1. Partial Fraction Expansion
All the polynomials in this note are assumed to be complex polynomials.
A rational function R(z) = P (z)/Q(z) is a quotient of two polynomials P, Q with Q(z) 6=
0. By Fundamental Theorem of algebra,
Q(z) = q(z − a1)m1· · · (z − ar)mr, where a1, · · · , ar are distinct complex numbers and mi ≥ 1.
Proposition 1.1. Suppose Q(z) = Q1(z)Q2(z) is a product of polynomials such that Q1(z), Q2(z) are relatively prime, i.e. the greatest common divisor of Q1(z), Q2(z) is 1.
Then there exist P1(z), P2(z) so that P (z)
Q(z) = P1(z)
Q1(z)+ P2(z) Q2(z).
Proof. Since Q1(z) and Q2(z) are relatively prime, using division algorithm, we can find u1(z), u2(z) so that Q1(z)u1(z) + Q2(z)u2(z) = 1. Then
P (z)
Q(z) = 1 · P (z) Q(z)
= Q1(z)u1(z)P (z) + Q2(z)u2(z)P (z) Q1(z)Q2(z)
= u2(z)P (z)
Q1(z) +u1(z)P (z) Q2(z)
= P1(z)
Q1(z) + P2(z) Q2(z),
where P1(z) = u2(z)P (z) and P2(z) = u1(z)P (z). We complete the proof. Corollary 1.1. Assume Q(z) = (z − a1)m1· · · (z − ar)mr where a1, · · · , ar are distinct.
Then
(1.1) P (z)
Q(z) = P1(z)
(z − a1)m1 + · · · + P2(z) (z − ar)mr.
Proof. We will prove the statement by induction on deg Q. Let Q1(z) = (z − a1)m1 and Q2(z) = (z − a2)m2· · · (z − ar)mr. By Lemma 1.1, we can find P1(z) and P10(z) so that
P (z)
Q(z) = P1(z)
(z − a1)m1 + P10(z) Q2(z). Since deg Q2 < deg Q, by induction hypothesis, we can write
P10(z)
Q2(z) = P2(z)
(z − a2)m2 + · · · + Pr(z) (z − ar)mr
for some polynomials P2(z), · · · , Pr(z). This implies (1.1). Let us study the case P (z)/(z − a)m first. We may assume that deg P ≤ m − 1. Using division algorithm, we can find a0, · · · , am so that
(1.2) P (z) = am+ am−1(z − a) + · · · + a1(z − a)m−1. Then we obtain that
P (z)
Q(z) = a1
z − a+ a2
(z − a)2 + · · · + am (z − a)m.
1
In reality, how do we compute a0, · · · , am? In fact, ai= 1
(i − 1)!
di−1 dzi−1
P (z)
Q(z)(z − a)m z=a
.
Or, we can compute these numbers a1, · · · , am using residues. First, we observe that 1
2πi I
Cr(z)
(z − a)m−1P (z)
Q(z)dz = am
where r is small. Denote
R1(z) = P (z)
Q(z) − am
(z − a)m = P (z) − am
(z − a)m . By (1.2),
P (z) − am = am−1(z − a) + · · · + a1(z − a)m−1. Hence we find
R1(z) = am−1+ am−2(z − a) + · · · + a1(z − a)m−2
(z − a)m−1 .
Denote P1(z) = am−1+ am−2(z − a) + · · · + a1(z − a)m−2 and Q1(z) = (z − a)m−1. Then we have
P1(z)
Q1(z) = a1
z − a+ a2
(z − a)2 + · · · + am−1
(z − a)m−1.
Similarly, we can compute am−1 using residue. By induction, we can compute all ai. In general, we consider P (z)/Q(z) of the form (1.1). Let f1(z) be the rational function
f1(z) = P2(z)
(z − a2)m2 + · · · + Pr(z) (z − ar)mr. Then we can write
P (z)
Q(z) = a11 z − a1
+ a12
(z − a1)2 + · · · + a1m
(z − a1)m1 + f1(z).
Notice that a1 is not a pole of f1(z). For r small enough, f (z) is holomorphic on Dr(a1).
For 0 ≤ i ≤ m − 1, one has
z→alim1
1 i!
di−1
dzi−1f1(z)(z − a1)m1 = 0.
Using the previous observation.
z→alim1
1 i!
di−1 dzi−1
P (z)
Q(z)(z − a1)m1 = a1i. Similarly, for 1 ≤ s ≤ r, one has
z→alims
1 i!
di−1 dzi−1
P (z)
Q(z)(z − as)ms = asi. Here we write
R(z) =
ms
X
i=1
asi
(z − as)i + fs(z), fs(z) =X
i6=s
Pi(z) (z − ai)mi.
Or using the residue computation, for r small enough so that Dr(ai) excludes all aj for j 6= i. Then
1 2πi
I
Cr(ai)
P (z)
Q(z)(z − ai)mi−1dz = aimi.
If we set Hi(z) =Q
j6=i(z − aj)mj, then P (z)
Q(z)(z − ai)mi−1 = 1 z − ai
P (z) Hi(z).
We know P (z)/Hi(z) is holomorphic on Dr(a). By Cauchy integral formula, we obtain 1
2πi I
Cr(a)
P (z)/Hi(z)
z − ai dz = P (ai) Hi(ai).
This gives aimi = P (ai)/Hi(ai). Now, subtract R(z) from aimi/(z − ai)mi, we obtain R1(z) = R(z) − aimi
(z − ai)mi =
mi−1
X
j=1
aij
(z − ai)j + fi(z).
Inductively, we can find all ais for 1 ≤ s ≤ mi.
