AC03-MGFs Djvu

ANALYTIC COMBINATORICS
PART TWO
Analytic
Combinatorics
Philippe Flajolet and

Robert Sedgewick
3. Combinatorial Parameters
and MGFs
http://ac.cs.princeton.edu
Analytic combinatorics overview
A. SYMBOLIC METHOD
1. OGFs
2. EGFs
3. MGFs
B. COMPLEX ASYMPTOTICS
SYMBOLIC METHOD
specification
4. Rational & Meromorphic
5. Applications of R&M
6. Singularity Analysis
7. Applications of SA
8. Saddle point
Analytic
Combinatorics
Robert Sedgewick
GF
equation
PART TWO
Analytic
Combinatorics

Robert Sedgewick
3. Combinatorial parameters and MGFs
• Basics
^Moment calculations
f©BGF examples
• Labelled classes
II.3a.MGFs.Basics
Natural questions about combinatorial parameters

What is the average number
of subsets in a random
set partition ?

of cycles in a random
permutation ?

of pacts in a random
composition ?
What is the average

number of pacts in a
random partition ?
What is the average coot

degcee of a random tree ?

of times each lettec appeacs
in a random M-word ?

of leaves in a random tree ?
Problem: Average-case results are sometime easy to derive but unsatisfying.
Example. Separate chaining hashing randomly assigns N keys to M lists.
Q. Average length of a list ?

A. N/M. < -
A trivial result that is not very useful because it

says nothing about the length of a particular list.
Ex: All the keys could be on one list.

Avg. length = (N + 0 + 0 + ... + 0 )/M = N/M
Section S.4
Solution: Find distribution (probability parameter value is k for all k )
Practical compromises. Ex: B 0unc j probability that list length deviates

significantly from average,
•compute average and variance
•compute average extrema! values * Ex: Compute average length of the longest list.
Goals for this lecture: Learn enough about parameters to be able to

•compute full distribution (in principle)
•compute moments and extremal values (in practice)
How many ways to partition

a set of N objects
into k subsets?
How many permutations of
size N have k cycles?
How many compositions
(sequences of positive integers

that sum to N ) have k parts?
How many partitions (sets

of positive integers that
sum to N ) have k parts?
How many trees with N

nodes have root degree k ?
How many letters

appear k times in an
M-word of length N ?
How many trees with N
nodes have k leaves ?
Basic definitions (combinatorial parameters for unlabelled classes)
Def. A combinatorial class is a set of combinatorial objects and an associated size

function that may have an associated parameter.
Def. The ordinary bivariate generating function (OBGF)

associated with a class is the formal power series
a(z,u) = z \ a \ u c ° st (A
size function
object name
a<EA
S v
parameter value
class name
Fundamental (elementary) identity
A(z) = z \ a \u cost ( a ) — y~^/\ N ^z N g /c
aeA N>0k>0
Terminology.
The variable z marks size
The variable u marks the parameter
Q. Flow many objects of size N with value k?
A. A Nk = [z N ][u k ]A(z,u)
Terminology.
BCF: bivariate CF.

MGF: multivariate GF
might add arbitrary

number of markers
With the symbolic method, we specify the class and at the same time characterize
the OBGF
Combinatorial enumeration: classic example
Q. How many binary strings with N bits?
£>i=2
0 0
0 1
1 0
1 1
£2 = 4
A. Bn= 2"
0 0 0
0 0 1
0 10
Oil
10 0
10 1
110
111
B 3 = 8
0 0 0 0
0 0 0 1
0 0 10
0 0 11
0 10 0
0 10 1
0 110
0 111
10 0 0
10 0 1
10 10
10 11
110 0
110 1
1110
1111
£ 4 =16
Combinatorial parameters: classic example
Q. How many A/-bit binary strings have k 0 bits?
fiio = 1
fin = 1
0 0
0 1
1 0
1 1
fi 2 o= 1
fi 2 i = 2
fi22= 1
A ' e “ = C)
0 0 0
0 0 1
0 10
Oil
10 0
10 1
110
111
fi30= 1
fisi = 3
fi32 = 3
fi33= 1
0 0
0 0
0 0
0 0
0 1
0 1
0 1
0 1
1 0
1 0
0
1
1 0
1 0
1 1
1 1
1 1
1 1
fi40
fi41
fi42
fi43
=
fi43
OBGF of binomial coefficients
= V(1 +u) N z N
N> 0
= E
k>0
(1 - z) k +'
(horizontal OGF)
(vertical OGF)
1 - z(1 + u)
(OBGF)
horizontal OGF
coefficients
N k '
0
1
■[u k ](1 + u) 7 =
10
The symbolic method for OBGFs (basic constructs)
Suppose that A and B are classes of unlabelled objects with OBCFs A(z,u) and B(z,u)
where z marks size and u marks a parameter value. Then
operation
notation
semantics
OGF
disjoint union
A + B
disjoint copies of objects from A and B
A(z, u ) + B(z , u )
Cartesian product
A x B
ordered pairs of copies of objects,

