Egorychev

Egorychev method and the evaluation of binomial
coefficient sums
Marko R. Riedel
June 23, 2016
The Egorychev method is from the book by G.P.Egorychev [Ego84]. We

collect several examples, the focus being on computational methods to produce
results. These are from posts to math.stackexchange.com and have retained the
question answer format from that site.
The crux of the method is the use of integrals from the Cauchy Residue
Theorem to represent binomial coefficients, exponentials and Iverson brackets.
We use three types of integrals:
• First binomial coefficient integral (B1 )
(1 + z)n
Z
n 1
= dz.
k 2πi |z|= z k+1
• Second binomial coefficient integral (B2 )

Z
n 1 1
= k+1
dz.
k 2πi |z|= (1 − z) z n−k+1
• Exponentiation integral (E)

Z
k! exp(nz)
nk = dz.
2πi |z|= z k+1
• First Iverson bracket (I1 )

1 + z + z2 + · · · + zn
Z
1
[[0 ≤ k ≤ n]] = dz
2πi |z|= z k+1
z n+1 − 1
Z
1
= dz.
2πi |z|= (z − 1)z k+1
• Second Iverson bracket (I2 )
zk
Z
1 1
[[0 ≤ k ≤ n]] = dz.
2πi |z|= z n+1 1−z
The residue at infinity is coded R.
1
Contents
1 Introductory example for the method (B1 ) 5
2 Introductory example for the method, convergence about zero

(B1 B2 ) 6
3 Introductory example for the method, an interesting substitu-

tion (B1 ) 7
4 Introductory example for the method, another interesting sub-

stitution (B1 ) 8
5 Introductory example for the method, yet another interesting

substitution (B2 ) 9
6 Introductory example for the method, a simple telescoping sum

(I1 ) 11
7 Verifying that a certain sum vanishes (B1 ) 11
8 Basic usage of exponentiation integral (B1 E) 14
9 Introductory example for the method, eliminating odd-even de-

pendence (B1 ) 16
10 Introductory example for the method, proving equality of two

double hypergeometrics (B1 ) 17
11 Evaluating a quadruple hypergeometric(B1 ) 19
12 Evaluating another quadruple hypergeometric(B1 ) 21
13 An identity by Strehl (B1 ) 24
14 Ternary strings that avoid 22 (B1 ) 25
15 Mixing the two types of binomial integrals (B1 B2 ) 28
16 Two companion identities by Gould (B1 ) 29
17 Exercise 1.3 from Stanley’s Enumerative Combinatorics (B1 B2 ) 32
18 Counting m-subsets (B1 I2 ) 34
19 Method applied to an iterated sum (B1 R) 36
20 A pair of two double hypergeometrics (B1 ) 39
2
21 A two phase application of the method (B1 ) 41
22 An identity from Mathematical Reflections (B1 ) 44
23 A triple Fibonacci-binomial coefficient convolution (B1 ) 45
24 Fibonacci numbers and the residue at infinity (B2 R) 47
25 Permutations containing a given subsequence (B1 I1 ) 49
26 Catalan numbers and Lagrange inversion (B1 ) 55
27 A binomial coefficient - Catalan number convolution (B1 ) 58
28 A new obstacle from Concrete Mathematics (Catalan numbers)

(B1 ) 60
29 Reducing the form of a double hypergeometric (B1 ) 62
30 Basic usage of the Iverson bracket (B1 I1 I2 R) 67
31 Basic usage of the Iverson bracket II (B1 I1 ) 71
32 Use of a double Iverson bracket (B1 I2 R) 73
33 Iverson bracket and an identity by Gosper (I1 R) 76
34 Iverson bracket and an identity by Gosper, generalized (I2 R) 78
35 A double hypergeometric sum (B1 ) 80
36 Factoring a triple hypergeometric sum (B1 ) 81
37 Factoring a triple hypergeometric sum II (B1 B2 ) 84
38 Factoring a triple hypergeometric sum III (B1 ) 86
39 Factoring a triple hypergeometric sum IV (B1 ) 87
40 A triple hypergeometric sum V (B1 ) 89
41 Basic usage of exponentiation integral to obtain Stirling number

formulae (E) 90
42 Advanced usage of the Iverson bracket with two instances (EI1 ) 92
43 Evaluation of a three-variable hypergeometric sum (B2 ) 95
44 Three phase application including Leibniz’ rule (B1 B2 R) 97
3
45 Symmetry of the Euler-Frobenius coefficient (B1 EI2 R) 101
46 A probability distribution with two parameters (B1 B2 ) 104
47 An identity involving Narayana numbers (B1 ) 107
48 Fibonacci, Tribonacci, Tetranacci (B1 ) 111
49 An identity involving two binomial coefficients and a fractional

term (B1 ) 114
50 Double chain of a total of three integrals (B1 B2 ) 117
51 Post Scriptum: Generating function of an initial segment of

Fibonacci numbers (I2 ) 120
4
1 Introductory example for the method (B1 )
Suppose we seek to evaluate
n
X n n+k k
Sj (n) = (−1)k
k k j
k=0
which is claimed to be

n n+j
n
(−1) .
j j
Introduce
(1 + z)n+k
Z
n+k 1
= dz
k 2πi |z|= z k+1
and
(1 + w)k
Z
k 1
= dw.
j 2πi |w|= wj+1
This yields for the sum
n
(1 + z)n 1 k k
Z Z
1 1 X k n (1 + z) (1 + w)
(−1) dw dz
2πi |z|= z 2πi |w|= wj+1 k zk
k=0
Z n Z n
1 (1 + z) 1 1 (1 + w)(1 + z)
= 1− dw dz
2πi |z|= z 2πi |w|= wj+1 z
(1 + z)n 1
Z Z
1 1 n
= (−1 − w − wz) dw dz
2πi |z|= z n+1 2πi |w|= wj+1
n Z n Z
(−1) (1 + z) 1 1 n
= (1 + w + wz) dw dz.
2πi |z|= z n+1 2πi |w|= wj+1
This is
n
(−1)n (1 + z)n 1
Z Z
1 X n q
w (1 + z)q dw dz.
2πi |z|= z n+1 2πi |w|= wj+1 q=0
q
Extracting the residue at w = 0 we get

(−1)n (1 + z)n n
Z
(1 + z)j dz
2πi |z|= z n+1 j
n (−1)n (1 + z)n+j
Z
= dz
j 2πi |z|= z n+1

n n+j
= (−1)n .
j n
thus proving the claim.
This is math.stackexchange.com problem 1331507.
5
2 Introductory example for the method, conver-
gence about zero (B1 B2 )
r
X r−k s+k
m n
k=0
where n ≥ s and m ≤ r.
Introduce

r−k
Z
1 1 1
= dz.
m 2πi |z|= z r−k−m+1 (1 − z)m+1
Note that this is zero when k > r − m so we may extend the sum in k to
k = ∞.
Introduce furthermore
(1 + w)s+k
Z
s+k 1
= dw.
n 2πi |w|= wn+1
(1 + w)s X k
Z Z
1 1 1 1
r−m+1 m+1
z (1 + w)k dw dz
2πi |z|= z (1 − z) 2πi |w|= wn+1
k≥0
(1 + w)s
Z Z
1 1 1 1 1
= r−m+1 m+1 n+1
dw dz.
2πi |z|= z (1 − z) 2πi |w|= w 1 − (1 + w)z
This is
(1 + w)s
Z Z
1 1 1 1 1
r−m+1 m+2
dw dz.
2πi |z|= z (1 − z) 2πi |w|= wn+1 1 − wz/(1 − z)
Extracting the inner residue we get

n
zq

X s
.
q=0
n − q (1 − z)q
Now Z
1 1 1 r+1
r−m−q+1
dz =
2πi |z|= z (1 − z)m+q+2 m+q+1
which yields for the sum
n
X s r+1
.
q=0
n−q m+q+1
6
This evaluates to
s+r+1
n+m+1
by inspection i.e. combinatorially.
However we can also evaluate it algebraically by re-indexing for
s
X s r+1
q=0
q m+n−q+1
where we have lowered the upper limit to s since the first binomial coefficient
is zero when q > s.
Using
(1 + z)r+1
Z
r+1 1
= dz
m+n−q+1 2πi |z|= z m+n−q+2
we thus obtain for the sum
s
(1 + z)r+1 X s q
Z
1
z dz
2πi |z|= z m+n+2 q=0 q
(1 + z)r+s+1
Z
1 s+r+1
= = .
2πi |z|= z m+n+2 n+m+1
This was math.stackexchange.com problem 928271.
3 Introductory example for the method, an in-

teresting substitution (B1 )
Suppose we seek to verify that
2m
X p−1+q 2m + 2p + q − 1 q p−1+m
(−1)q 2 = (−1)m .
q=0
q 2m − q m
Introduce

2m + 2p + q − 1
Z
1 1
= 2m−q+1
(1 + z)2m+2p+q−1 dz.
2m − q 2πi |z|= z
Observe that this controls the range being zero when q > 2m so we may
extend q to infinity to obtain for the sum
1
Z
1 X p − 1 + q
2m+2p−1
(1 + z) (−1)q 2q z q (1 + z)q dz
2πi |z|= z 2m+1 q
q≥0
Z
1 1 1
= (1 + z)2m+2p−1 dz
2πi |z|= z 2m+1 (1 + 2z(z + 1))p
7
Z
1 1 1
= (1 + z)2m+2p−1 dz
2πi |z|= z 2m+1 ((1 + z)2 + z 2 )p
Z
1 1 1
= (1 + z)2m−1 dz
2πi |z|= z 2m+1 (1 + z 2 /(1 + z)2 )p
Z
1 1 1 1
= (1 + z)2m dz.
2πi |z|= z 2m z(1 + z) (1 + z 2 /(1 + z)2 )p
Now put
z u 1
= u so that z= and dz = du
1+z 1−u (1 − u)2
to obtain for the integral

Z
1 1 1 1 1
du
2πi |u|= u u/(1 − u) × 1/(1 − u) (1 + u ) (1 − u)2
2m 2 p
Z
1 1 1
= du.
2πi |u|= u2m+1 (1 + u2 )p
This is

1 1 m m+p−1
[u2m ] = [v m
] = (−1) ,
(1 + u2 )p (1 + v)p m
as claimed.
4 Introductory example for the method, another

interesting substitution (B1 )
bm/2c
X n m − 2k + n − 1
(−1)k
k n−1
k=0
where m ≤ n and introduce

m − 2k + n − 1 m − 2k + n − 1
=
n−1 m − 2k
Z
1 1
= (1 + z)m−2k+n−1 dz
2πi |z|= z m−2k+1
which has the property that it is zero when 2k > m so we may set the upper
limit in the sum to n, getting
8
n
z 2k
Z
1 1 m+n−1
X n
m+1
(1 + z) (−1)k dz
2πi |z|= z k (1 + z)2k
k=0
n
z2
Z
1 1 m+n−1
= (1 + z) 1 − dz
2πi |z|= z m+1 (1 + z)2
Z
1 1
= (1 + z)m−n−1 (1 + 2z)n dz
2πi |z|= z m+1
(1 + z)m (1 + 2z)n
Z
1 1
= dz.
2πi |z|= zm z(1 + z) (1 + z)n
Now put
1 + 2z u−1 1 1+z 1
= u so that z=− , 1+z =− , = ,
1+z u−2 u−2 z u−1
1 (u − 2)2 1
= and dz = du
z(1 + z) u−1 (u − 2)2
to get for the integral
(u − 2)2 n
Z
1 1 1
u du
2πi |u−1|= (u − 1)m u − 1 (u − 2)2
Z
1 1
= un du.
2πi |u−1|= (u − 1)m+1
This is
n
m n m
X n n
q
[(u − 1) ]u = [(u − 1) ] (u − 1) = .
q=0
q m
This solution is more complicated than the obvious one (which can be found
at the stackexchange link) but it serves to illustrate the substitution aspect of
the method.
5 Introductory example for the method, yet an-

other interesting substitution (B2 )
The following proof uses complex variable techniques and improves the elemen-
tary one I posted earlier. It serves to demonstrate the method even though it
requires somewhat more of an effort.
n
X 2n
k .
n+k
k=0
9
Introduce
Z
2n 1 1 1
= dz.
n+k 2πi |z|= z n−k+1 (1 − z)n+k+1
Observe that this is zero when k > n so we may extend k to infinity to

obtain for the sum
zk
Z
1 1 1 X
k dz
2πi |z|= z n+1 (1 − z)n+1 (1 − z)k
k≥0
z/(1 − z)
Z
1 1 1
= n+1 (1 − z)n+1 (1 − z/(1 − z))2
dz
2πi |z|= z
Z
1 1 1 1
= dz.
2πi |z|= z n (1 − z)n (1 − 2z)2
Now put z(1 − z) = w so that
√
1 − 1 − 4w
z= and (1 − 2z)2 = 1 − 4w
2
and furthermore
1 1
dz = − × × (−4) × (1 − 4w)−1/2 dw = (1 − 4w)−1/2 dw
2 2
to get for the integral
Z Z
1 1 1 −1/2 1 1 1
n
(1 − 4w) dw = n
dw.
2πi |w|= w 1 − 4w 2πi |w|= w (1 − 4w)3/2
This evaluates by inspection to

n−2
4n−1 Y

n−1 n − 1 + 1/2 n−1 n − 1/2
4 =4 = (n − 1/2 − q)
n−1 n−1 (n − 1)! q=0
n−2
2n−1 Y 2n−1 (2n − 1)!
= (2n − 2q − 1) =
(n − 1)! q=0 (n − 1)! 2n−1 (n − 1)!
n2 2n

1 2n
= = n .
2n n 2 n
Here the mapping from z = 0 to w = 0 determines the choice of square root.
10
6 Introductory example for the method, a sim-
ple telescoping sum (I1 )
n n
X n! X n−k
(n − k)nk = n!nn nk−n .
k! k!
k=0 k=0
Introduce
zk
Z
1 1
nk−n = dz.
2πi |z|= z n+1 1 − z/n
Observe that this integral provides an Iverson bracket, as it vanishes when
k > n. Therefore we may extend k to infinity.
We get for the sum
1
Z
1 1 X n−k
n!nn n+1
z k dz
2πi |z|= z 1 − z/n k!
k≥0
 
Z
1 1 1 X 1
= n!nn n exp(z) − z z k−1  dz
2πi |z|= z n+1 1 − z/n (k − 1)!
k≥1
Z
1 1 n
= n!nn (n exp(z) − z exp(z)) dz
2πi |z|= z n+1 n − z
Z
1 1 1
= n!nn+1 exp(z) dz = n!nn+1 = nn+1 .
2πi |z|= z n+1 n!
This concludes the argument.
7 Verifying that a certain sum vanishes (B1 )

n n+1
X n X 1 a + bk k−n−1
.
m=0
m a + bk + 1 m k
k=0
Now we have
X m
a + bk m−q a + bk + 1
= (−1)
m q=0
q
m
X a + bk + 1
= (−1)m + (−1)m−q
q=1
q
11
and hence
m
(−1)m

1 a + bk X 1 m−q a + bk
= + (−1) .
a + bk + 1 m a + bk + 1 q=1 q q−1
Now from the first component we get in the main sum

n n+1
n X (−1)m

X k−n−1
m=0
m a + bk + 1 k
k=0
n+1 n
X 1 k−n−1 X n
= (−1)m = 0.
a + bk + 1 k m=0
m
k=0
We are thus left with the following sum:

n+1
X n m
k−n−1 X n X1 a + bk
(−1)m−q .
k m=0
m q=1 q q−1
k=0
Working with the inner sum we obtain

n Xm
X n 1 a + bk
(−1)m−q
m=1
m q=1 q q−1
n n
(−1)q a + bk X n
X
= (−1)m
q=1
q q − 1 m=q
m
n
X n − 1 1 a + bk
=
q=1
q−1 q q−1
n
1 X n a + bk
= .
n q=1 q q−1
Now put
Z
a + bk a + bk 1 1
= = a+bk−q+2
(1 + z)a+bk dz
q−1 a + bk − q + 1 2πi |z|= z
to get
Z n
1 1 1 a+bk
X n q
a+bk+2
(1 + z) z dz
n 2πi |z|= z q=1
q
Z
1 1 1
= (1 + z)a+bk (−1 + (1 + z)n ) dz
n 2πi |z|= z a+bk+2
The inner constant term does not contribute and we are left with
12
(1 + z)a+bk+n
Z
1 1 1 a + bk + n 1 a + bk + n
dz = = .
n 2πi |z|= z a+bk+2 n a + bk + 1 n n−1
Returning to the main sum we thus have

n+1
1 X k − n − 1 a + bk + n
n k n−1
k=0
n+1
X
1 −k a + b(n + 1) + n − bk
= .
n n+1−k n−1
k=0
Note that
n−k n−k
(−1)n−k+1 Y

−k 1 Y
= (−k − q) = (k + q)
n+1−k (n + 1 − k)! q=0 (n + 1 − k)! q=0
(−1)n−k+1

n! n
= = (−1)n−k+1 .
(n + 1 − k)! (k − 1)! k−1
This means for the main sum
n+1
(−1)n+1 X

n k a + b(n + 1) + n − bk
(−1)
n k−1 n−1
k=1
n
(−1)n X n

k a + bn + n − bk
= (−1) .
n k n−1
k=0
Introduce

a + bn + n − bk
Z
1 1
= (1 + z)a+bn+n−bk dz
n−1 2πi |z|= zn
We get for the sum

n
(−1)n 1
Z
1 a+bn+n
X n 1
n
(1 + z) (−1)k dz
n 2πi |z|= z k (1 + z)bk
k=0
n
(−1)n 1
Z
1 a+bn+n 1
= (1 + z) 1 − dz
n 2πi |z|= z n (1 + z)b
(−1)n 1 ((1 + z)b − 1)n
Z
1
= n
(1 + z)a+bn+n dz
n 2πi |z|= z (1 + z)bn
(−1)n 1
Z
1
= (1 + z)a+n ((1 + z)b − 1)n dz.
n 2πi |z|= z n
13
This is
(−1)n n−1
[z ](1 + z)a+n ((1 + z)b − 1)n .
n
Note however that
n
b n b b 2
((1 + z) − 1) = z+ z + ··· = bn z n + · · ·
1 2
so there is no coefficient on [z n−1 ] because the powered term starts at z n .

Therefore the end result of the whole calculation is
0.
Remark. We have made several uses of
X m
n n+1
= (−1)m−q .
m q=0
q
If this is not considered obvious we can prove it with the integral

Z
n+1 1 1
= (1 + z)n+1 dz
q 2πi |z|= z q+1
to get
Z m
1 1 X 1
(1 + z)n+1 (−1)m−q q dz
2πi |z|= z q=0
z
(−1)m 1 − (−1/z)m+1
Z
1
= (1 + z)n+1
2πi |z|= z 1 + 1/z
1 − (−1/z)m+1
Z
1
= (−1)m (1 + z)n+1 dz
2πi |z|= 1+z
Z
1
= (−1)m (1 + z)n (1 − (−1/z)m+1 ) dz
2πi |z|=

m m+1 n n
= −(−1) × (−1) = .
m m
8 Basic usage of exponentiation integral (B1 E)

p
X
pp q p!
(−1) (−1)q q p−r = .
q=r
q r r!
We use the integral representation
14
(1 + z)q
Z
q q 1
= = dz
r q−r 2πi |z|= z q−r+1
which is zero when q < r (pole vanishes) so we may extend q back to zero.
We also use the integral
(p − r)!
Z
exp(qw)
q p−r = p−r+1
dw.
2πi |w|=γ w
We thus obtain for the sum
(−1)p (p − r)!
Z
1
2πi w p−r+1
|w|=γ
p
(1 + z)q
Z
1 X p
× z r−1 (−1)q exp(qw) dz dw
2πi |z|= q=0
q zq
(−1)p (p − r)!
Z
1
= p−r+1
2πi |w|=γ w
Z p
1 1+z
× z r−1 1 − exp(w) dz dw
2πi |z|= z
(−1)p (p − r)!
Z
1
= p−r+1
2πi |w|=γ w
Z
1 1
× (− exp(w) + z(1 − exp(w)))p dz dw
2πi |z|= z p−r+1
(p − r)!
Z
1
= p−r+1
2πi |w|=γ w
Z
1 1
× (exp(w) + z(exp(w) − 1))p dz dw.
2πi |z|= z p−r+1
We extract the residue on the inner integral to obtain

(p − r)!
Z
1 p
p−r+1 p − r
exp(rw)(exp(w) − 1)p−r dw
2πi |w|=γ w
Z
p! 1 1
= exp(rw)(exp(w) − 1)p−r dw.
r! 2πi |w|=γ wp−r+1
It remains to compute
[wp−r ] exp(rw)(exp(w) − 1)p−r .
Observe that exp(w) − 1 starts at w so (exp(w) − 1)p−r starts at wp−r and
hence only the constant coefficient from exp(rw) contributes, the value being
one, which finally yields
p!
.
r!
15
9 Introductory example for the method, elimi-
nating odd-even dependence (B1 )
n
X n n−k k 2n + 1
2 = .
k bk/2c n
k=0
This is
n X n
X n n−2q 2q n n−2q−1 2q + 1
2 + 2 .
q=0
2q q q=0
2q + 1 q
We treat these in turn.

