large deviations random graphsdepts.washington.edu/ssproc/chunglecturers/varadhan_lecture.pdfwhat...
TRANSCRIPT
![Page 1: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/1.jpg)
Large Deviationsand
Random GraphsS.R.S. Varadhan
Courant Institute, NYU
UC Irvine
March 25, 2011
Large DeviationsandRandom Graphs – p.1/39
![Page 2: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/2.jpg)
Joint work with Sourav Chatterjee.
Large DeviationsandRandom Graphs – p.2/39
![Page 3: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/3.jpg)
Joint work with Sourav Chatterjee.
G is a graph
Large DeviationsandRandom Graphs – p.2/39
![Page 4: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/4.jpg)
Joint work with Sourav Chatterjee.
G is a graph
V (G) is the set of its vertices
Large DeviationsandRandom Graphs – p.2/39
![Page 5: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/5.jpg)
Joint work with Sourav Chatterjee.
G is a graph
V (G) is the set of its vertices
E(G) is the set of its edges.
Large DeviationsandRandom Graphs – p.2/39
![Page 6: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/6.jpg)
Joint work with Sourav Chatterjee.
G is a graph
V (G) is the set of its vertices
E(G) is the set of its edges.
Not oriented.
Large DeviationsandRandom Graphs – p.2/39
![Page 7: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/7.jpg)
Joint work with Sourav Chatterjee.
G is a graph
V (G) is the set of its vertices
E(G) is the set of its edges.
Not oriented.
|V (G)| is the number of vertices inG
Large DeviationsandRandom Graphs – p.2/39
![Page 8: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/8.jpg)
Joint work with Sourav Chatterjee.
G is a graph
V (G) is the set of its vertices
E(G) is the set of its edges.
Not oriented.
|V (G)| is the number of vertices inG
|E(G)| are the number of edges inG.
Large DeviationsandRandom Graphs – p.2/39
![Page 9: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/9.jpg)
GN is the set of all graphsG with N vertices.
Large DeviationsandRandom Graphs – p.3/39
![Page 10: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/10.jpg)
GN is the set of all graphsG with N vertices.
We think ofV (G) as{1, 2, . . . , N}
Large DeviationsandRandom Graphs – p.3/39
![Page 11: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/11.jpg)
GN is the set of all graphsG with N vertices.
We think ofV (G) as{1, 2, . . . , N}
The setE is all unordered pairs(i, j), i.e. the full setof edges.
Large DeviationsandRandom Graphs – p.3/39
![Page 12: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/12.jpg)
GN is the set of all graphsG with N vertices.
We think ofV (G) as{1, 2, . . . , N}
The setE is all unordered pairs(i, j), i.e. the full setof edges.
E(G) ⊂ E
Large DeviationsandRandom Graphs – p.3/39
![Page 13: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/13.jpg)
G is determined by its edge setE(G) ⊂ E
Large DeviationsandRandom Graphs – p.4/39
![Page 14: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/14.jpg)
G is determined by its edge setE(G) ⊂ E
|GN | = 2(N2)
Large DeviationsandRandom Graphs – p.4/39
![Page 15: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/15.jpg)
G is determined by its edge setE(G) ⊂ E
|GN | = 2(N2)
A random graph withN vertices is just a probabilitymeasure onGN , i.e. a collection of weights{p(G)}with
∑
G∈GNp(G) = 1
Large DeviationsandRandom Graphs – p.4/39
![Page 16: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/16.jpg)
What types of random graphs do we consider?
Large DeviationsandRandom Graphs – p.5/39
![Page 17: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/17.jpg)
What types of random graphs do we consider?
A pair of vertices are connected by an edge withprobabilityp
Large DeviationsandRandom Graphs – p.5/39
![Page 18: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/18.jpg)
What types of random graphs do we consider?
A pair of vertices are connected by an edge withprobabilityp
Different edges are independent.
Large DeviationsandRandom Graphs – p.5/39
![Page 19: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/19.jpg)
What types of random graphs do we consider?
A pair of vertices are connected by an edge withprobabilityp
Different edges are independent.
