ANALYTIC COMBINATORICS

PART TWO

Analytic

Combinatorics

Philippe Flajolet and

Robert Sedgewiek

8. Saddle-Point Asymptotics

Cambridge

http://ac.cs.princeton.edu

Analytic combinatorics overview

A. SYMBOLIC METHOD

1. OGFs

2. EGFs

3. MGFs

B. COMPLEX ASYMPTOTICS

4. Rational & Meromorphic

5. Applications of R&M

6. Singularity Analysis

7. Applications of SA
^ 8. Saddle point

specification

SYMBOLIC METHOD

Analytic

Combinatorics
Philippe Flajolet and
Robert Sedgewick

GF

equation

■ I

•»ill

COMPLEX ASYMPTOTICS

asymptotic

estimate

desired
result!

ANALYTIC COMBINATORICS

PART TWO

Analytic

Combinatorics
Philippe Flajolet and
Robert Sedgewick

CAMBRIDGE

8. Saddle-Point Asymptotics

• Modulus surfaces
^[^^oaadle point bounds

• Saddle point asymptotics

• Applications

http://ac.cs.princeton.edu

II.8a.Saddle.Surfaces

Warmup: 2D absolute value plots

Consider 2D plots of functions: all points (x, |f(x)| ) in a Cartesian plot.

4
Welcome to absolute-value-land!

Consider 3D versions of our plots of analytic functions.

A modulus surface is a plot of (x, y, \f(z )\) where z = x + yi

Example: 1 +4 z 2

1 — 4 z 2

2D version

Q. Can a modulus surface assume any shape ?

A. No.

A. (A surprise.) Only four types of points.

Modulus surface points type I: zeros

A zero is a point where f(z) = 0 and f'(z) ± 0.

Key point: All zeros have the same local behavior.

f(z) = f{z 0 ) + f\zo){z - z 0 ) + — (z - z 0 ) 2 + ... / r/ (z 0 )(z-z 0 )^

Ex. f(z) = 2z = 2re' 9 , |f(z)| = 2r

A zero of order p is a point where f (k Hz) = 0 for 0<k<p and fW(z) =*= 0.
zero of order 3

zero (order 1)

zero of order 2

Modulus surface points type II: poles

Quick in-class exercise

Q. What function is this?

A.
1 -z

Modulus surface points type III: ordinary points

An ordinary point is a point where f(z) =£ 0 and f\z) =*= 0.

All ordinary points have the same local behavior.

f(z) = f(z 0 ) + f'(z 0 )(z - z 0 ) + - z o) 2 + • • •

^ c

Modulus surface points type III: saddle points

A saddle point is a point where f(z) =*= 0 and f\z) = 0.

All saddle points have the same local behavior.

f(z) = f{z 0 ) + f'(zo)(z - z 0 ) + —jr~( z - z o) 2 + • • • ~c(z- Zp) 2 ^

(1-z)(2-z)

Basic characteristic

•Downwards-oriented parabola at one angle

•Upwards-oriented parabola at perpendicular angle

10

Modulus surface points: summary

f(z) f\z) local behavior

simple zero

not 0

~ c (z - Zo)

zero of order p > 1

~ c (z - Zo)P

saddle point

not 0

~ c(z- Zo ) 2

ordinary point

not 0

not 0

~ c

simple pole

~ c / (z - Zo)

poles at
0 + i/2 and 0 - //2

Maximum modulus principle : There are no other possibilities (!)

