AC05-PoleApps Djvu

ANALYTIC COMBINATORICS
Analytic
Combinatorics
Philippe Flajolet and

Robert Sedgewick
http://ac.cs.princeton.edu
5. Applications of
Rational and Meromor
Asymptotics
PART TWO
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
COMPLEX ASYMPTOTICS
asymptotic
estimate
desired
result !
Bottom line from last lecture
Analytic transfer for meromorphic GFs: f(z)/g(z ) ~ c $ N
• Compute the dominant pole cx (smallest real with g(z) = 0)
• Compute the residue h~] = -f(a)/g\a)
• Constant c is / 7-1 /cx.
• Exponential growth factor (3 is 1 /cx
Not order 1 if g\a) = 0.

Adjust to (slightly) more
complicated order M case.
This lecture: Numerous applications
PART TWO
Analytic
Combinatorics

Robert Sedgewick
Âpplications of
Rational and Meromorphic Asymptotics
• Bitstrings
• Other familiar examples
• Compositions
• Supercritical sequence schema
II.5a.RMapps.Bitstrings
Warmup: Bitstrings
How many bitstrings of length N ?
000
0000
000 1
00 10
0011
0100
001
0101
00
0 10
0 110
0 1
Oil
0 111
10
100
1000
11
101
1001
110
10 10
03
l\j
ll
111
1011
B 3 = 8
1100
1101
1110
1111
Ba= 16
counting sequence
Bn = 2
Y,2 N z" = '£[2z) N
N >0 N >0
OGF
1 — 2z
~ 1 — 2z
Warmup: Bitstrings
Asymptotics
B, the class of all bitstrings

B = E + (Z 0 + Zi)xB
Dominant singularity: pole at a — 1/2
> f
[z n ]B(z) = 2 n
Residue:
Coefficient of z N :
h -1 =
h -1
rsj -
f(z) 1
g'O) 2
Example 1: Bitstrings with restrictions on consecutive Os
How many bitstrings of length N have no two consecutive Os ?
To = 1
Ti = 2
01
10
11
T 2 = 3
Oil
0 10
101
110
111
T 3 = 5
0 10 1
0 110
0111
10 11
10 10
1101
1110
1111
T 4 = 8
0 10 11
01010
0 110 1
oino
0 1111
10 10 1
10110
10111
110 11
11010
1110 1
11110
11111
T 5 =1B
Boo, the class of all bitstrings having no 00

Boo = i + Zo + (Zo + Zo xZi ) x Boo
= ] -^- Z 2
[z n ]B 00 (z) = ^=4> n
Dominant singularity: pole at <fr
Residue: /?_i =
g'(4>)
Coefficient of z N \
hi n\ N
V5
C 2 P 2 with
f3 2 = 1.61803
c 2 = 1.17082
y/5 — 1
2
a/5 + 1
2
_ 1 +3
~ 1 + 20
h—i~4, N
0 + 20 2
00 = 1
0 2 = 0 + 1
B 4 , the class of all bitstrings having no 0 4
B4 = Z<4 (E + 1 B4)
> f
B 4 {z) = (1 +z + z 2 +z 3 )( 1 + zB 4 (z))
1 + z + z 2 + z 3
> f
Dominant singularity: pole at a
. /"(z) 1 -\- oi “t - ol2 -(- or 1
Residue: n_i = —=--- -—3 -—r
g(z) a + 2a + 3a 2 + 4a 3
Coefficient of z N : [z N ]BAz ) ~ f
[z N ]B 4 (z) ~ c 4/ 3^
with
/3 4 = 1.9276
c 4 = 1.0917
1 + Z
10
Information on consecutive Os in GFs for strings
[from AC Part I Lecture 5]
Bm(z ) = ^ z^ = ^{# of bitstrings of length N with no 0 M }z N
n> o
_ 1 +Z + Z 2 +...+z M -' _ 1 -Z M
- 1 - z - z 2 - ... z M “ 1 — 2z + z M+1
Bm(z/2) = ^ ({# of bitstrings of length N with no runs of M 0s}/2 N )z N
N> 0
Bm( 1/2) = ^{# of bitstrings of length N with no runs of M 0s}/2 N
N> 0
= ^ Pr {1 st N bits of a random bitstring have no runs of M Os}
N> 0
= ^ Pr (position of end of first 0 M is > N ] = Expected position of end of first 0

M
N> 0
Theorem. Probability that an A/-bit random bitstring has no 0 M : [z n ]Bm(z/ 2) ~

Cm(Pm/ 2) N
Theorem. Expected wait time for the first 0 M in a random bitstring: Bm( 1/2) = 2
m+1 -2
12
Autocorrelation
The probability that an A/-bit random bitstring does not contain 0000 is ~1.091 7 x
. 96328^
The expected wait time for the first occurrence of 0000 in a random bitstring is
30.
Q. Do the same results hold for 0001?