Th probabilityPN,p(G) of a graphG ∈ GN is givenby
Large DeviationsandRandom Graphs – p.5/39
![Page 20: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/20.jpg)
What types of random graphs do we consider?
A pair of vertices are connected by an edge withprobabilityp
Different edges are independent.
Th probabilityPN,p(G) of a graphG ∈ GN is givenby
p|E(G)|(1− p)(N2)−|E(G)|
Large DeviationsandRandom Graphs – p.5/39
![Page 21: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/21.jpg)
What types of random graphs do we consider?
A pair of vertices are connected by an edge withprobabilityp
Different edges are independent.
Th probabilityPN,p(G) of a graphG ∈ GN is givenby
p|E(G)|(1− p)(N2)−|E(G)|
If p = 12 , the distribution is uniform and we are
essentially counting the number of graphs.
Large DeviationsandRandom Graphs – p.5/39
![Page 22: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/22.jpg)
We want to calculate the probabilities of thefollowing types of events.
Large DeviationsandRandom Graphs – p.6/39
![Page 23: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/23.jpg)
We want to calculate the probabilities of thefollowing types of events.
Let Γ be a finite graph.
Large DeviationsandRandom Graphs – p.6/39
![Page 24: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/24.jpg)
We want to calculate the probabilities of thefollowing types of events.
Let Γ be a finite graph.
The number of different occurrences ofΓ in G is#G(Γ)
Large DeviationsandRandom Graphs – p.6/39
![Page 25: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/25.jpg)
We want to calculate the probabilities of thefollowing types of events.
Let Γ be a finite graph.
The number of different occurrences ofΓ in G is#G(Γ)
The number of different occurrences ofΓ in acomplete graph withN vertices#N(Γ)
Large DeviationsandRandom Graphs – p.6/39
![Page 26: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/26.jpg)
We want to calculate the probabilities of thefollowing types of events.
Let Γ be a finite graph.
The number of different occurrences ofΓ in G is#G(Γ)
The number of different occurrences ofΓ in acomplete graph withN vertices#N(Γ)
The ratiorG(Γ) =#G(Γ)#N (Γ)
Large DeviationsandRandom Graphs – p.6/39
![Page 27: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/27.jpg)
We consider mapsV (Γ) → V (G) that are one toone.
Large DeviationsandRandom Graphs – p.7/39
![Page 28: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/28.jpg)
We consider mapsV (Γ) → V (G) that are one toone.
If |V (G)| = N and|V (Γ)| = k there arep(N, k) = N(N − 1) · · · (N − k + 1) of them.
Large DeviationsandRandom Graphs – p.7/39
![Page 29: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/29.jpg)
We consider mapsV (Γ) → V (G) that are one toone.
If |V (G)| = N and|V (Γ)| = k there arep(N, k) = N(N − 1) · · · (N − k + 1) of them.
Of these a certain numberp(G,N, k) will map aconnected pair of vertices inΓ to connected ones inG.
Large DeviationsandRandom Graphs – p.7/39
![Page 30: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/30.jpg)
We consider mapsV (Γ) → V (G) that are one toone.
If |V (G)| = N and|V (Γ)| = k there arep(N, k) = N(N − 1) · · · (N − k + 1) of them.
Of these a certain numberp(G,N, k) will map aconnected pair of vertices inΓ to connected ones inG.
rG(Γ) =p(G,N,k)p(N,k) .
Large DeviationsandRandom Graphs – p.7/39
![Page 31: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/31.jpg)
There is some ambiguity here due to possiblemultiple counting.
Large DeviationsandRandom Graphs – p.8/39
![Page 32: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/32.jpg)
There is some ambiguity here due to possiblemultiple counting.
It is a multiple that depends only onΓ and willcancel out when we take the ratio.
Large DeviationsandRandom Graphs – p.8/39
![Page 33: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/33.jpg)
There is some ambiguity here due to possiblemultiple counting.
It is a multiple that depends only onΓ and willcancel out when we take the ratio.
What probabilities do we want to estimate ?