1 -4 Z 2 ,
1 + 4z 22

—8z(1 + 4z 2 ) - (1 - 4 z 2 )8z
( 1 + 4z 2 ) 2

16 z

“(1 — 4z 2 ) 2

Example: No local maxima

Quick in-class exercise

Q. Where are the saddle points?

1 + z + z 2 + z 3

A. Where 1 + 2z + 3z 2 = 0 ,

or

• zeros ( -1, -/, +/)

• saddle points

bottom view
Modulus surface plots for familiar AC GFs

Q-z-z 1 / 2—z 3 / 3 —z 4 /4—z s /5

1 - Z

13
ANALYTIC COMBINATORICS

PART TWO

Analytic

Combinatorics

Philippe Flajolet and

Robert Sedgewick

CAMBRIDGE

8. Saddle-Point Asymptotics

• Modulus surfaces
^[^^oaadle point bounds

• Saddle point asymptotics

• Applications

http://ac.cs.princeton.edu

II.8a.Saddle.Surfaces

ANALYTIC COMBINATORICS

PART TWO
Analytic

Combinatorics

Philippe Flajolet and

Robert Sedgewick

CAMBRIDGE

8. Saddle-Point Asymptotics

• Modulus surfaces

• Saddle point bounds

• Saddle point asymptotics

• Applications

http://ac.cs.princeton.edu

II.8b.Saddle.Bounds

Saddle-point bound for GFs: basic idea

Cauchy coefficient formula

Saddle point bound: „ „ „

1 zeta

•Saddle point at ^

•Use circle of radius t,

•Integrand is < G(Q/t ) N+] everywhere on circle

Note: G, is the solution to

C ; (z)
z N+^

Q zG\^Jg{^ = N +T )*

Example: e z /z

"saddle point equation"

Saddle-point bounds for GFs

Theorem. Saddle point bounds for CFs.

Let C(z), not a polynomial, be analytic at the origin with finite radius of
convergence R.

If G has nonnegative coefficients, then ^z N ]C(z) < c(0/C") where ^ is the saddle
point
closest to the origin, the unique real root of the saddle point equation = N + 1

Proof (sketch). By Cauchy coefficient formula

z N ]C(z)

I
dz

<

2ni J c G(z) z^

C r 2n r-, \ de

2n 1 (z) z«+i

Take C to be a

circle of radius X,
and change to
polar coordinates

c(0

C N

C(z) < G(Q/X n+ i on C

Example:

C(z)

C'(z)

C = 6

<

5! “ 6 5

e z /z 6
Saddle point GF bound example I: factorial/exponential

Goal. Estimate

N\

,N

JZ

G(z) = e

.z

Saddle point equation

z— = N + 1

nZ

Saddle point

C = N+1

Saddle point equation

C(z)

e z /z 6

Saddle point bound

e z = — <

,N+1

N! “ (N+ 1) N

Saddle point bound

z N ]C(z) < C(C)/C

,N

—)►

OGr-e
./V

Bound is too high by only a factor of v2nN, since — ~ — _

N\ N n V2^N

18

Saddle point GF bound example II: Catalan/central binomial

Goal. Estimate

2 N
N

[z N }( 1 +z)

2 N

C(z) = (1 +z)

2N

Saddle point equation 2Nz = (N + 1 )(1 + z)

Saddle point

Saddle point bound

c =

2/V
N

<

N + 1
N- 1

V N—1 )

IN

4N 2 \N

r/V+1 ' N
VN-1

N 2 - 1

» 4

Saddle point bound

z N ]C(z) < C(C)/C N

(1 +z) 10 /z 6

Bound is too high by only a factor of VnN , since (

2N\ 4 n
I ~ x

N / y/nN

ANALYTIC COMBINATORICS

PART TWO

Analytic

Combinatorics

Philippe Flajolet and

Robert Sedgewick

CAMBRIDGE

8. Saddle-Point Asymptotics

• Modulus surfaces

• Saddle point bounds

• Saddle point asymptotics

• Applications

http://ac.cs.princeton.edu
II.8b.Saddle.Bounds

ANALYTIC COMBINATORICS

PART TWO

Analytic

Combinatorics

Philippe Flajolet and

Robert Sedgewick

CAMBRIDGE

8. Saddle-Point Asymptotics

• Modulus surfaces
^[^^oaadle point bounds

• Saddle point asymptotics

• Applications

http://ac.cs.princeton.edu

II.8c.Saddle.Method

Saddle-point method for GFs: basic idea

Cauchy coefficient formula

[zN]c(z) = h / C c(z) ife

Saddle point bound:

•Saddle point at Z,
•Use circle of radius Z,

•Integrand is < G(Z)/Z, N+] everywhere on circle

Saddle point method :

•Focus on path near saddle point

•Bound “tail” contribution
•Use Laplace’s method

22

Saddle-point susceptibility

susceptibility. Technical conditions that enable us to unify saddle-point

approximations.