one from A and one from B
A(z, u)B(z, u )
sequence
SEQ(A)
sequences of objects from A
1 — A(z, u )
Construction immediately gives OBGF equation, as for enumeration.
Extends immediately to mark multiple parameters simultaneously with MGFs.
Proofs of correspondences
Ax B
SEQ(A)
A + B
z \c\ u cost(c)
c<EA+B
z l a li/ cosf ( a ) 4 z |6| a cost(b) — ^( Zj ^ _|_ b (^ z ^ ^
aeA beB
cEaxb
-Jc| yCOSt(c)
|a| + |b| yCost(a)+cost(b )
E
z \a\ u c °st(a)\
aEA b(E.B
aG/\ b(zB
A(z , u)B(z , a)
z |b| a cost(b)^
construction OCF
5£Q t (A) = A k
A(z,u) k
SEQ t (A) = A u + /\ f2 + /\ t3 + ...
where T = ti, t 2 , t 3 ,... is a subset of the integers
A(z , tv/ 1 + /A(z, i/) t2 4 /\(z, tv/ 3 4 ...
5£Q(A) = e + /\+A 2 +/\ 3 + ...
i +a( Z’ u ) +a (Z’ 4 + ---- 1 _ a(ZjU)
Combinatorial parameter example: 0 bits in bitstrings
Class B, the class of all binary strings

Size \b\, the number of bits in b
Parameter zeros(b), the number of 0 bits in b
OBCF B ( Z > U ) = Y1 Z |b| u zer0s(b) = B NkZ N U k
beB N>0k>0
Construction
OBGF equation
variable u “marks” the parameter

1
B = SEQ (uZo + Zi)
Expansion
B Nk = [u k }[z N ]B(z,u) = [u k ](1 +u) N = [z N ]
(1 -z) k +'
PART TWO
Analytic
Combinatorics

Robert Sedgewick
• Basics
f©BGF examples
II.3a.MGFs.Basics
PART TWO
Analytic
Combinatorics

Robert Sedgewick
• Moment calculations
f^©BGF examples
CAMBRIDGE
II.3b.MGFs.Moments
OBGF moment calculations
OBGF
P(z,u) = E z
size function
\p\ u cost(p)
object name
pEP
S' X
parameter value
^>") = EE PNkU k z N
N> 0 k >0
class name
Enumeration
P n = \z n ]P{z, 1) P(z, 1 ) = E z ' Pl
pev
number of objects of size N
PN = ^2pNk P(Z, 1)
k>0
total cost in objects

of size N
du
Mean cost of objects of size N Pn
[z N ]Pu(z, 1) Q
[z N ]P(z, 1) Pn
2 [z N ]Puu(z, 1 )
[z«]P(z,1)
+ Pn — P/v
E^ n = EEp^ n
N>0
N >0 Jt>0
Cumulated cost
Q n = [z N ]P„(z, 1) p„(z,1) = Ecost(p)z'Pl Q N = J2 k p Nk P„(z,1)

t pGP /c>0
dP(z , u)
( 7=1
Eq« zN = EE ><PNkZ N
N> 0
N>0k>0
Pn
= E
Pn/c
k>0
= E^-m :
k>0
Variance
Moments for 0 bits in bitstrings with OBGFs
Class
B, the class of all binary strings
Example
10111010001000
Size
\b\, the number of bits in b
OBCF
B{z,u) = ^ z l b\ u zeros(b)
Parameter
zeros(b), the number of 0 bits in b
beB
Construction
OBGF equation
Enumeration
Cumulated cost
Mean cost of objects of size N
B = SEQ (wZo +Z\)
B(z,u)
1 - z(1 + u)
[z N ]6(z,1) = [z N ] r i^
= 2
[z N ]^(z,1) = [z N ]
N 1
0 -2z)2
N2
N—1
[z N ]^(z,1)
B u (z,u)
(1 -z-zu) 2
Variance
(easier with horizontal GFs: stay tuned)
17
"Horizontal" and "vertical" OGFs
Y Y PNkZ N U k
N> 0 k>0
= p^(u)z N {horizontal OGF)
N> 0
— q/ c (z)a /c {vertical OGF)
k>0
horizontal OGF
coefficients
A/ ^0123456789
\
0
vertical OGF
coefficients
PS7
<-
MMu)
[z^lqsfz)
18
Moment calculations ("horizontal" OGF)
OBGF.
p( z ,u) = Y^ z
size function
\p\ u cost(p)
peP
s* \
object name class name
parameter value
GF for costs of objects