Large DeviationsandRandom Graphs – p.8/39
![Page 34: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/34.jpg)
There is some ambiguity here due to possiblemultiple counting.
It is a multiple that depends only onΓ and willcancel out when we take the ratio.
What probabilities do we want to estimate ?
For eachj ∈ {1, 2, . . . , k} we are given a finitegraphΓj and a numberrj ∈ [0, 1].
Large DeviationsandRandom Graphs – p.8/39
![Page 35: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/35.jpg)
There is some ambiguity here due to possiblemultiple counting.
It is a multiple that depends only onΓ and willcancel out when we take the ratio.
What probabilities do we want to estimate ?
For eachj ∈ {1, 2, . . . , k} we are given a finitegraphΓj and a numberrj ∈ [0, 1].
We are interested in estimating the probability
PN,p
[
∀j, |rG(Γj)− rj| ≤ ǫ]
Large DeviationsandRandom Graphs – p.8/39
![Page 36: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/36.jpg)
More precisely we are interested in calculating thefunction
Large DeviationsandRandom Graphs – p.9/39
![Page 37: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/37.jpg)
More precisely we are interested in calculating thefunction
ψp({Γj, rj}) given by
− limǫ→0
limN→∞
1(
N2
) logPN,p
[
∀j, |rG(Γj)− rj| ≤ ǫ]
Large DeviationsandRandom Graphs – p.9/39
![Page 38: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/38.jpg)
More precisely we are interested in calculating thefunction
ψp({Γj, rj}) given by
− limǫ→0
limN→∞
1(
N2
) logPN,p
[
∀j, |rG(Γj)− rj| ≤ ǫ]
ψp({Γj, rj}) = 0 if and only if rj = pE(Γj) forj = 1, 2, . . . , k.
Large DeviationsandRandom Graphs – p.9/39
![Page 39: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/39.jpg)
Let us consider the set
K = {f : [0, 1]2 → [0, 1]; f(x, y) = f(y, x)}
Large DeviationsandRandom Graphs – p.10/39
![Page 40: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/40.jpg)
Let us consider the set
K = {f : [0, 1]2 → [0, 1]; f(x, y) = f(y, x)}
Define
Hp(f) =
∫ 1
0
∫ 1
0
hp(f(x, y))dxdy
Large DeviationsandRandom Graphs – p.10/39
![Page 41: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/41.jpg)
Let us consider the set
K = {f : [0, 1]2 → [0, 1]; f(x, y) = f(y, x)}
Define
Hp(f) =
∫ 1
0
∫ 1
0
hp(f(x, y))dxdy
Where
hp(f) = f logf
p+ (1− f) log
1− f
1− p
Large DeviationsandRandom Graphs – p.10/39
![Page 42: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/42.jpg)
For anyf ∈ K, finite graphΓ with vertices{1, 2, . . . , k}, and edge setE(Γ) we define
rΓ(f) =
∫
[0,1]|V (Γ)|
Π(i,j)∈E(Γ)f(xi, xj)Πki=1dxi
Large DeviationsandRandom Graphs – p.11/39
![Page 43: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/43.jpg)
For anyf ∈ K, finite graphΓ with vertices{1, 2, . . . , k}, and edge setE(Γ) we define
rΓ(f) =
∫
[0,1]|V (Γ)|
Π(i,j)∈E(Γ)f(xi, xj)Πki=1dxi
For example ifΓ is the triangle, then
r∆(f) =
∫
[0,1]3f(x1, x2)f(x2, x3)f(x3, x1)dx1dx2dx3
Large DeviationsandRandom Graphs – p.11/39
![Page 44: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/44.jpg)
For ak cycle it is∫
[0,1]kf(x1, x2)f(x2, x3) · · · f(xk, x1)dx1dx2 · · · dxk
Large DeviationsandRandom Graphs – p.12/39
![Page 45: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/45.jpg)
For ak cycle it is∫
[0,1]kf(x1, x2)f(x2, x3) · · · f(xk, x1)dx1dx2 · · · dxk
The main result is.