Definition. Saddle-point susceptible contour integrals.

The contour integral J F(z)dz with F(z ) = e f(z) is susceptible to the saddle
point approximation if
C passes through a saddle point the unique real root of the saddle point equation
F'(z) = 0
(or f'(z) = 0) and C can be split into two parts T and Q such that

• Tails are negligable: J F(z)d

z — o

F(z)dz

A central quadratic approximation holds uniformly along Q: f(z)

f(0 + >(C)(z-0
• Tails can be completed back [details omitted].

to be expected unless
multiple saddle point
since f'(Q= 0

23

Saddle-point transfer theorem

= e^ (z) is susceptible to the

Theorem. If a contour integral / F(z)dz with F(z )

C 1 f F(()

saddle point approximation, then — / F(z)dz ~ — ,

2 mJc s/2nf"{ Q

a general technique for contour integration (not just for asymptotics)

Proof.

[ Similar to proof for SP bound; see text ]

24

Saddle-point transfer theorem

Theorem. If a contour integral J

saddle point approximation, then

F(z)dz with F(z) = e^ (z) is susceptible to

1

271 /

F(z)dz

f ( z )

J2nf"(0

Saddle-point transfer. Given a GF C(z), if the contour integral of G(z)/z N+]

along a path Cis susceptible to the saddle point approximation, then

where g(z) = InC(z) - (N + 1 )ln z and G, is the unique positive real root of
the saddle point equation g'(z) = 0.

Equivalent forms

SP equation

G'(z) _ N + 1
G(z) _ z

SP approximation

Q(C)

C N+1 \/2jrg"(£)

Proof. Take F(z) = G(z)/z N+] .

25
Saddle point transfer example I: factorial/exponential

Goal. Estimate —

N\

Saddle point

Saddle point
approximation

,N

JZ

C(z)

f{z ) = In G(z) — (N + 1) lnz

= z — (N + 1) lnz

Saddle point equation

e z =

C = N+1

r'(z) i N +1

f"( z) =

f\z) = 0

Saddle point approx

1

e N+1

[z N ]C(z) ~ _

1 C N+ V2irf"(C)

N\ (/v+ 1 )N+i v /2jc/(N+ 1

N N \/2n N

Important note : Need to check susceptibility, or use bound and sacrifice V2nN
factor.

tails are negligible, a central approximation holds, and tails can be completed
back

26
Saddle point method example I (susceptibility to saddle point)

Contour integral

2 71 /

I e z-(N+1) lnzj z

Cn

Switch to polar coordinates

N(e' e -1 -M) rfQ

z = /Ve' 61

Split into central and tail contours

Neglect tails

N!

1 / e _ \

2n (/\/^ (Qn + Tn)

Qn =

J_e^_

271 /V N
exponentially small for

0o = A/ a with a > -1 /2 [see text]

Note: Slightly shifting saddle point (from A/+1 to N) simplifies calculations.

27

Saddle point method example I (susceptibility to saddle point)

Approximate integrand

(e id - 1 - id) = -9 2 /2 + 0(6 3 )

Restrict Goto drop O-term

N02 ! 2 d0 for 6 q = N a with(g < —1 / 3 ^

Change of variable

Restrict Goto complete tails

Collect restrictions
1

Finish —

rsj

2k

P J tOq\/~N

y/N .