First sum. Observe that

n 2q n n−q
= .
2q q q q

n
X n n − q −2q
2n 2 .
q=0
q q
Introduce
(1 + z)n−q

n−q
Z
1
= dz
q 2πi |z|= z q+1
n
2n (1 + z)n X n −2q
Z
1
2 q (1 + z)q
dz
2πi |z|= z q=0
q z
n
2n (1 + z)n
Z
1
= 1+ dz
2πi |z|= z 4z(1 + z)
−n Z
(1 + 2z)2n

2 −n 2n n 2n
= dz = 2 2 = .
2πi |z|= z n+1 n n
Second sum. Observe that

n 2q + 1 n n−q
= .
2q + 1 q q q+1

n
X n n − q −2q
2n−1 2 .
q=0
q q+1
16
This time introduce
(1 + z)n−q

n−q
Z
1
= dz
q+1 2πi |z|= z q+2

n
2n−1 (1 + z)n X n −2q
Z
1
2
2 dz
2πi |z|= z q=0
q z (1 + z)q
q
n
2n−1 (1 + z)n
Z
1
= 1+ dz
2πi |z|= z2 4z(1 + z)
2−n−1 (1 + 2z)2n
Z
2n 2n
= dz = 2−n−1 2n+1 = .
2πi |z|= z n+2 n+1 n+1
Conclusion.
Collecting the two contributions we obtain

2n 2n 2n + 1
+ =
n n+1 n
as claimed.
10 Introductory example for the method, prov-

ing equality of two double hypergeometrics
(B1 )
Suppose we seek to verify that f1 (n, k) = f2 (n, k) where
n
X (2k + 2v)!
f1 (n, k) = 2−v
v=0
(k + v)! × v! × (2k + v)! × (n − v)!
and
bn/2c
X 1
f2 (n, k) = 2n−4m .
m=0
(k + m)! × m! × (n − 2m)!
Multiplying by (n + k)! we obtain

n
X n+k 2k + 2v
g1 (n, k) = 2−v
v=0
n−v v
and
17
bn/2c
X n+k n + k − m −4m
g2 (n, k) = 2n 2 .
m=0
m n − 2m
We will work with the latter two. Re-write the first sum as follows:
n
X n+k 2k + 2n − 2v v
2−n 2
v=0
v n−v
Introduce

2k + 2n − 2v
Z
1 1
= (1 + z)2k+2n−2v dz.
n−v 2πi |z|= z n−v+1
This integral is zero when v > n so we may extend v to infinity.
We get for g1 (n, k)
X n + k zv
Z
−n 1 1 2k+2n
2 (1 + z) 2v dz
2πi |z|= z n+1 v (1 + z)2v
v≥0
Z n+k
−n 1 1 2k+2n z
=2 n+1
(1 + z) 1+2 dz
2πi |z|= z (1 + z)2
Z
1 1 n+k
= 2−n n+1
1 + 4z + z 2 dz.
2πi |z|= z
For the second sum introduce

n+k−m
Z
1 1
= (1 + z)n+k−m dz.
n − 2m 2πi |z|= z n−2m+1
This is zero when 2m > n so we may extend m to infinity.
We get for g2 (n, k)
Z X n + k z 2m
n 1 1
2 (1 + z)n+k
2−4m dz
2πi |z|= z n+1 m (1 + z)m
m≥0
n+k
1 z2
Z
n1 1 n+k
=2 n+1
(1 + z) 1+ dz
2πi |z|= z 16 1 + z
Z n+k
n 1 1 1 2
=2 1+z+ z dz.
2πi |z|= z n+1 16
Finally put z = 4w in this integral to get
Z
1 1 n+k
2n n+1 n+1
1 + 4w + w2 4dw
2πi |w|= 4 w
Z
1 1 n+k
= 2−n n+1
1 + 4w + w2 dw.
2πi |w|= w
18
11 Evaluating a quadruple hypergeometric(B1 )
n X
n
X n+k−l k+l n n
(−1)k+l
n n k l
k=0 l=0
n n
X n X n+k−l k+l n
= (−1)k (−1)l .
k n n l
k=0 l=0
Evaluate the inner sum first and introduce

(1 + z)n+k−l

n+k−l
Z
1
= dz.
n 2πi |z|= z n+1
and
(1 + w)k+l
Z
k+l 1
= dw.
n 2πi |w|= wn+1
This yields for the inner sum
n
(1 + z)n+k 1 (1 + w)k X n (1 + w)l
Z Z
1
(−1)l dw dz
2πi |z|= z n+1 2πi |w|= w n+1 l (1 + z)l
l=0
n
(1 + z)n+k 1 (1 + w)k
Z Z
1 1+w
= 1− dw dz
2πi |z|= z n+1 2πi |w|= wn+1 1+z
(1 + z)k 1 (1 + w)k
Z Z
1
= (z − w)n dw dz.
2πi |z|= z n+1 2πi |w|= wn+1
Extracting the inner coefficient yields
n
X k n
(−1)n−q z q .
q=0
q n − q
The outer coefficient becomes

n
X k n n−q k
(−1)
q=0
q n−q n−q
n
X k n n−q k
= (−1) .
q=0
q q n−q
Call this S. By symmetry we have on re-indexing that

n
X k n k
2S = ((−1)q + (−1)n−q )
q=0
q q n−q
19
n
n
X k n q k
= (1 + (−1) ) (−1) .
q=0
q q n−q
This is zero when n is odd so the entire sum being evaluated vanishes when
n is odd and we may assume that n = 2m and get
2m
X k 2m k
(−1)q .
q=0
q q 2m − q
Substituting this into the outer sum yields

2m 2m
X 2m q
X 2m k k k
(−1) (−1) .
q=0
q k q 2m − q
k=0
We evaluate the inner sum with the integrals
(1 + z)k
Z
k 1
= dz.
q 2πi |z|= z q+1
and
(1 + w)k
Z
k 1
= dw
2m − q 2πi |w|= w2m−q+1
to get
Z Z 2m
1 1 1 1 X 2m
q+1
(−1)k (1 + z)k (1 + w)k dw dz
2πi |z|= z 2πi |w|= w2m−q+1 k
k=0
Z Z
1 1 1 1
= q+1 2m−q+1
(z + w + wz)2m dw dz
2πi |z|= z 2πi |w|= w
Z Z
1 1 1 1
= q+1 2m−q+1
(w(1 + z) + z)2m dw dz.
2πi |z|= z 2πi |w|= w
Extracting the coefficient we get for the inner term

2m
(1 + z)2m−q z q
2m − q
and for the outer integral
Z
2m 1 1 2m−q 2m
(1 + z) dz = .
2m − q 2πi |z|= z 2m − q
We are now ready to conclude and return to the main sum which has been
transformed into
2m
X 2m 2m
(−1)q
q=0
q 2m − q
20
which is
[v 2m ](1 − v)2m (1 + v)2m = [v 2m ](1 − v 2 )2m = [v m ](1 − v)2m

m 2m
= (−1) .
m
12 Evaluating another quadruple hypergeometric(B1 )

n −1
X
n+k 2k + 1 n n+k k k+m
(−1) = δmn .
n+k+1 k k m m
k=m
Here we may assume n ≥ m, the equality holds trivially otherwise.

Now we have
−1
n n+k n! k!n!
=
k k k!(n − k) (n + k)!
−1
n! n! 2n 2n
= = .
(n − k) (n + k)! n+k n
We get for the sum
n
X 2k + 1 2n k k+m 2n
(−1)n+k = δmn × .
n+k+1 n+k m m n
k=m
which is
n
X 2n + 1 k k+m
(−1)n+k (2k + 1)
n+k+1 m m
k=m

2n
= δmn × (2n + 1) × .
n
Introduce
Z
2n + 1 2n + 1 1 1
= = (1 + z)2n+1 dz.
n+k+1 n−k 2πi |z|= z n−k+1
Observe that this vanishes when k > n so we may extend k upward to
infinity.
Furthermore introduce
Z
k 1 1
= (1 + w)k dw.
m 2πi |w|=γ wm+1
21
Observe once again that the integral vanishes, this time when 0 ≤ k < m so
we may extend k back to zero.
We thus get for the sum
Z
n 1 1
(−1) n+1
(1 + z)2n+1
2πi |z|= z
Z
1 1 X k k+m k
× (−1) (2k + 1) z (1 + w)k dw dz.
2πi |w|=γ wm+1 m
k≥0
The inner sum yields two pieces, the first is

X k+m k 1
(−1)k z (1 + w)k =
m (1 + z + wz)m+1
k≥0
1 1
= m+1
.
(1 + z) (1 + wz/(1 + z))m+1
On extracting the residue for the integral in w we obtain
Z
1 1
(−1)n (1 + z)2n+1
2πi |z|= z n+1
zm

1 2m
× m+1
(−1)m dz
(1 + z) m (1 + z)m
Z
2m n+m 1
= (−1) n−m+1
(1 + z)2n−2m dz
m |z|= z

2m n+m 2n − 2m
= (−1) .
m n−m
The second piece from the sum is

X k+m k
2 (−1)k k z (1 + w)k .
m
k≥1
Write

k+m (k + m)! (k + m)!
k = = (m + 1)
m (k − 1)!m! (k − 1)!(m + 1)!

k+m
= (m + 1)
m+1
to get for the sum

X k + m k−1
2(m + 1)z(1 + w) (−1)k z (1 + w)k−1
m+1
k≥1
1
= −2(m + 1)z(1 + w)
(1 + z + wz)m+2
22
1 1
= −2(m + 1)z(1 + w) m+2
.
(1 + z) (1 + wz/(1 + z))m+2
Here we get two pieces, the first is
Z
1 z
−2(m + 1)(−1)n (1 + z)2n+1
2πi |z|= z n+1
zm

1 2m + 1
× m+2
(−1)m dz
(1 + z) m (1 + z)m
Z
2m + 1 n+m 1
= −2(m + 1) (−1) n−m
(1 + z)2n−2m−1 dz
m |z|= z
We have two cases, we get zero when n = m and when n > m we have

2m + 1 n+m 2n − 2m − 1
−2(m + 1) (−1) .
m n−m−1
The second piece is
Z
1 z
−2(m + 1)(−1)n n+1
(1 + z)2n+1
2πi |z|= z
z m−1

1 2m
× m+2
(−1)m−1 dz
(1 + z) m−1 (1 + z)m−1
Z
2m n+m 1
= 2(m + 1) (−1) n−m+1
(1 + z)2n−2m dz
m−1 |z|= z

2m n+m 2n − 2m
= 2(m + 1) (−1) .
m−1 n−m
Therefore when n = m we get

2n − 2m 2m 2m
(−1)m+n 2(m + 1) + .
n−m m−1 m
This simplifies to

2m 2m 2m
(−1) 2(m + 1) +
m−1 m

2m 2m 2m
= 2m + = (2m + 1) .
m m m
This is precisely the claim we were trying to prove. On the other hand when
n > m we obtain

2n − 2m
(−1)m+n
n−m

2m 2m 2m + 1 n − m
× 2(m + 1) + − 2(m + 1) .
m−1 m m 2n − 2m
23
The factor is

2m 2m + 1
(2m + 1) − (m + 1) = 0.
m m
Remark. For n = m we could have evaluated the single term in the initial
sum by expanding the four binomial coefficients and assumed n > m thereafter.
13 An identity by Strehl (B1 )

Suppose we seek to show that
n 3 n 2
X n X n 2k
= .
k k n
k=0 k=dn/2e
The RHS is
n
X n 2k k
.
k k n−k
k=dn/2e
Introduce
(1 + z)2k
Z
2k 1
= dz
k 2πi |z|= z k+1
and (this integral is zero when 0 ≤ k < dn/2e)
(1 + w)k
Z
k 1
= dw
n−k 2πi |w|= wn−k+1
to get for the RHS
n k
n w (1 + w)k (1 + z)2k
Z Z
1 1 1 1 X
dw dz
2πi |z|= z 2πi |w|= wn+1 k zk
k=0
n
w(1 + w)(1 + z)2
Z Z
1 1 1 1
= 1+ dw dz
2πi |z|= z 2πi |w|= wn+1 z
Z Z
1 1 1 1
= n+1 2πi n+1
(z + w(1 + w)(1 + z)2 )n dw dz
2πi |z|= z |w|= w
Z Z
1 1 1 1
= n+1 n+1
(z + w(z + 1))n (1 + w(z + 1))n dw dz.
2πi |z|= z 2πi |w|= w
Extracting first the residue in w in next the residue in z we get
Z n
1 1 X n n−q n
z (1 + z)q (1 + z)n−q dz
2πi |z|= z n+1 q=0
q n − q
24
n 2
(1 + z)n
Z
X n 1
= dz
q=0
q 2πi |z|= z q+1
n 3
X n
=
q=0
q
QED.
14 Ternary strings that avoid 22 (B1 )

Suppose we seek to enumerate ternary strings that avoid the string 22. Given
that the OP asked for inclusion-exclusion we present this method even though
it must be said that it is definitely not the simplest approach to this problem.
The computation is actually very similar to what we had at the following MSE
link.
The poset that we will use will have as its nodes the non-empty subsets S of
[n − 1] (call the set of these subsets P ) ordered by set inclusion. The elements
of a set S represent positions in the string where a 2 is followed by a 2. Let γ(S)
be the number of contiguous blocks of consecutive values that appear when we
list the elements of S in order. This has max |S| blocks and min one block.
Start from the inclusion-exclusion formula
X n−1
X X
(−1)|S| 3n−|S|−γ(S) = (−1)m 3n−m−γ(S)
S∈P,S6=∅ m=1 |S|=m
X n−m
n−1 X X
= (−1)m 3n−m−p .
m=1 p=1 |S|=m,γ(S)=p
Now to compute the cardinality of the innermost sum we use the generating
function
q
1 X w z w
v 1+v
1−z 1−w1−z 1−w
q≥0
where z represents spaces between blocks, w represents blocks and v counts

these. First extract [v p ], getting
p p−1
1 w z 1 w z w
+
1−z 1−w1−z 1−z 1−w1−z 1−w
wp zp wp z p−1
= + .
(1 − w)p (1 − z)p+1 (1 − w)p (1 − z)p
Extracting the coefficient on [wm ] we obtain
25
zp
p−1
m−p+p−1 m−p+p−1 z
p+1
+ .
p−1 (1 − z) p−1 (1 − z)p
The required coefficient on z is [z n−1−m ] and we obtain

m−1 n−1−m−p+p m − 1 n − 1 − m − (p − 1) + p − 1
+
p−1 p p−1 p−1

m−1 n−1−m m−1 n−1−m
= +
p−1 p p−1 p−1

m−1 n−m
= .
p−1 p
It follows that the desired answer is
n−1
X n−m
X m−1 n−m
3n + (−1)m 3n−m−p
m=1 p=1
p−1 p
n−1 n−m
X X m−1 n−m
= 3n + (−1)m 3n−m 3−p .
m=1 p=1
p−1 p
We are interested in a rational ordinary generating function of these numbers

so that we may establish a recurrence.
Note that we may extend the inner sum to p = 0 as the first binomial
coefficient vanishes there. Using

m−1
Z
1 1
= (1 + z)m−1 dz
p−1 2πi |z|= z p
we obtain for the inner sum
n−m
X
n − m −p −p
Z
1
(1 + z)m−1 3 z dz
2πi |z|= p=0
p
Z n−m
1 1
= (1 + z)m−1 1 + dz
2πi
|z|= 3z
Z
1 1 1
= n−m n−m
(1 + z)m−1 (1 + 3z)n−m dz.
3 2πi |z|= z
This yields for the outer sum
n−1
(1 + z)m−1 z m−1
Z
1 1 X
− n−1
(1 + 3z)n−1 (−1)m−1 dz
2πi |z|= z m=1
(1 + 3z)m−1
1 − (−1)n−1 (1 + z)n−1 z n−1 /(1 + 3z)n−1

Z
1 1
=− (1 + 3z)n−1 dz.
2πi |z|= z n−1 1 + (1 + z)z/(1 + 3z)
26
Fortunately the complicated term from the numerator of the fraction van-
ishes because the poles at z = 0 and z = −1/3 disappear by cancelation and we
may choose √ closer to the origin than the two poles of 1/(1 + 4z + z 2 ) which
are at −2 ± 3.
We are left with
Z
1 1 1
− (1 + 3z)n−1 dz
2πi |z|= z n−1 1 + (1 + z)z/(1 + 3z)
Z
1 1 1
=− (1 + 3z)n dz.
2πi |z|= z n−1 1 + 4z + z 2
To simplify this introduce
z w 1
=w or z = and dz = dw
1 + 3z 1 − 3w (1 − 3w)2
to get
Z
1 1 w 1 1
− dw
2πi |w|=γ w 1 − 3w 1 + 4w/(1 − 3w) + w /(1 − 3w) (1 − 3w)2
n 2 2
Z
1 1 1 1
=− dw
2πi |w|=γ wn−1 1 − 3w (1 − 3w) + 4w(1 − 3w) + w2
2
Z
1 1 1 1
=− dw.
2πi |w|=γ wn−1 1 − 3w 1 − 2w − 2w2
What we have here is
1 1 1 w2
−[wn−2 ] = −[w n
] .
1 − 3w 1 − 2w − 2w2 1 − 3w 1 − 2w − 2w2
It then follows that the generating function that solves the original problem
is given by
1 1 w2 1 1 − 2w − 2w2 − w2
− 2
=
1 − 3w 1 − 3w 1 − 2w − 2w 1 − 3w 1 − 2w − 2w2
1+w
= .
1 − 2w − 2w2
Substituting w = 1/v in the denominator we get 1 − 2/v − 2/v 2 = 0 or
2
v = 2v + 2 which is the characteristic equation of the corresponding recurrence
and we finally have
an+2 = 2an+1 + 2an

with a1 = 3 and a2 = 8.
27
15 Mixing the two types of binomial integrals
(B1 B2 )
b 2 2
X b n+j n
= .
j=0
j 2b b
where 0 ≤ b ≤ n.
Introduce Z
b 1 1 1
= dz
j 2πi |z|= z b−j+1 (1 − z)j+1
and Z
n+j 1 1
= 2b+1
(1 + w)n+j dw.
2b 2πi |w|= w
b
(1 + w)n 1 zj
Z Z
1 1 1 X b j
(1 + w) dz dw
2πi |w|= w2b+1 2πi |z|= z b+1 1 − z j=0 j (1 − z)j
b
(1 + w)n 1
Z Z
1 1 1 (1 + w)z
= 1+ dz dw
2πi |w|= w2b+1 2πi |z|= z b+1 1 − z 1−z
Z n Z
1 (1 + w) 1 1 1
= (1 + wz)b dz dw.
2πi |w|= w2b+1 2πi |z|= z b+1 (1 − z)b+1
The inner residue is
b
b q b−q+b
X
w .
q=0
q b
Substitute this into the outer integral to get

b
X b 2b − q n
.
q=0
q b 2b − q
Observe that

2b − q n (2b − q)! n!
=
b 2b − q b!(b − q)! (2b − q)!(n − 2b + q)!

(n − b)! n! n n−b
= = .
b!(b − q)! (n − b)!(n − 2b + q)! b b−q
28
Xb
n b n−b
.
b q=0 q b−q
which evaluates to 2
n
b
by inspection.
It can also be done with the integral
Z
1 1
(1 + z)n−b dz
2πi |z|= z b−q+1
which yields
Z b
n 1 1 n−b
X b q
(1 + z) z dz
b 2πi |z|= z b+1 q=0
q
Z 2
n 1 1 n n
= b+1
(1 + z) dz = .
b 2πi |z|= z b
A more general version of this identity is at section 37.
16 Two companion identities by Gould (B1 )

ρ
X 2x + 1 x − k
Q(x, ρ) =
2k ρ−k
k=0
where x ≥ ρ.
Introduce

x−k x−k
Z
1 1
= = (1 + z)x−k dz.
ρ−k x−ρ 2πi |z|= z x−ρ+1
Note that this controls the range being zero when ρ < k ≤ x so we can
extend the sum to x supposing that x > ρ. And when x = ρ we may also set
the upper limit to x.
We get for the sum
Z x
1 1 x
X 2x + 1 1
(1 + z) dz.
2πi |z|= z x−ρ+1 2k (1 + z)k
k=0
29
This is
Z 2x+1 2x+1 !
1 1 1 1 1
x−ρ+1
(1 + z)x 1+ √ + 1− √ dz
2 2πi |z|= z 1+z 1+z
√ √
Z
1 1 1 1
(1 + 1 + z)2x+1 + (1 − 1 + z)2x+1 dz.

= √
2 2πi |z|= z x−ρ+1
1+z
√
Observe that the second term in the parenthesis (i.e. 1 − 1 + z) has no
constant term and hence starts at z 2x+1 making for a zero contribution. This
leaves
√
Z
1 1 1 1
x−ρ+1
√ (1 + 1 + z)2x+1 dz.
2 2πi |z|= z 1+z
Now put 1 + z = w2 so that dz = 2w dw to get
Z
1 1 1
(1 + w)2x+1 w dw
2πi |w−1|= (w2 − 1)x−ρ+1 w
Z
1 1 1
= (1 + w)2x+1 dw
2πi |w−1|= (w − 1)x−ρ+1 (w + 1)x−ρ+1
Z
1 1
= (1 + w)x+ρ dw
2πi |w−1|= (w − 1)x−ρ+1
Z x+ρ
X x + ρ
1 1
= 2x+ρ−q (w − 1)q dw.
2πi |w−1|= (w − 1)x−ρ+1 q=0 q
This is
x+ρ
X
x + ρ x+ρ−q
[(w − 1)x−ρ ] 2 (w − 1)q
q=0
q

x + ρ x+ρ−(x−ρ) x + ρ 2ρ x + ρ 2ρ
= 2 = 2 = 2 .
x−ρ x−ρ 2ρ
We can also prove the companion identity from above. Suppose we seek to
evaluate
ρ
X 2x + 1 x − k
Q(x, ρ) =
2k + 1 ρ−k
k=0
where x ≥ ρ.
Introduce

x−k x−k
Z
1 1
= = x−ρ+1
(1 + z)x−k dz.
ρ−k x−ρ 2πi |z|= z
Note that this controls the range being zero when ρ < k ≤ x so we can
extend the sum to x supposing that x > ρ. And when x = ρ we may also set
the upper limit to x.
30
We get for the sum
Z x
1 1 x
X 2x + 1 1
(1 + z) dz.
2πi |z|= z x−ρ+1 2k + 1 (1 + z)k
k=0
This is
2x+1 2x+1 !
(1 + z)x √
Z
1 1 1 1
1+z 1+ √ − 1− √ dz
2 2πi |z|= z x−ρ+1 1+z 1+z
√ √
Z
1 1 1
(1 + 1 + z)2x+1 − (1 − 1 + z)2x+1 dz.

= x−ρ+1
2 2πi |z|= z
√
Observe that the second term in the parenthesis (i.e. 1 − 1 + z) has no
constant term and hence starts at z 2x+1 making for a zero contribution. This
leaves
√
Z
1 1 1
x−ρ+1
(1 + 1 + z)2x+1 dz.
2 2πi |z|= z
Now put 1 + z = w2 so that dz = 2w dw to get
Z
1 1
(1 + w)2x+1 w dw
2πi |w−1|= (w − 1)x−ρ+1
2
Z
1 1 1
= (1 + w)2x+1 w dw
2πi |w−1|= (w − 1)x−ρ+1 (w + 1)x−ρ+1
Z
1 1
= (1 + w)x+ρ w dw.
2πi |w−1|= (w − 1)x−ρ+1
Writing w = (w − 1) + 1 this produces two pieces, the first is
Z x+ρ
X x + ρ
1 1
2x+ρ−q (w − 1)q dw.
2πi |w−1|= (w − 1)x−ρ q=0 q
This is
x+ρ
X
x + ρ x+ρ−q
x−ρ−1
[(w − 1) ] 2 (w − 1)q
q=0
q

x+ρ x+ρ x + ρ 2ρ+1
= 2x+ρ−(x−ρ−1) = 22ρ+1 = 2 .
x−ρ−1 x−ρ−1 2ρ + 1
The second piece is
x+ρ
X
x + ρ x+ρ−q
[(w − 1)x−ρ ] 2 (w − 1)q
q=0
q
31

x + ρ x+ρ−(x−ρ) x + ρ 2ρ x + ρ 2ρ
= 2 = 2 = 2 .
x−ρ x−ρ 2ρ
Joining the two pieces we finally obtain

x−ρ x + ρ 2ρ
2× +1 × 2
2ρ + 1 2ρ

2x + 1 x + ρ 2ρ
= 2 .
2ρ + 1 2ρ
17 Exercise 1.3 from Stanley’s Enumerative Com-

binatorics (B1 B2 )
min(a,b)
X x+y+k x y x+a y+b
=
k b−k a−k b a
k=0
where we take y ≥ a and x ≥ b.

Observe that when we introduce the two integrals
Z
x x 1 1 1
= = dz
b−k x−b+k 2πi |z|= z b−k+1 (1 − z)x−b+k+1
and Z
y 1 1
= (1 + w)y dw
a−k 2πi |w|= wa−k+1
we get automatic range control so we may extend k to infinity to get for the
sum
(1 + w)y
Z Z
1 1 1 1
2πi |z|= z b+1 (1 − z) x−b+1 2πi |w|= wa+1
X x + y + k zk
× wk dw dz
k (1 − z)k
k≥0
(1 + w)y
Z Z
1 1 1 1
=
2πi |z|= z b+1 (1 − z)x−b+1 2πi |w|= wa+1
1
× dz dw
(1 − wz/(1 − z))x+y+1
(1 + w)y
Z Z
1 1 y+b 1
= (1 − z)
2πi |z|= z b+1 2πi |w|= wa+1
32
1
× dz dw.
(1 − z − wz)x+y+1
The integral in w is
(1 + w)y X x + y + q q
Z
1
z (1 + w)q dw
2πi |w|= wa+1 q
q≥0
X x + y + q y + q
= zq
q a
q≥0
which gives for the integral in z
X x + y + q y + q y + b
(−1)b−q .
q a b−q
q≥0
Note that

y+b y+q (y + b)! (y + q)!
=
b−q a (b − q)!(y + q)! a!(y + q − a)!
(y + b)! (y + b − a)!
=
(b − q)!(y + b − a)! a!(y + q − a)!

y+b y+b−a
= ,
a b−q
so we are done if we can show that
X x + y + q
b−q y + b − a x+a
(−1) = .
q b−q b
q≥0
To do this introduce
(1 + w)y+b−a
Z
1
dw
2πi |w|= wb−q+1
which once more provides range control so we get for the sum
(1 + w)y+b−a X x + y + q
Z
1 b
(−1) (−1)q wq dw
2πi |w|= wb+1 q
q≥0
(1 + w)y+b−a
Z
1 1
= (−1)b dw
2πi |w|= wb+1 (1 + w)x+y+1
Z
1 1 1
= (−1)b dw.
2πi |w|= wb+1 (1 + w)x−b+a+1
This yields

b b b+x−b+a x+a x+a
(−1) (−1) = =
x−b+a x−b+a b
as claimed.
33
18 Counting m-subsets (B1 I2 )
Permit me to contribute an algebraic proof.
n
X n n − 2q 2q 2n
2 = .
q=0
2q p−q 2p
Observe that the sum is

n
X n n−p+q q
4 .
q=0
p−q n−p−q
which is
p p
X n n−p+q q X n n−q
4 =4p
4−q .
q=0
p−q n−p−q q=0
q n + q − 2p
Introduce the Iverson bracket

zq
Z
1 1
[[0 ≤ q ≤ p]] = dz.
2πi |z|= z p+1 1−z
This provides range control so we may extend q to n.

