cs 598: spectral graph theory - ntuaakolla)_lec2.pdfcs 598: spectral graph theory: lecture 2 the...
TRANSCRIPT
![Page 1: CS 598: Spectral Graph Theory - NTUAAKolla)_Lec2.pdfCS 598: Spectral Graph Theory: Lecture 2 The Laplacian Today More on evectors and evalues The Laplacian, revisited Properties of](https://reader033.vdocuments.us/reader033/viewer/2022042423/5f3add299425c12f833b08d4/html5/thumbnails/1.jpg)
CS 598: Spectral Graph Theory: Lecture 2
The Laplacian
![Page 2: CS 598: Spectral Graph Theory - NTUAAKolla)_Lec2.pdfCS 598: Spectral Graph Theory: Lecture 2 The Laplacian Today More on evectors and evalues The Laplacian, revisited Properties of](https://reader033.vdocuments.us/reader033/viewer/2022042423/5f3add299425c12f833b08d4/html5/thumbnails/2.jpg)
Today
More on evectors and evalues
The Laplacian, revisited
Properties of Laplacian spectra, PSD matrices.
Spectra of common graphs.
Start bounding Laplacian evalues
![Page 3: CS 598: Spectral Graph Theory - NTUAAKolla)_Lec2.pdfCS 598: Spectral Graph Theory: Lecture 2 The Laplacian Today More on evectors and evalues The Laplacian, revisited Properties of](https://reader033.vdocuments.us/reader033/viewer/2022042423/5f3add299425c12f833b08d4/html5/thumbnails/3.jpg)
A Remark on Notation
![Page 4: CS 598: Spectral Graph Theory - NTUAAKolla)_Lec2.pdfCS 598: Spectral Graph Theory: Lecture 2 The Laplacian Today More on evectors and evalues The Laplacian, revisited Properties of](https://reader033.vdocuments.us/reader033/viewer/2022042423/5f3add299425c12f833b08d4/html5/thumbnails/4.jpg)
Evectors and Evalues
![Page 5: CS 598: Spectral Graph Theory - NTUAAKolla)_Lec2.pdfCS 598: Spectral Graph Theory: Lecture 2 The Laplacian Today More on evectors and evalues The Laplacian, revisited Properties of](https://reader033.vdocuments.us/reader033/viewer/2022042423/5f3add299425c12f833b08d4/html5/thumbnails/5.jpg)
The Laplacian: Definition Refresher
G = {V,E}
i jd1
d2
d3
di -1
…
dn
j
i
LG =
Where di is the degree of i-th vertex.For convenience, we have unweighted graphs
GGG ADL
• DG = Diagonal matrix of degrees• AG = Adjacency matrix of the graph•
otherwise
edgejiif
jiifd
jiL
i
G
0
),(1),(
![Page 6: CS 598: Spectral Graph Theory - NTUAAKolla)_Lec2.pdfCS 598: Spectral Graph Theory: Lecture 2 The Laplacian Today More on evectors and evalues The Laplacian, revisited Properties of](https://reader033.vdocuments.us/reader033/viewer/2022042423/5f3add299425c12f833b08d4/html5/thumbnails/6.jpg)
The Laplacian: Properties Refresher
• The constant vector 1 is an eigenvector with eigenvalue zero.
• Has n eigenvalues (spectrum)
• Second eigenvalue is called “algebraic connectivity”.G is connected if and only if
• We will see the further away from zero, the more connected G is.
> 0
01
GL
![Page 7: CS 598: Spectral Graph Theory - NTUAAKolla)_Lec2.pdfCS 598: Spectral Graph Theory: Lecture 2 The Laplacian Today More on evectors and evalues The Laplacian, revisited Properties of](https://reader033.vdocuments.us/reader033/viewer/2022042423/5f3add299425c12f833b08d4/html5/thumbnails/7.jpg)
Redefining the Laplacian
Let Le be the Laplacian of the graph on n vertices consisting of just one edge e=(u,v).
For a graph G with edge set E we now define
Many elementary properties of the Laplacian now follow from this definition as we will see next (prove facts for one edge and then add ).