/ e

J — Oq y/~N

y/N

y/ln/N for 6

( ° )

1 N 1

y/2nN

! 2 dt for 9 q — N a

0 = t/y/N
d9 — dt/y/N

28
Saddle-point asymptotics

Q. N 1 / 2 - 2 / 5

Aren’t we touching on N needing to be in the “galactic” range?

not relevant in this galaxy

A. Those estimates are in the exponent.

Ex. e~ N / = e~ 8 = .000335 when N is 2 30 (about 1 billion).

A. Methods extend to derive full asymptotic series to any desired precision.

A. Results are easy to validate numerically.

A. Towards goal of general schema cover whole families of combinatorial classes.

29

Saddle point transfer example II: Catalan/central binomia

Goal. Estimate

2 N
N

[z N }( 1 +z)

2 N
C(z) = (1 +z)

2N

Saddle point

c =

N+ 1

N- 1

Saddle point
approximation

[z N ]( 1 +z)" /v =

2N

2/V
N

f(z) = In G(z) — (N + 1) lnz

r'(z)

//

2Nln(1 +z) - (N + 1)ln

2N N + 1

1 + z z

Saddle point equation

2/V N+1

(1+z) 2 + z 2

f'(z) = 0

/ 2N \ 2N
\N -1 /

Saddle point approx

[z N ]C(z) - _

1 < N+ V2rf"«)

(^) N+i y^±j

Note : Slight shift of saddle point often simplifies calcuations (see next slide).

30

Saddle point transfer example II: Catalan/central binomial

Goal. Estimate

Saddle point

[z N ]( 1 + z) 2N
f(z) = In C(z) — (N + 1) lnz

2Nln(1 + z) — (N + 1) lnz
2 N N + 1

1 + z z

2N N + 1
“(1 +z) 2 +

Saddle point equation

f'(z) = 0

Saddle point
approximation

,/v

]G(z)

Saddle point approx

C(0

/■"(I) ~N/2

Important note : Need to check susceptibility, or use bound and sacrifice v^icN

factor.

tails are negligible, a central approximation holds, and tails can be completed
back

31
ANALYTIC COMBINATORICS

PART TWO

Analytic

Combinatorics

Philippe Flajolet and

Robert Sedgewick

CAMBRIDGE

8. Saddle-Point Asymptotics

• Modulus surfaces
^[^^oaadle point bounds

• Saddle point asymptotics

• Applications

http://ac.cs.princeton.edu

II.8c.Saddle.Method

ANALYTIC COMBINATORICS

PART TWO
Analytic

Combinatorics

Philippe Flajolet and

Robert Sedgewick

CAMBRIDGE

8. Saddle-Point Asymptotics

• Modulus surfaces
^[^^oaadle point bounds

• Saddle point asymptotics

• Applications

http://ac.cs.princeton.edu

II.8d.Saddle.Apps

Involutions

Q. How many different permutations of size N with no cycle lengths >2 ?

/i

k = 2

h = 4

U= 10
34

AC example with saddle-point asymptotics: Involutions

I, the class of involutions

I = SET(CYCi,2(Z))

/(z) = e

— o z + z V 2

Z N ]/(z)

N\[z n ]I(z)

3 N/2 + VN-1 /4

2 N N / 2 VnN
1 l-) N/2 e ^

</4e v e
Saddle-point transfer. Given a GF G(z), if the contour integral
is susceptible to the saddle point approximation, then

2 71 /

C(z)

c/z

Z N+1

2ni

C(z )

Z N+1

e g(0

\Z 2 *g "(0 y

where g(z) = InG(z) - (N + 1 )ln z and is the unique positive real root of the
saddle point equation g'(z ) = 0 (equivalently, G'(z)/G(z ) = (A/+l)/z ).

g(z) = z + z 2 /2 - (N+ 1) lnz

g'(z) = 1+ z-^±l

= 1 + N z t 1

C = -1 ^ v / TT4(NTTy

~ 1/2 + 0(1 /VN)

Important note: Need to check susceptibility.

• generally more difficult than for other transfer thms.

• option: use bound (sacrifice v2kN factor.