(1 + w)n−q

n−q
Z
1
= dw.
n + q − 2p 2πi |w|= wn+q−2p+1

n
4p (1 + w)n 1
Z Z
1 1 X n q 1
z q 4−q dz dw
2πi |w|= wn−2p+1 2πi |z|= z p+1 1 − z q=0 q w (1 + w)q
n
4p (1 + w)n 1
Z Z
1 1 1
= 1+z dz dw
2πi |w|= wn−2p+1 2πi |z|= z p+1 1−z 4w(1 + w)
4p−n
Z Z
1 1 1 1 n
= (4w(1 + w) + z) dz dw.
2πi |w|= w2n−2p+1 2πi |z|= z p+1 1 − z
We evaluate the inner integral using the negative of the residue of the pole
at z = 1 which yields
4p−n
Z
1
(4w + 4w2 + 1)n dw
2πi |w|= w2n−2p+1
4p−n
Z
1
= (2w + 1)2n dw
2πi |w|= w2n−2p+1
34

2n 2n
= 4p−n 22n−2p = .
2n − 2p 2p
If we want to be rigorous we need to verify that the contribution from the
residue at infinity of the last integral in z is zero when n ≥ p. We get for the
residue
1 p+1 1
−Resz=0 z (4w(1 + w) + 1/z)n
z2 1 − 1/z
1
= −Resz=0 z p (4w(1 + w) + 1/z)n
z−1
1 1
= −Resz=0 n−p (4zw(1 + w) + 1)n .
z z−1
This is clearly zero when n = p. For n > p we obtain
n−p−1
X
n q q
4 w (1 + w)q .
q=0
q
This polynomial has degree 2n − 2p − 2 but the integral in w extracts the

coefficient on 2n − 2p for a zero contribution.
Addendum. We can use the same method to prove the companion identity
n
X n n − 2q − 1 2q+1 2n
2 = .
q=0
2q + 1 p−q 2p + 1
The sum is
n
X n n−p+q
22q+1
q=0
p − q n − p − q − 1
which is
p p
X n n−p+q X n n−q
22q+1 = 22p+1 2−2q .
q=0
p − q n − p − q − 1 q=0
q n + q − 2p − 1
Using exactly the same substitution as before we obtain the integral
22p+1−2n
Z Z
1 1 1 1 n
2n−2p p+1
(4w(1 + w) + z) dz dw.
2πi |w|= w 2πi |z|= z 1−z
This time we get from the residue at the pole z = 1

2p+1−2n 2n 2n−2p−1 2n
2 2 = .
2n − 2p − 1 2p + 1
For the residue at infinity we are extracting the coefficient on w2n−2p−1 but
the inner term has degree 2n − 2p − 2, again for a contribution of zero.
Addendum II. We can actually eliminate the Iverson bracket starting from
35
p
X n n−q
4 p
4−q .
q=0
q n + q − 2p
and observing that this is

p
X n n−q
4p 4−q .
q=0
q 2p − 2q
Now introduce Z
1 1
(1 + z)n−q dz
2πi |z|= z 2p−2q+1
This is zero when q > p so it provides the range control, which we have now
obtained without the Iverson bracket.
We get for the sum
X n z 2q
Z
p 1 1 n −q
4 (1 + z) 4 dz
2πi |z|= z 2p+1 q (1 + z)q
q≥0
n
1 z2
Z
1p 1 n
=4 2p+1
(1 + z) 1 + dz
2πi |z|= z 41+z
Z n
1 1 1 2
= 4p 1 + z + z dz
2πi |z|= z 2p+1 4
Now put z = 2w to get
Z
1 1 n
4p 2p+1 2p+1
1 + 2w + w2 2dw
2πi |z|= 2 w
Z
1 1
= (1 + w)2n dw.
2πi |z|= w2p+1
This is
2n
2p
as claimed. This was math.stackexchange.com problem 1430202.
19 Method applied to an iterated sum (B1 R)

!  
n−1 k n
X X n 
X n  = 1 n 2n .
q=0
q q 2 n
k=0 q=k+1
36
Using the integral representation
(1 + z)n
Z
n n 1
= = dz
q n−q 2πi |z|= z n−q+1
we get for the first factor

k
(1 + z)n X q (1 + z)n 1 − z k+1
Z Z
1 1
n+1
z dz = dz
2πi |z|= z q=0
2πi |z|= z n+1 1−z
(1 + z)n z k+1
Z
1
= 2n − dz
2πi |z|= z n+1 1 − z
and for the second factor
(1 + z)n z k+1 − z n+1 (1 + z)n z k+1
Z Z
1 1
dz = dz.
2πi |z|= z n+1 1−z 2πi |z|= z n+1 1 − z
These add to 2n as they obviously should.

Summing from k = 0 to n − 1 we get a positive and a negative piece. The
positive piece is
n−1
(1 + z)n X z k
Z
1
2n dz
2πi |z|= zn 1−z
k=0
(1 + z) 1 − z n
n
Z
1
= 2n dz
2πi |z|= zn (1 − z)2
(1 + z)n
Z
1 1
= 2n dz.
2πi |z|= zn (1 − z)2
The negative piece is
n−1
(1 + z1 )n 1 (1 + z2 )n X k k
Z Z
1
z1 z2 dz2 dz1
2πi |z1 |= z1n (1 − z1 ) 2πi |z2 |= z2n (1 − z2 )
k=0
(1 + z1 )n 1 (1 + z2 )n 1 − z1n z2n
Z Z
1
= dz2 dz1
2πi |z1 |= z1n (1 − z1 ) 2πi |z2 |= z2n (1 − z2 ) 1 − z1 z2
n
(1 + z2 )n
Z Z
1 (1 + z1 ) 1 1
= dz2 dz1 .
2πi |z1 |= z1n (1 − z1 ) 2πi |z2 |= z2n (1 − z2 ) 1 − z1 z2
We evaluate the inner integral by taking the sum of the negatives of the
residues of the poles at z2 = 1 and z2 = 1/z1 instead of computing the residue
of the pole at zero by using the fact that the residues sum to zero.
Re-write the integral as follows.
(1 + z2 )n
Z
1 1
dz2
2πi |z2 |= z2n (z2 − 1) z1 z2 − 1
37
(1 + z2 )n
Z
1 1 1
= dz2 .
z1 2πi |z2 |= z2n (z2 − 1) z2 − 1/z1
Now the negative of the residue at z2 = 1 is
1 n 1 1
− 2 = 2n .
z1 1 − 1/z1 1 − z1
Substituting this into the outer integral we get
(1 + z1 )n
Z
1
2n dz1 .
2πi |z1 |= z1n (1 − z1 )2
We see that this piece precisely cancels the positive piece that we obtained
first.
Continuing the negative of the residue at z2 = 1/z1 is
1 (1 + 1/z1 )n 1 (1 + z1 )n (1 + z1 )n
− = − = − .
z1 1/z1n × (1/z1 − 1) z1 (1/z1 − 1) (1 − z1 )
We now substitute this into the outer integral flipping the sign because this
was the negative piece to get
(1 + z1 )2n
Z
1
dz1 .
2πi |z1 |= z1n (1 − z1 )2
Extracting the residue at z1 = 0 we get

n−1
X n−1
X 2n
2n
(q + 1) = (q + 1)
q=0
n−1−q q=0
n+q+1
n−1
X n−1
X 2n
2n
= −n + (n + q + 1)
q=0
n+q+1 q=0
n+q+1
n−1
X 2n − 1
1 2n 1 2n
= −n 2 − + 2n
2 2 n q=0
n+q

1 2n 1 2n 1
= −n 2 − + 2n 22n−1
2 2 n 2

1 2n
= n .
2 n
Remark. If we want to do this properly we also need to verify that the
residue at infinity of the inner integral is zero. We use the formula for the
residue at infinity

1 1
Resz=∞ h(z) = Resz=0 − 2 h
z z
38
which in the present case gives for the inner term in z2
1 (1 + 1/z2 )n 1
−Resz2 =0 2 n
z2 1/z2 × (1 − 1/z2 ) 1 − z1 /z2
1 (1 + z2 )n 1
= −Resz2 =0
z22 (1 − 1/z2 ) 1 − z1 /z2
(1 + z2 )n 1
= −Resz2 =0
(z2 − 1) z2 − z1
which is zero by inspection.
20 A pair of two double hypergeometrics (B1 )

We seek to show that
X n + k − 1n + k X k − 1k + 1
(1 − x)2k+1 xn = xj .
k k j−1 j
n≥0 j≥0
Suppose we start by evalutating the two sums in turn, where the parameter
k ≥ 1. For the first we will be using the following integral representation:
(1 + z)n+k
Z
n+k 1
= dz.
k 2πi |z|= z k+1
We seek
X n − 1 + k n + k
xn .
k k
n≥1
Using the integral we find
1
Z X n − 1 + k (1 + z)n+k
xn dz
2πi |z|= k z k+1
n≥1
(1 + z)k X n − 1 + k
Z
1
= (1 + z)n xn dz
2πi |z|= z k+1 k
n≥1
x(1 + z)k+1 X n − 1 + k
Z
1
= (1 + z)n−1 xn−1 dz
2πi |z|= z k+1 k
n≥1
x(1 + z)k+1
Z
1 1
= dz
2πi |z|= z k+1 (1 − x(1 + z))k+1
x(1 + z)k+1
Z
1 1
= k+1
dz
2πi |z|= z (1 − x − xz))k+1
39
x(1 + z)k+1
Z
1 1 1
= dz
(1 − x)k+1 2πi |z|= z k+1 (1 − xz/(1 − x)))k+1
k q
x X k+1 q+k x
=
(1 − x)k+1 q=0 k − q k 1−x
k q
x X k+1 q+k x
= .
(1 − x)k+1 q=0 q + 1 k 1−x
Applying the integral representation from the beginning a second time we

obtain for this sum
k q
k + 1 (1 + z)q+k
Z
x 1 X x
dz
(1 − x)k+1 2πi |z|= q=0 q + 1 z k+1 1−x
k q
(1 + z)k X k + 1
Z
x 1 q x
= (1 + z) dz
(1 − x)k+1 2πi |z|= z k+1 q=0 q + 1 1−x
k q+1
(1 + z)k−1 X k + 1
Z
1 1 q+1 x
= (1 + z) dz
(1 − x)k 2πi
|z|= z k+1 q=0
q+1 1−x
k+1 !
(1 + z)k−1
Z
1 1 x
= −1 + 1 + (1 + z) dz.
(1 − x)k 2πi |z|= z k+1 1−x
We have k + 1 − (k − 1) = 2, so the first component inside the parentheses

drops out, leaving
k+1
(1 + z)k−1
Z
1 1 x
k
1 + (1 + z) dz
(1 − x) 2πi |z|= z k+1 1−x
(1 + z)k−1
Z
1 1 k+1
= (1 − x + x(1 + z)) dz
(1 − x)2k+1 2πi |z|= z k+1
(1 + z)k−1
Z
1 1 k+1
= (1 + xz) dz.
(1 − x)2k+1 2πi |z|= z k+1
We need one more simplification on this and put z = 1/w, getting
(1 + 1/w)k−1
Z
1 1 k+1 1
2k+1 k+1
(1 + x/w) dw
(1 − x) 2πi |w|= (1/w) w2
Z k+1
1 1 2 k−1 w+x 1
= w (w + 1) dw
(1 − x)2k+1 2πi |w|= w w2
(w + 1)k−1
Z
1 1
= 2k+1
(w + x)k+1 dw.
(1 − x) 2πi |w|= wk+1
40
The reson this works is because we are essentially evaluating the residue at
infinity and the residues sum to zero. This concludes the evaluation of the first
sum. For the second we will be using the following integral representation:
(1 + z)k−1

k−1
Z
1
= dz.
j−1 2πi |z|= zj
We seek
X k + 1k − 1
xj .
j j−1
j≥1
Using the integral we find
1
Z X k + 1 (1 + z)k−1
xj dz
2πi |z|= j zj
j≥1
1
Z X k + 1 xj
= (1 + z)k−1 dz
2πi |z|= j zj
j≥1
Z
1
(1 + z)k−1 −1 + (1 + x/z)k+1 dz.

=
2πi |z|=
The entire component drops out, leaving

Z
1
(1 + z)k−1 (1 + x/z)k+1 dz
2πi |z|=
(1 + z)k−1
Z
1
= (z + x)k+1 dz.
2πi |z|= z k+1
This however is precisely the integral that we had for the first sum without
the factor in front, done.
The only infinite sum appearing here is the first one with convergence when
|(1 + z)x| < 1. Therefore choosing |x| < 1/Q and |z| < 1/Q with Q ≥ 2 we have
|(Q + 1)/Q/Q| = |1/Q2 + 1/Q| < 1 and get convergence of the first LHS integral
in a neighborhood of zero.
21 A two phase application of the method (B1 )

We seek to show that
bn/3c n
X
k n+1 2n − 3k X n+1 k
(−1) = .
k n k n−k
k=0 k=bn/2c
41
Note that the second binomial coefficient in both sums controls the range of
the sum, so we can write our claim like this:
n+1
X n+1
X n + 1 k
n+1 k 2n − 3k
(−1) = .
k n − 3k k n−k
k=0 k=0
To evaluate the LHS introduce the integral representation
(1 + z)2n−3k

2n − 3k
Z
1
= dz.
n − 3k 2πi |z|= z n−3k+1
We can check that this really is zero when k > bn/3c.

This gives for the sum the representation
n+1 k
(1 + z)2n X n + 1 z3
Z
1 k
(−1) dz
2πi |z|= z n+1 k (1 + z)3
k=0
n+1
(1 + z)2n z3
Z
1
= 1− dz
2πi |z|= z n+1 (1 + z)3
Z
1 1 1 n+1
= 3z 2 + 3z + 1 dz
2πi |z|= z n+1 (1 + z)n+3
Z n+1
X n + 1
1 1 1
= 3q z q (1 + z)q dz
2πi |z|= z n+1 (1 + z)n+3 q=0 q
Z n+1
X
1 n + 1 q q−n−1
= 3 z (1 + z)q−n−3 dz
2πi |z|= q=0 q
Z n+1
X
1 n+1 q 1 1
= 3 n+1−q dz.
2πi |z|= q=0 q z (1 + z)n+3−q
Computing the residue we find

n+1
X
n+1 q n−q n − q + n + 2 − q
3 (−1)
q=0
q n+2−q
n+1
X
n+1 q 2n − 2q + 2
= 3 (−1)n−q .
q=0
q n−q+2
Now introduce the integral representation
(1 + z)2n−2q+2

2n − 2q + 2
Z
1
= dz
n−q+2 2πi |z|= z n−q+3
42
which gives for the sum the integral
n+1 q
(1 + z)2n+2 X n + 1 q
Z
1 n−q z
3 (−1) dz
2πi |z|= z n+3 q=0
q (1 + z)2
n+1
(1 + z)2n+2
Z
1 3z
=− − 1 dz
2πi |z|= z n+3 (1 + z)2
Z
1 1
=− (−1 + z − z 2 )n+1 dz.
2πi |z|= z n+3
Put w = −z which just rotates the small circle to get
Z
1 1
(−1 − w − w2 )n+1 dw
2πi |w|= (−w)n+3
Z
1 1
= (1 + w + w2 )n+1 dw.
2πi |w|= wn+3
We get for the final answer
[wn+2 ](1 + w + w2 )n+1
but we have 2n + 2 − n − 2 = n and thus exploiting the symmetry of 1 + w + w2

we get
[wn ](1 + w + w2 )n+1 .
To evaluate the RHS introduce the integral representation
(1 + z)k
Z
k 1
= dz.
n−k 2πi |z|= z n−k+1
This gives for the sum the representation

Z n+1
X
1 1 n+1 k
((1 + z)z) dz
2πi |z|= z n+1 k
k=0
Z
1 1
= (1 + z(1 + z))n+1 dz.
2πi |z|= z n+1
The answer is
[z n ](1 + z + z 2 )n+1 ,
the same as the LHS, and we are done.
43
22 An identity from Mathematical Reflections
(B1 )
b(m+n)/2c
X n m + n − 2k
(−1)k .
k n−1
k=0
Observe that in the second binomial coefficient we must have m + n − 2k ≥

n − 1 in order to avoid hitting the zero value in the product in the numerator
of the binomial coefficient, so the upper limit for the sum is in fact m + 1 ≥ 2k
with the sum being
b(m+1)/2c
X n m + n − 2k
(−1)k .
k n−1
k=0
Introduce
(1 + z)m+n−2k

m + n − 2k m + n − 2k
Z
1
= = dz.
n−1 m + 1 − 2k 2πi |z|= z m+2−2k
This integral correctly encodes the range for k being zero when k is larger
than b(m + 1)/2c. Therefore we may let k go to infinity in the sum and obtain
for n > m
(1 + z)m+n X n z 2k
Z
1
m+2
(−1)k dz
2πi |z|= z k (1 + z)2k
k≥0
n
(1 + z)m+n z2
Z
1
= 1− dz
2πi |z|= z m+2 (1 + z)2
Z
1 1
= (1 + 2z)n dz.
2πi |z|= (1 + z)n−m z m+2
This produces the closed form
m+1
X
n q m+1−q+n−m−1
2 (−1)m+1−q
q=0
q n−m−1
m+1
X
n n−q
= (−1)m+1 (−1)q 2q .
q=0
q n−m−1
This is
m+1
X
m+1 n q q n−q
(−1) (−1) 2 .
q=0
q m+1−q
44
Introduce
(1 + z)n−q

n−q
Z
1
= dz
m+1−q 2πi |z|= z m+2−q
which once more correctly encodes the range with the pole at z = 0 disap-
pearing when q > m + 1. Therefore we may extend the range to n to get
n
(−1)m+1 (1 + z)n X n zq
Z
q q
(−1) 2 dz
2πi |z|= z m+2 q=0 q (1 + z)q
n
(−1)m+1 (1 + z)n
Z
z
= 1−2 dz
2πi |z|= z m+2 1+z
(−1)m+1 (1 + z)n (1 − z)n
Z
= dz
2πi |z|= z m+2 (1 + z)n
m+1
(1 − z)n
Z
(−1)
= m+2
dz
2πi |z|= z

n n
= (−1)m+1 (−1)m+1
= .
m+1 m+1
23 A triple Fibonacci-binomial coefficient con-

volution (B1 )
Here is a proof using complex variables. We seek to show that
n n
X n n+k X n n+k
Fk+1 = (−1)n−k F2k+1 .
k k k k
k=0 k=0
Start from Z
n+k 1 1
= (1 + z)n+k dz.
k 2πi |z|=1 z k+1
This yields the following expression for the sum on the LHS
n
ϕk+1 − (−1/ϕ)k+1
Z
1 X n 1
k+1
(1 + z)n+k √ dz
2πi |z|=1 k z 5
k=0
This simplifies to
n k k !
(1 + z)n X n
Z
1 1 1+z 1 1 1+z
√ ϕ ϕ + − dz
5 2πi |z|=1 z k
k=0
z ϕ ϕ z
45
This finally yields
n n
(1 + z)n
Z
1 1 1+z 1 1 1+z
√ ϕ 1+ϕ + 1− dz
5 2πi |z|=1 z z ϕ ϕ z
or
n
(1 + z)n
Z
1 1 n 1 1
√ ϕ (z + ϕ(1 + z)) + z− (1 + z) dz
5 2πi |z|=1 z n+1 ϕ ϕ
Continuing we have the following expression for the sum on the RHS
n
ϕ2k+1 − (−1/ϕ)2k+1
Z
1 X n 1
(−1)n−k k+1 (1 + z)n+k √ dz
2πi |z|=1 k z 5
k=0
This simplifies to
(1 + z)n
Z
1 1
√
5 2πi |z|=1 z
n k k !
X n n−k 21 + z 1 1 1+z
× (−1) ϕ ϕ + dz
k z ϕ ϕ2 z
k=0
This finally yields

n n
(1 + z)n
Z
1 1 1+z 1 1 1+z
√ ϕ −1 + ϕ2 + −1 + 2 dz
5 2πi |z|=1 z z ϕ ϕ z
or
n
(1 + z)n
Z
1 1 2
n 1 1
√ ϕ −z + ϕ (1 + z) + −z + 2 (1 + z) dz
5 2πi |z|=1 z n+1 ϕ ϕ
Apply the substitution z = 1/w to this integral to obtain (the sign to correct
the reverse orientation of the circle is canceled by the minus on the derivative)
Z n
1 1 1
√ 1+ wn+1
5 2πi |w|=1 w
n n
1 2 1 1 1 1 1 1
× ϕ − + ϕ (1 + ) + − + 2 (1 + ) dw
w w ϕ w ϕ w w2
which is Z n
1 1 1 1
√ 1+
5 2πi |w|=1 w w
n
n 1 1
× ϕ −1 + ϕ2 (w + 1) + −1 + 2 (w + 1) dw
ϕ ϕ
which finally yields
(1 + w)n
Z
1 1
√
5 2πi |w|=1 wn+1
46
n
n 1 1
× ϕ −1 + ϕ2 (w + 1) + −1 + 2 (w + 1) dw
ϕ ϕ
This shows that the LHS is the same as the RHS because
−1 + ϕ2 (w + 1) = −1 + (1 + ϕ)(w + 1) = w + ϕ(w + 1)
and
1 1
−1 + 2
(w + 1) = −1 + (1 − )(w + 1)
ϕ ϕ
1 1
= −1 + (w + 1) − (w + 1) = w − (w + 1).
ϕ ϕ
24 Fibonacci numbers and the residue at infin-

ity (B2 R)
Suppose we seek to evaluate in terms of Fibonacci numbers
X n − pn − q
.
q p
p,q≥0
We use the integrals

n−p
Z
1 1
= dz
q 2πi |z|= (1 − z)q+1 z n−p−q+1
and
n−q
Z
1 1
= dw.
p 2πi |w|= (1 − w)p+1 wn−p−q+1
These correctly control the range so we may let p and q go to infinity to get
for the sum
z p+q wp+q
Z Z
1 1 1 1 X
dw dz
2πi |z|= (1 − z)z n+1 2πi |w|= (1 − w)w n+1 (1 − w)p (1 − z)q
p,q≥0
Z Z
1 1 1 1
=
2πi |z|= (1 − z)z n+1 2πi |w|= (1 − w)wn+1
1 1
× dw dz
1 − zw/(1 − w) 1 − zw/(1 − z)
Z Z
1 1 1 1 1 1
= dw dz
2πi |z|= z n+1 2πi |w|= wn+1 1 − w − zw 1 − z − zw
Z Z
1 1 1 1 1 1
= dw dz.
2πi |z|= z n+2 (1 + z) 2πi |w|= wn+1 w − 1/(1 + z) w − (1 − z)/z
47
We evaluate the inner integral using the fact that the residues of the function
in w sum to zero. We have two simple poles. We get for the first pole at
w = (1 − z)/z
z n+1 1 z n+1 z(1 + z)

n+1
= n+1
(1 − z) (1 − z)/z − 1/(1 + z) (1 − z) (1 − z)(1 + z) − z
z n+2 1+z
= .
(1 − z)n+1 (1 − z)(1 + z) − z
Substituting this expression into the outer integral we see that the pole at
z = 0 is canceled making for a contribution of zero.
For the second pole at w = 1/(1 + z) we get
1 z(1 + z)
(1 + z)n+1 = (1 + z)n+1 .
1/(1 + z) − (1 − z)/z z − (1 − z)(1 + z)
This yields the contribution (taking into account the sign flip from the sum
of residues)
Z
1 1 z(1 + z)
(1 + z)n+1 dz
2πi |z|= z n+2 (1 + z) 1 − z − z2
Z
1 1 1
= (1 + z)n+1 dz.
2πi |z|= z n+1 1 − z − z2
We evaluate this using again the fact that the residues sum to zero. There
are simple poles at z = −ϕ and z = 1/ϕ.
These yield
n+1 n+1
1−ϕ 1 1 + 1/ϕ 1
+
−ϕ −1 + 2ϕ 1/ϕ −1 − 2/ϕ
1 1 1
= √ 2n+2 − √ ϕ2n+2 .
5 ϕ 5
Taking into account the sign flip this is obviously Binet / de Moivre for
F2n+2 .

residue at infinity in both cases is zero. For example in the first application we
use the formula for the residue at infinity

1 1
z z
which in the present case gives for the inner term in w
1 n+1 1 1
−Resw=0 w
w2 1/w − 1/(1 + z) 1/w − (1 − z)/z
48
1 1
= −Resw=0 wn+1
1 − w/(1 + z) 1 − w(1 − z)/z
25 Permutations containing a given subsequence

(B1 I1 )
The WZ machinery is very powerful but it is also an incentive to evaluate these
sums manually e.g. by using the Egorychev method which I hope will make for
a rewarding read. Suppose we are trying to evaluate
n
X r + n − 1 3n − r
.
r=0
n−1 n
Put
(1 + w)3n−r