of size N
"Horizontal" OGF
Enumeration
Cumulated cost
[z N ]P(u,z) = p N (u) = Y u cosf(p)
pEP and size(p)=N
Pn( i)= 1 =p
PEVn
PnO) = E cost (p) = Q
PEVn
Hn
PnO) _ Q
Pn( i) p.
a N —
pM
Pn{ 1)
+ pN ~ PN^
p N kU k z N
N> 0 k>0
Pn(u) = YP NkLjk
k>0
Pn( 1) = Y,P Nk =
k>0
p'nO) = Y k P Nk =
k>0
Variance
0 bits in bitstrings with a "horizontal" OGF
OBGF
B(z,u) = , /'ll \
1 — z(1 + U)
"Horizontal" OGF
b N (u) = [z n ]B(z, u) = (1 + u) N
Enumeration
M 1)=2 N
Cumulated cost
6^(1) = N2 n “ 1
Average # 1-bits in a
random N-b\t string
*n( 1)/M1) = /V2 n_1 /2 n =(N/2) /
Variance
6"(1 )/b N (1) + N/2 - (N/2) 2 = N/4
concentrated: a n — VN/2 (stay tuned)
20
Moment calculations ("vertical" OGF)
OBGF.
p( z ,u) = E z
size function
\p\ u cost(p)
object name
pEP
S' V
parameter value
class name
GF for costs of objects

of cost k
^) = EE PNkU k Z N
N> 0 k >0
"Vertical" OGF
Enumeration
Cumulated cost.
[u K \P(u,z) = q k {z) = Y, z ' Pl
pEP and cost(p)=k
P N = [z N ]P(z , 1)
[ Z N iE%( z ) = Qn
Pn —
P,
q k (z) = Y,PNkZ N
N> 0
E%w = EE !<PNkZ N
k N> 0
= E(EW) zN
N> 0 k \
Qn
E kpnk
k>0
Variance
(omitted)
21
0 bits in bitstrings with a "vertical" OGF
OBGF
"Vertical" OGF
Enumeration
Cumulated cost
B(z,u)
1 -z( 1 +tv)
q k (z) = [u k ]B(z, u) =
z
(1 -z) k +'
P n = [z n ]B{z, 1) = 2 n
Average # 1-bits in a
random A/-bit string
= N2 n ~'
Pn/Qn =(n}2) /
^Tkr k ~i
yk
V/, z _
/ 1 _ -7^+1 /"I _ -r \2
(1 -r)
Z E*;
Jc-1
t 0-z)*+’ (1 -z) 2 ^-(i -z;
k -1
(1 -z) 2 0 -T^l) 2
(1 -2z)2
22
Moment inequalities and concentration
Let Xn be the value of a parameter for a random object of size N with mean p/v and
std dev cjn.
Markov inequality. Pr{X^ > < 1 ft
Chebyshev inequality. Pr{|X^ — /i N \ > ta^} < /t 2
“The probability of being much larger

than the mean must decay, and an upper
bound on the rate of decay is measured
in units given by the standard deviation.”
Def. A distribution is concentrated if on — o(/^n).
Proposition. If a distribution is concentrated,
then 1 in probability: lim Pr{1
N—7-00
e <
Xn_
flN
< 1 +e} = 1
When a distribution is concentrated, the expected value is “typical”.

Example: 100,000,000 random bits
Expected # 1 bits N/2 50,000,000
Standard deviation \fN/2 5,000
Probability X/v is between 49,900,000 and 50,100,000 .9975
23
Moments for letters in M-words with OBGFs
Class
Wm, the class of all M-words
Example
Size
|w|, the number of letters in w
OBGF
Parameter
occ{w), # of occurrences of a given letter in w
43552411
W M (z,u) = ^ 2
w£ W M
Construction
OBGF equation
Enumeration
Cumulated cost
Mean # of occurences of a
given letter in a random
M-word with N letters
Variance
B = SEQ (uZ + (M- 1)Z)

W M (z,u)
[z N ]l/l/(z, 1) = [z N ]
1 - (M- 1 +u)z
N-i
1 — Mz
[z N ]W u (z, 1) = [z™]
N-\
(1 - Mz) 2
= NM
•N -1
Bn —
[z N ]W u ( z,1)
[z N ]W(z, 1)
2 r ^Ni W uty (z, 1) 2 Kil\/i2
° z n = [n
[z N ]W(z, 1)
+ BN - Bn = N / M - N/M 2
[z N ]W u (z } 1)
[z N ]W uu (z, 1)
2 3
.|w| ^occ(w)
N(N- 1)/V/ N “ 2
Standard deviation
a N = - N/M 2
concentrated for fixed M
24
Application: Hashing algorithms
Goal: Provide efficient ways to
• Insert key-value pairs in a symbol table .
• Search the table for the pair corresponding to a given key.
Algorithms
The Art of
Computer
Programming
Sorting and Searching

Second Edition
DONALD E. KNUTH
Strategy
• Develop a hash function that maps each

key into value between 0 and M- 1.
• Maintain M lists of key-value pairs
Q. Average list length for N keys?
A. N/M < -Trivial
Q. Typical list length for N keys, for fixed IVR