35

Set partitions

Q. How many ways to partition a set of size of N ?

Si = 1

{1}

{2}

{1}

{2}

{1}
{2

{1

2}

{2}

{1

S 2 =

= 2

{3}

{1

{1}

{2}

S 3

= 5

{1} {2}

{3}

■ {4}

{1} {2 3

4}

{2} {1 3

4}

{3} {1 2

4}

{4} {1 2

3}
UJ

{1 2}

{3}

{4}

3}

{1 3}

{2}

{4}

3}

{1 4}

{2}

{3}

2}

{2 3}

{1}

{4}

{3}

{2 4}

{1}

{3}

{3 4}

{1}

{2}

{1 2}

{3

4}

{1 3}

{2

4}
{1 4}

{2

3}

{1 2

3 4}

S 4 =

= 15

36

AC example with saddle-point asymptotics: Set partitions

S, the class of set partitions

S = SET(SET>o(Z))

S(z)

[complex expression: use bound]

.N—1

Sn < N\

(In N)

&V^NTe
Saddle-point transfer. Given a GF G(z), if the contour integral
is susceptible to the saddle point approximation, then

2ni

dz

Z N +1

e g( 0 N

\/ 2 *g"(C) )

where g(z) = InG(z) - (N + 1 )ln z and ^ is the unique positive real root of the
saddle point equation g'(z) = 0 (equivalently, C'(z)/C(z) = (A/+l)/z ).

g(z )
g'V)

e z — 1 — (N + *\) In z
N + 1

e z —

g"(z) = e z +

N+ 1

^2

= N + 1

£ ~ In N — In In N

SP bound

[z N ]S(z) < ^

37
Saddle point: summary of combinatorial applications

construction CF saddle point bound

urns

U = SET(Z )

e z

e N
^ N n

central binomial

[z N ]( 1 +z) 2N

->• 4 n

involutions

1 = SET( CYO, 2(Z))

e z+z2 /2

p N/2+>/N-1/4

< A/'
" V2N n ^ 2

set partitions

S = SET(SET>o(Z))

e 6 " -1

pN-1

< N>

- (lnN) w

fragmented

permutations

F = SET(SEQ>o(Z ))

e z /0~ z )

< N!e 2v ^“ 1/2

integer partitions

P= MSET(SEQ>o(Z))

e z/(1-z)+z 2 /2(1-z 2 )+...

< e n^/2N/3

coefficient

asymptotics

N\ ~ N N V2nN

VnN

„N/2+VN- 1/4

/V'-

2 N N/2 VnN

< N\

,2\//V—1 /2
2 \/7t

e 7i^2W/3

4N\/3

amateurs

38

ANALYTIC COMBINATORICS

PART TWO

Analytic

Combinatorics

Philippe Flajolet and

Robert Sedgewick

CAMBRIDGE

8. Saddle-Point Asymptotics

• Modulus surfaces
^[^^oaadle point bounds

• Saddle point asymptotics

• Applications

http://ac.cs.princeton.edu

II.8d.Saddle.Apps

ANALYTIC COMBINATORICS

PART TWO

Analytic

Combinatorics

Philippe Flajolet and

Robert Sedgewick

CAMBRIDGE

http://ac.cs.princeton.edu

8. Saddle-Point Asymptotics

• Modulus surfaces
^[^^oaadle point bounds

• Saddle point asymptotics

• Applications

• AC wrapup

II.8e.Saddle.Summary

Analytic combinatorics overview

A. SYMBOLIC METHOD

1. OGFs

2. EGFs

3. MGFs

B. COMPLEX ASYMPTOTICS

4. Rational & Meromorphic

5. Applications of R&M

6. Singularity Analysis

7. Applications of SA

8. Saddle point

specification

SYMBOLIC METHOD

Analytic

Combinatorics

Philippe Flajolet and

Robert Sedgewick

GF

equation

■ I

•»ill

COMPLEX ASYMPTOTICS

asymptotic
estimate

desired
result!