3n − r
Z
1
= dw
n 2πi |w|= wn+1
and furthermore introduce
1 + z + z2 + · · · + zn
Z
1
[[0 ≤ r ≤ n]] = dz
2πi |z|= z r+1
z n+1 − 1
Z
1
= dz.
2πi |z|= (z − 1)z r+1
This second integral controls the range so that we may extend the sum to
infinity to get
∞
(1 + w)3n 1 z n+1 − 1 X r + n − 1
Z Z
1 1
dz dw.
2πi |w|= wn+1 2πi |z|= (z − 1)z r=0 n−1 z (1 + w)r
r
This simplifies to
(1 + w)3n 1 z n+1 − 1
Z Z
1 1
dz dw
2πi |w|= wn+1 2πi |z|= (z − 1)z (1 − 1/z/(1 + w))n
(1 + w)3n 1 z n+1 − 1 zn
Z Z
1
= dz dw.
2πi |w|= wn+1 2πi |z|= (z − 1)z (z − 1/(1 + w))n
Computing the contributions from the pole at z = 1/(1 + w) (there is no
pole at zero anymore) we get two pieces, call them A and B, A is
(1 + w)3n 1 z n−1
Z Z
1 1
n+1
dz dw.
2πi |w|= w 2πi |z|= 1 − z (z − 1/(1 + w))n
49
We have
1 1 1
= =
1−z 1 − 1/(w + 1) − (z − 1/(w + 1)) w/(w + 1) − (z − 1/(w + 1))
1 1
=
w/(w + 1) 1 − (z − 1/(1 + w))/(w/(w + 1))
w+1 1
= .
w 1 − (z − 1/(1 + w))(w + 1)/w
This gives for the inner integral from A
 
Z q
w+1 1  (z − 1/(1 + w))q (w + 1) 
X
w 2πi |z|= wq
q≥0
n−1
!
X n−1

1 1
p
× (z − 1/(1 + w)) n−1−p
dz.
p=0
p (1 + w) (z − 1/(1 + w))n
Computing the residue we get

n−1
w + 1 X n − 1 (1 + w)n−1−p

1
w p=0 p wn−1−p (1 + w)n−1−p
n−1
w+1 X n−1 1
= n−1−p
w p=0 p w
n−1
(1 + w)n

w+1 1
= 1+ = .
w w wn
Substituting this into A we get
(1 + w)3n (1 + w)n (1 + w)4n

Z Z
1 1 4n
dw = dw = .
2πi |w|= wn+1 wn 2πi |w|= w2n+1 2n
Returning to B which is
(1 + w)3n 1 z 2n
Z Z
1 1
− n+1
dz dw.
2πi |w|= w 2πi |z|= 1 − z (z − 1/(1 + w))n
we get for the inner integral

 
Z q
w+1 1 X
 (z − 1/(1 + w))q (w + 1) 
w 2πi |z|= wq
q≥0
2n !
X 2n p 1 1
× (z − 1/(1 + w)) dz.
p=0
p (1 + w)2n−p (z − 1/(1 + w))n
50
Computing the residue we get
n−1
w + 1 X 2n (1 + w)n−1−p

1
w p=0 p wn−1−p (1 + w)2n−p
n−1
X 2n
w+1 1
= wp
w (1 + w)n+1 × wn−1 p=0 p
n−1
X 2n
1
= wp .
(1 + w)n wn p=0 p
Substituting this into B we get

n−1
(1 + w)3n X 2n
Z
1 1
− wp dw
2πi |w|= wn+1 (1 + w)n wn p=0 p
n−1
(1 + w)2n X 2n p
Z
1
=− w dw.
2πi |w|= w2n+1 p=0 p
This is
n−1
X
2n 2n
− .
p=0
p 2n − p
By symmetry this evaluates to

2 X 2n !
1 2n 2n 2n
− − + .
2 n p=0
p 2n − p
This last sum is

4n
[x2n ](1 + x)2n (1 + x)2n = [x2n ](1 + x)4n = .
2n
so we finally have for piece B the value
2 !
1 2n 4n
− − + .
2 n 2n
We may now conclude, collecting the contributions from A and B to obtain

2 ! 2 !
4n 1 2n 4n 1 4n 2n
− − + = + .
2n 2 n 2n 2 2n n
Addendum. The calculation above is not quite rigorous in that the infinite
series that appears does not converge in a neighborhood of zero for z. This can
be fixed by using a different Iverson bracket.
51
Suppose as before that we are trying to evaluate
n
X r + n − 1 3n − r
S=
r=0
n−1 n
which is
2n 2n
X r + n − 1 3n − r X r + n − 1 3n − r
S2 − S1 = − .
r=0
n−1 n r=n+1
n−1 n
Start by evaluating S2 . Put

3n − r
Z
1 1 1
= 2n−r+1
dw.
n 2πi |w|= w (1 − w)n+1
and use the following Iverson bracket

zr
Z
1 1
[[0 ≤ r ≤ 2n]] = 2n+1
dz.
2πi |z|= z 1−z
This second integral controls the range so that we may extend the sum to
infinity to get
∞
1 X r+n−1 r r
Z Z
1 1 1 1 1
z w dz dw.
2πi |w|= w2n+1 (1 − w)n+1 2πi |z|= z 2n+1 1 − z r=0 n−1
This simplifies to
Z Z
1 1 1 1 1 1 1
dz dw.
2πi |w|= w2n+1 (1 − w)n+1 2πi |z|= z 2n+1 1 − z (1 − wz)n
We evaluate the inner integral using the fact that the residues at the three
poles sum to zero. The residue at z = 0 is the sum S2 which we are trying to
compute. The residue at z = 1 yields
Z
1 1 1 2n + 2n 4n
− dw = − =− .
2πi |w|= w2n+1 (1 − w)2n+1 2n 2n
For the residue at z = 1/w re-write the integral as follows:
(−1)n
Z Z
1 1 1 1 1 1
3n+1 n+1 2n+1
dz dw.
2πi |w|= w (1 − w) 2πi |z|= z 1 − z (z − 1/w)n
We require a derivative which we compute using Leibniz’ rule:

(n−1)
1 1 1
(n − 1)! z 2n+1 1 − z
52
n−1
X n − 1
1 (2n + q)! (n − 1 − q)!
= (−1)q
(n − 1)! q=0 q (2n)! × z 2n+1+q (1 − z)1+n−1−q
n−1
X
2n + q 1 1
= (−1)q 2n+1+q .
q=0
2n z (1 − z)n−q
Evaluate at z = 1/w to get

n−1
X wn−q

2n + q
(−1)q w2n+1+q .
q=0
2n (w − 1)n−q
Substitute this back into the integral in w to obtain

n−1
X (−1)n w3n+1
Z
2n + q 1 1
(−1)q 3n+1 n+1
dw
q=0
2n 2πi |w|= w (1 − w) (w − 1)n−q
n−1
X Z
2n + q 1 1
= dw = 0.
q=0
2n 2πi |w|= (1 − w)2n−q+1
We have shown that

4n
S2 = .
2n
Continuing with S1 we see that
n−1
X n−1
X
r + 2n 2n − 1 − r r + 2n 2n − 1 − r
S1 = = .
r=0
n−1 n r=0
n−1 n−1−r
For this sum no Iverson bracket is needed as the second binomial coefficient
controls the range via the following integral:
(1 + w)2n−1−r

2n − 1 − r
Z
1
= dw.
n−1−r 2πi |w|= wn−r
(1 + z)r+2n
Z
r + 2n 1
= dz.
n−1 2πi |z|= zn
This gives the integral
(1 + w)2n−1 1 (1 + z)2n X wr (1 + z)r

Z Z
1
dz dw
2πi |w|= wn 2πi |z|= zn (1 + w)r
r≥0
(1 + w)2n−1 1 (1 + z)2n
Z Z
1 1
= dz dw
2πi |w|= wn 2πi |z|= z n 1 − w(1 + z)/(1 + w)
53
(1 + w)2n 1 (1 + z)2n
Z Z
1 1
= dz dw
2πi |w|= wn 2πi |z|= zn 1 + w − w(1 + z)
2n
(1 + z)2n
Z Z
1 (1 + w) 1 1
= dz dw.
2πi |w|= wn 2πi |z|= zn 1 − wz
Extracting the inner residue we obtain
n−1
X
2n
wq
q=0
n−1−q
which yields
n−1
X (1 + w)2n
Z
2n 1
dw
q=0
n − 1 − q 2πi |w|= wn−q
n−1
X
2n 2n
= .
q=0
n−1−q n−1−q
This is
n−1
X 2
2n
q=0
q
which may be evaluated by inspection as in the first version and we are done.
Remark. To be fully rigorous we must also show that the residue at infinity
of
Z
1 1 1 1
dz
2πi |z|= z 2n+1 1 − z (1 − wz)n
is zero. Recall the formula for the residue at infinity

1 1
z z
which in this case yields

1 2n+1 1 1
−Resz=0 2
z
z 1 − 1/z (1 − w/z)n
1 1
= −Resz=0 z 2n
z − 1 (1 − w/z)n
1 1
= −Resz=0 z 3n
z − 1 (z − w)n
54
26 Catalan numbers and Lagrange inversion (B1 )
Suppose we seek the series for
1 p
1 − x − y − 1 − 2x − 2y − 2xy + x2 + y 2 .
2
Introduce u = xy and v = x + y to get
1 p 1 p
1 − v − 1 − 2v + v 2 − 4u = 1 − v − (1 − v)2 − 4u .
2 2
Lagrange inversion now asks us to compute the integral
Z
1 1 1 p
2 − 4u du.
1 − v − (1 − v)
2πi |u|= un+1 2
Put w2 = (1 − v)2 − 4u so that 4u = (1 − v)2 − w2 and du = − 12 w dw.

This yields
4n+1
Z
1 1
− (1 − v − w)w dw
4 2πi |w−(1−v)|= ((1 − v)2 − w2 )n+1
4n
Z
1 1
=− w dw.
2πi |w−(1−v)|= ((1 − v) − w) ((1 − v) + w)n+1
n
This has two pieces, piece B1 is

4n
Z
1 1
− dw
2πi |w−(1−v)|= ((1 − v) − w) ((1 − v) + w)n
n
and piece B2 is
4n
Z
1 1
(1 − v) dw
2πi |w−(1−v)|= ((1 − v) − w)n ((1 − v) + w)n+1
Observe that
1 1
=
((1 − v) + w)n (w − (1 − v) + 2(1 − v))n
1 1
=
2n (1
− v) ((w − (1 − v))/2/(1 − v) + 1)n
n
X q + n − 1 q
1 q (w − (1 − v))
= n (−1) .
2 (1 − v)n n−1 2q (1 − v)q
q≥0
It follows that B1 is

n 1
n 2n − 2 1
−4 (−1) n (−1)n−1 n−1
2 (1 − v)n n − 1 2 (1 − v)n−1
55

2n − 2 1
=2 .
n − 1 (1 − v)2n−1
Similarly we have for B2 the series expansion
X q + n q
1 q (w − (1 − v))
(−1) .
2n+1 (1 − v)n+1 n 2q (1 − v)q
q≥0
It follows that B2 is

n n 1 2n − 1 1
(1 − v)4 (−1) n+1 (−1)n−1 n−1
2 (1 − v)n+1 n 2 (1 − v)n−1

2n − 1 1
=− .
n (1 − v)2n−1
Collecting the two contributions we finally obtain for the two pieces

2n − 1 2n − 2 1 1 2n − 2 1
− +2 2n−1
= .
n n − 1 (1 − v) n n − 1 (1 − v)2n−1
We recognize the Catalan numbers here which are OEIS A000108.

We have thus obtained the following series
X 1 2n − 2 1
un .
n n − 1 (1 − v)2n−1
n≥1
Observe that
1 1 1
=
(1 − x − y)2n−1 (1 − x)2n−1 (1 − y/(1 − x))2n−1
1 X q + 2n − 2 y q X q + 2n − 2 yq
= 2n−1 q
= .
(1 − x) 2n − 2 (1 − x) 2n − 2 (1 − x)q+2n−1
q≥0 q≥0
ν
It follows that the coefficient on [y ] is
ν
X 1 2n − 2 ν + n − 2
n 1
x .
n=1
n n−1 2n − 2 (1 − x)ν+n−1
This yields for the coefficient on [xµ y ν ]

min(µ,ν)
X 1 2n − 2 ν + n − 2 µ + ν − 2
.
n=1
n n−1 2n − 2 ν+n−2
The inner term here is

1 (2n − 2)! (ν + n − 2)! (µ + ν − 2)!
n (n − 1)! × (n − 1)! (2n − 2)! × (ν − n)! (ν + n − 2)! × (µ − n)!
56
which is
1 1 1 1
(µ + ν − 2)!
n (n − 1)! × (n − 1)! (ν − n)! (µ − n)!
1 1 1 1 1
= (µ + ν − 1)!
µ+ν−1 n (n − 1)! × (n − 1)! (ν − n)! (µ − n)!

1 µ+ν−1
= .
µ + ν − 1 n, n − 1, µ − n, ν − n
It remains to evaluate the sum in the multinomial coefficient since we already
have the correct factor in front. This gives
min(µ,ν) Z Z Z
X 1 1 1 1 1 1
n=1
2πi |z1 |= z1n 2πi |z2 |=
µ−n+1 2πi
z2 ν−n+1
|z3 |= z3
×(1 + z1 + z2 + z3 )µ+ν−1 dz3 dz2 dz1 .

Observe that the poles in z2 and z3 disappear when n > µ or n > ν so they
effectively control the range and we may extend the sum to infinity, obtaining
Z Z Z
1 1 1 1 1
µ+1 ν+1
2πi |z1 |= 2πi |z2 |= z2 2πi |z3 |= z3
X znzn
2 3
×(1 + z1 + z2 + z3 )µ+ν−1 dz3 dz2 dz1 .
z1n
n≥1
This is Z Z Z
1 1 1 1 1
2πi |z1 |= 2πi |z2 |= z2µ+1 2πi |z3 |= z3ν+1
z2 z3 /z1
×(1 + z1 + z2 + z3 )µ+ν−1 dz3 dz2 dz1 .
1 − z2 z3 /z1
or Z Z Z
1 1 1 1 1
µ
2πi |z1 |= 2πi |z2 |= z2 2πi |z3 |= z3ν
1/z1
×(1 + z1 + z2 + z3 )µ+ν−1 dz3 dz2 dz1 .
1 − z2 z3 /z1
or Z Z Z
1 1 1 1 1
µ
2πi |z1 |= 2πi |z2 |= z2 2πi |z3 |= z3ν
1
×(1 + z1 + z2 + z3 )µ+ν−1 dz3 dz2 dz1 .
z1 − z2 z3
First treat the pole at z1 = z2 z3 to get
Z Z
1 1 1 1
µ (1 + z2 z3 + z2 + z3 )µ+ν−1 dz3 dz2
2πi |z2 |= z2 2πi |z3 |= z3ν
57
Z Z
1 1 1 1
= (1 + z2 )µ+ν−1 (1 + z3 )µ+ν−1 dz3 dz2 .
2πi |z2 |= z2µ 2πi |z3 |= z3ν
This factors into
Z Z
1 1 µ+ν−1 1 1
µ (1 + z2 ) dz2 × (1 + z3 )µ+ν−1 dz3 .
2πi |z2 |= z2 2πi |z3 |= z3ν
Extracting coefficients from this we obtain

µ+ν−1 µ+ν−1
.
µ−1 ν−1
This concludes the argument since we have established the value

1 µ+ν−1 µ+ν−1
µ+ν−1 µ−1 ν−1
which may be rewritten as

1 µ+ν−1 µ+ν−1
.
µ+ν−1 ν µ
27 A binomial coefficient - Catalan number con-

volution (B1 )
n+1
X 1 2r m + n − 2r m+n
= .
r=1
r+1 r n+1−r n
We will assume familiarity with the generating function of the Catalan num-
bers (which seems like a reasonable assumption). This generating function is
given by √
X 1 2r 1 − 1 − 4z
r
z =
r+1 r 2z
r≥0
so that √
1 1 − 1 − 4z
Z
1 2r 1
= dz.
r+1 r 2πi |z|= z r+1 2z
(1 + w)m+n−2r

m + n − 2r
Z
1
= dw.
n+1−r 2πi |w|= wn+2−r
Observe carefully that this last integral is zero when r > n + 1, so we may
extend the range of the sum to infinity.
58
√
(1 + w)m+n 1 1 1 − 1 − 4z X wr
Z Z
1
n+2
dz dw
2πi |w|= w 2πi |z|= z 2z (1 + w)2r z r
r≥1
√
(1 + w)m+n 1 1− 1 − 4z w/(1 + w)2 /z
Z Z
1
= dz dw
2πi |w|= wn+2 2πi |z|= 2z 2 1 − w/(1 + w)2 /z
m+n
√
1 − 1 − 4z
Z Z
1 (1 + w) 1 1
= dz dw.
2πi |w|= wn+2 2πi |z|= 2z 2 z(1 + w)2 /w − 1
Observe that with the principal branch of the logarithm
√
1 − 1 − 4z = 2z + 2z 2 + 4z 3 + · · ·
and
1 (1 + w)2 2 (1 + w)
4
= −1 − z − z − ··· .
z(1 + w)2 /w − 1 w w2
so that the contribution from the pole at z = 0 is
(1 + w)m+n 1
Z
1 m+n
× (−2) dw = − .
2πi |w|= wn+2 2 n+1
On the other hand the contribution from the simple pole at z = w/(1 + w)2
which is inside the contour is
p
(1 + w)m+n 1 − 1 − 4w/(1 + w)2
Z
1 w
dw
2πi |w|= wn+2 2w2 /(1 + w)4 (1 + w)2
p
(1 + w)m+n−2 (1 + w)4 − (1 + w)3 (1 + w)2 − 4w
Z
1
= dw
2πi |w|= wn+1 2w2
(1 + w)m+n−2
Z
1
= ((1 + w)4 − (1 − w)(1 + w)3 ) dw
2πi |w|= 2wn+3
(1 + w)m+n−2
Z
1
= (1 + w)3 × (2w) dw
2πi |w|= 2wn+3
(1 + w)m+n+1
Z
1
= dw.
2πi |w|= wn+2
which yields
m+n+1
.
n+1
Collecting the two contributions we obtain

m+n+1 m+n m+n+1 m m+n
− = −
n+1 n+1 n+1 n+1 n
59

m+n
=
n
as claimed.
Addendum. In fact the above admits considerable simplification.
Write n+1
m+n X 1 2rm + n − 2r
− +
n+1 r=0
r+1 r n+1−r
and use the same integral as before for the binomial coefficient to obtain
(1 + w)m+n X 1 wr
Z
1 2r
n+2
dw
2πi |w|= w r + 1 r (1 + w)2r
r≥0
which becomes
p
(1 + w)m+n 1 − 1 − 4w/(1 + w)2
Z
1
dw
2πi |w|= wn+2 2w/(1 + w)2
p
(1 + w)m+n 1 + w − (1 + w)2 − 4w
Z
1
= dw
2πi |w|= wn+2 2w/(1 + w)
(1 + w)m+n+1
Z
1 1 p
2

= 1 + w − (1 − w) dw.
2 2πi |w|= wn+3
Now with w in a neighborhood of zero w − 1 would produce the negative
root so the root is in fact 1 − w which finally yields
(1 + w)m+n+1
Z
1 1
(2w) dw
2 2πi |w|= wn+3
(1 + w)m+n+1
Z
1
= dw
2πi |w|= wn+2

m+n+1
= .
n+1
We may then conclude as before. This was math.stackexchange.com problem
563307.
28 A new obstacle from Concrete Mathematics

(Catalan numbers) (B1 )
X n + k 2k (−1)k
m + 2k k k+1
k≥0
60
where m, n ≥ 0. In fact we may assume that n ≥ m because if m > n
when counting down from the non-negative value n + k with m + 2k terms we
invariably hit zero and the sum vanishes.
2n−m+q

Furthermore observe that when k = n−m+q with q > 0 we obtain 2n−m+2q
which is zero by the same argument. This gives
n−m
X 2k (−1)k

n+k
.
n−m−k k k+1
k=0
Introduce
Z
n+k 1 1
= (1 + z)n+k dz.
n−m−k 2πi |z|= z n−m−k+1
Observe that this is zero when k > n − m so we may extend k to infinity to

get for the sum
1
Z
1 X 2k (−1)k
n
(1 + z) z k (1 + z)k dz.
2πi |z|= z n−m+1 k k+1
k≥0
Here we recognize the generating function of the Catalan numbers

√
X 2k 1 1 − 1 − 4w
k
w =
k k+1 2w
k≥0
where the branch cut of the logarithm is on the negative real axis and hence
the branch cut of the square root term is (1/4, ∞) so we certainly have analyt-
icity in a neighborhood of zero. We obtain
p
n1 −
Z
1 1 1 + 4z(1 + z)
− n−m+1
(1 + z) dz
2πi |z|= z 2z(1 + z)
Z
1 1 1 n−1
p
2

=− (1 + z) 1 − (1 + 2z) dz.
2 2πi |z|= z n−m+2
Now with z in a neighborhood of zero the square root produces the positive
root so we finally have
Z
1 1 1
− n−m+2
(1 + z)n−1 (−2z) dz
2 2πi |z|= z
Z
1 1
= (1 + z)n−1 dz
2πi |z|= z n−m+1
n−1

which evaluates by inspection to n−m which is

n−1
.
m−1
This problem has not yet appeared at math.stackexchange.com.
61
29 Reducing the form of a double hypergeomet-
ric (B1 )
n−2 n
XX k+q 2n − q − k − 1
S(n) = .
q=0 k=1
k n−k+1
Introduce
(1 + z)k+q
Z
k+q 1
= dz
k 2πi |z|= z k+1
and
(1 + w)2n−q−k−1

2n − q − k − 1
Z
1
= dw.
n−k+1 2πi |w|= wn−k+2
n−2 n
(1 + z)k+q (1 + w)2n−q−k−1
Z Z
1 1 XX
dw dz.
2πi |z|= 2πi |w|= q=0 k=1 z k+1 wn−k+2
Initial sum simplification. The inner sums simplify to

n n−2
X (1 + z)k (1 + w)2n−k−1 X (1 + z)q
z k+1 wn−k+2 q=0
(1 + w)q
k=1
n
(1 + w)2n−1 (1 + z)n−1 /(1 + w)n−1 − 1 X (1 + z)k wk
= n+2 k
zw (1 + z)/(1 + w) − 1 z (1 + w)k
k=1
n
(1 + w)2n−1 (1 + z)n /(1 + w)n−1 − (1 + z) X (1 + z)k−1 wk
= 2 n+2 k−1
z w (1 + z)/(1 + w) − 1 z (1 + w)k
k=1
n
(1 + w)2n−1 (1 + z)n /(1 + w)n−1 − (1 + z) X (1 + z)k−1 wk−1
=
z 2 wn+1 (1 + z) − (1 + w) z k−1 (1 + w)k−1
k=1
2n−1 n n−1
(1 + w) (1 + z) /(1 + w) − (1 + z) (1 + z)n wn /z n /(1 + w)n − 1
=
z 2 wn+1 z−w (1 + z)w/z/(1 + w) − 1
(1 + w)2n (1 + z)n /(1 + w)n−1 − (1 + z) (1 + z)n wn /z n /(1 + w)n − 1
=
zwn+1 z−w (1 + z)w − (1 + w)z
(1 + w)2n (1 + z)n /(1 + w)n−1 − (1 + z) (1 + z)n wn /z n /(1 + w)n − 1
= .
zwn+1 z−w w−z
This has four components, the first is
1 (1 + w)(1 + z)2n wn
− .
zwn+1 (z − w)2 zn
62
The second component is
1 (1 + w)n (1 + z)n+1 wn
.
zwn+1 (z − w)2 zn
The third component is
1
(1 + w)n+1 (1 + z)n .
zwn+1 (z − w)2
The fourth component is
1
− (1 + w)2n (1 + z).
zwn+1 (z − w)2
We need to set the order of integration before we proceed. We will use the
integral in w as the inner integral and the one in z as the outer one. Having
fixed the order we can no longer make use of the symmetry in the first and
fourth and the second and third component.
First component.
Re-write the first component as
1 1
− (1 + w)(1 + z)2n = − n+3 (1 + w)(1 + z)2n .
wz n+1 (z − w)2 wz (1 − w/z)2
This yields for the residue at w = 0 the value
(1 + z)2n
− .
z n+3
For the pole at w = z we get
(1 + z)2n (1 + z)2n

1
− − = ,
z n+1 z2 z n+3
which cancels the previous contribution for a result of zero.
Second component.
Re-write the second component as
1 1
(1 + w)n (1 + z)n+1 = (1 + w)n (1 + z)n+1 .
wz n+1 (z − w)2 wz n+3 (1 − w/z)2
We get for the residue at w = 0 the value
(1 + z)n+1
z n+3
again for a contribution of zero.
The residue at w = z is
(1 + z)n+1

1 n 1 n−1
− 2 (1 + z) + n(1 + z)
z n+1 z z
63
(1 + z)2n+1 (1 + z)2n
=− n+3
+n
z z n+2
for a contribution of

2n 2n + 1
n − .
n+1 n+2
Third component.
Re-write the third component as follows
1
(1 + w)n+1 (1 + z)n .
z 3 wn+1 (1 − w/z)2
Extracting the residue at w = 0 for the third component we obtain
n n
(1 + z)n X n + 1 q + 1 (1 + z)n X

n 1
= (n + 1)
z3 q=0
n − q z q z 3
q=0
n − q z q
(1 + z)n (1 + z)n (1 + z)2n

= (n + 1) 3 n
= (n + 1) .
z z z n+3
For the residue at w = z we obtain
n
(1 + z)n (1 + z)2n

1 1
(n + 1) 1 + × − 2 = −(n + 1)
z z z z n+3
and this cancels the previous contribution for a result of zero.
Fourth component.
Re-write the fourth component as
1
− (1 + w)2n (1 + z)
z 3 wn+1 (1 − w/z)2
Extracting the residue at w = 0 we obtain
n
1+z X 2n q + 1
− 3
z q=0 n − q zq
for a zero contribution.