ψp({Γj, rj}) = inf{f : ∀j, rΓj (f)=rj}
Hp(f)
Large DeviationsandRandom Graphs – p.12/39
![Page 46: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/46.jpg)
What is it good for?
Large DeviationsandRandom Graphs – p.13/39
![Page 47: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/47.jpg)
What is it good for?
Let us analyzeψp(Γ∆, c).
Large DeviationsandRandom Graphs – p.13/39
![Page 48: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/48.jpg)
What is it good for?
Let us analyzeψp(Γ∆, c).
Calculus of variations. The infimum is attained.Proof later. Compactness and continuity.
Large DeviationsandRandom Graphs – p.13/39
![Page 49: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/49.jpg)
What is it good for?
Let us analyzeψp(Γ∆, c).
Calculus of variations. The infimum is attained.Proof later. Compactness and continuity.
Euler equation
logf(x, y)
1− f(x, y)−log
p
1− p= β
∫ 1
0
f(x, z)f(y, z)dx
Large DeviationsandRandom Graphs – p.13/39
![Page 50: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/50.jpg)
Subject to∫
[0,1]3f(x, y)f(y, z)f(z, x)dxdydz = c
Large DeviationsandRandom Graphs – p.14/39
![Page 51: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/51.jpg)
Subject to∫
[0,1]3f(x, y)f(y, z)f(z, x)dxdydz = c
If |c− p3| << 1, thenf(x, y) = c13 is the only
solution and so is optimal.
Large DeviationsandRandom Graphs – p.14/39
![Page 52: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/52.jpg)
Subject to∫
[0,1]3f(x, y)f(y, z)f(z, x)dxdydz = c
If |c− p3| << 1, thenf(x, y) = c13 is the only
solution and so is optimal.
For anyc, if p << 1, then a clique
f = 1[0,c
13 ](x)1
[0,c13 ](y)
is a better option thanf ≡ c13 .
Large DeviationsandRandom Graphs – p.14/39
![Page 53: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/53.jpg)
A slight increase or decrease fromp3 in theproportion of triangles is explained by acorresponding deviation in the number of edgesfrom p to c
13 .
Large DeviationsandRandom Graphs – p.15/39
![Page 54: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/54.jpg)
A slight increase or decrease fromp3 in theproportion of triangles is explained by acorresponding deviation in the number of edgesfrom p to c
13 .
This is not the case ifp is small butc is not.
Large DeviationsandRandom Graphs – p.15/39
![Page 55: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/55.jpg)
A slight increase or decrease fromp3 in theproportion of triangles is explained by acorresponding deviation in the number of edgesfrom p to c
13 .
This is not the case ifp is small butc is not.
A similar story whenc is small butp is not.
Large DeviationsandRandom Graphs – p.15/39
![Page 56: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/56.jpg)
A slight increase or decrease fromp3 in theproportion of triangles is explained by acorresponding deviation in the number of edgesfrom p to c
13 .
This is not the case ifp is small butc is not.
A similar story whenc is small butp is not.
A bipartite graph is a better option.
Large DeviationsandRandom Graphs – p.15/39
![Page 57: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/57.jpg)
What is the general Large Deviations setup and howdo we apply it here?
Large DeviationsandRandom Graphs – p.16/39
![Page 58: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/58.jpg)
What is the general Large Deviations setup and howdo we apply it here?
A metric spaceX and a sequencePn of probabilitydistributions.
Large DeviationsandRandom Graphs – p.16/39
![Page 59: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/59.jpg)
What is the general Large Deviations setup and howdo we apply it here?
A metric spaceX and a sequencePn of probabilitydistributions.
Pn → δx0.
Large DeviationsandRandom Graphs – p.16/39
![Page 60: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/60.jpg)
What is the general Large Deviations setup and howdo we apply it here?
A metric spaceX and a sequencePn of probabilitydistributions.
Pn → δx0.
If A is such thatd(x0, A) > 0 Pn(A) → 0 asn→ ∞.