1 -1
-1 1
uLe =
otherwise
versaviceorvjuiif
vuijiif
jiLe
0
,,1
,,1
),(
v
u v
Ee
eG LL
][zeros
![Page 8: CS 598: Spectral Graph Theory - NTUAAKolla)_Lec2.pdfCS 598: Spectral Graph Theory: Lecture 2 The Laplacian Today More on evectors and evalues The Laplacian, revisited Properties of](https://reader033.vdocuments.us/reader033/viewer/2022042423/5f3add299425c12f833b08d4/html5/thumbnails/8.jpg)
Laplacian of an edge, contd.
1 -1
-1 1
uLe =
v
u v
][zeros eigenvalue
eigenvector
Since evalues are zero and 2, we see that Le is P.S.D. Moreover,
![Page 9: CS 598: Spectral Graph Theory - NTUAAKolla)_Lec2.pdfCS 598: Spectral Graph Theory: Lecture 2 The Laplacian Today More on evectors and evalues The Laplacian, revisited Properties of](https://reader033.vdocuments.us/reader033/viewer/2022042423/5f3add299425c12f833b08d4/html5/thumbnails/9.jpg)
Review of Positive Semidefiniteness
nT RxMxx 0
![Page 10: CS 598: Spectral Graph Theory - NTUAAKolla)_Lec2.pdfCS 598: Spectral Graph Theory: Lecture 2 The Laplacian Today More on evectors and evalues The Laplacian, revisited Properties of](https://reader033.vdocuments.us/reader033/viewer/2022042423/5f3add299425c12f833b08d4/html5/thumbnails/10.jpg)
Review of Positive Semidefiniteness
nT RxMxx 0
![Page 11: CS 598: Spectral Graph Theory - NTUAAKolla)_Lec2.pdfCS 598: Spectral Graph Theory: Lecture 2 The Laplacian Today More on evectors and evalues The Laplacian, revisited Properties of](https://reader033.vdocuments.us/reader033/viewer/2022042423/5f3add299425c12f833b08d4/html5/thumbnails/11.jpg)
More Properties of LaplacianFrom the definition using edge sums, we get:
(PSD-ness)The Laplacian of any graph is PSD.
(Connectivity) G is connected iff λ2 positive or alternatively, the null space is 1-d and spanned by the vector 1.
Corollary: The multiplicity of zero as an eigenvalue equals
the number of connected components of the graph.
![Page 12: CS 598: Spectral Graph Theory - NTUAAKolla)_Lec2.pdfCS 598: Spectral Graph Theory: Lecture 2 The Laplacian Today More on evectors and evalues The Laplacian, revisited Properties of](https://reader033.vdocuments.us/reader033/viewer/2022042423/5f3add299425c12f833b08d4/html5/thumbnails/12.jpg)
More Properties of Laplacian (Edge union)If G and H are two graphs on the same
vertex set, with disjoint edge set then
If a vertex is isolated, the corresponding row and column of Laplacian are zero
(Disjoint union) Together these imply that for the disjoint union of graphs G and H
![Page 13: CS 598: Spectral Graph Theory - NTUAAKolla)_Lec2.pdfCS 598: Spectral Graph Theory: Lecture 2 The Laplacian Today More on evectors and evalues The Laplacian, revisited Properties of](https://reader033.vdocuments.us/reader033/viewer/2022042423/5f3add299425c12f833b08d4/html5/thumbnails/13.jpg)
More Properties of Laplacian (Edge union)If G and H are two graphs on the same
vertex set, with disjoint edge set then
If a vertex is isolated, the corresponding row and column of Laplacian are zero
(Disjoint union) Together these imply that for the disjoint union of graphs G and H
(Disjoint union spectrum)If LG has evectors v1,…, vn
with evalues λ1,…, λn and LH has evectors w1,…, wn
with evalues µ1,…, µn then has evectors
with evalues
![Page 14: CS 598: Spectral Graph Theory - NTUAAKolla)_Lec2.pdfCS 598: Spectral Graph Theory: Lecture 2 The Laplacian Today More on evectors and evalues The Laplacian, revisited Properties of](https://reader033.vdocuments.us/reader033/viewer/2022042423/5f3add299425c12f833b08d4/html5/thumbnails/14.jpg)
The Incidence Matrix: Factoring the Laplacian
1 2 3