For the residue at w = z we obtain

1+z 1 2n 1 2n−1
− −(n + 1) n+2 (1 + z) + 2n n+1 (1 + z)
z z z
(1 + z)2n+1 (1 + z)2n
= (n + 1) n+3
− 2n
z z n+2

2n + 1 2n
(n + 1) − 2n .
n+2 n+1
64
Conclusion. Finally collecting the four pieces we have

2n + 1 2n 2n 2n + 1
(n + 1) − 2n +n −
n+2 n+1 n+1 n+2

2n + 1 2n 2n + 1
= (n + 1) −n −
n+2 n+1 n+2

2n + 1 n + 2 2n + 1 2n
= (n + 1) −n − .
n−1 n−1 n−1 n+2
This is
2n
n×
n+2
as claimed.
Alternate solution. As before we start trying to evaluate
n−2 n
XX k+q 2n − q − k − 1
S(n) =
q=0 k=1
k n−k+1
which we re-write as
n−2
X 2n − q − 1 n−2 X n+1
X n + 1 + q n−2 X k + q 2n − q − k − 1
− − + .
q=0
n+1 q=0
n+1 q=0
k n−k+1
k=0
Call these pieces up to sign from left to right S1 , S2 and S3 .

The two pieces in front cancel the quantities introduced by extending k to
include the values zero and n + 1.
Evaluation of S1 . Introduce
(1 + z)2n−q−1

2n − q − 1 2n − q − 1
Z
1
= = dz.
n+1 n−q−2 2πi |z|= z n−q−1
This vanishes when q > n − 2 so we may extend the sum to infinity to get
(1 + z)2n−1 X zq
Z
1
dz
2πi |z|= z n−1 (1 + z)q
q≥0
(1 + z)2n−1
Z
1 1
= n−1
dz
2πi |z|= z 1 − z/(1 + z)
(1 + z)2n
Z
1
= dz
2πi |z|= z n−1

2n
= .
n−2
65
(1 + z)n+1+q
Z
n+1+q 1
= dz.
n+1 2πi |z|= z n+2

n−2
(1 + z)n+1 X
Z
1
(1 + z)q dz
2πi |z|= z n+2 q=0
(1 + z)n+1 (1 + z)n−1 − 1
Z
1
= dz
2πi |z|= z n+2 1+z−1
(1 + z)n+1
Z
1
= ((1 + z)n−1 − 1) dz
2πi |z|= z n+3

2n
= .
n+2
A more efficient evaluation is to notice that when we re-index q as n − 2 − q
in S2 we obtain
n−2
X n−2
X
n+1+n−2−q 2n − q − 1
=
q=0
n+1 q=0
n+1
which is S1 .
(1 + z)2n−q−k−1

2n − q − k − 1
Z
1
= dz.
n−k+1 2πi |z|= z n−k+2
This effectively controls the range so we can let k go to infinity to get

n−2
(1 + z)2n−1 X X k + q zk
Z
1
n+2
dz
2πi |z|= z q=0
q (1 + z)q+k
k≥0
n−2
(1 + z)2n−1 X
Z
1 1 1
= dz
2πi |z|= z n+2 q=0
(1 + z) q (1 − z/(1 + z))q+1
n−2
(1 + z)2n X
Z
1 1 1
= n+2 q+1
dz
2πi |z|= z q=0
(1 + z) (1 − z/(1 + z))q+1
(1 + z)2n
Z
1
= × (n − 1) × dz
2πi |z|= z n+2

2n
= (n − 1) × .
n+1
66
Finally collecting the three contributions we obtain

2n 2n 2n 2n
(n − 1) × −2 = (n + 2) −2
n+1 n+2 n+2 n+2

2n
=n× .
n+2
30 Basic usage of the Iverson bracket (B1 I1 I2 R)

l
X q+k l−q
S(k, l) = .
q=0
k k
Introduce
(1 + w)l−q

l−q
Z
1
= dw.
k 2πi |w|= wk+1
1 + z + z2 + · · · + zl
Z
1
[[0 ≤ q ≤ l]] = dz
2πi |z|= z q+1
z l+1 − 1
Z
1
= dz
2πi |z|= (z − 1)z q+1
which controls the range so we may let q go to infinity to obtain for the sum
z l+1 − 1 1 (1 + w)l X q + k
Z Z
1 1
dw dz
2πi |z|= (z − 1)z 2πi |w|= wk+1 q (1 + w)q z q
q≥0
z l+1 − 1 1 (1 + w)l
Z Z
1 1
= dw dz
2πi |z|= (z − 1)z 2πi |w|= wk+1 (1 − 1/(1 + w)/z)k+1
l+1
−1 1 (1 + w)l+k+1
Z Z
1 z 1
= dw dz.
2πi |z|= (z − 1)z 2πi |w|= wk+1 (1 + w − 1/z)k+1
In the evaluation of the inner integral we will use the residue at infinity
instead of computing the residues at w = 0 and w = (1 − z)/z. The residue at
infinity of a function h(z) is given by the formula

1 1
z z
which in the present case gives for the inner term in w
67
1 (1 + 1/w)l+k+1 1
−Resw=0
w2 (1/w)k+1 (1 + 1/w − 1/z)k+1
1 (1 + w)l+k+1 1
= −Resw=0 2 l
w w (1 + 1/w − 1/z)k+1
(1 + w)l+k+1 1
= −Resw=0 l+2
w (1 + 1/w − 1/z)k+1
(1 + w)l+k+1 1
= −Resw=0 l+1−k
.
w (w(z − 1)/z + 1)k+1
We extract the residue where we may omit the minus sign because residues
sum to zero, getting
l−k
(z − 1)q

X l+k+1 q+k
(−1)q .
q=0
l−k−q k zq
Substituting this into the outer integral we get two pieces, the first is
l−k
z l+1 X l + k + 1 q + k (z − 1)q
Z
1
(−1)q dz
2πi |z|= (z − 1)z q=0 l − k − q k zq
Z l−k
1 1 X l+k+1 q+k
= (−1)q (z − 1)q z l−q dz.
2πi |z|= (z − 1) q=0 l − k − q k
There is no contribution from this integral.

The second piece is
l−k
(z − 1)q
Z
1 1 X l+k+1 q+k
− (−1)q dz.
2πi |z|= (z − 1)z q=0 l − k − q k zq
We now have two cases. Suppose k = l so the only contribution originates

with q = 0 giving
Z
1 1 l+k+1 k l+k+1
− dz = ,
2πi |z|= (z − 1)z l−k k 2k + 1
which is the formula we are in the process of proving.
If on the other hand l > k there is an additional contribution of
l−k
(z − 1)q−1
Z
1 X l+k+1 q+k
− (−1)q dz
2πi |z|= q=1 l−k−q k z q+1
which is zero by inspection because [z q ](z − 1)q−1 = 0.

This concludes the proof, QED.
68
If you are reading all three articles in this thread. The Iverson
bracket was essential here because the binomial coefficient l−q

k starts producing
nonzero values when q > l.
Alternate solution. To get convergence in a neighborhood of zero of the
infinite sum that appears we use the alternate Iverson bracket valid for q ≥ 0
zq
Z
1 1
[[0 ≤ q ≤ l]] = dz
2πi |z|= z l+1 1 − z
This gives for the sum
(1 + w)l 1 zq
Z Z
1 1 1 X q+k
dw dz
2πi |w|= wk+1 2πi |z|= z l+1 1 − z q (1 + w)q
q≥0
(1 + w)l 1
Z Z
1 1 1 1
= dw dz
2πi |w|= wk+1 2πi |z|= z l+1 1 − z (1 − z/(1 + w))k+1
Z l+k+1 Z
1 (1 + w) 1 1 1 1
= dw dz.
2πi |w|= wk+1 2πi |z|= z l+1 1 − z (1 + w − z)k+1
We evaluate the inner integral by taking the negative of the sum of the
residues at z = 1 and at z = 1 + w and z = ∞.
The negative of the residue at z = 1 is
1
wk+1
which when substituted into the outer integral yields
(1 + w)l+k+1
Z
1 l+k+1
dw = ,
2πi |w|= w2k+2 2k + 1
which is the formula we are trying to establish.

Next we prove that the residue at infinity is zero. This is given by
1 l+1 1 1 1 z k+1
−Resz=0 2
z k+1
= −Resz=0 z l
z 1 − 1/z (1 + w − 1/z) z − 1 (z(1 + w) − 1)k+1
1 1 z l+k+1
=− Res z=0 .
(1 + w)k+1 z − 1 (z − 1/(1 + w))k+1
This is zero by inspection, which leaves the residue at z = 1 + w. Write
(−1)k+1 (1 + w)l+k+1 1
Z Z
1 1 1
k+1 l+1
dw dz.
2πi |w|= w 2πi |z|= z 1 − z (z − (1 + w))k+1
We require the derivative

(k) k
1 1 1 1 X k (l + q)! (k − q)!
= (−1)q
k! z l+1 1 − z k! q=0 q l! × z l+1+q (1 − z)1+k−q
69
k
X l+q 1 1
= (−1)q l+1+q 1+k−q
.
q=0
q z (1 − z)
Evaluate this at z = 1 + w to get

k
X l+q 1 1
(−1)q l+1+q
q=0
q (1 + w) (−w)1+k−q
and substitute into the outer integral to obtain

k
(−1)k+1 (1 + w)l+k+1 X l + q
Z
1 1
(−1)q dw
2πi |w|= wk+1 q=0
q (1 + w) l+1+q (−w)1+k−q
k
(1 + w)l+k+1 X l + q
Z
1 1 1
= k+1 l+1+q 1+k−q
dw
2πi |w|= w q=0
q (1 + w) w
k
(1 + w)k−q
Z
X l+q 1
= dw.
q=0
q 2πi |w|= w2k+2−q
[w2k+1−q ](1 + w)k−q .
But we have 2k + 1 − q ≥ k + 1 while k − q ≤ k so these terms are zero, thus

concluding the proof.
Simplified solution. As observed elsewhere this can be done without the
Iverson bracket.
l
X q+k l−q
S(k, l) =
q=0
k k
where l ≥ k.
Introduce

l−q
Z
1 1 1
= l−q−k+1
dz.
k 2πi |z|= z (1 − z)k+1
This controls the range becoming zero when q > l − k so we may extend q
to infinity. We obtain for the sum
1
Z
1 1 X q + k
z q dz
2πi |z|= z l−k+1 (1 − z)k+1 k
q≥0
Z
1 1 1 1
= dz
2πi |z|= z l−k+1 (1 − z)k+1 (1 − z)k+1
70
Z
1 1 1
= dz.
2πi |z|= z l−k+1 (1 − z)2k+2
This evaluates by inspection to

l − k + 2k + 1 l+k+1
= .
2k + 1 2k + 1
31 Basic usage of the Iverson bracket II (B1 I1 )

Suppose we seek to compute
n
X m+k
S(n, m) = k .
m+1
k=0
Introduce
Z
m+k 1 1
= (1 + z)m+k dz
m+1 2πi |z|= z m+2
as well as the Iverson bracket
wk
Z
1 1
[[0 ≤ k ≤ n]] = dw.
2πi |w|=γ wn+1 1 − w
Z Z
1 1 1 1 1 X k
m+2
(1 + z)m kw (1 + z)k dw dz.
2πi |z|= z 2πi |w|=γ wn+1 1−w
k≥0
For this to converge we must have |w(1 + z)| < 1. We get
Z Z
1 1 1 m 1 1 w(1 + z)
(1 + z) dw dz
2πi |z|= z m+2 2πi |w|=γ wn+1 1 − w (1 − w(1 + z))2
Z Z
1 1 1 1 1 1
= m+2
(1 + z)m+1 dw dz.
2πi |z|= z 2πi |w|=γ wn 1 − w (1 − w(1 + z))2
We evaluate the inner integral using the fact that the residues at the poles
sum to zero. The residue at w = 1 produces
Z Z
1 1 1 1 1
− m+2
(1 + z)m+1 dz = − (1 + z)m+1 dz = 0.
2πi |z|= z (−z)2 2πi |z|= z m+4
For the residue at w = 1/(1 + z) we re-write the inner integral to get
71
Z
1 1 1 1 1
2
dw.
(1 + z) 2πi |w|=γ w 1 − w (w − 1/(1 + z))2
n
We thus require
0
1 1
n
w 1−w

w=1/(1+z)

−n 1 1 1
= n+1
+ n 2

w 1 − w w (1 − w) w=1/(1+z)
= −n(1 + z)n+1 (1 + z)/z + (1 + z)n (1 + z)2 /z 2 .

Substituting this into the outer integral and flipping signs we get two pieces
which are
Z
1 1
m+2
(1 + z)m−1 n(1 + z)n+2 /z dz
2πi |z|= z
Z
n 1 n+m+1 n+m+1
= (1 + z) dz = n × .
2πi |z|= z m+3 m+2
The second piece is
Z
1 1
− m+2
(1 + z)m−1 (1 + z)n+2 /z 2 dz
2πi |z|= z
Z
1 1 n+m+1 n+m+1
=− (1 + z) dz = − .
2πi |z|= z m+4 m+3
It follows that our answer is

n−1 n+m+1 nm + 2n + 1 n + m + 1
n− = .
m+3 m+2 m+3 m+2
Remark. Being rigorous we also verify that the residue at infinity in the
calculation of the inner integral is zero. We get
1 n 1 1
−Resw=0 w
w2 1 − 1/w (1 − (1 + z)/w)2
w w2 wn+1 1
= −Resw=0 wn−2 2
= −Resw=0 .
w − 1 (w − (1 + z)) w − 1 (w − (1 + z))2
There is certainly no pole at zero here and the residue is zero as claimed
(the term 1 + z rotates in a circle around the point one on the real axis and with
< 1 it is never zero). This last result could also be obtained by comparing
degrees of numerator and denominator.
72
32 Use of a double Iverson bracket (B1 I2 R)
n
X k
Y (n) = 2n−k ,
bk/2c
k=1
by considering
bn/2c b(n−1)/2c
X
n−2k 2k X
n−2k−1 2k + 1
Y1 (n) = 2 and Y2 (n) = 2 .
k k
k=0 k=0
We will use the following Iverson bracket:
zk
Z
1 1
[[0 ≤ k ≤ n]] = dz.
2πi |z|= z n+1 1 − z
Evaluation of Y1 (n). Introduce

Z
2k 1 1
= k+1
(1 + w)2k dw.
k 2πi |w|= w
With the Iverson bracket controlling the range we can extend k to infinity
to get for the sum
2n 1 X −2k k (1 + w)2k
Z Z
1 1 1
2 z dz dw.
2πi |w|= w 2πi |z|= z bn/2c+1 1 − z wk
k≥0
We can instantiate these contours to get convergence of the series. We thus

obtain
2n
Z Z
1 1 1 1 1
bn/2c+1
dz dw
2πi |w|= w 2πi |z|= z 1 − z 1 − z(1 + w)2 /w/4
2n+2
Z Z
1 1 1 1 1
= dz dw.
2πi |w|=
2
(1 + w) 2πi |z|= z bn/2c+1 z − 1 z − 4w/(1 + w)2
We evaluate the inner piece by computing the negative of the sum of the
residues at z = 1, z = 4w/(1 + w)2 and z = ∞. We get for z = 1
1 (1 + w)2 (1 + w)2
2
= 2
=
1 − 4w/(1 + w) (1 + w) − 4w (1 − w)2
for a zero contribution.

We get for z = ∞
1 1 1 1
−Resz=0
z 2 1/z bn/2c+1 1/z − 1 1/z − 4w/(1 + w)2
73
1 1
= −Resz=0 z bn/2c+1
1 − z 1 − 4wz/(1 + w)2
again for a zero contribution.
Finally for z = 4w/(1 + w)2 we get
(1 + w)2bn/2c+2 (1 + w)2
− .
22bn/2c+2 × wbn/2c+1 (1 − w)2
Substitute into the outer integral to obtain
2n mod 2 (1 + w)2bn/2c+2
Z
1
− bn/2c+1
dw.
2πi |w|= w (1 − w)2
Extracting the negative of the residue we get the sum

bn/2c
X 2bn/2c + 2
2n mod 2
(bn/2c − q + 1).
q=0
q
This yields

n mod 2 1 2bn/2c+2 2bn/2c + 2
2 (bn/2c + 1) 2 −
2 bn/2c + 1
bn/2c

n mod 2 2bn/2c + 1
X
−2 (2bn/2c + 2)
q=1
q−1

n mod 2 1 2bn/2c+2 2bn/2c + 2
=2 (bn/2c + 1) 2 −
2 bn/2c + 1

2bn/2c + 1
−2n mod 2 (bn/2c + 1) 22bn/2c+1 − 2
bn/2c

1 2bn/2c + 2 2bn/2c + 1
= 2n mod 2 (bn/2c + 1) 2 −
2 bn/2c + 1 bn/2c

2bn/2c + 1
= 2n mod 2 (bn/2c + 1) .
bn/2c
Evaluation of Y2 (n). This is obviously very similar to the first case. We
get the integral
2n+1
Z Z
1 1 1 1 1
b(n−1)/2c+1
dz dw.
2πi |w|= 1 + w 2πi |z|= z z − 1 z − 4w/(1 + w)2
There is no contribution from z = 1 and z = ∞ as before which leaves

2(n+1) mod 2 (1 + w)2b(n−1)/2c+3
Z
1
− b(n−1)/2c+1
dw.
2πi |w|= w (1 − w)2
74
Extracting the negative of the residue we obtain
b(n−1)/2c
X 2b(n − 1)/2c + 3
2(n+1) mod 2 (b(n − 1)/2c − q + 1).
q=0
q
This yields
n−1 !
n−1 1 2b n−1 2b 2 c + 3
2(n+1) mod 2
(b c + 1) × 2 2 c+3 −2
2 2 b n−1
2 c+1
n−1
b 2 c
n−1 X 2b n−1 c + 2
(n+1) mod 2 2
−2 (2b c + 3)
2 q=1
q −1
n−1 !
n−1 1 2b n−1 2b 2 c + 3
=2 (n+1) mod 2
(b c + 1) × 2 2 c+3 −2
2 2 b n−1
2 c+1
n−1
−2(n+1) mod 2
c + 3)
(2b
2
n−1 n−1 !
1 2b n−1 c+2 2b 2 c + 2 2b 2 c + 2
× 2 2 −2 − .
2 b n−1
2 c b n−1
2 c+1
Evaluation of Y (n). Keeping in mind that Y (n) does not include a term
for k = 0 we get for n = 2p the contributions

2p + 1 2p + 1
−22p + (p + 1) + p 22p+1 − 2
p p

2p 2p
−(2p + 1) 22p − 2 −
p−1 p

2p
= −22p+1 + (4p + 2) .
p
On the other hand for n = 2p + 1 we obtain

2p+1 2p + 1 1 2p+3 2p + 3
−2 + 2(p + 1) + (p + 1) 2 −2
p 2 p+1

1 2p + 2 2p + 2
− (2p + 3) 22p+2 − 2 −
2 p p+1

2p + 1
= −22p+2 + (4p + 5) .
p
Joining the two formulae we get the compact closed form

n+1 n
−2 + (2n + 2 + (n mod 2)) .
bn/2c
I would conjecture that with the closed form being this simple now that it
has been computed we can probably find a much simpler proof.
75
33 Iverson bracket and an identity by Gosper
(I1 R)
n
X n+k
(xn+1 (1 − x)k + (1 − x)n+1 xk ) = 1.
k
k=0
We use an Iverson bracket to control the range of k so we can let it range

from zero to infinity, which is
1 + v + · · · + vn
Z
1
[[0 ≤ k ≤ n]] = dv
2πi |v|= v k+1
v n+1 − 1
Z
1
= dv.
2πi |v|= (v − 1)v k+1
We evaluate this using the formula for the residue at infinity

1 1
z z
which in this case yields (omit the minus sign as the residues sum to zero)
1 1/v n+1 − 1 1/v n+1 − 1

Resv=0 = Resv=0
v 2 (1/v − 1) × 1/v k+1 (1 − v) × 1/v k
k
vk

v 1
= Resv=0 − .
v n+1 1 − v 1 − v
With the additional assumption that k ≥ 0 which is the case here this yields
vk vk
Z
1 1 1
Resv=0 n+1 = dv
v 1−v 2πi |v|= v n+1 1 − v
which could have been obtained by inspection.

This yields for the second component of the sum
Z
1 1 1 X n+k
(1 − x)n+1 xk v k dv
2πi |v|= v n+1 1 − v n
k≥0
Z
1 1 1 1
= (1 − x)n+1 dv.
2πi |v|= v n+1 1 − v (1 − xv)n+1
We will evaluate this not by evaluating the residue at zero but the sum of
the negatives of the residues at v = 1 and v = 1/x given that the residues sum
to zero.
76
For the residue at v = 1 re-write the integral as follows:
Z
n+1 1 1 1 1
−(1 − x) dv.
2πi |v|= v n+1 v − 1 (1 − xv)n+1
The residue at v = 1 here is

1
−(1 − x)n+1 = −1
(1 − x)n+1
1
upon negation.
For the residue at v = 1/x re-write the integral as follows:
(1 − x)n+1 1
Z
1 1 1
n+1 n+1
dv
x 2πi |v|= v 1 − v (1/x − v)n+1
(1 − x)n+1 1
Z
1 1 1
= (−1)n+1 dv.
xn+1 2πi |v|= v n+1 1 − v (v − 1/x)n+1
Use Leibniz’ rule to differentiate the two terms in v to get
(n) n
1 X n (−1)k (n + k)! (n − k)!