Large DeviationsandRandom Graphs – p.16/39
![Page 61: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/61.jpg)
Want a lower semi continuos functionI(x) such that
1
nlogPn[S(x, ǫ)] = −I(x) + o(ǫ) + oǫ(n)
Large DeviationsandRandom Graphs – p.17/39
![Page 62: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/62.jpg)
Want a lower semi continuos functionI(x) such that
1
nlogPn[S(x, ǫ)] = −I(x) + o(ǫ) + oǫ(n)
lim supǫ→0
lim supn→∞
1
nlogPn(S(x, ǫ)) ≤ −I(x)
Large DeviationsandRandom Graphs – p.17/39
![Page 63: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/63.jpg)
Want a lower semi continuos functionI(x) such that
1
nlogPn[S(x, ǫ)] = −I(x) + o(ǫ) + oǫ(n)
lim supǫ→0
lim supn→∞
1
nlogPn(S(x, ǫ)) ≤ −I(x)
lim infǫ→0
lim infn→∞
1
nlogPn(S(x, ǫ)) ≥ −I(x)
Large DeviationsandRandom Graphs – p.17/39
![Page 64: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/64.jpg)
If K is compact
lim supn→∞
1
nlogPn(K) ≤ − inf
x∈KI(x)
Large DeviationsandRandom Graphs – p.18/39
![Page 65: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/65.jpg)
If K is compact
lim supn→∞
1
nlogPn(K) ≤ − inf
x∈KI(x)
If G is open
lim infn→∞
1
nlogPn(G) ≥ − inf
x∈GI(x)
Large DeviationsandRandom Graphs – p.18/39
![Page 66: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/66.jpg)
For anyℓ the set{x : I(x) ≤ ℓ} is compact.
Large DeviationsandRandom Graphs – p.19/39
![Page 67: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/67.jpg)
For anyℓ the set{x : I(x) ≤ ℓ} is compact.
For anyℓ <∞, there is a setKℓ such thatC ∩Kℓ = ∅ implies
lim supn→∞
1
nlogPn(C) ≤ −ℓ
Large DeviationsandRandom Graphs – p.19/39
![Page 68: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/68.jpg)
For anyℓ the set{x : I(x) ≤ ℓ} is compact.
For anyℓ <∞, there is a setKℓ such thatC ∩Kℓ = ∅ implies
lim supn→∞
1
nlogPn(C) ≤ −ℓ
It now follows that for any closed setC
lim supn→∞
1
nlogPn[C] ≤ − inf
x∈CI(x)
Large DeviationsandRandom Graphs – p.19/39
![Page 69: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/69.jpg)
Contraction Principle.
Large DeviationsandRandom Graphs – p.20/39
![Page 70: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/70.jpg)
Contraction Principle.
{Pn} satisfies LDP with rateI(x) onX ,
Large DeviationsandRandom Graphs – p.20/39
![Page 71: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/71.jpg)
Contraction Principle.
{Pn} satisfies LDP with rateI(x) onX ,
F : X → Y is a continuous map.
Large DeviationsandRandom Graphs – p.20/39
![Page 72: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/72.jpg)
Contraction Principle.
{Pn} satisfies LDP with rateI(x) onX ,
F : X → Y is a continuous map.
Qn = PnF−1 satisfies an LDP onY
Large DeviationsandRandom Graphs – p.20/39
![Page 73: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/73.jpg)
Contraction Principle.
{Pn} satisfies LDP with rateI(x) onX ,
F : X → Y is a continuous map.
Qn = PnF−1 satisfies an LDP onY
With rate function
J(y) = infx:F (x)=y
I(x)
Large DeviationsandRandom Graphs – p.20/39
![Page 74: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/74.jpg)
Let us turn to our case. The probability measures areon graphs withN vertices.
Large DeviationsandRandom Graphs – p.21/39
![Page 75: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/75.jpg)
Let us turn to our case. The probability measures areon graphs withN vertices.
The space keeps changing.
Large DeviationsandRandom Graphs – p.21/39
![Page 76: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/76.jpg)
Let us turn to our case. The probability measures areon graphs withN vertices.
The space keeps changing.
Need to put them all on the same space.