Basic ideas of analytic combinatorics (summary)

1. Combinatorial specifications provide succinct definitions of a wide range of

discrete structures.

2. The symbolic method transforms specifications to equations that define

generating functions.

3. Complexification treats generating functions as analytic objects, giving

estimates of coefficients.

Cauchy’s coefficient formula gives coefficient asymptotics when singularities are

poles.
Singularity analysis provides a general approach to analyzing GFs with essential
singularities.

Saddle-point asymptotics is effective for functions with no singularities.

4. Combinatorial classes fall into general schema that are governed by universal
asymptotic laws.

42

Constructions and symbolic transfers

The symbolic method for unlabeled objects (summary)

operation

notation
semantics

disjoint

union

A + B

disjoint copies of objects

from A and B

Cartesian

product

Ax B

ordered pairs of copies of objects,

one from A and one from B

sequence

SEQ(A)

sequences of objects from A

powerset

PSET(A)

finite sets of objects from A

(no repetitions)

no +z '

n> 1

multiset

MSET(A)

finite sets of objects from A

(with repetitions)

n (T

n> 1 v

Additional constructs are

available (and still being invented)-

—one exa

OCF
A(z) + B(z)

A(z)B(z)

The symbolic method for labelled classes (transfer theorem)

Theorem. Let A and B be combinatorial classes of labelled objects with EGFs A(z)
and B(z). Then

construction

notation

semantics

EGF

disjoint union

A + B

disjoint copies of objects from A and B

A(z) + B(z)

labelled product

A ★ B

ordered pairs of copies of objects,

one from A and one from B

A(z)B(z)

SEQk(A) or A k

k- sequences of objects from A

A(z) k

sequence

SEQ(A)

sequences of objects from A

1 -A(z)

set

SET k (A)
/c-sets of objects from A

A(zf/k\

SET( A)

sets of objects from A

e A(z)

CYCk(A)

/(-cycles of objects from A

A(z) k /k

cycle

CYC {A)

cycles of objects from A

hl 1-A(z)

43

Explicit analytic transfers

meromorphic?

standard

scale?

Meromorphic Transfer
(see Lecture 4)

> f
Standard Scale Transfe
(see Lecture 6)

square root?
logarithmic?

no singularities?

Singularity Analysis
(see Lecture 6)

> r

Saddle Point

Schemas

Combinatorial problems can be organized into broad schemas, covering infinitely

many combinatorial types and governed by simple asymptotic laws.

Theorem. Asymptotics of exp-log labelled sets.

Suppose that a labelled set class F = SET<t>(G) is exp-log(cx, (3, p)

1 1

with G(z) ~ ctlog --— b (3 . Then F(z) ~ e fl (-- —

1 -z/p 1- z/p

and
[z n ]F(z)

e 13 ,1

Theorem. Asymptotics of supercriti cal sequences. If F = SEQ(G) is a strongly

aperiodic
supercritical sequence class, then M z N ]F(z) ~ where A is the root of G(\) = 1 in
(0, p).

v--- y /

r(a) V

(-) N

1 —Of

Theorem. If C is an irreducible context-free class, then its generating function

C(z) has a
square-root singularity at it s radius of convergence p. If C(z) is aperiodic, then
the dominant

singularity is unique and f[z N ]F(z) ~ } _ (—) A/ N~ 3 ^ 2 J where a is a

computable real.

\_ ymt yr _ J

Theorem. If a simple variety of trees with GF F(z) = Z(f(F(z)) is

A-invertible (where A is the positive real root of 4>(u) = U(f f (u) )
then

[z N ]F(z)

\/2jt<?i"(A)/</>(A)

(</>'( A)) N N“ 3/2

The discovery of such schemas and of the

associated universality properties constitues
the very essence of analytic combinatorics.
Theorem. Asymptotics of implicit tree-like classes.