A. N/M, concentrated «-Useful
key hash a
S 2 0
E 0
A 0 2
R 4
C 4
H 4
E 0
X 2
A 0
M 4
P 3
L 3
E 0 12
25
PART TWO
Analytic
Combinatorics

Robert Sedgewick
f^©BGF examples
CAMBRIDGE
II.3b.MGFs.Moments
PART TWO
Analytic
Combinatorics

Robert Sedgewick

• OBGF examples
CAMBRIDGE
II.3c.MGFs.0BGFs
Number of parts in compositions
Q. How many compositions of N have k parts?
1 + 1
1 + 1+1
1 + 2
2 + 1
C 21 = 1
C 22 = 1
C 31 = 1
cumulated cost: 3
average: 1.5
C 32 = 2
C 33 = 1
cumulated cost: 8
average: 2
(ro
1 + 1 + 1 + 1
1 + 1 + 2
1 + 2 + 1
1+3
2 + 1+1
2 + 2
3 + 1
C 4 1 = 1
Caz = 3
C 43 = 3
C 4 4 = 1
cumulated cost: 20
average: 2.5
C41 + 2C42 + 3C43 +4C44 = 20
l+l+l+l + l
1 + 1 + 1 + 2
1 + 1 + 2 + 1
1 + 1+3
1 + 2 + 1 + 1
1 + 2 + 2
1+3 + 1
1 + 4
2 + 1 + 1 + 1
2 + 1 + 2
2 + 2 + 1
2 + 3
3 + 1 + 1
3 + 2
4 + 1
C51 = 1
C52 = 4
C53 = 6
C54 = 4
C55 = 1
cumulated cost: 48
average: 3
Number of parts in compositions
Class
C, the class of all compositions
Example
1+3 + 1+ 5 + 2 = 12
Size
|c|, the number of «s in c
• 1 ••• 1 • 1 ••••• | •• - ••••••••••••••••
C(z,u) = ^z |c| u pa ' ts( c)

cec
Parameter
parts(c), the number of parts in c
OBCF
Construction
OBGF equation from symbolic method
"Horizontal" OGF for parts in a composition of N
Enumeration
Cumulated cost
C(z, u ) =
C = SEQ ( u SEQ> 0 (Z))
1 1 -z
1 — u
1 -z
1 — z(u + 1)
Qv(tv) = [z N ]C(z, a) = (u + 1 ) N — (u + 1 ) N_1
c/v(1) = 2 N — 2 n_1 = 2 n_1
c^(1) = N2 n_1 - (N - 1 )2 n_2 = (N + 1 )2 n “ 2
Average # parts in a random composition of W
c n(1 )/ c n(1 )
29
Tree parameters
Q. What is the expected root degree of a random tree with N nodes ?

Q. How many leaves in a random tree with N nodes ?
30
Leaves in a random tree
Q. How many leaves in a random tree with N nodes ?
31
Leaves in random trees
Q. How many trees with N nodes and k leaves ?
•i :
<7-10=1 <7-21 = 1
O 31 = 1 ?
Ch 2=1 °
cumulated cost: 3 X
average: 1.5
0 - 4 1 = 1
042 = 3
043 = 1
A. N/2 (next slide)
cumulated cost: 1 0
average: 2
A A A
l IA
a 5] = i
052 = 3
Oss = 6
O 54 = 1
cumulated cost: 35
average: 2.5
32
Leaves in random trees
Class C, the class of all ordered trees
Size \g\, the number of «s in g
Parameter leaves(g), the number of leaves in g
OBGF
Construction
Enumeration OGF
Cumulated cost OGF
Average # leaves in a random tree
G L = uZ+ZxSEQ> 0 ( G L )
~ L , x zG [ (z, u)
C (z,u) = zu + 1 _ G t (Z;0)
C L {z, 1) = C(z)
c " (z ’ 1) = f (1 + 7rbl
[z N ]C[,(z, 1) /N
[z N ]C(z) = -
1 f2N — 2"
[z N ]
v/1 -4z
N V N- 1
[z N ]C(z) =^2 lf ° rW>2/
^concentrated: o/v is O(VN)
33
Root degree in random trees
Q. How many trees with N nodes and root degree k ?
•i :
C D io = 0 C D 21 = 1
C D 31 = 1 J
C D 32=1 °
cumulated cost: 3 X
average: 1.5
C°4 i = 2
C°42 = 2
C D 4B = 1
A. (next slide)
cumulated cost: 9
average: 1.8
A A A
l IA
A
C D si = 5
C D 52 = 5
Cd 53 = 3
C D 54 = 1
cumulated cost: 28
average: 2
34
Root degree in random trees
Class
Size
C, the class of all ordered trees