Large DeviationsandRandom Graphs – p.21/39
![Page 77: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/77.jpg)
Let us turn to our case. The probability measures areon graphs withN vertices.
The space keeps changing.
Need to put them all on the same space.
Every graph is an adjacency matrix.
Large DeviationsandRandom Graphs – p.21/39
![Page 78: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/78.jpg)
Let us turn to our case. The probability measures areon graphs withN vertices.
The space keeps changing.
Need to put them all on the same space.
Every graph is an adjacency matrix.
Random graph is a random symmetric matrix.X = {xi,j}, xi,i = 0, xi,j ∈ {0, 1}
Large DeviationsandRandom Graphs – p.21/39
![Page 79: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/79.jpg)
0 x1,2 · · · x1,Nx2,1 0 · · · x2,N. . . · · · · · · · · ·
xN,1 xN,2 · · · 0
Large DeviationsandRandom Graphs – p.22/39
![Page 80: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/80.jpg)
Imbed inK. Simple functions constant on smallsquares.
− −− −− −− −− −− −− −− −
| 0 | x1,2 | · · · | x1,N |
− −− −− −− −− −− −− −− −
| x2,1 | 0 | · · · | x2,N |
−− −− −− −− −− −−
| · · · | · · · | · · · | · · · |
− −− −− −− −− −− −− −− −
| xN,1 | xN,2 | · · · | 0 |
− −− −− −− −− −− −− −− −Large DeviationsandRandom Graphs – p.23/39
![Page 81: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/81.jpg)
Measures{QN,p} onK.
Large DeviationsandRandom Graphs – p.24/39
![Page 82: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/82.jpg)
Measures{QN,p} onK.
The spaceK needs a topology. Weak is good. Nicecompact space.
Large DeviationsandRandom Graphs – p.24/39
![Page 83: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/83.jpg)
Measures{QN,p} onK.
The spaceK needs a topology. Weak is good. Nicecompact space.
QN,p ⇒ δp
Large DeviationsandRandom Graphs – p.24/39
![Page 84: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/84.jpg)
Measures{QN,p} onK.
The spaceK needs a topology. Weak is good. Nicecompact space.
QN,p ⇒ δp
Lower Bound. Letf be a nice function inK.
Large DeviationsandRandom Graphs – p.24/39
![Page 85: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/85.jpg)
Measures{QN,p} onK.
The spaceK needs a topology. Weak is good. Nicecompact space.
QN,p ⇒ δp
Lower Bound. Letf be a nice function inK.
Create a random graph with probabilityf( iN, jN) of
connectingi andj
Large DeviationsandRandom Graphs – p.24/39
![Page 86: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/86.jpg)
By law of large numbers
QfN ⇒ δf
in the weak topology onK.
Large DeviationsandRandom Graphs – p.25/39
![Page 87: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/87.jpg)
By law of large numbers
QfN ⇒ δf
in the weak topology onK.
The new measureQfN onK has entropy
H(QfN , QN,p) ≃
(
N
2
)
Hp(f)
Large DeviationsandRandom Graphs – p.25/39
![Page 88: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/88.jpg)
Standard tilting argument
P (A) =
∫
A
dP
dQdQ
= Q(A)1
Q(A)
∫
A
e− log dQdP dQ
≥ Q(A) exp[−1
Q(A)
∫
A
logdQ
dPdQ]
= exp[−H(Q;P ) + o(H(Q,P ))]
Large DeviationsandRandom Graphs – p.26/39
![Page 89: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/89.jpg)
Upper Bound. Cramér.
2
N2logEQN,p[
N2
2
∫
J(x, y)f(x, y)dxdy]
→
∫ 1
0
∫ 1
0
log[peJ(x,y) + (1− p)]dxdy
Large DeviationsandRandom Graphs – p.27/39
![Page 90: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/90.jpg)
Upper Bound. Cramér.
2
N2logEQN,p[
N2
2
∫
J(x, y)f(x, y)dxdy]
→
∫ 1
0
∫ 1
0
log[peJ(x,y) + (1− p)]dxdy
Tchebychev. Half-plane. For small balls, optimize.
