RANDOM WALK ON THE RADO GRAPH

WITH
SOURAV CHATTERJEE LAURENT Micco

1 TUBE6iN_LET T BE INDEPENDENT GO.lk 6AhAHs

WHAT IS THE CHANCE THAT I L

5mn11

How small
E n 1217 SOE lo IF nooo

NOW LET 2 0

THEOREM WITH PROBABILITY l f If

Phooey HERE is A PROPERTY THAT A GrantaMIGHT HAVE

tf n M X yn y ym Distinct VERTICES

3 3
xing Kien
Yi Xz leg Em

X Y Hn
y Yy Yz

Y
FACT 1 A Random f Glo 44 Has PROPERTY a S

FAILS FOR Trill 3 3

PI THE CHHNLE THAT
w
3W is
l Yum
CHHNLE A knits Kun FIXED Xi 4 is 0
until countably mnw4 44 obj

f
FAIL IF I BOTH SATISFY It THEN I I 52
KONTH ARGUMENT TV SEQUENTIALLY
PI USE THE Back AND Buicaf
ENUMERATE VENTRES OF fi I 442 Gg
suppose
fn Xi yn facial fun
constantTED IS AN Isomorphism
IF N EVEN LET M BE Smallest INDEX
OF A Point NOT IN Domain of fn

LET
EXTEND
fn to Xm

Us Neighbors of Xm AMONG X xn
EY Yall U
A Potential IMAGE OF 4am MUST BE A NEIGHBOR

of EACH fad x all NOT OF flx.iq all

BECHASE 92 satisfies THONE
IS SACHA Y
IF n is ODD WORK BACKWARD LET M BE
SMNIEST INDEX WITH Ym NOT IN LARGE OFfa
EXTEND
fntv fat USING PROPERTY For S
D
 THUS WE MAY TALK ABOUT

THE RANDOM Grant R THERMOGRAPH

2 EhtE5

YT IF R FROM R BY
DELETING ANY FINITE A VERTICES

0h EDGES

THEN R E R

Pi toLE
IF YOU DIVIDE THE VERTEX SETINTO
the Boks THEN THE INDUCED GRAPH
OF VERTICES IN ONE BOX IN
This CHANNTENIZES R

a
Univers AWI FINITE 0h COUNTABLE

GANNA AMEND AS AN INDUCED SUBGRAPH

HomoGts GIVEN AN Isomonditism DEFINED

Foh ANY Y xn THERE is AN EXTENSION

To All OF R

DIAMETER 2 AND EACH VERTEX IS CONNECTED

TO HALF THE
OTHERS
 3 OTHER constitutions THERE ARE MANY

NADI OW N 1444 3
CONNECT ios if ith Bit OF j is A 1
u
Bit z
I bio

SO 0 CONNECTED to All ODD NUMBERS

a 1 CONNECTED TO All I 2,3 MOD4 Awa c

i

CLEAN4 N HRS PROPERTY

NumBERTHEot ON primes I vD4 connect

thy E Ig 1
GROUP THEON LET Hinkle die 3 Keef gaff
jeep
429 b Ellis 3 s Hythists'txg'T
HEISENBERG GROUP CENTER 9933

FORM A GRAPH with VENICE's vow centhal ELEMENTS

X 93 4.4 I IF THEYCOMMUTE

HIP contains R AS AN INDUCED SUBERLAH

WORK WITH MAMANTHE

MalliARAS

4 011 LAW A FIRST ORDER PhuPEATY Holds in R

E it HOLDS in Almost All FINITE Chants wit'tPbb

a KONNELTED DIAMETER2

Not Hamiltonian Cihcut CHHUMATE h

GREAT ARTICE PETER CAMERON THE RANDOM GRAPH

 MARKOVCHAIRS
5 UNDEnstanDMk R VIA FIX A VERTEX SET
ET Q Iso EQ 1 1 BE A PROBABILITY on V
vav

Ktv v
Qd't IF V V
Ball wack

FAI Ktv V't is REVENSIBLE WITH

64 41
Tcl
z

Qalcelwall 461 k 41
PF IxUklxoH ITCHthe x
Z ja Z

SI
Kew if
do
I

WHAT is THE ANTE OF CONVERGENCE

FAST L TI CONCENTHATED NEAR 0 DIAMETER2

EVERY VERTEX Connected to 42 REST a

How Fast BOUNDED

f A LOWER Bound THE STARTING STATE MATTERS

WORK in Boccian MODEL i j s ith Bitolj is l

Q41 GEICO
c
Since to and I let 12 G
Tizz

Set 21h 2 ExponentintED h TIMES

2 1 241 2 I s 5 4 213 222 16

 START THE CHAIN At 21h1
00061 4
is 21h so 2
THE 0h4 SmallEh connected j
THE HIGHER I HAVE SUDEH EXPONEN Small PROB