Example 1.1. Find the partial fraction expansion of R(z) = 3z2+ z + 1
(z − 1)2(z − 2)3. Let f1(z) and f2(z) be rational functions
f1(z) = A
z − 1+ B
(z − 1)2, f2(z) = C
z − 2+ D
(z − 2)2 + E (z − 2)3 so that R(z) = f1(z) + f2(z). Multiplying R(z) by z − 1, we find
R(z)(z − 1) = 1
z − 1·3z2+ z + 1
(z − 2)3 = A + B
z − 1+ f2(z)(z − 1).
Let r be small enough so that 2 does not belong to Dr(1) and hence f2(z) is holomorphic on Dr(1). By Cauchy-integral formula,
B = 1 2πi
I
Cr(1)
R(z)(z − 1)dz = 3 · 12+ 1 + 1 (1 − 2)3 = −5.
Subtracting R(z) from −5/(z − 2)2, we obtain R(z) − −5
(z − 2)2 = A
z − 1+ f2(z).
By simple calculation, R(z) − −5
(z − 2)2 = 3z2+ z + 1 + 5(z − 2)3
(z − 1)2(z − 2)3 = 5z3− 27z2+ 61z − 39
(z − 1)2(z − 2)3 = 5z2− 22z + 39 (z − 1)(z − 3)2. By Cauchy integral formula and by the fact that f2(z) is holomorphic on Dr(1),
A = 1 2πi
I
Cr(1)
5z2− 22z + 39
(z − 1)(z − 2)3dz = 5 · 1 − 22 · 1 + 39
(1 − 2)3 = −22.
On the other hand, f2(z) =
R(z) − −5 (z − 1)2
− −22
z − 1 = 5z2− 22z + 39
(z − 1)(z − 2)3 − −22 z − 1. Now, we know
f2(z) = 5z2− 22z + 39
(z − 1)(z − 2)3 − −22
z − 1 = 22z2− 105z + 137 (z − 2)3 .
Denote P (z) = 22z2− 105z + 137. Note that P (2) = 15, P0(2) = −17 and P00(2)/2! = 22.
We have (Taylor expansion of P (z) at z = 2)
P (z) = 15 − 17(z − 2) + 22(z − 2)2. Hence we find
f2(z) = P (z)
(z − 2)3 = 22
z − 2+ −17
(z − 2)2 + 15 (z − 2)3. We conclude that
3z2+ z + 1
(z − 1)2(z − 2)3 = −22
z − 1+ −5 (z − 1)2
+
22
z − 2+ −17
(z − 2)2 + 15 (z − 2)3
Proposition 1.2. (Special Case) Let a1, · · · , ar be distinct complex numbers and Q(z) = (z − a1) · · · (z − ar). Then
P (z) Q(z) =
r
X
i=1
P (ai)/Q0(ai) z − ai
. Proof. Let us assume that
P (z)
Q(z) = c1
z − a1 + · · · + cr z − ar. Then we know
(z − ai)P (z) Q(z) = c1
z − ai
z − a1 + · · · + ci−1
z − ai
z − ai−1 + ci+ ci+1
z − ai
z − ai+1+ · · · + cr
z − ai
z − an. Taking limit z → ai, we obtain
z→alimi
(z − ai)P (z) Q(z) = ci. On the other hand, by Q(ai) = 0,
z→alimi
Q(z)
z − ai = lim
z→ai
Q(z) − Q(ai)
z − ai = Q0(ai).
As long as we can show that Q0(ai) 6= 0,
z→alimi
(z − ai)P (z)
Q(z) = P (ai) Q0(ai). Notice that
Q(z) z − ai =Y
j6=i
(z − aj) ⇒ Q0(ai) = lim
z→ai
Q(z) z − ai =Y
j6=i
(ai− aj) 6= 0.
We prove that ci= P (ai)/Q0(ai).
In the above case, since Q(z) has simple pole at ai, by residue calculus, P (ai)
Q0(ai) = Resai
P (z)
Q(z) = P (ai) Q
j6=i(ai− aj). Example 1.2. Find the partial fraction expansion of
R(z) = z2+ z + 1 (z − 1)(z − 2)(z − 3).
Write
R(z) = A
z − 1+ B
z − 2+ C z − 3.
Then A, B, C are residues of R(z) at 1, 2, 3 respectively. Choose r small enough so that Dr(1) does not contain 2, 3 and Dr(2) does not contain 1, 3 and Dr(3) does not contain 1, 2.
By Cauchy integral formula, A = 1
2πi I
Cr(1)
R(z)dz = 12+ 1 + 1 (1 − 2)(1 − 3) = 3
2 and
B = 1 2πi
I
Cr(2)
R(z)dz = 22+ 2 + 1
(2 − 1)(2 − 3) = −7 and
C = 1 2πi
I
Cr(3)
R(z)dz = 32+ 3 + 1
(3 − 1)(3 − 2) = 13 2 . Example 1.3. Find the partial fraction expansion of
R(z) = 1 z4+ 1.
The polynomial z4 + 1 has 4 distinct roots. We will denote them by {z1, z2, z3, z4} = {eiπ/4, eiπ/2, ei3π/4, ei3π/2}. Let us consider the partial fraction expansion of R(z) :
1
z4+ 1 = A1 z − z1
+ A2 z − z2
+ A3 z − z3
+ A4 z − z4
. The zero zi of z4+ 1 is a simple pole of R(z), and thus
Ai = Reszi(R(z)) = lim
z→zi
z − zi z4+ 1 = 1
4z3i.
Since zi4 = −1, zi3= −1/zi. This gives Ai = −zi/4. It is very interesting to see that if D is a closed disk containing all zi, then
1 2πi
I
C
R(z)dz =
4
X
i=1
Reszi(R(z)) = −z1+ z2+ z3+ z4
4 = 0
using Vieta’s formula. Here C is the boundary of D.