1 1 1
= .
n! v n+1 1 − v n! k n! × v n+1+k (1 − v)n−k+1
k=0
Evaluate this at v = 1/x including the factor in front to get for the residue
n
(1 − x)n+1 1 X n (−1)k (n + k)! (n − k)!
(−1)n+1
xn+1 n! k n! × (1/x)n+1+k (1 − 1/x)n−k+1
k=0
n
(1 − x)n+1 X (−1)k (n + k)! 1
= (−1)n+1 n+1
x n! × k! × (1/x)n+1+k (1 − 1/x)n−k+1
k=0
n
n + k (−1)k

X 1
= (−1)n+1 (1 − x)n+1
k (1/x)k (x − 1)n−k+1 /xn−k+1
k=0
n
n + k (−1)k

X 1
=
k (1/x)k (x − 1)−k /xn−k+1
k=0
n
X n+k
= xn+1 (1 − x)k .
k
k=0
Upon negation this becomes the negative of the first component of the sum.
Hence adding the three pieces (first component, one, negative of first compo-
nent) we obtain a sum of
1.
77
residue at infinity of the integral in v is zero.
In the present case this becomes
1 1 1 1
−Resv=0
v 2 (1/v)n+1 1 − 1/v (1 − x/v)n+1
1 v n+1 × v n+1 1
= −Resv=0
v 2 1 − 1/v (v − x)n+1
1 v 2n+2 1
= −Resv=0
v v − 1 (v − x)n+1
v 2n+1 1
= −Resv=0
v − 1 (v − x)n+1
34 Iverson bracket and an identity by Gosper,

generalized (I2 R)
m−1
X n−1
X m − 1 + q
n−1+q n
x (1 − x)q + xq (1 − x)m = 1
q=0
q q=0
q
where n, m ≥ 1.
We will evaluate the second term by a contour integral and show that is
equal to one minus the first term which is the desired result.
zq 1
Z
1
[[0 ≤ q ≤ n − 1]] = dz.
2πi |z|= z n 1 − z
With this bracket we may extend the sum in q to infinity to get

1 1 X m−1+q q q
Z
1
z x (1 − x)m dz
2πi |z|= z n 1 − z q
q≥0
m

(1 − x) 1 1 X m−1+q q q
Z
= n
z x dz
2πi |z|= z 1 − z q≥0 q
(1 − x)m
Z
1 1 1
= n 1 − z (1 − xz)m
dz.
2πi |z|= z
Now we have three poles here at z = 0 and z = 1 and z = 1/x and the
residues at these poles sum to zero, so we can evaluate the residue at zero by
computing the negative of the residues at z = 1 and z = 1/x.
78
Observe that the residue at infinity is zero as can be seen from the following
computation:
1 1 1
−Resz=0 2 z n
z 1 − 1/z (1 − x/z)m
1 n z zm
−Resz=0 2
z
z z − 1 (z − x)m
1 1
−Resz=0 z n+m−1 = 0.
z − 1 (z − x)m
Returning to the main thread the residue at z = 1 as seen from
(1 − x)m
Z
1 1 1
− n m
dz.
2πi |z|= z z − 1 (1 − xz)
is
1
−(1 − x)m = −1.
(1 − x)m
For the residue at z = 1/x we consider
(1 − x)m
Z
1 1 1
dz
x × 2πi |z|= z 1 − z (1/x − z)m
m n
(−1)m (1 − x)m
Z
1 1 1
= m n m
dz.
x × 2πi |z|= z 1 − z (z − 1/x)
and use the following derivative:

(m−1)
1 1 1
(m − 1)! z n 1 − z
m−1
X m − 1 (−1)q (n + q − 1)! (m − 1 − q)!
1
=
(m − 1)! q=0 q (n − 1)!z n+q (1 − z)m−q
m−1
X 1 (−1)q (n + q − 1)! 1
=
q=0
q! (n − 1)!z n+q (1 − z)m−q
m−1
X n + q − 1 (−1)q

1
= .
q=0
q z n+q (1 − z)m−q
Evaluate this at z = 1/x and multiply by the factor in front to get

m−1
(−1)m (1 − x)m X n + q − 1 1
m
× (−1)q xn+q
x q=0
q (1 − 1/x)m−q
m−1
(−1)m (1 − x)m X n + q − 1 xm−q
= m
× (−1)q xn+q
x q=0
q (x − 1)m−q
79
m−1
X
n+q−1 1
= (−1)m (1 − x)m × (−1)q xn (−1)m−q
q=0
q (1 − x)m−q
m−1
X
n+q−1 n
= x (1 − x)q .
q=0
q
This yields for the second sum term the value

m−1
X
n+q−1 n
1− x (1 − x)q
q=0
q
showing that when we add the first and the second sum by cancellation the
end result is one, as claimed.
35 A double hypergeometric sum (B1 )

X n (n−l)/2
X nn − ln − l − r
2n
=
n r=0
l r r
l=0
which is
n (n−l)/2
X n − ln − l − r
X n
.
l r=0
r r
l=0
Start by computing the inner sum.

Introduce

n−l n−l
Z
1 1
= = (1 + z)n−l dz.
r n−l−r 2πi |z|= z n−l−r+1
and

n−l−r n−l−r
Z
1 1
= = (1 + w)n−l−r dw.
r n − l − 2r 2πi |w|= wn−l−2r+1
Observe carefully that this second integral is zero when 2r ≥ n − l + 1 so it

controls the range and we may extend r to infinity.
This yields for the inner sum
(1 + w)n−l 1 (1 + z)n−l X w2r

Z Z
1
n−l+1
z r dz dw
2πi |w|= w 2πi |z|= z n−l+1 (1 + w)r
r≥0
(1 + w)n−l 1 (1 + z)n−l
Z Z
1 1
= dz dw.
2πi |w|= wn−l+1 2πi |z|= z n−l+1 1 − zw2 /(1 + w)
80
Substitute this into the outer sum to get
(1 + w)n 1 (1 + z)n
Z Z
1 1
2πi |w|= wn+1 2πi |z|= z n+1 1 − zw2 /(1 + w)
n
X n wl zl
× dz dw
l (1 + w)l (1 + z)l
l=0
(1 + w)n 1 (1 + z)n
Z Z
1 1
= n+1 n+1 2
2πi |w|= w 2πi |z|= z 1 − zw /(1 + w)
n
w z
× 1+ dz dw
1+w1+z
Z Z
1 1 1 1 1
= (1 + w + z + 2wz)n dz dw
2πi |w|= wn+1 2πi |z|= z n+1 1 − zw2 /(1 + w)
Z Z
1 1 1 1 1
= n+1 n+1 2
(1 + w + z(1 + 2w))n dz dw.
2πi |w|= w 2πi |z|= z 1 − zw /(1 + w)
Extracting the inner residue we obtain
n
w2n−2q

X n
n−q
(1 + w)n−q (1 + 2w)q
q=0
(1 + w) q
n
X n
= (1 + 2w)q w2n−2q = (1 + 2w + w2 )n = (1 + w)2n .
q=0
q
We conclude in extracting the residue in w to get

Z
1 1 2n n 2n 2n
(1 + w) dw = [w ](1 + w) = .
2πi |w|= wn+1 n
36 Factoring a triple hypergeometric sum (B1 )

n
X 1+p+q p+n−k q+n−k
(−1)k
k n−k n−k
k=0
p q
.
n n
Introduce
(1 + z1 )p+n−k

p+n−k
Z
1
= dz1
n−k 2πi |z1 |= z1n−k+1
81
and
(1 + z2 )q+n−k

q+n−k
Z
1
= dz2 .
n−k 2πi |z2 |= z2n−k+1
Observe that these integrals vanish when k > n and we may extend k to
infinity.
We thus obtain for the sum
(1 + z1 )p+n 1 (1 + z2 )q+n
Z Z
1
n+1
2πi |z1 |= z1 2πi |z2 |= z2n+1
X 1 + p + q z1k z2k
× (−1)k dz2 dz1 .
k (1 + z1 )k (1 + z2 )k
k≥0
This is
(1 + z1 )p+n 1 (1 + z2 )q+n
Z Z
1
2πi |z1 |= z1n+1 2πi |z2 |= z2n+1
p+q+1
z1 z2
× 1− dz2 dz1
(1 + z1 )(1 + z2 )
or
(1 + z1 )n−q−1 1 (1 + z2 )n−p−1
Z Z
1
n+1 (1 + z1 + z2 )p+q+1 dz2 dz1
2πi |z1 |= z1 2πi |z2 |= z2n+1
Supposing that p ≥ n and q ≥ n this may be re-written as

Z Z
1 1 1 1
2πi |z1 |= z1n+1 (1 + z1 )q+1−n 2πi |z2 |= z2n+1 (1 + z2 )p+1−n
×(1 + z1 + z2 )p+q+1 dz2 dz1

Put z2 = (1 + z1 )z3 so that dz2 = (1 + z1 ) dz3 to get
Z
1 1
n+1
2πi |z1 |= z1 (1 + z1 )q+1−n
Z
1 1
× n+1
2πi |z2 |= (1 + z1 )n+1 z3 (1 + (1 + z1 )z3 )p+1−n
×(1 + z1 )p+q+1 (1 + z3 )p+q+1 (1 + z1 ) dz3 dz1
which is
(1 + z1 )p 1
Z Z
1 1
2πi |z1 |= z1n+1 2πi n+1
|z2 |= z3 (1 + z3 + z1 z3 )p+1−n
×(1 + z3 )p+q+1 dz3 dz1

(1 + z1 )p 1 (1 + z3 )n+q
Z Z
1
= n+1 n+1 dz3 dz1
2πi |z1 |= z1 2πi |z2 |= z3 (1 + z1 z3 /(1 + z3 ))p+1−n
82
Extracting the residue for z1 first we obtain
n
(1 + z3 )n+q k + p − n z3k

X p
n+1 (−1)k .
n−k z3 k (1 + z3 )k
k=0
The residue for z3 then yields

n
X
k p k+p−n n−k+q
(−1) .
n−k k n−k
k=0
The sum term here is

p! × (p + k − n)! × (q + n − k)!
(n − k)!(p + k − n)! × k!(p − n)! × (n − k)!q!
which simplifies to
p! × n! × (q + n − k)!
(n − k)! × n! × k!(p − n)! × (n − k)!q!
which is
n p q+n−k
k n q
so we have for the sum
n
X
p n k q+n−k
(−1) .
n k q
k=0
To evaluae the remaining sum we introduce
(1 + v)q+n−k

q+n−k
Z
1
= dv
q 2πi |v|= v q+1
getting for the sum

n
(1 + v)q+n X n
Z
p 1 1
q+1
(−1)k dv
n 2πi |v|= v k (1 + v)k
k=0
n
(1 + v)q+n
Z
p 1 1
= 1 − dv
n 2πi |v|= v q+1 1+v
(1 + v)q
Z
p 1 p q
= dv =
n 2πi |v|= v q−n+1 n q−n
which is
p q
.
n n
83
37 Factoring a triple hypergeometric sum II (B1 B2 )
X p q n + k
k k p+q
k≥0
n n
.
p q
Z
p 1 1 1
= dz1
k 2πi |z1 |= z1p−k+1 (1 − z1 )k+1
and Z
q 1 1 1
= dz2 .
k 2πi |z2 |= z2q−k+1 (1 − z2 )k+1
These two effectively control the range their product being zero when k >
min(p, q) so that we may extend the sum to infinity.
We also use
Z
n+k 1 1
= p+q+1
(1 + w)n+k dw
p+q 2πi |w|= w

Z Z Z
1 1 n 1 1 1 1 1 1
(1 + w)
2πi |w|= wp+q+1 2πi |z1 |= z11+p 1 − z1 2πi |z2 |= z21+q 1 − z2
X (1 + w)k z1k z2k

× dz2 dz1 dw
(1 − z1 )k (1 − z2 )k
k≥0
Z Z Z
1 1 n 1 1 1 1 1 1
= (1 + w)
2πi |w|= wp+q+1 2πi |z1 |= z11+p 1 − z1 2πi |z2 |= z21+q 1 − z2
1
× dz2 dz1 dw
1 − (1 + w)z1 z2 /(1 − z1 )/(1 − z2 )
Z Z Z
1 1 n 1 1 1 1
= p+q+1
(1 + w) 1+p 1+q
2πi |w|= w 2πi |z1 |= z1 2πi |z2 |= z2
1
× dz2 dz1 dw.
(1 − z1 )(1 − z2 ) − (1 + w)z1 z2
1
1 − z1 − z2 + z1 z2 − z1 z2 − wz1 z2
84
1 1 1
= = .
1 − z1 − z2 − wz1 z2 1 − z1 1 − (1 + wz1 )z2 /(1 − z1 )
Extracting the residue in z2 then yields
1 (1 + wz1 )q
Z Z
1 1 n 1 1
p+q+1
(1 + w) 1+p dz1 dw
2πi |w|= w 2πi |z1 |= z1 1 − z1 (1 − z1 )q
1 (1 + wz1 )q
Z Z
1 1 n 1
= p+q+1
(1 + w) 1+p dz1 dw.
2πi |w|= w 2πi |z1 |= z1 (1 − z1 )q+1
By symmetry of the initial sum we may suppose that p ≤ q, getting for the
inner integral
p
q m p+q−m
X
w .
m=0
m q
The outer integral now yields
p
X q p+q−m n
.
m=0
m q p+q−m
The sum term here is

n!
(q − m)! × m! × (p − m)! × (n + m − p − q)!

n p! × (n − p)!
=
p (q − m)! × m! × (p − m)! × (n + m − p − q)!

n p n−p
= .
p m q−m
It thus remains to show that
p
X p n−p n
=
m=0
m q − m q
which may be done combinatorially or by inspecting the integral

Z p
1 1 X p m
q+1
(1 + v)n−p v dv
2πi |v|= v m=0
m
Z
1 1
= q+1
(1 + v)n−p (1 + v)p dv
2πi |v|= v
Z
1 1 n n
= (1 + v) dv = .
2πi |v|= v q+1 q
This was math stackexchange problem 280481.
A simpler version of this identity is at section 15.
85
38 Factoring a triple hypergeometric sum III
(B1 )
n 2
X n pn − n pn + k pn
= .
k k k n
k=0
(1 + z)pn−n

pn − n
Z
1
= dz
k 2πi |z|= z k+1
and
(1 + w)pn+k
Z
pn + k 1
= dw.
k 2πi |w|= wk+1
n
(1 + z)pn−n 1 (1 + w)pn X n (1 + w)k
Z Z
1
dw dz
2πi |z|= z 2πi |w|= w k z k wk
k=0
n
(1 + z)pn−n 1 (1 + w)pn
Z Z
1 1+w
= 1+ dw dz
2πi |z|= z 2πi |w|= w zw
(1 + z)pn−n 1 (1 + w)pn
Z Z
1
= (1 + w + zw)n dw dz.
2πi |z|= z n+1 2πi |w|= wn+1
Expanding the binomial in the inner sum we get
n
X n
wq (1 + z)q
q=0
q
which yields
n
(1 + z)pn−n+q
Z
X n 1 pn
dz
q=0
q 2πi |z|= z n+1 n−q
n
X n pn − n + q pn
= .
q=0
q n n−q
The inner term is

n pn − n + q pn
q n pn − n + q
(pn)!
=
q! × (n − q)! × (pn − 2n + q)! × (n − q)!
86

pn n! × (pn − n)!
=
n q! × (n − q)! × (pn − 2n + q)! × (n − q)!

pn n pn − n
= .
n q n−q
Thus it remains to show that
n
X n pn − n pn
= .
q=0
q n − q n
This can be done combinatorially or using the integral

n
(1 + v)pn−n X n q
Z
1
v dv
2πi |v|= v n+1 q=0
q
(1 + v)pn−n
Z
1
= (v + 1)n dv
2πi |v|= v n+1
(1 + v)pn
Z
1 pn
= n+1
= .
2πi |v|= v n
39 Factoring a triple hypergeometric sum IV

(B1 )
min{m,n,p}
X m n p+m+n−r p+m p+n
= .
r=0
r r m+n m n
Introduce
Z
n n 1 1
= = (1 + z)n dz
r n−r 2πi |z|= z n−r+1
and
p+m+n−r p+m+n−r
=
m+n p−r
Z
1 1
= p−r+1
(1 + w)p+m+n−r dw.
2πi |w|= w
Observe carefully that the first of these is zero when r > n and the second
one when r > p so we may extend the range of r to infinity.
87
(1 + z)n 1 (1 + w)p+m+n X m r wr
Z Z
1
z dw dz
2πi |z|= z n+1 2πi |w|= wp+1 r (1 + w)r
r≥0
m
(1 + z)n 1 (1 + w)p+m+n
Z Z
1 zw
= 1+ dw dz
2πi |z|= z n+1 2πi |w|= wp+1 1+w
(1 + z)n 1 (1 + w)p+n
Z Z
1
= (1 + w + zw)m dw dz.
2πi |z|= z n+1 2πi |w|= wp+1
The inner integral is
m
(1 + w)p+n X m
Z
1
(1 + z)q wq dw
2πi |w|= wp+1 q=0
q
with residue
min(m,p)
X m p+n
(1 + z)q
q=0
q p−q
which in combination with the outer integral yields

min(m,p)
X m p+n n+q
.
q=0
q n+q n
Now note that

p+n n+q (p + n)! (n + q)!
=
n+q n (p − q)!(n + q)! q!n!

(p + n)! p! p+n p
= = .
(p − q)!p! q!n! n q
Therefore we just need to verify that
min(m,p)
X m p p+m
=
q=0
q p−q m
which follows by inspection.

It can also be done with the integral
(1 + w)p
Z
p 1
= dw
p−q 2πi |w|= wp−q+1
which is zero when q > p so we can extend q to infinity to get for the sum
(1 + w)p X m q
Z
1
w dw
2πi |w|= wp+1 q
q≥0
88
(1 + w)p+m
Z
1
= dw
2πi |w|= wp+1

p+m
= .
m
40 A triple hypergeometric sum V (B1 )

l X p
q m−p n m−n l m−n
X
(−1) =2
p=0 q=0
m−l q p−q l
where m ≥ n and m − n ≥ l.
This is
l p
X m−p X q n m−n
(−1) .
p=0
m−l q=0
q p−q
Now introduce the integral

m−n
Z
1 1
= (1 + z)m−n dz.
p−q 2πi |z|= z p−q+1
Note that this vanishes when q > p so we may extend the range of q to
infinity, getting for the sum
l
m−p 1
Z
X 1 m−n
X
q n
p+1
(1 + z) (−1) z q dz
p=0
m − l 2πi |z|= z q
q≥0
l
m−p 1
Z
X 1
= p+1
(1 + z)m−n (1 − z)n dz.
p=0
l − p 2πi |z|= z

m−p
Z
1 1
= l−p+1
(1 + w)m−p dw.
l−p 2πi |w|=γ w
This too vanishes when p > l so we may extend p to infinity, getting
Z
1 1
l+1
(1 + w)m
2πi |w|=γ w
1
Z
1 X wp 1
× (1 + z)m−n (1 − z)n dz dw.
2πi |z|= z z p (1 + w)p
p≥0
89
The geometric series converges when |w/z/(1 + w)| < 1. We get
Z
1 1
l+1
(1 + w)m
2πi |w|=γ w
Z
1 1 1
× (1 + z)m−n (1 − z)n dz dw
2πi |z|= z 1 − w/z/(1 + w)
Z
1 1
= l+1
(1 + w)m
2πi |w|=γ w
Z
1 1
× (1 + z)m−n (1 − z)n dz dw.
2πi |z|= z − w/(1 + w)
Now from the convergence we have |w/(1 + w)| < |z| which means the pole
at z = w/(1 + w) is inside the contour |z| = . Extracting the residue yields (the
pole at zero has disappeared)
Z m−n n
1 1 m w w
l+1
(1 + w) 1+ 1− dw
2πi |w|=γ w 1+w 1+w
Z
1 1
= l+1
(1 + 2w)m−n dw
2πi |w|=γ w

l m−n
=2 .
l
41 Basic usage of exponentiation integral to ob-

tain Stirling number formulae (E)
n
X n
(n − 2q)k .
q=0
2q + 1
We observe that
Z
k!
k 1
(n − 2q) = exp((n − 2q)z) dz.
2πi |z|= z k+1

Z n
k! 1 X n
exp((n − 2q)z) dz
2πi |z|= z k+1 q=0 2q + 1
Z n
k! exp((n + 1)z) X n
= exp((−2q − 1)z) dz
2πi |z|= z k+1 q=0
2q + 1
90
which is Z
1 k! exp((n + 1)z)
2 2πi
|z|= z k+1
n n
!
X n X n q
× exp(−qz) − (−1) exp(−qz) dz.
q=0
q q=0
q
This yields two pieces, call them A1 and A2 . Piece A1 is

Z
1 k! exp((n + 1)z)
(1 + exp(−z))n dz
2 2πi |z|= z k+1
Z
1 k! exp(z)
= (exp(z) + 1)n dz
2 2πi |z|= z k+1
and piece A2 is
Z
1 k! exp((n + 1)z)
(1 − exp(−z))n dz
2 2πi |z|= z k+1
Z
1 k! exp(z)
= (exp(z) − 1)n dz.
2 2πi |z|= z k+1
Recall the species equation for labelled set partitions:
P(UP≥1 (Z))
which yields the bivariate generating function of the Stirling numbers of the
second kind
exp(u(exp(z) − 1)).
This implies that
X n z n (exp(z) − 1)q
=
q n! q!
n≥q
and
X n z n−1 (exp(z) − 1)q−1
= exp(z).
q (n − 1)! (q − 1)!
n≥q
Now to evaluate A1 proceed as follows:

Z
1 k! exp(z)
(2 + exp(z) − 1)n dz
2 2πi |z|= z k+1
Z n
1 k! exp(z) X n n−q
= 2 (exp(z) − 1)q dz
2 2πi |z|= z k+1 q=0 q
n
exp(z) (exp(z) − 1)q
Z
X n n−q 1 k!
= 2 × q! × dz.
q=0
q 2 2πi |z|= z k+1 q!
91
Recognizing the differentiated Stirling number generating function this be-
comes
n
X n n−q−1 k+1
2 × q! × .
q=0
q q+1
Now observe that when n > k + 1 the Stirling number for k + 1 < q ≤ n
is zero, so we may replace n by k + 1. Similarly, when n < k + 1 the binomial
coefficient for n < q ≤ k + 1 is zero so we may again replace n by k + 1. This
gives the following result for A1 :
k+1
X
n n−q−1 k+1
2 × q! × .
q=0
q q+1
Moving on to A2 we observe that when k < n the contribution is zero because

the series for exp(z) − 1 starts at z. This integral is simple and we have
1 k! × n! exp(z) (exp(z) − 1)n

Z
k+1
dz.
2 2πi |z|= z n!
Recognizing the Stirling number this yields

1 k+1
× n! × .
2 n+1
which correctly represents the fact that we have a zero contribution when
k < n.
This finally yields the closed form formula
k+1
X
n n−q−1 k+1 1 k+1
2 × q! × − × n! × .
q=0
q q+1 2 n+1
confirming the previous results.