A. NO!
0001 occurs much

earlier than 0000
10111110100101001100111000100111110110110100000111100001
Observation. Consider first occurrence of 000.
•0000 and 0001 equally likely, BUT
•mismatch for 0000 means 0001, so need to wait four more bits
•mismatch for 0001 means 0000, so next bit could give a match.
Q. What is the probability that an A/-bit random bitstring does not contain 0001?
Q. What is the expected wait time for the first occurrence of 0001 in a random
bitstring?
13
Constructions for strings without specified patterns
Cast of characters:
p — a pattern p 101001010
S P — binary strings that do not contain p Sp 10111110101101001100110000011111
Tp binary strings that end in p 7 loinnoiOllOlOOllOOllOlOlOOlOlO
and have no other occurrence of p
First construction
• Sp and Tp are disjoint
• the empty string is in S P
• adding a bit to a string in S P gives a string in S P or T p
Sp + T p = E + Sp x {Zq + Zi }
14
Constructions for bitstrings without specified patterns
Every pattern has an autocorrelation polynomial
• slide the pattern to the left over itself.
• for each match of / trailing bits with the leading bits include a term z \p\-'
101001010
101001010
101001010
101001010
101001010
101001010
101001010
101001010
101001010
101001010
autocorrelation
polynomial
Cioiooioio(z) — 1 + z 5 + z 7
15
Constructions for bitstrings without specified patterns
Second construction
• for each 1 bit in the autocorrelation of any string in T P add a “tail”
• result is a string in S P followed by the pattern
p 101001010
a string in T p 10111110101101001100110101001010
f 10111110101101001100110 101001010
strings in S P ^—►1011111010110100110011010100101001010
l 10111110101101001100110 101001010 1001010
first tail
is null
101001010
101001010
101001010
101001010
101001010
101001010
10100101001010
101001010
1010010101001010
101001010
Sp x (pi — x
C/# 0
16
Bitstrings without specified patterns
How many A/-bit strings do not contain a specified pattern pi
Sp — the class of binary strings with no
OCFs
s p ( z ) = Z z ' s|
sESp
Tp — the class of binary strings that end in p

and have no other occurence
M z ) = Z z ' s|
sETp
Constructions
OGF equations
Solution
Sp + T p — E + S p x {Zq + Zi}
5 p (z) + 7p(z) = 1 + 2 zSp(z)
Sp{z)
Sp x {p} — T p x N ' {t/}
c,/0
5 p (z)z p = Tp(z)c p (z)

c P (z)
z p + (1 -2 z)Cp(z)
„ w , ( 8 D is the dominant root of z p + (1 —2z)c D (z)
Extract cofficients z N S„(z) ~ Cd / 3" where { ^ p , , , , v ; pW
1 Jpw Pp Icp = [explicit formula available]
17
Bitstrings without specified patterns
Asymptotics
18
PART TWO
Analytic
Combinatorics

Robert Sedgewick
Âpplications of
• Bitstrings
• Compositions
II.5a.RMapps.Bitstrings
PART TWO
Analytic
Combinatorics

Robert Sedgewick
Âpplications of
• Bitstrings
• Compositions
CAMBRIDGE
II. 5b.RMapps.Exam pies
Example 2: Derangements
How many permutations of size N have no singleton cycles ?
Di = 0
®©
Di = 1
©©©
Db = 2
©©©©
J©
J©
©©(
®@i
©© ( 2 )
© (4 s
©®
©
©J® ©IJ©
©^ ©
^ ©J® ©J©
©®
®®
®/®X ®/®\|
©J© ©J©
©s
©s
, ©I V ©^
©J© ©J®|
Da = 9
D, the class of all permutations