Suppose that F is an implicit tree-like class with associated GF F(z) = 4>(z, F(z))
that
is aperiodic and smooth-implicit^, s), so that G(r, s) = s and G w (r, s) = 1. Then
F(z) converges at z= r where it has a square-root singularity with

f(z) ~ s - ayj 1 - z/r and([z N ]F(z)

2y/n v r

(l)V- 3 Awhere a = ' 2r **' S )

45
"If you can specify it, you can analyze it"

( Specification )

ir

Symbolic transfer

GF equation)

IT

Analytic transfer

Example 1: Bitstrings with restrictions on consecutive Os

_L

Example 3: Suriections

( Specification]

VT J

Symbolic trans

Example 6: Cycles in derangements

T J
D, the class of all derangements
D = SET(CYC > 0 (Z))

Example 5: Cycles in permutations

P, the class of all permutations

P = SET(CYC(Z))

PS 2 >

:®\ © r®

Cl

Ana

Mappings

raphs

Example

Spei

Symlj

Symbolic transfer

M, the class of all mappings

M = SET ( Y )

from previous slide

■-V

the class of 2-regular graphs

• V * • *

i V

* v ^ 4

■'v'-z...

r = set(ucyc>2(;

a a n

Analytic

Combinatorics

Theorem. Asymptotics of exp-log labelled sets.

Suppose that a labelled set class F = SETo(G) is exp-log(a, 3, p)

with G(z) ~ a log -——— + f3 . Then F{z) ~ e^(-—

1-z/p V 1 -z/p'

and

(J

M(z) = e Y ^s

v ' / Theorem. Asymptotics of exp-log labelled sets.

1 1 *

Y(z) ~ — In 1 Suppose that a labelled set class F = SET»(G) is exp-log(ot, 3, p)

AC example with saddle-point asymptotics: Set partitions

AC example with saddle-point asymptotics: Involutions

) =e xp(^l n l“

Analytic transfer

~T~

N\[z‘

Example 9. Labelled hierarchic

L, the class of

L = i

L(z)=z +

S, the class of set partitions

S = SET(SET>o(Z))

5(z) = e 1

,e z —1

I, the class of involutions

I = SET(CYCi. 2 (Z))

(7)

©®

¥
— o z + z2 / 2

g'( z ) =

[complex expression: use bound] g"(z) = e z +

Asymptotics ) . SP bound

{2 3} {5 7 9} {4} {1 8}

Saddle-point transfer. Given a GF G(z), if the contour integral f G(z)

2 tit J £

is susceptible to the saddle point approximation, then

( flew = -1 ,[ ~ ^

_ 2m J c Z y/2*g"(0 J

where g(z) = InG(z) - (/V+ l)lnz and G, is the unique positive real root of the
saddle point equation g’(z) = 0 (equivalently, G\z)/G(z) = ( N+])/z ).

g(z) = e z - 1 - (N + 1) In z

N + 1 Ce c = N + 1

Saddle-point transfer. Given a GF G(z), if the contour integral zr—. f G(z)-^nr

2ni J c /rv+i

is susceptible to the saddle point approximation, then

( |/'|G(z) = — / C{z)Z!Z- ~ c, '° A

V 2 Wr y/2ip(0 J

where g(z) = InG(z) - (W + 1 )ln z and G, is the unique positive real root of the
saddle point equation g'(z) = 0 (equivalently, C'(z)/C(z) = (/V+l)/z ).

g(z) = z + z 2 /2 - (N + 1) lnz

N + 1 c 2 +C-(n+i) = o

z N ]l(z)

e N/2+VN-1/4

2 N N / 2 V^
g'(z) = 1+z--

g"( z ) = 1 + ^

Q = -\ + l^ + 4(N + 1 )

~ VN - 1 /2 + 0(1 /Vn)

Important note: Need to check susceptibility.

• generally more difficult than for other transfer thms.

• option: use bound (sacrifice V2nN factor.