![Page 15: CS 598: Spectral Graph Theory - NTUAAKolla)_Lec2.pdfCS 598: Spectral Graph Theory: Lecture 2 The Laplacian Today More on evectors and evalues The Laplacian, revisited Properties of](https://reader033.vdocuments.us/reader033/viewer/2022042423/5f3add299425c12f833b08d4/html5/thumbnails/15.jpg)
Spectra of Some Common Graphs
![Page 16: CS 598: Spectral Graph Theory - NTUAAKolla)_Lec2.pdfCS 598: Spectral Graph Theory: Lecture 2 The Laplacian Today More on evectors and evalues The Laplacian, revisited Properties of](https://reader033.vdocuments.us/reader033/viewer/2022042423/5f3add299425c12f833b08d4/html5/thumbnails/16.jpg)
The Complete Graph
K5
![Page 17: CS 598: Spectral Graph Theory - NTUAAKolla)_Lec2.pdfCS 598: Spectral Graph Theory: Lecture 2 The Laplacian Today More on evectors and evalues The Laplacian, revisited Properties of](https://reader033.vdocuments.us/reader033/viewer/2022042423/5f3add299425c12f833b08d4/html5/thumbnails/17.jpg)
The Complete Graph
K5
![Page 18: CS 598: Spectral Graph Theory - NTUAAKolla)_Lec2.pdfCS 598: Spectral Graph Theory: Lecture 2 The Laplacian Today More on evectors and evalues The Laplacian, revisited Properties of](https://reader033.vdocuments.us/reader033/viewer/2022042423/5f3add299425c12f833b08d4/html5/thumbnails/18.jpg)
The Ring Graph R10
![Page 19: CS 598: Spectral Graph Theory - NTUAAKolla)_Lec2.pdfCS 598: Spectral Graph Theory: Lecture 2 The Laplacian Today More on evectors and evalues The Laplacian, revisited Properties of](https://reader033.vdocuments.us/reader033/viewer/2022042423/5f3add299425c12f833b08d4/html5/thumbnails/19.jpg)
The Ring Graph
Let z(u) be the point (x_k(u), y_k(u)) on the plane.
Consider the vector z(u-1) - 2 z(u) + z(u+1). By the reflection symmetry of the picture, it is parallel to z(u)
Let z(u-1) - 2 z(u) + z(u+1) = λz(u). By rotational symmetry, the constant λ is independent of u.
To compute λ consider the vertex u=1.
Verify details as excercise
Spectral embedding for k=3
R10
![Page 20: CS 598: Spectral Graph Theory - NTUAAKolla)_Lec2.pdfCS 598: Spectral Graph Theory: Lecture 2 The Laplacian Today More on evectors and evalues The Laplacian, revisited Properties of](https://reader033.vdocuments.us/reader033/viewer/2022042423/5f3add299425c12f833b08d4/html5/thumbnails/20.jpg)
The Path Graph
R10
P5
P5
![Page 21: CS 598: Spectral Graph Theory - NTUAAKolla)_Lec2.pdfCS 598: Spectral Graph Theory: Lecture 2 The Laplacian Today More on evectors and evalues The Laplacian, revisited Properties of](https://reader033.vdocuments.us/reader033/viewer/2022042423/5f3add299425c12f833b08d4/html5/thumbnails/21.jpg)
The Path Graph
R10
P5
![Page 22: CS 598: Spectral Graph Theory - NTUAAKolla)_Lec2.pdfCS 598: Spectral Graph Theory: Lecture 2 The Laplacian Today More on evectors and evalues The Laplacian, revisited Properties of](https://reader033.vdocuments.us/reader033/viewer/2022042423/5f3add299425c12f833b08d4/html5/thumbnails/22.jpg)
Graph Products
![Page 23: CS 598: Spectral Graph Theory - NTUAAKolla)_Lec2.pdfCS 598: Spectral Graph Theory: Lecture 2 The Laplacian Today More on evectors and evalues The Laplacian, revisited Properties of](https://reader033.vdocuments.us/reader033/viewer/2022042423/5f3add299425c12f833b08d4/html5/thumbnails/23.jpg)
Graph Products
![Page 24: CS 598: Spectral Graph Theory - NTUAAKolla)_Lec2.pdfCS 598: Spectral Graph Theory: Lecture 2 The Laplacian Today More on evectors and evalues The Laplacian, revisited Properties of](https://reader033.vdocuments.us/reader033/viewer/2022042423/5f3add299425c12f833b08d4/html5/thumbnails/24.jpg)
Graph Products: Grid Graph
Immediately get spectra from path.