with no singleton cycles
D = SET(CYC>i(Z)
> f
D(z)
N\[z n ]D(z) ~
N\
Dominant singularity: pole at 1
Residue: h _i
m =e -i
g'O)
[z N ]D(z)
h ~^N = 1
1 e
N\/e
Dn
.7357...
2.2072...
8.8291...
9
5
44.1455...
44
estimates are extremely accurate

even for small N
22
Dm, the class of all permutations

with no cycles of length < M
Dm = SET(CYC>m(Z)
> f
D m (z)
> f
N\[z n ]D(z) ~
N\
Dominant singularity: pole at 1
Residue: /?_i
^0 ) _ e -H M
g'O) "
[ Z N ]D(Z)
h -^N = _L
1 e HM
23
24
e -z-z 2 /2-z 3 /3-z 4 /4-z 5 /5-z 6 /6-z 7 /7-z 8 /8-z 9 /9-z 10 /10
25
Example 3: Surjections
How many words of length N are M-surjections for some M ?
1 2
R} = 1 2 1
Ri = 3
12 1 13 2
1 2 2 2 1 3
2 11 2 3 1
2 12 3 12
2 2 1 3 2 1
Kb = 13
"coupon collector sequences"
For some M, each of the first M letters appears at least once.
1111
1112
112 1
12 11
2 111
112 2
12 12
2 112
2 12 1
2 2 11
12 2 1
12 2 2
2 12 2
2 2 12
2 2 2 1
Ra = 75
12 3 1
13 2 1
2 13 1
2 3 11
3 12 1
3 2 11
12 13
13 12
2 113
3 112
112 3
113 2
12 3 3
13 2 3
2 13 3
2 3 13
3 12 3
3 2 13
13 3 2
2 3 3 1
3 13 2
3 2 3 1
3 3 12
3 3 2 1
12 3 2
13 2 2
2 13 2
2 3 12
3 12 2
3 2 12
12 2 3
2 12 3
2 3 2 1
3 2 2 1
2 2 13
2 2 3 1
12 3 4
13 2 4
2 13 4
2 3 14
3 12 4
3 2 14
12 4 3
13 4 2
2 14 3
2 3 4 1
3 14 2
3 2 4 1
14 2 3
14 3 2
2 4 13
2 4 3 1
3 4 12
3 4 2 1
4 12 3
4 13 2
4 2 13
4 2 3 1
4 3 12
4 3 2 1
26
R, the class of all surjections
R = SEQ(SET> 0 (Z))
-+-
Dominant singularity: pole at z = In 2
[z n ]R(z)
1
2(ln2) N + 1
Residue:
1 1
in 2 )
A/!/2(ln 2) w+1
Rn
B.0027...
12.9962...
13
74.9987...
75

even for small N
How many words of length N are double surjections for some M ?
11 111 1111 1122
12 12
Ri = 1 fh = 1 2 112
2 12 1
2 2 11
12 2 1
/? 4 = 7
"double coupon collector sequences"
For some M, each of the first M letters appears at least twice.
11111
1112 2
112 12
112 2 1
12 112
12 12 1
12 2 11
2 1112
2 112 1
2 12 11
2 2 111
112 2 2
12 2 12
2 12 2 1
2 2 112
1 2 2 2 1
2 2 2 1 1
2 12 12
12 12 2
2 112 2
2 2 12 1
R.5 = 21
28
R, the class of all double surjections
R = SEQ(SET>i(Z)
R{z)
1 — (e z — z — 1)
2 + z — e 2
1 N\
p + 1 p N+1
Singularities where e z = z + 2
Dominant singularity: pole at p =
i 1 1
Residue: =
g'(p) eP - 1
1.14619
" pTT
29
2 + z + z 2 / 2 — e z
2 + z + z 2 /2 + z 3 / 6 — e z
30
Example 4: Alignments
How many sequences of labelled cycles of size N ?
O i = 1
®©
©©
0 2 = B
©@©
©©©
©0®
s©
©©©
©©©
@ §
@ 0
©@©
Ob = 14
OOOO x 24
x 12
°©>
32
Example 3: Alignments
O, the class of all alignments

O = SEQ(CYC(Z)
> f
O(z)
1 — In
Singularities where In--
1 - z
Dominant singularity: pole at
Residue: /?_! = ——-—
g'O - 1/e.
z = 1
1
On ~
N\
A/!/e(l -1 /e) N+1
On
e(1 — 1 /e) N+1
2.9129...

even for small N
1 B.8247...
14
87.4816...
88
Example 4: Set partitions
Q. How many ways to partition an N-element set into r subsets ?