N+1

£ ~ In N — In In N

Sn < N\

a N -1

(In NY

(In/V) ' /27tN/e [z N ]S(z)<^

, the class of all bracketings

S = Z + SEQ >i(S)

$(z,w) = z-1 +e w - w <5> z (r, s) = 1

N\[z n ]L(z) ~N\ 4>z(z,w) = 1 $ ww (r,s)=2

iVnTfi' r>
with r = 2 In 2 — 1

<t>„,(z, w) = e w — 1
$mv(z, w) = e w
%/21n2 - 1 )

"T

4\/7r

4 n N 3 /2

d>'(u) =
<t>"(u) =

V2k

(i - up
i

(TAj 1

1 - A (1 - A) 2

_ 1/2

«i(A) = 2
<t>'{ A) = 4
gi"(A) = 16

otics

F -7
Note that the specification is the most succinct of all the descriptions

S(z)=z +

1 -S(z)

1 - S(z)

Analytic transfer

Theorem. Asymptotics of implicit tree-like classes.

Suppose that F is an implicit tree like class with associated GF F(z) = 4>(z, F(z))
that
is aperiodic and smooth-implicit(r, s), so that C(r, s) = s and G w (r, s) = 1.
Then
F(z) converges at z= r where it has a square-root singularity with

F(z) ~ s - f*v/l -z/r and^^]F(z) ~ 2 ^( 7 -)^^ 3 ^ where a = ^

2 r^,(r,s)
^ww(r, s)

[z~]5(z)

,^ = (l) N N -3/2
with r = 3 - 2 V 2

details left for exercise ]

Dn

7.. . 1

2 .. . 2

1... 9

even for small N

5 44.1455... 44
46
What is "Analytic combinatorics"?

[ In case someone asks... ]

Analytic combinatorics aims to enable precise quantitative predictions of the

properties
of large combinatorial structures. The theory has emerged over recent decades as
essential both for the analysis of algorithms and for the study of scientific
models in other
discliplines, including statistical physics, computational biology, and information
theory.

47

What's next?

Suggestions for further study in Analytic Combinatorics

• Additional constructions and associated symbolic transfers

• Applications to paths in lattices and many other types

• Details of SA proofs

• Periodicity, irreducibility, algebraic functions

• Additional schema

• Drmota-Llaley-Woods theorem

• Technical conditions for SP approximations

• Multivariate asymptotics and limit laws

• Applications, applications, applications, applications

Available as "postscript"
to this course

For an overview of Flajolet's work and current research in AC, watch the lecture
"If you can specify it you can analyze it": the lasting legacy of Philippe Flajolet

48

Shameless plugs

Books

Booksites

TO THE

Analysis

Algorithms

ROBERT SEDGEWICK
PHILIPPE FLAJOLET

Analytic

Combinatorics

Philippe Flajotet and

Robert Sedgewick

Online courses

Princeton University

Algorithms, Part I

with Kevin Wayne & Robert Sedgewick

Princeton University

Analytic Combinatorics, Part I

with Robert Sedgewick

Princeton University

Algorithms, Part II

with Kevin Wayne & Robert Sedgewick

Princeton University

Analytic Combinatorics, Part II

with Robert Sedgewick

And,

especially for students in

this course . . .

see AC booksite
for details

49
Now this is not the end. It is not even the beginning of the end.
But it is, perhaps, the end of the beginning. "

Winston Churchill, 1942

ANALYTIC COMBINATORICS

PART TWO

Analytic

Combinatorics
Philippe Flajolet and
Robert Sedgewick

CAMBRIDGE

http://ac.cs.princeton.edu

8. Saddle-Point Asymptotics

• Modulus surfaces
^[^^oaadle point bounds

• Saddle point asymptotics

• Applications

• AC wrapup

II.8e.Saddle.Summary

ANALYTIC COMBINATORICS

PART TWO

Analytic

Combinatorics

Philippe Flajolet and

Robert Sedgewiek

8. Saddle-Point Asymptotics

Cambridge

http://ac.cs.princeton.edu