\g\, the number of «s in g
Parameter deg(g), the degree of the root of g
OBCF
Example
C L (z, u) =
gee
Construction
G D = Z x SEQ> 0 ( uG D )
N 3
N + 1
Enumeration OGF
Cumulated cost OGF
Average # leaves in a random tree
C D (z, u)
1 — uG{z)
C D (z, 1) =
C D u {z, 1) = -
C(z)
zC{z)
— G(z)) 2
(1
1
1.5
1.8
2 /
[z N ]C„ (z, 1) G n+1

[z N ]C(z) C N
- 1
wTtQ = 2N(2N-\)N
b CnZ) (N+1)NN
N+ 1
35
Rhyming schemes
Q. How many ways to rhyme a poem ?
There was a small boy of Quebec A

Who was buried in snow to his neck A
When they said, "Are you friz?" B
He replied, " Yes, I is — B
But we don't call this cold in Quebec! A
TWO roads diverged in a yellow wood, A

And sorry I could not travel both B
And be one traveler, long I stood A
And looked down one as far as I could A
To where it bent in the undergrowth; B
36
Rhyming schemes
Q. How many ways to rhyme an N-line poem with k rhymes ?
Si i = 1
A B
A A
521 = 1
522 = 1
ABC
ABB
ABA
A A B
AAA
531 = 1
5 32 = 3
S3 3 = 1
A B C D
A B C C
A B C B
A B B C
A B C A
ABAC
A A B C
A A B B
A B A B
ABBA
A B B B
A B A A
A A B A S 4 i = 1
A A A B *2 = 7
A A A A *3 = 6
*4= 1
37
Rhyming schemes
Class
S, the class of all rhyming patterns
Example
ABCADABE
Size
number of lines
Parameter
number of rhymes with k lines
OBCF
S(Z,U) = ^ z l s\ u rhymes(s)
ses
"Vertical" construction Za x SEQ ( Za) x Zb x SEQ ( Za + Zb) x Zc x SEQ ( Za + Zb +

Zc) x ...
Vertical OGF
5/c(z)
(1 — z)(1 — 2z)... (1 — kz)
"Stirling numbers of the 2nd kind " (stay tuned)

-yN. .k
Z U ~
/c!
Average # /(-rhyming patterns in an AMine poem
N>0 ^>0
details omitted
(see page 63)
OBGF of Stirling numbers of the 2nd kind (partition numbers)
N>0k >0
z N u k
= Bn
r N
N> 0
(horizontal OGF)
"Bell polynomials"
N k -
= E
k>0
(1 — z)( 1 — 2z)... (1 — kz)

U (vertical OGF)
vertical OGF
coefficients
1 I
3 1
horizontal OGF
coefficients
-*-
15
25
10
1
31
90
65
15
63
301
350
140
[u k ]B 4 (u)
21 1
(1 -z)(1 — 2z)(1 — 3z)
39
PART TWO
Analytic
Combinatorics

Robert Sedgewick

• OBGF examples
CAMBRIDGE
II.3c.MGFs.0BGFs
PART TWO
Analytic
Combinatorics

Robert Sedgewick
f^GBGF examples
CAMBRIDGE
II.3d.MGFs.EBGFs
Basic definitions (combinatorial parameters for labelled classes)
Def. A labelled combinatorial class is a set of labelled combinatorial objects and

an associated
size function that may have an associated parameter.
Def. The exponential bivariate generating function (EBGF)

associated with a labelled class is the power series A(z, u )
object name
size function
z 1
cost(a)
a£A
S' v
a || parameter value
class name
Fundamental (elementary) identity
Mz,u) = J2
aeA
yCost(a)
E E w zN “ k
N>0k >0
Terminology.
The variable z marks size
The variable u marks the parameter

Q. How many objects of size N with value k ?
A. A Nk = N\[z N ][u k ]A(z,u)
Terminology.
BCF: bivariate CF.

MGF: multivariate GF
might add arbitrary

number of markers
With the symbolic method, we specify the class and at the same time characterize
the EBGF
42
The symbolic method for BGFs (basic constructs)
Suppose that A and B are classes of unlabelled objects with EBCFs A(z,u) and B(z,u)
where z marks size and u marks a parameter value. Then
operation
notation
semantics
OGF
disjoint union
A + B
disjoint copies of objects from A and B
A(z, u ) + B(z , u )
labelled product
A* B
ordered pairs of copies of objects,