< - see Lecture 3
Sni = 2 W -1 ABC
S 33 = 1
only B B B... B
disallowed
Application: rhyming schemes
A B
A B
B
A
A B
A
A
A B
A A
B
B
A B
C
A
S 43
A
C
Ss3= 25
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; g
34
Example 4: Set partitions
Sr, the class of all poems with r rhymes
Sr = Za x SEQ (Z A ) x Zb x SEQ(Za + Zb) x

Zc x SEQ ( Za + Zb + Zc) x ...
> f
(1 z) (1 — 2z)... (1 -rz)
> f
[z N ]S\z) ~
r!
Singularities at 1, 1/2, 1/3, ... ]/r
Dominant singularity: pole at ]/r
Residue: /7_i
fQA) _ i
g'OA) r ‘ r!
35
PART TWO
Analytic
Combinatorics
Robert Sedgewick
Âpplications of
• Bitstrings
• Compositions
CAMBRIDGE
II. 5b.RMapps.Exam pies
PART TWO
Analytic
Combinatorics

Robert Sedgewick
Âpplications of
• Bitstrings
• Compositions
CAMBRIDGE
II.5c.RMapps.Compositions
Example 5: Compositions
Q. How many ways to express N as a sum of positive integers?
/i = 1
l + l
2
li = 2
A. In =2 n-i
1 +
1
+
1 +
+ 1
+
3
+ 1
+ 1
1 +
2 +
1
2
k =
2
+
1
3
B + 2
4 + 1
5
Is = 16
38
/, the class of all positive integers

/ = SEQ> 0 (Z)
> f
In = 1 for N > 0
Singularity: pole at 1
Residue: /7_ n
f(1)
g'(1)
= 1
39
C, the class of all compositions
C= SEQ(/)
40
Q. How many ways to express N as a sum of 1 s and 2s ?
1 + 1
+ 1
1 +
1 +
+
= 1
1 +
+ 1
2 +
+ 2
f 2
+ 1
Fb =
3
2 +
f 4 =
A. Fibonacci numbers
2
+
2
+
Fs = 8
41
F, the class of all compositions

composed of 1 s and 2s
F = SEQ(Z + Z 2 )
F n ~
<j) N
7 !
— = .4472 and 6= 1.618
V5
Dominant singularity: pole at (f)
Residue: Ai_i =
Coefficient of z N \
f {4>) _ 1
1+20
/ j _! /'\\ N +1
lN
1 +20
00 = 1
0 ^ — 0+1
1 + 20 =
42
Q. How many ways to express N as a sum of primes ?
Pi = 1
2 + 2
P 4 = 1
P3 = 1
2 + 3
B + 2
5
Ps = 3
2 + 2
+ 2
3 +
2
2
Pe =
Pi = 6
+
2
2 +
+ 2
2
2
+
5
Ps
Pg = 10
43
P, the class of all compositions
composed of primes
P = SEQ(Z 2 + + Z5 + Z 7 + . . .)
"] _ ^2 _ ^3 _ ^5 _ ^7 _ 1 _
Dominant singularity: pole at 1 / (3 = .6774
> f
[z N ]P{z) ~ A/3 n with
P = 1.4762
A = .3037
Note: periodic oscillations are present in the next term
interesting
calculations
omitted
(see text)
pp. 298-299
44
Example 6: Denumerants (partitions from a fixed set)
Q. How many ways to make change for N cents?

l + l+ l + l + l + l + l + l + l+ l + l+ l + l
5 + 1 + 1+1 + 1+1 + 1 + 1 + 1 + 1
5 + 5+1 + 1 + 1 + 1
10 + 1 + 1 + 1 + 1
Qm = 4
l + l + l + l + l + l + l + l + l + l + l + l + l + l
5 + 1 + 1 + 1 + 1 + 1 + 1 + 1+1 + 1+1
5 + 5+1 + 1 + 1 + 1 + 1
5 + 5 + 5
10 + 1 + 1 + 1 + 1 + 1
10 + 5
Qi 5 = 6
45
Example 6: Denumerants (partitions from a fixed set)
Q, the class of all partitions

composed of]s, 5s, 10s, 25s
Q = MSET(Z+ Z5 + Z 1( > + Z25 )
Q ^ = (1 -z)(1 -z 5 )(1 -z 10 )( 1 -z 25 )
Dominant
na(z) ~
N 3
1 -5 • 10-25 - 3 !
N 3
7500
Residue:
singularity: po/e of order 5 1
4 = lim(1 — z) 4 Q(z)
“ 1 -5-10-25
Iim -- = -t t" = -
z->1 1 + Z + Z 2 + . . . + Z^ 1 f
46
PART TWO
Analytic
Combinatorics