SO Whit Goes IN Ih IhLg Ek et

pair Piya LXe 2 l z

cont Il Klan AHN T ITIet

cold IT TAKES AT LEAST log Li STEPS TO CONVENE

STARTING Farm i FOR SOME i

Fun in 65356 265356
leyte 5
YOU'll HAVE TO WAIT FOR LAMENTS TALK TO HEADREST

7 FINITE VERSIONS R is THE FRAISSE Limit OF

THE SET OF All FINITE GRAAHS

A LIMITING OBJECT IS SUPPOSED TO SAY SOMETHING ABOUT

RACE FINITE N THIS ONE SORT OF Stinks

QUESTIONL Pick fi INDEPENDENTLY FROM 6th 4

WHAT IS SIZE OF LARGEST INDUCED ISOMONAHIC

SUBGRAPH

ANS e
ABOUT 4 login t EXTRELY CONCENTRATED
 WITHCHATTERJEE Let 5 BE INDEPENDENT Gln g
THEIN
AND I Gaap

LET Lq BE THE SIZE OF THE LARGEST INDUCEDISOMORPHICSUBGANAH

THEN WITH PROBABILITY TENDING to 1 As n Nd

Ln is EitHEh Lin Ea oh Lynten WITH PROBABILITY 91

Xn 4login 2laglagg
Llagdle
ta 4Chynik
IF 222 EnE's so Lya ta Listen DIFFER BY AT Most 1 45481119 Same

ASYMPTOTES kick in FAST

2 31 Xn 15.08 En 0.23 A simulationGAVE

14 th SAYS Ln 9113,14
COMPUTING
Ln is p complete BUT HAVE thin Amont OF DATA 145
ITS RIGHT 0W MME't

J Stone I was SURE IomoBODY knows THEOREM ABOVE I ASKED

EXPERTS Wb PE BUT THEYHAD RELATED QUESTIONS

HOW SHOULD n t W RELATE SOTHAT

9 Ella k is Isometric to SUBGHANH OFGCNf 2

AWS n About 2 log N
WHEN SUBGRAPH ISOMORPHISM IS REALLY HARD AND WHY I MATTERS FOR
INTENEGENCE RESEARCH 2018
GRAPH DATABASE J Uh Ahtieicial

THEOREM ET Pln wl BE PROBABILITY THAT

f A Eln LL is contained in AIGNAN.tl
THEN AS NAO
Pl Lau Ent N I

PL L YnetEnt N 0

GW Lhgywtii tw luhg.nl

EI D 150
Yw 15.46 Ew _0.19
l
THE 0h0m 5h45 Pll 5 led P 16,150 0
SEE PICTURES ABOVE

THE PROOF OF BOTH THEOREMS IS

DETAILED TECHNICALITIES
ROUGHLY
LET Kleis Yli.it BE THE
INDICATORS OF AN EDGE in f I

LET f BE THE SET OF ORDERED

n TUPLES in N

FOR It Be A SAY AI B if Nia a Tlbi is

All If it is E N

LET W I A BE A HEB

w O
so LN za E

UPPER BOUND 0W W BY FIRST MOMENT METHUD

Pcw o Elwin e E w
NT
E w LAY It Won2
l
choose a 50THIS IS SUN

LOWER BOUND ON W BY 2ndMOMENTMETHOD

Plwso z WIT
E 4
THE DETAILS ARE QUITE TECHNICALd
16 Summary
INTRODUCED YOU TO R
THE Random Graph
PROPERTIES UNIVENSNITY HOMOGENEITY ZEHHONELKWS

INTRODUCED A SIMPLE MARKOV CHAIN

ON R AND SHOWED WORSTCASE

IT TAKES At least STEPS TO

lag n
CONVERGE STARTING From n

SHOWED How ATEMPTS TO UNDERSTAND

FINITE ANALOGS OF A LEADS TO

NICE MATH

REFERENCES

CHATTERJEE S Diaconis P.cz Gh ISOMORPHISMS BETWEEN

RANDOM Grunts

2024 COMPLEXITY ANDRANDOMNESS

DIACONIS P AND MAllinnis th
IN THE HEISENBERG GROUPS AND BEYOND HZ John

CAMERON D Kolb THE RANDOM GRAPH

J 20181
Me CREESH E PROSSER P SANON E ANDTRIMBLE
WHEN SUBGRANH ISOMORPHISM IS REALLY HARD
AND WHY IT MATTERS FOR GRAPH DATABASE
3 ARTIE INTEL 61