one from A and one from B
A(z, u)B(z, u )
sequence
SEQ(A)
sequences of objects from A
1 — A(z, u )
Construction immediately gives BGF equation, as for enumeration.
Extends immediately to mark multiple parameters simultaneously with MGFs.
43
Number of different letters in 3-words
Q. How many different letters in a 3-word of length N ?
3 Wu = 3
cumulated cost: 3
average: 1.5
1 1
1 2
1 3
2 1
2 2
2 3
3 1
32 W21 = 3
33 I/I/2 2 = 6
cumulated cost: 1 5
average: 1.667
111 211
112 2 12
113 2 13
12 1 2 2 1
1 2 2 2 2 2
1 2 3 2 2 3
13 1 2 3 1
1 3 2 2 3 2
1 3 3 2 3 3
3 11
3 12
3 13
3 2 1
3 2 2
3 2 3
3 3 1
3 3 2
3 3 3
cumulated cost: 57
average: 2.111
14/bi = 3
H/32 = 1 8
W33 = 6
44
Number of different letters in M-words
Class
Wm, the class of all M -words
Example
314641223441
Size
\w\, the length of w
y\w\
W M (z,u)= V
kv !
wE Wm
Parameter
lets(w), the # of different letters in w
EBGF
Construction
Wm = SEQm (E+ u SET>o(Z) )
EBGF equation from symbolic method
W M (u,z) = (1 + u(e z - 1))
Enumeration EGF
Cumulated cost EGF
W M (1,z) = e
zM
W 0 (1,z) = Me z(M "'>(e z - 1) = Me zM - Me
z(M- 1)
,»- 'SSSrn =^(. - (. - bn
N![z N ] W(1, z)
N mn
2
B
1.667
2.111
Average # different letters in

a random M -word of length A/
45
Number of different letters with a given frequency in M-words
Class
Wm, the class of all M -words
Example
314641223441
Size
| w\, the length of w
7 m
W M (z,u)= T -^u 4(w)
^ \w !
W(E W/v;
Parameter
fk(w), the # of different letters in w
EBGF
Construction
Enumeration EGF
Cumulated cost EGF
Average # letters that appear k times

in a random M-word of length N
Wm = SEQm ( SET±k ( Z) + u SETk(Z) )
W M (u,z) = (e z + (u -
W M 0,z) = e zM
7 k
W u (1,z) = Me z(M_1) —
k\
/V![z N ]lV u (1,z) (N\ l_ k _

N![z N ]W(1,z) 1 M 1
occupancy /■
distribution y
46
Cycles in random permutations
Q. How many permutations of N elements have k cycles
P21 = 1
P22 = 1
cumulated cost: 3
average: 1.5
Pb 1 = 2
Pb 2 = B
Pb b= 1
cumulated cost: 1 1
average: 1.8333
P41 = 6
P42 = 1 1
P43 = 6
P44 = 1
cumulated cost: 50
average: 2.0833
47
Cycles in random permutations
Class P, the class of all permutations
Size \p\, the length of p
Parameter cyc(p), the number of cycles in p
Example
EBCF
Construction
Enumeration EGF
Cumulated cost EGF
Average # cycles in a random permutation
P = SET ( u CYC ( Z) )
„i„ 1
P(z. u) = e 1 — z = (1 — z)
P u(z, 1) = —In 1
— U
N\[z N \P u (z, 1) /TT\

N![z N ]P(z, 1)
N H
1.5
B 1.8BB
4 2.083
concentrated: c tn is 0(^log N)
48
EBGF of Stirling numbers of the 1st kind (cycle numbers)
N > 0 k >0
k
rN
N\
z N
u(u + 1)...(t; + N—1) — (horizontal EGF)
/ \ •
N >0
1 \/c k
Vl(ln—)
^ /d v 1 -z y
k >0
(1 z) u
(vertical EGF)
(EBGF)
A/
[l/^]l/(l/ + 1 )(l/ -h2 )(l/ + 3)—* 4
1
1
24
vertical OGF
coefficients
I I
3 1
II 6 1 <-
50 35 10
120 274 225 85 15
7 720 1764 1624 735 175
1 1 ^3
horizontal OGF
coefficients
21 1
49
Number of cycles of a given length in random permutations
Class P, the class of all permutations
Size \p\, the length of p
Parameter cyc r (p), # of cycles of length r in p
Example
EBCF
Construction

Enumeration EGF
Cumulated cost EGF
P = SET ( CYC*r(Z) + u CYC=r(Z))
1 -,r ,,-,r
, . z uz
In --1-
P(z,u) =e 1 -z r r
PC 1 )
(u-iy/r
1 -z
1 -Z
N\[z N ]P u (z, 1) _/T\

