2b2 . . . S
0 0 0 0 a4c0b4c1b3c2b2c3b1 . . . S
0 0 0 0 0 . . . S
. . . S
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
,
where the coefficients c0, c1, c2, c3, c4, . . . are given recursively by
¢
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¤
c0 a0
b0
c1 a1c0b1
b0
c2 a2c0b2c1b1
b0
c3 a3c0b3c1b2c2b1
b0
c4 a4c0b4c1b3c2b2c3b1
b0
or
¢
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¤
0 a0c0b0
0 a1c0b1c1b0
0 a2c0b2c1b1c2b0
0 a3c0b3c1b2c2b1c3b0
0 a4c0b4c1b3c2b2c3b1c4b0
.
6
4. Kid Examples
Example 1. Writefzas a power series centered at the origin, where fz 1
1z.
Solution. By employing the division algorithm for power series (long division algorithm), we find
1 1z
z 1 z z2 z3 z4 z5 z2
z3 z4
z5
.
[As a mere remark, we notice that the left column below the horizontal line, the column of the remainders, originates from the very short computations
¢
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¤
z 1 11z z2 z z1z
z3 z2 z21z z4 z3 z31z
z5 z4 z41z
,
where 1,z,z2,z3, and z4 indicate monomials appearing in the line of the quotient, the second line on the right of the vertical line. The polynomial 1 is in the position of the dividend while the polynomial 1z is the divisor.]
Hence, we have the development 1
1z 1zz2z3z4z5.
As is well known, this development is valid for all z such that SzS1¥
Example 2. Writefzas a power series centered at the origin, where fz 1z2
1z .
Solution. By employing the division algorithm for power series (long division algorithm), we find
1z2 1z
zz2 1 z 2z2 2z3 2z4 2z5 2z2
2z3 2z4
2z5
.
[As a mere remark, we notice that the left column below the horizontal line, the column of the remainders, originates from the short computations
¢
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¤
zz2 1z2 11z 2z2 zz2 z1z
2z3 2z2 2z21z 2z4 2z3 2z31z
z5 z4 2z41z
,
where1,z,2z2,2z3,and2z4indicate monomials appearing inthe line of the quotient, the second line on the right of the vertical line. The polynomial 1z2 is the dividend while the polynomail 1z is the divisor.]
Hence, we have the development 1z2
1z 1z2z22z32z42z5.
This development is valid for all z such that SzS1. Please, check ¥
Example 3. Writefzas a power series centered at the origin, where fz 1z
1z2.
Solution. By employing the long division algorithm, we find
1z 1z2
zz2 1 z z2 z3 z4 z5 z6 z7 z8 z9 z10
z2z3
z3z4 z4z5 z5z6
z6z7
z7z8 z8z9 z9z10
z10z11
.
[The column of the remainders originates from
¢
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¨
¤
zz2 1z 11z2
z2z3 zz2 z1z2
z3z4 z2z3 z21z2 z4z5 z3z4 z31z2 z5z6 z4z5 z41z2
z6z7 z5z6 z51z2
z7z8 z6z7 z61z2 z8z9 z7z8 z71z2 z9z10 z8z9 z81z2
z10z11 z9z10 z91z2
,
where 1,z,z2,z3,z4,z5,z6,z7,z8,z9 and z10indicate monomials in the quotient. Yet, 1z2 is the dividend and 1z is the divisor.]
Hence, we have the development (a convergent one, for all SzS1).
1z
1z2 1zz2z3z4z5z6z7z8z9z10 ¥
5. Examples
Example 4. Write down the first five non null terms of the power series, centered at the origin, of the tangent function
tanz sinz cosz.
Solution. By employing the division algorithm for power series (the long division algorithm), we find
z z3!3 z5!5 z7!7 z9!9 1z2!2 z4!4 z6!6 z8!8
z z2!3 z4!5 z6!7 z8!9 z 3!2z3 165!z5 2727! z7 79369! z9 0 23!z3 45!z5 67!z7 89!z9
2z3
3!
20z5
5!
70z7
7!
168z9
9!
0 165!z5 64z7
7!
160z9
9!
16z5
5!
336z7
7!
2016z9
9!