Robert Sedgewick
Âpplications of
• Bitstrings
• Compositions
CAMBRIDGE
II.5c.RMapps.Compositions
PART TWO
Analytic
Combinatorics

Robert Sedgewick
CAMBRIDGE
Âpplications of
• Bitstrings
• Restricted compositions
II.5d.RMapps.SeqSchema
Sequence schema
Terminology. A schema is a treatment that unifies the analysis of a family of

classes.
Definition. A class that admits a construction of the form F = SEQ( G), where G is
any class
(labelled or unlabelled) is said to be a sequence class, which falls within the
sequence schema.
Enumeration:
F = SEQ( G)
A = 1 fN = [z N }F(z)
1 ~ C ( Z ) gN = [Z N ]C(Z)
unlabelled case: number of structures is fw

labelled case: number of structures is N ! f/v
Parameters: mark number of G components with u
F = SEQi u G) — F(z, u) = - - E—
1 — uG(z)
. f*(z, U )
mark number of Gk components with u

F = SEQ(u Gk + G\ Gk) —
1 - {C{z) + (u - ^g k z k )
49
Supercritical sequence classes
Supercriticality. A technical condition that enables us to unify the analysis of

sequence classes.
Definition. Supercritical sequence classes.
A sequence class F = SEQ(G) is said to be supercritical if G( p) > 1 where G(z) is

the
generating function associated with G and p>0 is the radius of convergence of C(z).
Example: GF for integers: /(z) = ^ Z
radius of convergence: P= 1—e for any e>0
supercriticality test: /(I - e) =-1 > 1 for e < 1/2
Therefore, the class of compositions C = SEQ(I) is supercritical.
Note: For simplicity, we ignore periodicities in GFs in this lecture:
Definition. Strong aperidoicity. A GF C(z) is said to be strongly aperiodic when

there does not exist an integer d> 1 such that G(z) = h(z d ) for some h(z)
analytic at 0.
50
Transfer theorem for supercritical sequence classes
Theorem. Asymptotics of supercritical sequences. If F = SEQ(G) is a strongly

aperiodic
supercritical sequence class, then l[z N ]F(z)
G'(A) A n +t
jwhere A is the root of C(A) = 1 in (0, p).
radius of
convergence of C(z)
Proof sketch:
• G(z) increases from C(0) = 0 to C(p)>l, so X is well defined.
• At X, G(z) admits the series expansion G(z) = 1 + G'(A)(z - A) + G // (A)(z - A)

2 /2! H-
1 1 1
• Therefore, F(z) = 1/(1 -G(z)) has a simple pole at X, andF(z) ~ ——
G'(A)(z - A) AG'(A) 1 -z/A
51
Transfer theorem for supercritical sequence classes
Theorem. Asymptotics of supercritical sequences. If F
1 1
supercritical sequence class, then [z N ]F(z)
G'(A) A N+1
construction
F(z)
surjections R = SEQ (SET> o( Z ))
2 - e 2
alignments O = SEQ (CYC( Z ))
1 — In
1 -z
compositions C = SEQ( I )
1-2
SEQ(G) is a strongly aperiodic