N\[z N ]P(z, 1) \rj
Average # r-cycles in a random permutation

Set partitions
Q. How many ways to partition a set of size of N?
{ 1 } { 2 } { 3 } { 4 }
{ 1 } (2 3 4 }
{ 2 } {1 3 4 }
{ 3 } {1 2 4 }
{ 4 } {1 2 3 }
{ 1 }
{ 2 }
{ 3 }
{1
2 }
{ 3 }
{ 4 }
{ 1 }
{ 2 }
{ 1 }
{2
3 }
{1
3 }
{ 2 }
{ 4 }
{1
2 }
{ 2 }
{1
3 }
{1
4 }
{ 2 }
{ 3 }
52 =
= 2
{ 3 }
{1
2 }
{2
3 }
{ 1 }
{ 4 }
{ 1 }
{ 2 }
{ 3 }
{2
4 }
{ 1 }
{ 3 }
5 b
= 5
{3
4 }
{ 1 }
{ 2 }
{1 2 } {3 4 }
{1 3 } {2 4 }
{1 4 } {2 3 }
{ 1234 }
5 4 = 1 5
51
Set partitions
Q. How many ways to partition a set of size of N into k subsets ?
{ 1 } { 2 }
W {1 2}
S22 = 1
cumulated cost: 3
average: 1.5
{ 1 } { 2 } { 3 }
{ 1 } (2 3 }
{ 2 } {1 3 }
{ 3 } {1 2 }
{1 2 3 }
531 = 1
5 3 2 = 3
533 = 1
cumulated cost: 1 1
average: 2
{ 1 } { 2 } { 3 } { 4 }
{ 1 }
{2 3
4 }
{ 2 }
{1 3
4 }
{ 3 }
{1 2
4 }
{ 4 }
{1 2
3 }
{1 2 }
{ 3 }
{ 4 }
{1 3 }
{ 2 }
{ 4 }
{1 4 }
{ 2 }
{ 3 }
{2 3 }
{ 1 }
{ 4 }
{2 4 }
{ 1 }
{ 3 }
{3 4 }
{ 1 }
{ 2 }
{1 2 } {3
4 }
{1 3 } {2
4 }
{1 4 } {2
3 }
{ 1234 }
S 4 i = 1
S 4 2= 7
543 = 6
544 = 1
cumulated cost: 37
average: 2.466
52
Number of subsets in set partitions
Class
S, the class of all set partitions
Example
{1} {2 5 6} {3 7 8} {4}
Size
size of the set
5(z, u) = ^j|u su6sets(s)
5<ES 5 '
Parameter
number of subsets in the partition
EBGF
Construction
Enumeration EGF
Cumulated cost EGF
Average # subsets in a random set partition
5 = SET ( u SET> 0 (Z))
5(z, u ) — e a ( e — 1)
5(z,1) = e eZ - 1
5„(z, 1) = (e z - 1)e( e
N'.[ zN ]S u (z, 1) _
N\[z N ]S(z, 1)
1 )
need complex asymptotics

(stay tuned)
EBGF of Stirling numbers of the 2nd kind (partition numbers)
N> 0 k>0
N) z
k ) ~N\ U
= J2 Bn (°)
N> 0
= 5]( ez
k>0
(horizontal EGF)
"Bell polynomials"
(vertical EGF)
= e U ^ Z 1 ) (EBGF)
horizontal EGF
coefficients
/C^
6 7
vertical EGF
coefficients
1
4
<-
[(/]5 4 (t/)
15
25
10
31
90
65
15
63
301
350
140
21 1
N![z w ]l(e^-1) 3
54
Mappings
Def. A mapping is a function from the set of integers from 1 to N onto itself.
Example
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
32 33 34 35 36 37
9 12 29 33 5 20 30 37 26 20 13 8 2 33 29 2 35 37 33 9 35 21 18 2 25 1 20 33 23 18
29 5 5 9 11 5 11
55
Mapping EGFs (see lecture on EGFs)
Combinatorial class C, the class of Cayley trees < -labelled, rooted, unordered
Construction
C = Z * (557(C)) «— "a tree is a root connected to a set of trees"
EGF equation
C(z) = ze c<yZ )
Combinatorial class Y, the class of mapping components
Construction
y — CYC(C) <— "a mapping component is a cycle of trees"
EGF equation
Y(Z) - ln 1 - C(z)
Combinatorial class C, the class of Cayley trees

Construction
M — SET(CYC(C)) <— "a mapping is a set of components"
EGF equation
MW-expOn z) )_ z)
56
Mapping parameters
are available via EBGFs based on the same constructions

Ex 1. Number of components
Construction M = SET(uCYC(C))
1 1
EGF equation M{z) = exp (u In - _ ) =
Ex 2. Number of trees (nodes on cycles)
Construction M = SET(CYC(uC))
1 1
EGF equation M(z) = exp(ln -—= - ——
v 1 - uC{z)' 1 — uC(z)
Q. Moments? Coefficients? Other parameters?
A. Stay tuned for general theorems from complex aysmptotics.
57
"We shall now stop supplying examples that could be multiplied ad libitum, since
such
calculations greatly simplify when interpreted in the light of asymptotic analysis"
— Philippe Flajolet, 2007
PART TWO
Analytic
Combinatorics

Robert Sedgewick
f^GBGF examples
CAMBRIDGE
II.3d.MGFs.EBGFs
PART TWO
o^pmbinalorial parafn^ters and MGFs
s r©BGF examples
• Exercises
Analytic
Combinatorics