This was math.stackexchange.com problem 1353963
42 Advanced usage of the Iverson bracket with

two instances (EI1 )
bn/2c
X n
S1 = (−1)q (n − 2q)n .
q=0
q
Introduce
Z
n! 1
(n − 2q)n = exp((n − 2q)z) dz
2πi |z|= z n+1
92
1 + w + w2 + · · · + wbn/2c
Z
1
[[0 ≤ q ≤ bn/2c]] = dw
2πi |w|= wq+1
wbn/2c+1 − 1
Z
1
= dw.
2πi |w|= (w − 1)wq+1
This is an Iverson bracket that ensures that we may extend the range of the
sum from bn/2c to n.
We get for the sum
wbn/2c+1 − 1
Z Z
n! 1 1
2πi |z|= z n+1 2πi |w|= (w − 1)w
n
X n 1
× (−1)q exp((n − 2q)z) dw dz
q=0
q wq
wbn/2c+1 − 1
Z Z
n! exp(nz) 1
= n+1 2πi
2πi |z|= z |w|= (w − 1)w
n
X n
1
× (−1)q exp(−2qz) q dw dz.
q=0
q w
This is
n
wbn/2c+1 − 1
Z Z
n! exp(nz) 1 exp(−2z)
1 − dw dz
2πi |z|= z n+1 2πi |w|= (w − 1)w w
wbn/2c+1 − 1
Z Z
n! exp(nz) 1 n
= (w − exp(−2z)) dw dz.
2πi |z|= z n+1 2πi |w|= (w − 1)wn+1
Call this integral J1 .
An alternate representation of the sum is
n
n
X n
S2 = (−1) (−1)q (2q − n)n .
q
q=bn/2c+1
The Iverson bracket now becomes

wbn/2c+1 + · · · + wn
Z
1
[[bn/2c + 1 ≤ q ≤ n]] = dw
2πi |w|= wq+1
wn−bn/2c − 1
Z
1
= wbn/2c+1 dw.
2πi |w|= (w − 1)wq+1
We get for the sum
wn−bn/2c − 1
Z Z
n! 1 1
(−1)n n+1
wbn/2c+1
2πi |z|= z 2πi |w|= (w − 1)w
93
n
X n 1
× (−1)q exp((2q − n)z) dw dz
q=0
q wq
n−bn/2c
−1
Z Z
n! n exp(−nz) 1 bn/2c+1 w
= (−1) n+1
w
2πi |z|= z 2πi |w|= (w − 1)w
n
X n
1
× (−1)q exp(2qz) q dw dz.
q=0
q w
This is
wn−bn/2c − 1
Z Z
n! exp(−nz) 1
(−1)n n+1
wbn/2c+1
2πi
|z|= z 2πi |w|= (w − 1)w
n
exp(2z)
× 1− dw dz
w
n−bn/2c
−1
Z Z
n! exp(−nz) 1 bn/2c+1 w
= (−1)n n+1
w n+1
2πi |z|= z 2πi |w|= (w − 1)w
n
× (w − exp(2z)) dw dz.
Call this integral J2 .
By cancellation of the wn+1 factor this is
Z Z
n! exp(−nz) 1 1 n
−(−1)n n+1
(w − exp(2z)) dw dz.
2πi |z|= z 2πi |w|= (w − 1)wn−bn/2c
Now put z = −v to get

Z Z
n! exp(nv) 1 1
(−1)n
2πi |v|= (−1)n+1 v n+1 2πi |w|= (w − 1)wn−bn/2c
n
× (w − exp(−2v)) dw dv
Z Z
n! exp(nv) 1 1
=−
2πi |v|= v n+1 2πi |w|= (w − 1)wn−bn/2c
n
× (w − exp(−2v)) dw dv.
It follows that
J1 + J2 = 2S
Z Z
n! exp(nz) 1 1 n
=− (w − exp(−2z)) dw dz.
2πi |z|= z n+1 2πi |w|= (w − 1)wn+1
This is Z Z
n! exp(nz) 1 1
−
2πi |z|= z n+1 2πi |w|= (w − 1)wn+1
n
X n
× (w − 1)n−q (1 − exp(−2z))q dw dz.
q=0
q
94
Now we have two cases namely q = n and q < n. When q = n we get the
integral
Z Z
n! exp(nz) 1 1
− n+1
(1 − exp(−2z))n dw dz.
2πi |z|= z 2πi |w|= (w − 1)wn+1
The series for 1 − exp(−2z) starts with 2z so that on extracting the residue
in z we obtain
n! × 2n
Z
1
− n+1
dw
2πi |w|= (w − 1)w
n! × 2n
Z
1
= n+1
dw = n! × 2n .
2πi |w|= (1 − w)w
It remains to show that the following integral is zero:
Z Z
n! exp(nz) 1 1
−
2πi |z|= z n+1 2πi |w|= (w − 1)wn+1
n−1
X
n
× (w − 1)n−q (1 − exp(−2z))q dw dz.
q=0
q
This is Z Z
n! exp(nz) 1 1
−
2πi |z|= z n+1 2πi |w|= wn+1
n−1
X n
× (w − 1)n−q−1 (1 − exp(−2z))q dw dz
q=0
q
which is seen to be zero by inspection since the maximum power of the w − 1

term is n − 1 and we are extracting the wn coefficient.
We have shown that 2S = 2n × n! and hence
S = 2n−1 × n!
as claimed.
This is problem math.stackexchange.com problem 131826 where it is shown
that this problem admits considerable simplification.
43 Evaluation of a three-variable hypergeomet-

ric sum (B2 )
The sum
X p+q p+r q+r
p+q+r=n
p r q
with p, q, r ≥ 0
95
is claimed to be
n
X 2q
,
q=0
q
and is equal to
n n−p
X X p + q p + n − p − q q + n − p − q
p=0 q=0
q p q
which is
n n−p
X X p + q n − q n − p
.
p=0 q=0
q p q
Re-write this as
X p + q n − q n − p
.
0≤p,q
q p q
p+q≤n
Introduce the integral representations

n−q
Z
1 1
= dz1
p 2πi |z1 |= (1 − z1 )p+1 z1n−p−q+1
and
n−p
Z
1 1
= dz2 .
q 2πi |z2 |= (1 − z2 )q+1 z2n−p−q+1
Observe carefully that these integrals are zero when p + q > n so we may
extend the summation in p and q to infinity.
We get for the sum
Z Z
1 1 1 1
2πi |z1 |= (1 − z1 )z1 2πi |z2 |= (1 − z2 )z2n+1
n+1
X p + q z1p+q z2p+q
× dz2 dz1 .
q (1 − z1 )p (1 − z2 )q
p,q≥0
The inner term is

X z p z p X q + p z q z q X zpzp 1
1 2 1 2 1 2
p q
=
(1 − z1 ) p (1 − z2 ) (1 − z1 ) (1 − z1 z2 /(1 − z2 ))p+1
p
p≥0 q≥0 p≥0
1 1
=
1 − z1 z2 /(1 − z2 ) 1 − z1 z2 /(1 − z1 )/(1 − z1 z2 /(1 − z2 ))
1
=
1 − z1 z2 /(1 − z2 ) − z1 z2 /(1 − z1 )
96
(1 − z1 )(1 − z2 )
=
(1 − z1 )(1 − z2 ) − z1 z2 (2 − z1 − z2 )
(1 − z1 )(1 − z2 )
= .
(1 − z1 z2 )/(1 − z1 − z2 )
Substituting this into the integral yields
Z Z
1 1 1 1 1
dz2 dz1 .
2πi |z1 |= z1n+1 2πi |z2 |= z2n+1 (1 − z1 z2 )/(1 − z1 − z2 )
This is
Z Z
1 1 1 1 1 1
dz2 dz1 .
2πi |z1 |= z1n+1 1 − z1 2πi |z2 |= z2n+1 (1 − z1 z2 )/(1 − z2 /(1 − z1 ))
Extracting the inner residue now yields

Z n
1 1 1 X q 1
n+1 1 − z z1 dz1
2πi |z1 |= z1 1 q=0 (1 − z1 )n−q
n Z
X 1 1 1
= n−q+1 (1 − z )n+1−q dz1
q=0
2πi z
|z1 |= 1 1
n Z
X 1 1 1
= q+1 (1 − z )q+1 dz1 .
q=0
2πi |z1 |= z1 1
This is
n n
X q+q X 2q
= ,
q=0
q q=0
q
which was to be shown, QED.

44 Three phase application including Leibniz’

rule (B1 B2 R)
n
X 2n m+q−1 n
q = m × 4n−m ×
q=0
n + q 2m − 1 m
where n ≥ m.
Z
2n 1 1 1
= dz.
n+q 2πi |z|= z n−q+1 (1 − z)n+q+1
97
and
(1 + w)m+q−1

m+q−1
Z
1
= dw.
2m − 1 2πi |w|= w2m
Observe that the first integral is zero when q > n so we may extend q to
infinity.
(1 + w)m−1 X z q (1 + w)q
Z Z
1 1 1 1
q dw dz
2πi |z|= z n+1 (1 − z)n+1 2πi |w|= w2m (1 − z)q
q≥0
Z
1 1 1
=
2πi |z|= z n+1 (1 − z)n+1
(1 + w)m−1 z(1 + w)/(1 − z)
Z
1
× dw dz
2πi |w|= w2m (1 − z(1 + w)/(1 − z))2
Z
1 1 1
= n+1
2πi |z|= z (1 − z)n+1
(1 + w)m−1 z(1 + w)(1 − z)
Z
1
× dw dz
2πi |w|= w2m (1 − z − z(1 + w))2
Z
1 1 1
=
2πi |z|= z n (1 − z)n
(1 + w)m
Z
1 1
× dw dz.
2πi |w|= w2m (1 − 2z − zw)2
We evaluate the inner integral using the negative of the residue at the pole
at w = (1 − 2z)/z, starting from
(1 + w)m
Z Z
1 1 1 1 1
n+2 n 2m
dw dz.
2πi |z|= z (1 − z) 2πi |w|= w (w − (1 − 2z)/z)2
Differentiating we have
(1 + w)m−1 (1 + w)m (1 + w)m−1
m − 2m = (w − 2(1 + w)) m
w2m w2m+1 w2m+1
(1 + w)m−1
= (−w − 2)m .
w2m+1
The negative of this evaluated at w = (1 − 2z)/z is
1 (1 − z)m−1 z 2m+1
×m× m−1
×
z z (1 − 2z)2m+1
which finally yields
Z
m 1 1 1
dz.
2πi |z|= z n−m+1 (1 − z)n−m+1 (1 − 2z)2m+1
98
We have that the residues at zero, one and one half sum to zero with the
first one being the sum we are trying to compute. Therefore we evaluate these
in turn. We will restore the front factor of m at the end.
For the residue at zero we have using the Cauchy product that
n−m
X
n − m + q n−m−q 2m + n − m − q
2
q=0
q n−m−q
n−m
X
n − m + q n−m−q m + n − q
= 2 .
q=0
q 2m
For the residue at one we have that

(n−m)
(−1)n−m+1

1 1
(n − m)! z n−m+1 (1 − 2z)2m+1
n−m
(−1)n−m+1 X n − m

(n − m + q)!
= (−1)q
(n − m)! q=0 q (n − m)! × z n−m+1+q
(2m + n − m − q)!
×2n−m−q
(2m)! × (1 − 2z)2m+1+n−m−q
n−m
(−1)n−m+1 2n−m X n − m (n − m + q)!
= (−1)q
(n − m)! q=0
q (n − m)! × z n−m+1+q
(m + n − q)!
×2−q .
(2m)! × (1 − 2z)m+1+n−q
Evaluate this at one to get
n−m
X
n − m + q −q m + n − q
2n−m 2 .
q=0
q 2m
The residue at one evaluates to the sum we seek just like the residue at zero.
This leaves the residue at one half, where we find
(2m)
(−1)2m+1

1 1
(2m)! × 22m+1 z n−m+1 (1 − z)n−m+1
2m
(−1)2m+1 X 2m

(n − m + q)!
= (−1)q
(2m)! × 22m+1 q=0 q (n − m)! × z n−m+1+q
(n − m + 2m − q)!
×
(n − m)! × (1 − z)n−m+1+2m−q
2m
(−1)2m+1 X 2m

(n − m + q)!
= (−1)q
(2m)! × 22m+1 q=0 q (n − m)! × z n−m+1+q
99
(n + m − q)!
× .
(n − m)! × (1 − z)n+m+1−q
Evaluate this at one half to get
2m
1 X n−m+q n + m − q n+m+1−q
− (−1)q 2n−m+1+q 2
22m+1 q=0
q 2m − q
2m
2n−2m+1
X n−m+q q n+m−q
= −2 (−1) .
q=0
q 2m − q
For this last sum use the integral

n+m−q n+m−q
Z
1 1 1
= = dv.
2m − q n−m 2πi |v|= v 2m−q+1 (1 − v)n−m+1
This controls the range so we can let q go to infinity in the sum to get
1
Z
1 1 X n − m + q
(−1)q v q dv
2πi |v|= v 2m+1 (1 − v)n−m+1 q
q≥0
Z
1 1 1 1
= dv
2πi |v|= v 2m+1 (1 − v) n−m+1 (1 + v)n−m+1

n−m+m
Z
1 1 1 n
= 2m+1 2 n−m+1
dv = = .
2πi |v|= v (1 − v ) m m
We have shown that

n
2S − m × 2 × 22n−2m × =0
m
and hence may conclude that

n
S = m × 4n−m × .
m
residue at infinity of the integral in w is zero. Recall the formula for the residue
at infinity

1 1
z z
In the present case this becomes
1 (1 + 1/w)m 1
−Resw=0 2 2m
w 1/w (1 − 2z − z/w)2
(1 + 1/w)m 1
= −Resw=0
1/w2m (w(1 − 2z) − z)2
100
1
= −Resw=0 (1 + w)m wm
(w(1 − 2z) − z)2
The same procedure applied to the main integral yields
1 n−m+1 1 1
−Resz=0 z
z2 (1 − 1/z)n−m+1 (1 − 2/z)2m+1
1 n−m+1 z n−m+1 z 2m+1
= −Resz=0 z
z2 (z − 1)n−m+1 (z − 2)2m+1
1 1
= −Resz=0 z 2n+1
(z − 1)n−m+1 (z − 2)2m+1
which is zero as well.
45 Symmetry of the Euler-Frobenius coefficient

(B1 EI2 R)
Suppose we have the coefficient of the Euler-Frobenius polynomial
k
X n+1
bnk = (−1)k−l ln
k−l
l=1
and we seek to show that bnk = bnn+1−k where 0 ≤ k ≤ n + 1.

First re-write this as
k
X n+1
(−1)l (k − l)n .
l
l=0

zl
Z
1 1
[[0 ≤ l ≤ k]] = dz
2πi |z|= z k+1 1−z
and the exponentiation integral

Z
n! 1
(k − l)n = exp((k − l)w) dw.
2πi |w|= wn+1
to get for the sum (extend the summation to n + 1 since the Iverson bracket
controls the range)
Z Z n+1
n! 1 1 1 1 X n+1
exp(kw) (−1)l z l exp(−lw) dz dw
2πi |w|= wn+1 2πi |z|= z k+1 1 − z l
l=0
101
Z Z
n! 1 1 1 1
= exp(kw) (1 − z exp(−w))n+1 dz dw.
2πi |w|= wn+1 2πi |z|= z k+1 1−z
Evaluate this using the residues at the poles at z = 1 and at infinity. We
obtain for z = 1
Z
n! 1
− exp(kw)(1 − exp(−w))n+1 dw,
2πi |w|= wn+1
note however that 1 − exp(−w) starts at w so the power starts at wn+1

making for a zero contribution.
We get for the residue at infinity
1 k+1 1
−Resz=0 2
z (1 − exp(−w)/z)n+1
z 1 − 1/z
1
= −Resz=0 z k (1 − exp(−w)/z)n+1
z−1
1 zk
= Resz=0 (z − exp(−w))n+1 .
1−z z n+1
We need to flip the sign on this one more time since we are exploiting the
fact that the residues at the three poles sum to zero. Actually extracting the
coefficient we get
n−k
X
n+1
− (−1)n+1−q exp(−(n + 1 − q)w).
q=0
q
Substitute this into the integral in w to get

n−k
X Z
n + 1 n! 1
− n+1
exp(kw)(−1)n+1−q exp(−(n + 1 − q)w) dw
q=0
q 2πi |w|= w
n−k
X
n+1
=− (−1)n+1−q (−1)n (n + 1 − k − q)n
q=0
q
n−k
X
n+1
= (−1)q (n + 1 − k − q)n .
q=0
q
Using the fact that n + 1 − k − q is zero at q = n + 1 − k we finally obtain

n+1−k
X
n+1
(−1)q (n + 1 − k − q)n
q=0
q
which is precisely bnn+1−k by definition, QED.

Addendum. An alternate proof (variation on the theme from above) starts
from the unmodified definition and introduces
102
Z
n+1 1 1
= (1 + z)n+1 dz.
k−l 2πi |z|= z k−l+1
This controls the range so we may extend l to infinity. Introduce furthermore
Z
n n! 1
l = n+1
exp(lw) dw.
2πi |w|= w
These two yield for the sum
(−1)k
Z Z
n! 1 1 X
n+1 k+1
(1 + z)n+1 (−1)l z l exp(lw) dz dw
2πi |w|= w 2πi |z|= z
l≥0
(−1)k
Z Z
n! 1 1 1
= n+1 k+1
(1 + z)n+1 dz dw
2πi |w|= w 2πi |z|= z 1 + z exp(w)
(−1)k
Z Z
n! exp(−w) 1 1
= n+1
(1 + z)n+1 dz dw.
2πi |w|= w 2πi |z|= z k+1 z + exp(−w)
We evaluate this using the negatives of the residues at z = − exp(−w) and
at infinity. We get for z = − exp(−w)
(−1)k
Z
n! exp(−w)
n+1 k+1
(1 − exp(−w))n+1 dw
2πi |w|= w (−1) exp(−(k + 1)w)
Z
n! exp(kw)
=− (1 − exp(−w))n+1 dw.
2πi |w|= wn+1
As before the exponentiated term starts at wn+1 so there is no coefficient
on wn for a contribution of zero.
We get for the residue at infinity (starting from the next-to-last version of
the integral)
n+1
1 k k+1 (1 + z) 1
−Resz=0 (−1) z
z2 z n+1 1 + exp(w)/z
1 (1 + z)n+1 z/ exp(w)
= −Resz=0 2
(−1)k z k+1
z z n+1 1 + z/ exp(w)
(1 + z)n+1 exp(−w)
= −Resz=0 (−1)k z k .
z n+1 1 + z/ exp(w)
Doing the sign flip and simplifying we obtain
(1 + z)n+1 1
exp(−w)(−1)k × Resz=0 n−k+1
.
z 1 + z/ exp(w)
103
Extract the residue to get
n−k
X
k n+1
exp(−w)(−1) (−1)n−k−q exp(−(n − k − q)w)
q=0
q
Substitute into the integral in w to obtain

n−k
X Z
n + 1 n! 1
(−1)n−q exp(−(n + 1 − k − q)w) dw
q=0
q 2πi |w|= wn+1
n−k
X
n+1
= (−1)n−q (−1)n (n + 1 − k − q)n
q=0
q
n−k
X
n+1
= (−1)q (n + 1 − k − q)n .
q=0
q
We have obtained bnn+1−k

as before.
46 A probability distribution with two parame-

ters (B1 B2 )
A sum of binomial coefficients CLXVII
Suppose we have a random variable X where
−1
N N −k k−1
P[X = k] =
2n + 1 n n
for k = n + 1, . . . , N − n and zero otherwise.
We seek to show that these probabilities sum to one and compute the the
mean and the variance.
Sum of probabilities. This is given by
−n
−1 NX
N N −k k−1
.
2n + 1 n n
k=n+1
Introduce

N −k
Z
1 1 1
= dz
n 2πi |z|= z N −n−k+1 (1 − z)n+1
and
(1 + w)k−1

k−1
Z
1
= dw.
n 2πi |w|= wn+1
104
Observe carefully that the first integral is zero when k > N − n and the
second one when 1 ≤ k ≤ n so we may extend the range of the sum to 1 ≤ k.
This gives for the sum (without the scalar)
Z Z
1 1 1 1 1 X k−1
N −n n+1 n+1
z (1 + w)k−1 dw dz
2πi |z|= z (1 − z) 2πi |w|= w
k≥1
Z Z
1 1 1 1 1 1
= dw dz.
2πi |z|= z N −n (1 − z)n+1 2πi |w|= wn+1 1 − z(1 + w)
Z
1 1 1 1
dw
1 − z 2πi |w|= wn+1 1 − wz/(1 − z)
which yields for the integral in z

zn
Z
1 1 1
N −n n+1
dz
2πi |z|= z (1 − z) (1 − z)n+1
which is
N − 2n − 1 + 2n + 1 N
= .
2n + 1 2n + 1
This confirms that the probabilities sum to one.
Expectation. This is given by
−n
−1 NX
N N −k k−1
E[X] = k .
2n + 1 n n
k=n+1
Introduce

k−1 k! k
k = = (n + 1)
n n! × (k − 1 − n)! n+1
(1 + w)k
Z
1
= (n + 1) dw.
2πi |w|= wn+2
The range control from this integral produces zero when 0 ≤ k ≤ n so we
may extend the sum to zero, getting
Z Z
1 1 1 1 1 X
(n + 1) N −n+1 n+1
z k (1 + w)k dw dz.
2πi |z|= z (1 − z) 2πi |w|= wn+2
k≥0
Z
1 1 1
dw
2πi |w|= wn+2 1 − z(1 + w)
105
Z
1 1 1 1
= dw
1 − z 2πi |w|= wn+2 1 − wz/(1 − z)
which yields for the integral in z including the factor in front
z n+1
Z
1 1 1
(n + 1) N −n+1 n+1
dz
2πi |z|= z (1 − z) (1 − z)n+2
which is

N − 2n − 1 + 2n + 2 N +1
(n + 1) = (n + 1) .
2n + 2 2n + 2
We will scale this at the end, same as the variance.
Variance. Start by computing
−n
−1 NX
N N −k k−1
E[(X + 1)X] = (k + 1)k .
2n + 1 n n
k=n+1
Introduce

k−1 (k + 1)!
(k + 1)k =
n n! × (k − 1 − n)!
(1 + w)k+1
Z
k+1 1
= (n + 2)(n + 1) = (n + 2)(n + 1) dw.
n+2 2πi |w|= wn+3
The range control from this integral produces zero when 0 ≤ k ≤ n as before
so we may extend the sum to zero, getting
Z
1 1 1
(n + 2)(n + 1) N −n+1
2πi |z|= z (1 − z)n+1
Z
1 1+w X k
× z (1 + w)k dw dz.
2πi |w|= wn+3
k≥0
Z
1 1+w 1
n+3
dw
2πi |w|= w 1 − z(1 + w)
Z
1 1 1+w 1
= dw
1 − z 2πi |w|= wn+3 1 − wz/(1 − z)
which yields for the integral in z including the factor in front
z n+2 z n+1
Z
1 1 1
(n + 2)(n + 1) + dz
2πi |z|= z N −n+1 (1 − z)n+1 (1 − z)n+3 (1 − z)n+2
which is

N − 2n − 2 + 2n + 3 N − 2n − 1 + 2n + 2
(n + 2)(n + 1) +
2n + 3 2n + 2
106

N +1 N +1
= (n + 2)(n + 1) + .
2n + 3 2n + 2
Simplification for ease of interpretation.
We get for the expectation
(N + 1)! (N − 2n − 1)!(2n + 1)!

E[X] = (n + 1)
(N − 2n − 1)!(2n + 2)! N!
1
= (N + 1).
2
We obtain furthermore
E[(X + 1)X] = (n + 2)(n + 1)

(N + 1)! (N + 1)! (N − 2n − 1)!(2n + 1)!
× +
(N − 2n − 2)!(2n + 3)! (N − 2n − 1)!(2n + 2)! N!

1 N − 2n − 1
= (N + 1)(n + 2) +1
2 2n + 3
1 n+2
= (N + 2)(N + 1) .
2 2n + 3
This yields for the variance
Var[X] = E[X 2 ] − E[X]2

1 n+2 1 1
= (N + 2)(N + 1) − (N + 1) − (N + 1)2 .
2 2n + 3 2 4
which simplifies to
1 N − 2n − 1
Var[X] = (N + 1) .
4 2n + 3
47 An identity involving Narayana numbers (B1 )

Suppose we have the Narayana number

1 n n
N (n, m) =
n m m−1
and let
X k
Y
A(n, k, l) = N (it , jt + 1)
i0 +i1 +···+ik =n t=0
j0 +j1 +···+jk =l
107
where the compositions for n are regular and the ones for l are weak and we
seek to verify that

k+1 n n
A(n, k, l) = .
n l l+k+1
Introducing
1
X p p X
G(z, u) = zp uq
p q+1 q
p≥1 q≥0
X 1 X p p
= zp uq
p q+1 q
p≥1 q≥0
we have by inspection that
A(n, k, l) = [z n ][ul ]G(z, u)k+1 .
To evaluate this introduce for the inner sum term

Z
p p 1 1
= = (1 + w)p dw.
q+1 p−q−1 2πi |w|= wp−q
We get for the inner sum
1
Z
1 X p
p
(1 + w) uq wq dw
2πi |w|= wp q
q≥0
Z
1 1
= (1 + w)p (1 + uw)p dw
2πi |w|= wp
Z
1 1
= (1 + w(1 + u + uw)))p dw.
2πi |w|= wp
Extracting the coefficient from this we get
p
p−1
X p
[w ] wq (1 + u + uw)q
q=0
q
p−1
X p
= [wp−1−q ](1 + u + uw)q
q=0
q
p−1
X p q
= up−1−q (1 + u)2q+1−p .
q=0
q p−1−q
This is
p−1
X p p−1−q q
u (1 + u)p−1−2q
q=0
p−1−q q
108
p−1
X p p−1−q q
= u (1 + u)p−1−2q .
q=0
q + 1 q
Now observe that

1 p p−1−q 1 p−1 p−1−q
=
p q+1 q q+1 q q

1 p−1 p−1−q 1 p − 1 2q
= = .
q+1 p−1−q q q + 1 2q q
where
1 2q
Cq =
q+1 q
is a Catalan number.
p−1
X
p
X p−1
z Cq uq (1 + u)p−1−2q
q=0
2q
p≥1
p
X
p
X p
=z z Cq uq (1 + u)p−2q
q=0
2q
p≥0
X X p
q −2q
=z Cq u (1 + u) z p (1 + u)p
2q
q≥0 p≥q
X X p
=z Cq uq (1 + u)−2q z p (1 + u)p
2q
q≥0 p≥2q
X X p + 2q
=z Cq uq (1 + u)−2q (1 + u)2q z 2q z p (1 + u)p
2q
q≥0 p≥0
X 1
=z Cq uq z 2q .
(1 − z(1 + u))2q+1
q≥0
Using the generating function of the Catalan numbers

√
X
q 1 − 1 − 4w
Q(w) = Cq w =
2w
q≥0
which has functional equation
Q(w) = 1 + wQ(w)2
we obtain
2
uz 2 uz 2 uz 2

Q =1+ Q
(1 − z(1 + u))2 (1 − z(1 + u))2 (1 − z(1 + u))2
109
which is
1 − z(1 + u)
G(z, u) = 1 + uG(z, u)2 .
z
Extract the coefficient in z first. We get from the functional equation
G(z, u)
z= .
uG(z, u)2 + (1 + u)G(z, u) + 1
The coefficient extractor integral is

Z
1 1
[z n ]G(z, u) = G(z, u)k+1 dz.
2πi |z|= z n+1
which becomes with G(z, u) = v
(uv 2 + (1 + u)v + 1)n+1