0 2727!z7 18569!z9
272z7
7!
9792z9
9!
0 79369!z9
.
Or the following short presentation, simply built by omitting all the subtrac- tions indicated in the remainders column that is right above,
z z3!3 z5!5 z7!7 z9!9 1z2!2 z4!4 z6!6 z8!8
0 23!z3 45!z5 67!z7 89!z9 z 3!2z3 165!z5 2727! z7 79369! z9 0 165!z5 647!z7 1609!z9
0 2727!z7 18569!z9 0 79369!z9
.
We may also write
tanz z 1 3z3 2
15z5 17
315z7 62
2835z9¥
Example 5. Write down the sixth non null term of the power series, centered at the origin, of the tangent function
tanz sinz cosz.
Solution. Following Example 4 and taking advantage that sinz is an odd func- tion we may conveniently highlight the monomialsz, z3, z5, z7, . . .in its expansion.
Hence, one may also write the division table for tanzin the following way.
z z3 z5 z7 z9 z11
1 3!1 5!1 7!1 9!1 11!1 1z2!2 z4!4 z6!6 z8!8 z10!10
1 2!1 4!1 6!1 8!1 10!1 z 23!z3 165!z5 2727!z7 79369!z9 35379211!z11 0 3!2
4
5!
6
7!
8
9!
10
11!
2
3!
20
5!
70
7!
168
9!
330
11!
0 165! 647! 1609! 32011!
16
5!
336
7!
2016
9!
7392
11!
0 2727! 18569! 707211!
272
7!
9792
9!
89760
11!
0 79369!
82688
11!
7936
9!
436480
11!
0 35379211!
.
We may also write tanz z 1
3z3 2
15z5 17
315z7 62
2835z9 1382
155925z11. Thus, the sixth non null term of the power series expansion of tanz is
353792
11! z11 1382 155925z11¥
Example 6. Write down the first four non null terms of the power series, centered at the origin, of the function
fz z sinz
z
z z3!3 z5!5 z7!7 . [The very importantsynchronic function is defined bysinz~z.]
Solution. We remark that in order to employ the long division algorithm we may divide, or we may not divide, the numerator and the denominator by z1. Let us choose not to divide.
Computing factorials, we write sinz z z3
6 z5 120
z7
5040 .
By employing the division algorithm for power series we find
z zz63 120z5 5040z7
z z63 120z5 5040z7 1 z62 7360z4 1512031z6 0 z63 120z5 5040z7
z3
6
z5
36
z7
720
0 7360z5 6z7
5040
7z5
360
7z7
2160
0 1512031z7
.
Thus, we arrive at the expression z
sinz 1z2 6
7z4 360
31z6
15120¥
Example 7. Write down the first four non null terms of the series expansion, in powers of z and centered at the origin, of the hyperbolic cosecant function
fz cosechz 1 sinhz. Solution. First, we notice that
sinhz ezez
2 z z3 3!
z5 5!
z7 7!
z9 9! . Thus,
1 sinhz
1 z
1
1 z3!2 z5!4 z7!6 z9!8
.
Moreover, we have sinh0 0, which implies that we won’t find a power series expansion. Instead, we will need the power z1, a negative power of z.
Putting z into evidence in the denominator and by employing the division algorithm for power series (the long division algorithm), we find
1 1 z62 120z4 5040z6
1 z62 z4
120
z6
5040 1 z62
7z4 360
31z6
15120
0 z62 120z4 5040z6
z2
6
z4
36
z6
720
0 7360z4 840z6
7z4
360
7z6
2160
0 1512031z6
31z6
15120
0
.
Finally, we obtain the expansion (also called a Laurent series, since it contains negative powers ofz)
cosechz 1 z
z 6
7z3 360
31z5 15120.
The domain of convergence. We notice that we have sinhz 0 if and only if ezez eze2z1 0. Writing z xiy, with x and y real numbers, we have
sinhz 0e2x2yi 1 e02πie2xe2yi 1e2πi.
Thus, z xiy satisfies sinhz 0 if and only if x 0 and y nπ, for some n>Z. Hence, the expansion that we found converges on the punctured disk
z>C0SzSπ.