Robert Sedgewick
CAMBRIDGE
II.3e.MGFs.Exercises
Note 111.17
Leaves in Cayley trees
Analytic
Combinatorics

Robert Sedgewick
> III-17- Leaves and node-degree profile in Cayley trees. For Cayley trees, the
bivariate EGF
with u marking the number of leaves is the solution to
T(z,u) = uz + z(e T{z ' u) - 1).
(By Lagrange inversion, the distribution is expressible in terms of Stirling

partition numbers.)
The mean number of leaves in a random Cayley tree is asymptotic to ne ~ 1 . More
generally, the
mean number of nodes of outdegree k in a random Cayley tree of size n is asymptotic
to
Degrees are thus approximately described by a Poisson law of rate 1. <1
61
Note 111.21
After Bhaskara Acharya
Analytic
Combinatorics

Robert Sedgewick
> 111 . 21 . After Bhaskara Acharya ( circa 1150 ad). Consider all the numbers
formed in decimal
with digit 1 used once, with digit 2 used twice.with digit 9 used nine times. Such
numbers
all have 45 digits. Compute their sum S and discover, much to your amazement that S
equals
458755596000061532190847692863 999999999999999 54124440399993846780915230713600000.
This number has a long run of nines (and further nines are hidden!). Is there a
simple explana¬
tion? This exercise is inspired by the Indian mathematician Bhaskara Acharya who
discovered
multinomial coefficients near 1 150ad. <1
45875559600006153219084769286399999999999999954124440399993846780915230713600000
62
Assignments
1. Read pages 151-219 in text.
2. Write up solutions to Notes III. 1 7 and 111.21.
Analytic
Combinatorics

Robert Sedgewick
3. Programming exercise.
Program III. 1 . Write a program that generates 1000 random

permutations of size N for N = 1 0 3 , 1 0 4 , ... (going as far as you
can) and plots the distribution of the number of cycles, validating
that the mean is concentrated at Hn.
63
PART TWO
Analytic
Combinatorics

Robert Sedgewick
3. Combinatorial Parameters
and MGFs

AC03-MGFs Djvu

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

AC03-MGFs Djvu

Hochgeladen von

Copyright:

Verfügbare Formate

ANALYTIC COMBINATORICS

Philippe Flajolet and

Analytic combinatorics overview

4. Rational & Meromorphic

Philippe Flajolet and

3. Combinatorial parameters and MGFs

Natural questions about combinatorial parameters

What is the average number

What is the average number

What is the average

What is the average coot

What is the average number

What is the average number

Natural questions about combinatorial parameters

Problem: Average-case results are sometime easy to derive but unsatisfying.

Example. Separate chaining hashing randomly assigns N keys to M lists.

Q. Average length of a list ?

A trivial result that is not very useful because it

Ex: All the keys could be on one list.

Solution: Find distribution (probability parameter value is k for all k )

Practical compromises. Ex: B 0unc j probability that list length deviates

•compute average and variance

Goals for this lecture: Learn enough about parameters to be able to

•compute moments and extremal values (in practice)

Natural questions about combinatorial parameters

How many ways to partition

How many permutations of

size N have k cycles?

How many compositions

(sequences of positive integers

How many partitions (sets

How many trees with N

How many letters

Basic definitions (combinatorial parameters for unlabelled classes)

Def. A combinatorial class is a set of combinatorial objects and an associated size

Def. The ordinary bivariate generating function (OBGF)

Fundamental (elementary) identity

A(z) = z \ a \u cost ( a ) — y~^/\ N ^z N g /c

The variable z marks size

The variable u marks the parameter

Q. Flow many objects of size N with value k?

BCF: bivariate CF.

might add arbitrary

Combinatorial enumeration: classic example

Q. How many binary strings with N bits?

Combinatorial parameters: classic example

Q. How many A/-bit binary strings have k 0 bits?

OBGF of binomial coefficients

The symbolic method for OBGFs (basic constructs)

disjoint copies of objects from A and B

ordered pairs of copies of objects,

sequences of objects from A

Construction immediately gives OBGF equation, as for enumeration.

Extends immediately to mark multiple parameters simultaneously with MGFs.

z l a li/ cosf ( a ) 4 z |6| a cost(b) — ^( Zj ^ _|_ b (^ z ^ ^

|a| + |b| yCost(a)+cost(b )

SEQ t (A) = A u + /\ f2 + /\ t3 + ...

where T = ti, t 2 , t 3 ,... is a subset of the integers

A(z , tv/ 1 + /A(z, i/) t2 4 /\(z, tv/ 3 4 ...

5£Q(A) = e + /\+A 2 +/\ 3 + ...

i +a( Z’ u ) +a (Z’ 4 + ---- 1 _ a(ZjU)

Combinatorial parameter example: 0 bits in bitstrings

Class B, the class of all binary strings