where A is the root of C(A) = 1 in (0, p).
G(z)
coefficient
asymptotics
e z — 1
In 2
N!
2(ln2) A/+1
In 1
1 -z
N!
e(1 - 1 /e) N+1
2^-i
1 -z
Analytic transfer
Parts in compositions
Q. How many parts in a random composition of size N ?
1 + 1
2
1.5
1 +
+
1
1 +
+ 1
2
1
+ 1
+
2
4
3
+
2
2.5
B + 2
4 + 1
5
53
Components in surjections
What is the expected value of M in a random surjection of size N ?
1 i
1 2
2 1
(1 + 2-2)/3 = 1.666
12 1
12 2
2 11
2 12
2 2 1
1 2 3
1 3 2
2 13
2 3 1
3 1 2
3 2 1
(1 +2-6 + 3-6)/l B = 2.B84
For some M, each of the first M letters appears at least once.
1111
1112
112 1
12 11
2 111
112 2
12 12
2 112
2 12 1
2 2 11
12 2 1
12 2 2
2 12 2
2 2 12
2 2 2 1
12 3 1
13 2 1
2 13 1
2 3 11
3 12 1
3 2 11
12 13
13 12
2 113
3 112
112 3
113 2
12 3 3
13 2 3
2 13 3
2 3 13
3 12 3
3 2 13
13 3 2
2 3 3 1
3 13 2
3 2 3 1
3 3 12
3 3 2 1
12 3 2
13 2 2
2 13 2
2 3 12
3 12 2
3 2 12
12 2 3
2 12 3
2 3 2 1
3 2 2 1
2 2 13
2 2 3 1
(1 + 2-14 + 3-36 + 4- 24)/75 =
12 3 4
13 2 4
2 13 4
2 3 14
3 12 4
3 2 14
12 4 3
13 4 2
2 14 3
2 3 4 1
3 14 2
3 2 4 1
14 2 3
14 3 2
2 4 13
2 4 3 1
3 4 12
3 4 2 1
4 12 3
4 13 2
4 2 13
4 2 3 1
4 3 12
4 3 2 1
3.106
54
Components in alignments
How many cycles in a random alignment of size N ?

(1 + 2 • 2)/3 = 1.666
©@©
©8
nj
8 ©
nj
©©©
© 0 ®
©8
nT
(T)
p©
nT
©©©
©@©
©8
(YJ
8 ©
(YJ
©@©
(1 -2 + 2-6 + 3 - 6)/14 = 2.286
OOOO x 24
OujO
x 6
( 1-12 + 2-16 + 3-36 + 4 - 24)/88 = 2.818
55
A poster child for analytic combinatorics
Parts in compositions
Q. How many parts in a random composition of size N ?
l+l+l+l+l
Components in surjections
l + l
2
1.5
What is the expected value of M in a random surjection of size N ?
1 i
1 2
2 1
(1 + 2-2)/3 = 1.666
12 1 13
1 2 2 2 1
2 11 2 3
2 12 3 1
2 2 1 3 2
(1 +2-6 + 3 • 6)/l 3 = 2
1111 17 3 1 1233 1 7 3 4
Components in alignments
How many cycles in a random alignment of size N ?
For some M, each of the first M letters appears at le
(1 -2 + 2-6 + 3 * 6)/1 4 = 2.286
OOOO x 24
OO0 oQo Qoo X12
(1-12 + 2-16 + 3-36 + 4-24)/88 = 2.818
Such questions can be answered immediately via general transfer theorems
56
Number of components in supercritical sequence classes
Corollary. Number of components in supercritical sequence classes. If F = SEQ(G) is

a strongly aperiodic
supercritical sequence class, then the expected number of C -components in a random
F-component of
size N is
N + 1
AC'(A)
C"(A)
C'(A)2
with variance
AC"(A) + C'(A)-C'(A)\ ,
A 2 C'(A) 3
A is the root of
C( X) = 1 in (0, p)
Proof idea:
1 [ 7 Nl C ( z )
f N [ J (1-C(Z))2
[further details omitted]
57
Number of components in supercritical sequence classes
Corollary. Number of components in supercritical sequence classes. If F = SEQ(G) is

a strongly aperiodic
supercritical sequence class, then the expected number of G -components in a random

F-component of
N+1 G"( A) n 2 AG"(A) + G'(A) — G'(A) 2
slze N15 ~ xo(X) + cw ■ Wlth vanance ° N ~-Zcw- N ■
construction
F(z)
C(z)
expected number
of components
compositions
C = SEQ( I )
1 z
1 -z
surjections
R = SEQ (SET> o( Z ))
e z — 1
In 2
N
2 - e z
' 2 In 2
alignments
O = SEQ (CYC( Z ))
1 In 1 1
1 — z
In 1
1 -z
i-l
rsj -
e - 1
A is the root of
C( A) = 1 in (0, p)
Same idea extends to give profile of component sizes.
58
PART TWO
Analytic
Combinatorics

Robert Sedgewick
CAMBRIDGE
Âpplications of
• Bitstrings
II.5d.RMapps.SeqSchema
PART TWO
Analytic
Combinatorics