Z
1
2πi |v|= v n+1

k+1 1 v
×v − (2uv + (1 + u)) dv
uv 2 + (1 + u)v + 1 (uv 2 + (1 + u)v + 1)2
(uv 2 + (1 + u)v + 1)n−1
Z
1
= (1 − uv 2 ) dv.
2πi |v|= v n−k
This is
(1 + v)n−1 (1 + uv)n−1
Z
1
(1 − uv 2 ) dv.
2πi |v|= v n−k
Extracting the coefficient on [ul ] we get two pieces which are, first piece A
(1 + v)n−1 v l

n−1 1 n−1 n−1
Z
dv =
l 2πi |v|= v n−k l n−k−l−1
which is

n−1 n−1 n−1 k+l+1 n
=
l k+l l n k+l+1

k+l+1 n n
= (n − l) .
n2 l k+l+1
and piece B which is
(1 + v)n−1 v l−1 2

n−1 1 n−1 n−1
Z
− v dv = −
l − 1 2πi |v|= v n−k l−1 n−k−l−2
which is

n−1 n−1 n−1 n−k−l−1 n
− =−
l−1 k+l+1 l−1 n k+l+1
110

n−k−l−1 n n
= −l .
n2 l k+l+1
Collecting the two pieces we finally obtain

k+l+1 −n + k + l + 1 n n
(n − l) +l
n2 n2 l k+l+1

k+l+1 −n n n
= n +l 2
n2 n l k+l+1

k+1 n n
=
n l k+l+1
as claimed, QED.
Remark. The closed form of G(z, u) can be computed as follows:
p
z 1 − 1 − 4uz 2 /(1 − z(1 + u))2
1 − z(1 + u) 2uz 2 /(1 − z(1 + u))2
p
z 1 − z(1 + u) − 1 − 2z(1 + u) + z 2 (1 + u)2 − 4uz 2
=
(1 − z(1 + u))2 2uz 2 /(1 − z(1 + u))2
p
1 − z(1 + u) − 1 − 2z(1 + u) + z 2 (1 + u)2 − 4uz 2
= .
2uz
The above material incorporates data from OEIS A055151 and from OEIS
A001263 on Narayana numbers.
This was math.stackechange.com problem 1498014.
48 Fibonacci, Tribonacci, Tetranacci (B1 )

Suppose we seek to evaluate the following sum (with a condition on the binomial
coefficient)
n Xk
X
q k n − 1 − qm
G(n, m) = (−1) .
q=0
q k−1
k=0
Now when n − 1 − qm < 0 we usually get a non-zero value for the binomial
coefficient but this is not wanted here. Therefore we have
n b(n−k)/mc
X X k
q n − 1 − qm
G(n, m) = (−1) .
q=0
q k−1
k=0
If we have lost any values for q above b(n − k)/mc these would render the
second binomial coefficient zero. If we have added in any values for q above k
the first binomial coefficient is zero there.
Now with the integral
111
(1 + z)n−1−qm

n − 1 − qm n − 1 − qm
Z
1
= = dz
k−1 n − k − qm 2πi |z|= z n−k−qm+1
we get range control because the pole vanishes when q > (n − k)/m and we
may extend q to infinity. We thus obtain for the inner sum
(1 + z)n−1 X z qm
Z
1 q k
(−1) dz
2πi |z|= z n−k+1 q (1 + z)qm
q≥0
k
(1 + z)n−1 zm
Z
1
= 1− dz
2πi |z|= z n−k+1 (1 + z)m
This yields for the outer sum
−1
(1 + z)n−1 zm
Z
1
1−z 1−
2πi |z|= z n+1 (1 + z)m
m
n+1 !
z
× 1 − z n+1 1 − dz
(1 + z)m
which is
(1 + z)n+m−1
Z
1 −1
n+1
(1 − z)(1 + z)m + z m+1
2πi |z|= z
n+1 !
zm

× 1 − z n+1 1− dz
(1 + z)m
Extracting the second component from the difference we get
n+1
zm
Z
1 −1
− (1 + z)n+m−1 (1 − z)(1 + z)m + z m+1 1− dz
2πi |z|= (1 + z)m
The pole at zero has vanished. We now have non-zero poles at z = −1 and
from the inverted term. These depend on m and we can certainly choose small
enough so that none of them are inside the contour. Therefore this term does
not contribute, leaving only
(1 + z)n+m−1
Z
1 1
n+1
dz.
2πi |z|= z (1 − z)(1 + z)m + z m+1
The generating function f (w) of these numbers is thus given by
n
X X
n n+m−1 1
f (w) = w [z q ] .
q=0
n − q (1 − z)(1 + z)m + z m+1
n≥0
112
This is
X 1 X n + m − 1
[z q ] wn
(1 − z)(1 + z)m + z m+1 n−q
q≥0 n≥q

1 n n+m−1+q
X X
q q
= w [z ] w
(1 − z)(1 + z)m + z m+1 n
q≥0 n≥0
q
1 X w 1
= [z q ] .
(1 − w)m (1 − w)q (1 − z)(1 + z)m + z m+1
q≥0
What we have here is an annihilated coefficient extractor that simplifies to
1 1
f (w) =
(1 − w) (1 − w/(1 − w))(1 + w/(1 − w))m + (w/(1 − w))m+1
m
1 1
=
(1 − w) (1 − 2w)/(1 − w)/(1 − w)m + wm+1 /(1 − w)m+1
m
1−w
= .
1 − 2w + wm+1
Now observe that
1 − 2w + wm+1 = (1 − w)(1 − w − w2 − · · · − wm−1 − wm )
so we finally have
m
!−1
X
q 1
f (w) = 1− w = .
q=1
1−w− w2− · · · − wm
We see that by the basic theory of linear recurrences what we have here is
a Fibonacci, Tribonacci, Tetranacci etc. recurrence. The question is what are
the initial values.
Observe however that [w0 ]f (w) = 1 and for 1 ≤ q ≤ m we have
1−w 1 1
[wq ] m+1
= [wq ] m+1
− [wq−1 ] .
1 − 2w + w 1 − 2w + w 1 − 2w + wm+1
But
1 1 X
= = 2n wn (1 − wm /2)n
1 − 2w + wm+1 1 − 2w(1 − wm /2)
n≥0
With the condition on q and n ≥ 1 only the constant term from the term
(1 − wm /2)n contributes because the degree would be more than m otherwise.
This produces just one matching term with coefficient 2q .
This yields for f (w)
[wq ]f (w) = 2q − 2q−1 = 2q−1 .
113
Therefore we get for the intial terms starting at q = 0
m
X
1, 1, 2, 4, 8, 16, . . . , 2m−1 with recurrence fn = fn−q .
q=1
This recurrence also shows (by subtraction) that the sequence may be pro-
duced starting from m − 1 zero terms followed by one.
The OEIS has the Fibonacci numbers, OEIS A000045
1, 2, 3, 5, 8, 13, 21, 34, 55, 89, . . .
and the Tribonacci numbers, OEIS A000073
1, 2, 4, 7, 13, 24, 44, 81, 149, 274, . . .
and the Tetranacci numbers, OEIS A000078
1, 2, 4, 8, 15, 29, 56, 108, 208, 401, . . .
and more.
This was math.stackexcange.com problem 1626949.
49 An identity involving two binomial coefficients

and a fractional term (B1 )
m
X q pk + q pm − pk mp + q
= .
pk + q k m−k m
k=0
Observe that

pk + q pk + q pk + q − 1
=
k k k−1
so that

pk + q pk + q − 1 q pk + q − 1 q pk + q
−p = = .
k k−1 k k−1 pk + q k
This yields two pieces for the sum, call them S1
m
X pk + q pm − pk
k m−k
k=0
and S2
m
X pk + q − 1 pm − pk
−p .
k−1 m−k
k=0
114
For S1 introduce the integrals
(1 + z)pk+q
Z
pk + q 1
= dz
k 2πi |z|=γ z k+1
and
(1 + w)pm−pk

pm − pk
Z
1
= dw.
m−k 2πi |w|= wm−k+1
The second one controls the range of the sum because the pole at zero
vanishes when k > m so we may extend k to infinity, getting for the sum
(1 + w)pm 1 (1 + z)q X wk (1 + z)pk

Z Z
1
dz dw
2πi |w|= wm+1 2πi |z|=γ z z k (1 + w)pk
k≥0
(1 + w)pm 1 (1 + z)q
Z Z
1 1
= dz dw
2πi |w|= wm+1 2πi |z|=γ z 1 − w(1 + z)p /z/(1 + w)p
(1 + w)pm+p 1
Z Z
1 1
= (1 + z)q dz dw.
2πi |w|= wm+1 2πi z(1 +
|z|=γ w)p − w(1 + z)p
p

Suppose || < |γ| which makes w(1+z)
z(1+w)p < 1 so that we have convergence

of the geometric series and suppose we can prove that z = w is the only pole
inside the contour and it is simple. We have
0
((1 + w)p z − w(1 + z)p ) = (1 + w)p − pw(1 + z)p−1
= (1 + w)p−1 (1 + w − wp).
We can choose || small enough such that |1 + w − wp| > 0 so the pole is
order one which yields
(1 + w)pm+p
Z
1 1 1
m+1
(1 + w)q p−1
dw
2πi |w|= w (1 + w) 1 + w − pw
(1 + w)pm+q+1
Z
1 1
= m+1
dw.
2πi |w|= w 1 + w − pw
Following exactly the same procedure we obtain for S2
(1 + w)pm+q
Z
1 1
−p m
dw.
2πi |w|= w 1 + w − pw
Adding these two pieces now yields
(1 + w)pm+q 1 + w
Z
1 1
m
− p dw
2πi |w|= w w 1 + w − pw
115
(1 + w)pm+q
Z
1
= dw
2πi |w|= wm+1

pm + q
= .
m
Remark Mon Jan 25 2016.
An alternate proof which is completely rigorous and does not depend on
assumptions about the poles of a bivariate complex function proceeds from the
integral
(1 + w)pm X wk (1 + z)q
Z Z
1 1
(1 + z)pk dz dw
2πi |w|= wm+1 (1 + w)pk 2πi |z|=γ z k+1
k≥0
Now put
z
u= and introduce g(u) = z.
(1 + z)p
We then have

1 z u pu
du = − p dz = − dz
(1 + z)p (1 + z)p+1 g(u) 1 + g(u)
and
1 g(u)(1 + g(u))
dz = du.
u 1 + g(u) − pg(u)
This yields
(1 + w)pm X wk
Z
1
2πi |w|= wm+1 (1 + w)pk
k≥0
Z
1 1 1 g(u)(1 + g(u))
× (1 + g(u))q du dw
2πi |u|=γ g(u)uk u 1 + g(u) − pg(u)
or
(1 + w)pm
Z
1 q 1 + g(u)
m+1
(1 + g(u)) dw.
2πi |w|= w 1 + g(u) − pg(u) u=w/(1+w)p
Now observe that g(w/(1 + w)p ) = w by definition so we get
(1 + w)pm
Z
1 1+w
m+1
(1 + w)q dw
2πi |w|= w 1 + w − pw
(1 + w)pm+q+1
Z
1 1
= dw.
2πi |w|= wm+1 1 + w − pw
This is exactly the same as before and the rest of the proof continues un-
changed.
116
50 Double chain of a total of three integrals
(B1 B2 )
n−1
X q 2n − 2k − 2 2k − q − 1 2n − q − 2
= .
k n−k−1 k−1 n−1
k=q
This is the same as

n
X q 2n − 2k 2k − q − 1 2n − q
= .
k n−k k−1 n
k=q
which is equivalent to
n n
X q − k 2n − 2k 2k − q − 1 X 2n − 2k 2k − q − 1
+
k n−k k−1 n−k k−1
k=q k=q

2n − q
= .
n
Now
q − k 2k − q − 1 q − k (2k − q − 1)!
=
k k−1 k (k − 1)!(k − q)!

(2k − q − 1)! 2k − q − 1
=− =− .
k!(k − q − 1)! k
It follows that what we have is in fact
n
X 2n − 2k 2k − q − 1 2k − q − 1 2n − q
− =
n−k k−1 k n
k=q
or alternatively
n
X 2n − 2k 2k − q − 1 2k − q − 1 2n − q
− = .
n−k k−q k−q−1 n
k=q
There are two pieces here, call them A and B. We use the integral repre-
sentation
(1 + z)2n−2k

2n − 2k
Z
1
= dz
n−k 2πi |z|= z n−k+1
which is zero when k > n (pole vanishes) so we may extend k to infinity. We
also use the integral
117
(1 + w)2k−q−1

2k − q − 1
Z
1
= dw
k−q 2πi |w|=γ wk−q+1
which is zero when k < q so we may extend k back to zero. We obtain for
piece A
wq−1 (1 + z)2n X zk (1 + w)2k

Z Z
1 1
dz dw
2πi |w|=γ (1 + w)q+1 2πi |z|= z n+1 (1 + z)2k wk
k≥0
wq−1 (1 + z)2n
Z Z
1 1 1
= dz dw
2πi |w|=γ (1 + w)q+1 2πi |z|= z n+1 1 − z(1 + w)2 /w/(1 + z)2
q
(1 + z)2n+2
Z Z
1 w 1 1
= dz dw
2πi |w|=γ (1 + w)q+1 2πi |z|= z n+1 w(1 + z)2 − z(1 + w)2
q−1
(1 + z)2n+2
Z Z
1 w 1 1
= dz dw.
2πi |w|=γ (1 + w)q+1 2πi |z|= z n+1 (z − w)(z − 1/w)
The derivation for piece B is the same and yields
wq (1 + z)2n+2
Z Z
1 1 1
q+1 n+1
dz dw.
2πi |w|=γ (1 + w) 2πi |z|= z (z − w)(z − 1/w)
The difference of these two is
wq−1 (1 + z)2n+2 1−w

Z Z
1 1
q+1 n+1
dz dw.
2πi |w|=γ (1 + w) 2πi |z|= z (z − w)(z − 1/w)
Using partial fractions by residues we get

1−w 1−w 1 1−w 1
= +
(z − w)(z − 1/w) w − 1/w z − w 1/w − w z − 1/w
w(1 − w) 1 w(1 − w) 1 w 1 w 1
= 2
+ 2
=− +
w −1 z−w 1 − w z − 1/w 1 + w z − w 1 + w z − 1/w
1 1 w2 1
= − .
1 + w 1 − z/w 1 + w 1 − wz
At this point we can see that there will be no contribution from the second
term but this needs to be verified. We get for the residue in z
n
w2 X 2n + 2 n−p

− w
1 + w p=0 p
There is no pole at zero in the outer integral for a contribution of zero.

Continuing with the first term we get
118
n
1 X 2n + 2 1
1 + w p=0 p wn−p
which yields
n
wq−1
Z
X 2n + 2 1 1
q+2 n−p
dw
p=0
p 2πi |w|=γ (1 + w) w
n Z
X 2n + 2 1 1 1
= q+2 w n−q−p+1
dw
p=0
p 2πi |w|=γ (1 + w)
n
X 2n + 2 n−q−p n−p+1
= (−1) .
p=0
p q+1
This is
n
X 2n + 2 n−q−p n−p+1
(−1) .
p=0
p n−p−q
The last integral we will be using is
(1 + v)n−p+1

n−p+1
Z
1
= dv.
n−p−q 2πi |v|=γ v n−p−q+1
Observe that this is zero when p ≥ n so we may extend p to infinity, getting
(1 + v)n+1 X 2n + 2 vp
Z
1 n−q−p
(−1) dv
2πi |v|=γ v n−q+1 p (1 + v)p
p≥0
2n+2
(1 + v)n+1
Z
1 v
= (−1)n−q 1 − dv
2πi|v|=γ v n−q+1 1+v
Z
1 1 1
= (−1)n−q dv
2πi |v|=γ v n−q+1 (1 + v)n+1

n−q+n 2n − q
= (−1)n−q (−1)n−q = .
n n
This is the claim. QED.
119
51 Post Scriptum: Generating function of an
initial segment of Fibonacci numbers (I2 )
Suppose we are given the OGF f (z) of an interesting sequence and want to
extract the generating function of the first n terms.
Thus we wish to compute
n
X X
g(w) = wk [z k ]f (z) = [[0 ≤ k ≤ n]]wk [z k ]f (z).
k=0 k≥0

vk
Z
1 1
[[0 ≤ k ≤ n]] = dv
2πi |v|= v n+1 1 − v
and the integral

Z
1 1
[z k ]f (z) = f (z) dz.
2πi |z|= z k+1
Z Z
1 1 1 1 1 X
g(w) = n+1
f (z) v k wk /z k dz dv
2πi |v|= v 1 − v 2πi |z|= z
k≥0
Z Z
1 1 1 1 1 1
= f (z) dz dv
2πi |v|= v n+1 1 − v 2πi |z|= z 1 − vw/z
Z Z
1 1 1 1 1
= f (z) dz dv.
2πi |v|= v n+1 1 − v 2πi |z|= z − vw
We need to examine convergence here. We no longer have a pole at z = 0
but in order to get convergence of the series on the first line we need |vw| < |z|
which means that the pole at z = vw is now inside the contour. Computing the
residue we find
Z
1 1 1
n+1
f (vw) dv.
2πi |v|= v 1−v
Extracting coefficients we indeed obtain
n
X n
X n
X
1 × [v n−k ]f (vw) = [v k ]f (vw) = wk [v k ]f (v).
k=0 k=0 k=0
If we have a special case and we know what the properties of f (z) are we
can also evaluate this using the residues at the poles at v = 1 and v = ∞ and
potential additional poles (the sum then has the sign flipped). As an example
suppose we are working with Fibonacci numbers so that
120
z
f (z) =
1 − z − z2
This has simple poles at z = −ϕ and z = 1/ϕ. Using partial fractions by
residues we obtain
1 ϕ 1 1/ϕ
f (z) = − .
z + ϕ 1 − 2ϕ z − 1/ϕ 1 + 2/ϕ
Examining the poles other than zero in turn we get for the pole at v = 1 the
residue −f (w). For the other two finite poles we have
1 ϕ 1 1/ϕ
f (vw) = −
vw + ϕ 1 − 2ϕ vw − 1/ϕ 1 + 2/ϕ
1 1 ϕ 1 1 1/ϕ
= − .
w v + ϕ/w 1 − 2ϕ w v − 1/ϕ/w 1 + 2/ϕ
We thus obtain for the pole at v = −ϕ/w
1 ϕ 1 ϕ 1
(−w/ϕ)n+1 = (−w/ϕ)n+1
w 1 − 2ϕ 1 + ϕ/w 1 − 2ϕ w+ϕ
and for the pole at v = 1/ϕ/w
1 1/ϕ 1 1/ϕ 1
− (wϕ)n+1 =− (wϕ)n+1 .
w 1 + 2/ϕ 1 − 1/ϕ/w 1 + 2/ϕ w − 1/ϕ
Finally the residue at infinity is
1 n+1 1 1 w/v
−Resv=0 v f (w/v) = −Resv=0 v n
v2 1 − 1/v v − 1 1 − w/v − w2 /v 2
1 wv 1 w
= −Resv=0 v n = −Resv=0 v n+1 = 0.
v − 1 v 2 − wv − w2 v − 1 v 2 − wv − w2
We have discovered the following expression for the initial segment of length
n of the generating function of the Fibonacci numbers:
ϕ 1 1/ϕ 1
g(w) = f (w) − (−w/ϕ)n+1 + (wϕ)n+1 .
1 − 2ϕ w + ϕ 1 + 2/ϕ w − 1/ϕ
Now to verify this we must look at the coefficient [wm ]g(w) in this formula.
Note however that the two series terms start at [wn+1 ] and hence the initial
segment from [w0 ] to [wn ] for 0 ≤ m ≤ n is not affected and we preserve the
terms up to degree n with no change, which is what we wanted.
For m ≥ n + 1 we get
ϕ 1 1/ϕ 1
Fm − (−1/ϕ)n+1 [wm−n−1 ] + (ϕ)n+1 [wm−n−1 ]
1 − 2ϕ w + ϕ 1 + 2/ϕ w − 1/ϕ
121
1 1
= Fm − (−1/ϕ)n+1 [wm−n−1 ]
1 − 2ϕ w/ϕ + 1
1 1
+ (ϕ)n+1 [wm−n−1 ]
1 + 2/ϕ ϕw − 1
1 1
= Fm − (−1/ϕ)n+1 (−1/ϕ)m−n−1 − (ϕ)n+1 (ϕ)m−n−1
1 − 2ϕ 1 + 2/ϕ
1 1
= Fm + √ (−1/ϕ)m − √ (ϕ)m .
5 5
In conclusion note that by Binet’s formula we have
1 1
Fm = √ ϕm − √ (−1/ϕ)m
5 5
We see that for m ≥ n + 1 we have [wm ]g(w) = 0 exactly as claimed. This
completes the sample computation.
References
[Ego84] G.P. Egorychev. Integral Representation and the Computation of Com-
binatorial Sums. American Mathematical Society, 1984.
122

Egorychev

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Egorychev

Hochgeladen von

Copyright:

Verfügbare Formate

Egorychev method and the evaluation of binomial

The Egorychev method is from the book by G.P.Egorychev [Ego84]. We

• Second binomial coefficient integral (B2 )

• Exponentiation integral (E)

• First Iverson bracket (I1 )

The residue at infinity is coded R.

2 Introductory example for the method, convergence about zero

3 Introductory example for the method, an interesting substitu-

4 Introductory example for the method, another interesting sub-

5 Introductory example for the method, yet another interesting

6 Introductory example for the method, a simple telescoping sum

7 Verifying that a certain sum vanishes (B1 ) 11

8 Basic usage of exponentiation integral (B1 E) 14

9 Introductory example for the method, eliminating odd-even de-

10 Introductory example for the method, proving equality of two

11 Evaluating a quadruple hypergeometric(B1 ) 19

12 Evaluating another quadruple hypergeometric(B1 ) 21

13 An identity by Strehl (B1 ) 24

14 Ternary strings that avoid 22 (B1 ) 25

15 Mixing the two types of binomial integrals (B1 B2 ) 28

16 Two companion identities by Gould (B1 ) 29

17 Exercise 1.3 from Stanley’s Enumerative Combinatorics (B1 B2 ) 32

18 Counting m-subsets (B1 I2 ) 34

19 Method applied to an iterated sum (B1 R) 36

20 A pair of two double hypergeometrics (B1 ) 39

22 An identity from Mathematical Reflections (B1 ) 44

23 A triple Fibonacci-binomial coefficient convolution (B1 ) 45

24 Fibonacci numbers and the residue at infinity (B2 R) 47

25 Permutations containing a given subsequence (B1 I1 ) 49

26 Catalan numbers and Lagrange inversion (B1 ) 55

27 A binomial coefficient - Catalan number convolution (B1 ) 58

28 A new obstacle from Concrete Mathematics (Catalan numbers)

29 Reducing the form of a double hypergeometric (B1 ) 62

30 Basic usage of the Iverson bracket (B1 I1 I2 R) 67

31 Basic usage of the Iverson bracket II (B1 I1 ) 71

32 Use of a double Iverson bracket (B1 I2 R) 73

33 Iverson bracket and an identity by Gosper (I1 R) 76

34 Iverson bracket and an identity by Gosper, generalized (I2 R) 78

35 A double hypergeometric sum (B1 ) 80

36 Factoring a triple hypergeometric sum (B1 ) 81

37 Factoring a triple hypergeometric sum II (B1 B2 ) 84

38 Factoring a triple hypergeometric sum III (B1 ) 86

39 Factoring a triple hypergeometric sum IV (B1 ) 87

40 A triple hypergeometric sum V (B1 ) 89

41 Basic usage of exponentiation integral to obtain Stirling number

42 Advanced usage of the Iverson bracket with two instances (EI1 ) 92

43 Evaluation of a three-variable hypergeometric sum (B2 ) 95

44 Three phase application including Leibniz’ rule (B1 B2 R) 97

46 A probability distribution with two parameters (B1 B2 ) 104

47 An identity involving Narayana numbers (B1 ) 107

48 Fibonacci, Tribonacci, Tetranacci (B1 ) 111

49 An identity involving two binomial coefficients and a fractional

50 Double chain of a total of three integrals (B1 B2 ) 117

51 Post Scriptum: Generating function of an initial segment of

Extracting the residue at w = 0 we get

This yields for the sum

Extracting the inner residue we get

3 Introductory example for the method, an in-

to obtain for the integral

4 Introductory example for the method, another

where m ≤ n and introduce

5 Introductory example for the method, yet an-

Observe that this is zero when k > n so we may extend k to infinity to