![Page 25: CS 598: Spectral Graph Theory - NTUAAKolla)_Lec2.pdfCS 598: Spectral Graph Theory: Lecture 2 The Laplacian Today More on evectors and evalues The Laplacian, revisited Properties of](https://reader033.vdocuments.us/reader033/viewer/2022042423/5f3add299425c12f833b08d4/html5/thumbnails/25.jpg)
Start Bounding Laplacian Eigenvalues
![Page 26: CS 598: Spectral Graph Theory - NTUAAKolla)_Lec2.pdfCS 598: Spectral Graph Theory: Lecture 2 The Laplacian Today More on evectors and evalues The Laplacian, revisited Properties of](https://reader033.vdocuments.us/reader033/viewer/2022042423/5f3add299425c12f833b08d4/html5/thumbnails/26.jpg)
Sum of Eigenvalues, ExtremalEigenvalues
xx
AxxAxx
T
T
x
T
x 011 minmin
xx
AxxAxx
T
T
xvx
T
vxx 0,,12
11
minmin
xx
AxxAxx
T
T
x
T
x 01max maxmax
![Page 27: CS 598: Spectral Graph Theory - NTUAAKolla)_Lec2.pdfCS 598: Spectral Graph Theory: Lecture 2 The Laplacian Today More on evectors and evalues The Laplacian, revisited Properties of](https://reader033.vdocuments.us/reader033/viewer/2022042423/5f3add299425c12f833b08d4/html5/thumbnails/27.jpg)
Courant-Fischer. Courant-Fischer Formula: For any nxn symmetric matrix A,
Proof: see blackboard
xx
AxxT
T
SxkofSk
maxmin
dim
xx
AxxT
T
SxknofSk
minmax
1dim
![Page 28: CS 598: Spectral Graph Theory - NTUAAKolla)_Lec2.pdfCS 598: Spectral Graph Theory: Lecture 2 The Laplacian Today More on evectors and evalues The Laplacian, revisited Properties of](https://reader033.vdocuments.us/reader033/viewer/2022042423/5f3add299425c12f833b08d4/html5/thumbnails/28.jpg)
Courant-Fischer Formula:xx
AxxT
T
SxknofSk
minmax
1dim
![Page 29: CS 598: Spectral Graph Theory - NTUAAKolla)_Lec2.pdfCS 598: Spectral Graph Theory: Lecture 2 The Laplacian Today More on evectors and evalues The Laplacian, revisited Properties of](https://reader033.vdocuments.us/reader033/viewer/2022042423/5f3add299425c12f833b08d4/html5/thumbnails/29.jpg)
Courant-Fischer Formula:
xx
AxxT
T
SxknofSk
minmax
1dim
![Page 30: CS 598: Spectral Graph Theory - NTUAAKolla)_Lec2.pdfCS 598: Spectral Graph Theory: Lecture 2 The Laplacian Today More on evectors and evalues The Laplacian, revisited Properties of](https://reader033.vdocuments.us/reader033/viewer/2022042423/5f3add299425c12f833b08d4/html5/thumbnails/30.jpg)
Courant-Fischer. Courant-Fischer Formula: For any nxn symmetric matrix A,
Proof: see blackboard
Definition (Rayleigh Quotient): The ratio is called the Rayleigh Quotient of x with respect to A.
Next lecture we will use it to bound evalues of Laplacians.
xx
AxxT
T
SxkofSk
maxmin
dim
xx
AxxT
T
SxknofSk
minmax
1dim
xx
AxxT
T