Robert Sedgewick
Âpplications of
• Bitstrings

• Summary
II.5e.RMapps.Summary
AC via meromophic asymptotics: summary of classic applications
class
specification
generating function
coefficient asymptotics
bitstrings
B = E + (Zo + Zi ) X B
1 — 2z
2 /v
derangements
D = SET(CYC>o( Z ))
e _z
1 -z
N\
rxj -
surjections
R = SEQ(SET>o(Z))
2 - e z
2(ln2) N+1
alignments
O = SEQ (CYC( Z ))
N\
e(1 — 1 /e) N+1
set partitions
Sr= ZxSEQ(Z)xZxSEQ(Z+Z)x...
z r
(1 - z)... (1 -rz)
r N
r!
integers
1 = SEQ> 0 ( Z ))
1 -z
compositions
C = SEQ(I)
1 z
1 1 -z
2^-i
61
AC via meromophic asymptotics: summary of classic applications (variants)
class
specification
generating function
coefficient
asymptotics
bitstrings
4 = z<4 ( + 2 l B 4 )
1 + z + z 2 + z 3
1.092(1 .928) n
with no 0000
1 — z — z 2 — z 3 — z 4
generalized
derangements
D = SET(CYC>m( Z ))
,2 7 3 7 M
0 Z 2 3 M
N\
1 -z
qHm
double
R = SEQ(SET>i(Z))
N\
400S
surjections
2 + z - e z
• 1 vJ U 9 . \ \ /
(1.146) N
compositions
F = SEQ(Z + Z2)
.4472(1.61 8) n
of 1 s and 2s
1 - z - z 2
compositions
of primes
P = SEQ(Z 2 + Z3 + Z5 + . . .)
.3037(1.476) n
1 — z 2 — z 3 — z 5 — z 7 — . . .
denumerants
Q = MSET(Z + Z5 + Z 10 + Z25)
(1 — z)(1 -z 5 )(1 -z 10 )(1 -z 25 )
N 3
7500
62
"If you can specify it, you can analyze it"

1. The transfer theorem for
meromorphic GFs enables
immediate analysis of a
variety of classes.
2. Variations are handled

just as easily.
class
specification
generating function
coefficient asymptotics
bitstrings
B = E + (Zo + Zi)xB
1 — 2z
2 n
derangements
D = SET(CYC>o( Z ))
e~ z
1 -z
N\
surjections
R = SEQ(SET>o(Z))
2 — e z
2(ln2) N+l
alignments
O = SEQ ( CYC( Z ))
’-■“rh
Nl
e(1 — 1 /e) N+1
class
specification
generating function
coefficient
asymptotics
bitstrings
with no 0000
B 4 = Z< 4 (E + ZiB 4 )
1 + z + z 2 + z 3
1 - z - z 2 - z 3 - z 4
1.092(1,928)~
generalized
derangements
D = SET(CYC>m( Z ))
0 z 2 i •- M
1 -z
N! L
double
R = SEQ(SET>i(Z))
1065 M
surjections
2 + z - e 7
' 1065 (1.146) n
compositions
of 1 s and 2s
F = SEQ(Z + Z 2 )
1 - z - z 2
.4472(1.61 8) n
compositions
of primes
P = SEQ(Z 2 + Z3 + Z5 + . . .)
1 - z 2 - z 3 - z 5 - z 7 - .. .
.3037(1.476) n
denumerants
Q = MSET(Z + Z* + Z'o + Z 2 5)
N 3
(1 -z)(1 -z 5 )(1 -z'0)(1 -z 25 )
7500
3. The supercritical sequence schema unifies the analysis for

an entire family of classes, including analysis of parameters.
Note: Several other schemas have been developed (see text).
Next: GFs that are not meromorphic (singularities are not poles).
63
PART TWO
Analytic
Combinatorics

Robert Sedgewick
Âpplications of
• Bitstrings
• Summary
II.5e.RMapps.Summary
PART TWO
Âpplications
CAMBRIDGE
• Bitstrings
• Exercises
Analytic
Combinatorics

Robert Sedgewick
II.5f.RMapps.Exercises
Web Exercise V.l
Patterns in strings.
Analytic
Combinatorics

