An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

54
An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles

Transcript of An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

Page 1: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

An introduction to expander families and Ramanujan

graphs

Tony ShaheenCSU Los Angeles

Page 2: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

Before we get started on expander graphs I want to give a definition that we will use in this talk.

A graph is regular if every vertex has the same degree (the number of edges at that vertex).

A 3-regular graph.

Page 3: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

Think of a graph as a communications network.

Now for the motivation behind expander families…

Page 4: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

Two vertices can communicatedirectly with one another iff they are connected by an edge.

Page 5: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

Communication is instantaneousacross edges, but there may bedelays at vertices.

Page 6: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

Edges are expensive.

Page 7: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

Our goal: Let d be a fixed integer with d > 1.

Create an infinite sequence of d-regular graphs

where

1. the graphs are getting bigger and bigger (the number of vertices of goes to infinity asn goes to infinity)

2. each is as good a communications network as possible.

Page 8: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

Questions:

1. How do we measure if a graph is a good communications network?

2. Once we have a measurement, can we find graphs that are optimal withrespect to the measurement?

Page 9: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

Let’s start with the first question.

Questions:

1. How do we measure if a graph is a good communications network?

2. Once we have a measurement, howgood can we make our networks?

Page 10: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

Consider the following graph:

Page 11: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

Let’s look at the set of vertices that we can reach after n steps, starting at the top vertex.

Page 12: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

Here is where we can get to after 1 step.

Page 13: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

Here is where we can get to after 1 step.

Page 14: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

We would like to have many edges going outward from there.

Page 15: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

Here is where we can get to after 2 steps.

Page 16: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.
Page 17: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.
Page 18: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.
Page 19: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.
Page 20: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.
Page 21: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.
Page 22: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

Take-home Message #1:

The expansion constantis one measure of howgood a graph is as acommunications network.

Page 23: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

We want h(X) to be BIG!

Page 24: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

We want h(X) to be BIG!If a graph has small degreebut many vertices, this is noteasy.

Page 25: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

𝐶3 𝐶4 𝐶5 𝐶6

Consider the cycles graphs:

Page 26: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

𝐶3 𝐶4 𝐶5 𝐶6

Consider the cycles graphs:

Each is 2-regular.

Page 27: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

𝐶3 𝐶4 𝐶5 𝐶6

Consider the cycles graphs:

Each is 2-regular.The number of vertices goes to infinity.

Page 28: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

Let S be the bottom half.

Page 29: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.
Page 30: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

We say that a sequence of regular graphsis an expander family if

• All the graphs have the same degree

• The number of vertices goes to infinity

• There exists a positive lower bound r such that the expansion constant is alwaysat least r.

Page 31: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

We just saw that expander families of degree 2 do not exist.

Page 32: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

We just saw that expander families of degree 2 do not exist.

What is amazing is that if d > 2 then expander families of degree d exist.

Page 33: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

We just saw that expander families of degree 2 do not exist.

What is amazing is that if d > 2 then expander families of degree d exist.

Existence: Pinsker 1973First explicit construction: Margulis 1973

Page 34: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

So far we have looked at the combinatorial way of looking at expander families.

Let’s now look at it from an algebraic viewpoint.

Page 35: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

We form theadjacency matrixof a graph as follows:

Page 36: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

Facts about eigenvalues of a d-regulargraph G:

Facts about the eigenvalues of a d-regular connected graph G with n vertices:

Page 37: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

Facts about eigenvalues of a d-regulargraph G:

● They are all real.

Facts about eigenvalues of a d-regular gra G with n vertices:Facts about the eigenvalues of a d-regular connected graph G with n vertices:

Page 38: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

● They are all real.● The eigenvalues satisfy

… = d

Facts about the eigenvalues of a d-regular connected graph G with n vertices:

Page 39: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

● The second largest eigenvalue

Facts about eigenvalues of a d-regulargraph G:

● They are all real.

(Alon-Dodziuk-Milman-Tanner)

Facts about the eigenvalues of a d-regular connected graph G with n vertices:

● The eigenvalues satisfy

… = d satisfies

Page 40: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

(Alon-Dodziuk-Milman-Tanner)

Page 41: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

(Alon-Dodziuk-Milman-Tanner)

Page 42: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

(Alon-Dodziuk-Milman-Tanner)

Page 43: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

Take-home Message #2:

Page 44: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.
Page 45: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

The red curve has a horizontal asymptote at 2√𝑑−1

Page 46: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

In other words, is asymptotically

the smallest that can be.

2√𝑑−1𝜆1

Page 47: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

We say that a d-regular graph X is Ramanujan if all the non-trivial eigenvalues of X (the ones that aren’t equal to d or -d) satisfy

𝜆

Page 48: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

Hence, if X is Ramanujan then

Page 49: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

.

Page 50: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

Take-home Message #3:

Ramanujan graphs essentiallyhave the smallest possible 𝜆1

Page 51: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

A family of d-regular Ramanujan graphs is an expander family.

Page 52: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.
Page 53: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.
Page 54: An introduction to expander families and Ramanujan graphs Tony Shaheen CSU Los Angeles.

Shamelessself-promotion!!!

Expander families and Cayley graphs – A beginner’s guide

by Mike Krebs and Anthony Shaheen