Robert Sedgewick
Web Exercise V.l. Give an asymptotic expression for the number of

strings that do not contain the pattern 0000000001. Do the same for
0101010101 .
66
Web Exercise V.2
Variants of supercritical sequence classes.
Analytic
Combinatorics

Robert Sedgewick
Web Exercise V.2. Give asymptotic expressions for the number of

objects of size N and the number of parts in a random object of size
N for the following classes: compositions of 1 s, 2s, and Bs, triple
surjections, and alignments with no singleton cycles.
67
Assignments
1. Read pages 289-300 (Applications of R&M Asymptotics) in text. Skim pages 301-
375.
Usual caveat: Try to get a feeling for what's there, not understand every detail.
2. Write up solutions to Web exercises V.l and V.2.
3. Programming exercise.
Program V.l . In the style of the plots in the lectures slides, plot the
GFs for the set of bitstrings having no occurrence of the pattern
000000000. Do the same for 01 01 01 01 01. (See Web Exercise V.l).
Analytic
Combinatorics

Robert Sedgewick
68
Analytic
Combinatorics

Robert Sedgewick
5. Applications of
Rational and Meromor
Asymptotics
PART TWO

AC05-PoleApps Djvu

Transféré par

Informations du document

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

AC05-PoleApps Djvu

Transféré par

Droits d'auteur :

Formats disponibles

ANALYTIC COMBINATORICS

Philippe Flajolet and

Analytic combinatorics overview

4. Rational & Meromorphic

Philippe Flajolet and

Bottom line from last lecture

Analytic transfer for meromorphic GFs: f(z)/g(z ) ~ c $ N

• Compute the dominant pole cx (smallest real with g(z) = 0)

• Compute the residue h~] = -f(a)/g\a)

• Constant c is / 7-1 /cx.

• Exponential growth factor (3 is 1 /cx

Not order 1 if g\a) = 0.

Philippe Flajolet and

Rational and Meromorphic Asymptotics

• Other familiar examples

• Supercritical sequence schema

How many bitstrings of length N ?

Y,2 N z" = '£[2z) N

B, the class of all bitstrings

Dominant singularity: pole at a — 1/2

Example 1: Bitstrings with restrictions on consecutive Os

How many bitstrings of length N have no two consecutive Os ?

Example 1: Bitstrings with restrictions on consecutive Os

Boo, the class of all bitstrings having no 00

Dominant singularity: pole at <fr

Example 1: Bitstrings with restrictions on consecutive Os

B 4 , the class of all bitstrings having no 0 4

Dominant singularity: pole at a

. /"(z) 1 -\- oi “t - ol2 -(- or 1

Residue: n_i = —=--- -—3 -—r

Example 1: Bitstrings with restrictions on consecutive Os

Information on consecutive Os in GFs for strings

[from AC Part I Lecture 5]

Bm(z ) = ^ z^ = ^{# of bitstrings of length N with no 0 M }z N

Bm(z/2) = ^ ({# of bitstrings of length N with no runs of M 0s}/2 N )z N

Bm( 1/2) = ^{# of bitstrings of length N with no runs of M 0s}/2 N

= ^ Pr (position of end of first 0 M is > N ] = Expected position of end of first 0

Theorem. Probability that an A/-bit random bitstring has no 0 M : [z n ]Bm(z/ 2) ~

[from AC Part I Lecture 5]

Q. Do the same results hold for 0001?

0001 occurs much

Observation. Consider first occurrence of 000.

•0000 and 0001 equally likely, BUT

[from AC Part I Lecture 5]

S P — binary strings that do not contain p Sp 10111110101101001100110000011111

Tp binary strings that end in p 7 loinnoiOllOlOOllOOllOlOlOOlOlO

and have no other occurrence of p

• Sp and Tp are disjoint

• the empty string is in S P

• adding a bit to a string in S P gives a string in S P or T p

Constructions for bitstrings without specified patterns

[from AC Part I Lecture 5]

Every pattern has an autocorrelation polynomial

• slide the pattern to the left over itself.

Constructions for bitstrings without specified patterns

[from AC Part I Lecture 5]

• for each 1 bit in the autocorrelation of any string in T P add a “tail”

• result is a string in S P followed by the pattern

l 10111110101101001100110 101001010 1001010

Bitstrings without specified patterns

[from AC Part I Lecture 5]

How many A/-bit strings do not contain a specified pattern pi

Sp — the class of binary strings with no