Large DeviationsandRandom Graphs – p.27/39
![Page 91: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/91.jpg)
Upper Bound. Cramér.
2
N2logEQN,p[
N2
2
∫
J(x, y)f(x, y)dxdy]
→
∫ 1
0
∫ 1
0
log[peJ(x,y) + (1− p)]dxdy
Tchebychev. Half-plane. For small balls, optimize.
I(f) = Hp(f)
Large DeviationsandRandom Graphs – p.27/39
![Page 92: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/92.jpg)
Upper Bound. Cramér.
2
N2logEQN,p[
N2
2
∫
J(x, y)f(x, y)dxdy]
→
∫ 1
0
∫ 1
0
log[peJ(x,y) + (1− p)]dxdy
Tchebychev. Half-plane. For small balls, optimize.
I(f) = Hp(f)
Are we done!
Large DeviationsandRandom Graphs – p.27/39
![Page 93: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/93.jpg)
NO!, Why?
Large DeviationsandRandom Graphs – p.28/39
![Page 94: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/94.jpg)
NO!, Why?
The object of interest is the map
F = {rΓj(f)}; K → [0, 1]k
Large DeviationsandRandom Graphs – p.28/39
![Page 95: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/95.jpg)
NO!, Why?
The object of interest is the map
F = {rΓj(f)}; K → [0, 1]k
They are not continuous unless no two edges inconsistsΓ share a common vertex.
Large DeviationsandRandom Graphs – p.28/39
![Page 96: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/96.jpg)
NO!, Why?
The object of interest is the map
F = {rΓj(f)}; K → [0, 1]k
They are not continuous unless no two edges inconsistsΓ share a common vertex.
Well. Change the topology toL1
Large DeviationsandRandom Graphs – p.28/39
![Page 97: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/97.jpg)
NO!, Why?
The object of interest is the map
F = {rΓj(f)}; K → [0, 1]k
They are not continuous unless no two edges inconsistsΓ share a common vertex.
Well. Change the topology toL1
No chance. Even the Law of large numbers fails.
Large DeviationsandRandom Graphs – p.28/39
![Page 98: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/98.jpg)
NO!, Why?
The object of interest is the map
F = {rΓj(f)}; K → [0, 1]k
They are not continuous unless no two edges inconsistsΓ share a common vertex.
Well. Change the topology toL1
No chance. Even the Law of large numbers fails.
In between topology! Cut topology.
Large DeviationsandRandom Graphs – p.28/39
![Page 99: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/99.jpg)
d(f, g) = sup‖a‖≤1‖b‖≤1
∫ ∫
[f(x, y)− g(x, y]a(x)b(y)dxdy
= supA,B
∫
A
∫
B
[f(x, y)− g(x, y]dxdy
Large DeviationsandRandom Graphs – p.29/39
![Page 100: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/100.jpg)
d(f, g) = sup‖a‖≤1‖b‖≤1
∫ ∫
[f(x, y)− g(x, y]a(x)b(y)dxdy
= supA,B
∫
A
∫
B
[f(x, y)− g(x, y]dxdy
In the cut topologyF is continuous. Half the battle!
Large DeviationsandRandom Graphs – p.29/39
![Page 101: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/101.jpg)
d(f, g) = sup‖a‖≤1‖b‖≤1
∫ ∫
[f(x, y)− g(x, y]a(x)b(y)dxdy
= supA,B
∫
A
∫
B
[f(x, y)− g(x, y]dxdy
In the cut topologyF is continuous. Half the battle!
Law of large numbers?
Large DeviationsandRandom Graphs – p.29/39
![Page 102: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/102.jpg)
Enough to takeA andB as unions of intervals of theform [ j
N, j+1
N]
Large DeviationsandRandom Graphs – p.30/39
![Page 103: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/103.jpg)
Enough to takeA andB as unions of intervals of theform [ j
N, j+1
N]
For eachA× B it is only the ordinary LLN forindependent random variables.
Large DeviationsandRandom Graphs – p.30/39
![Page 104: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/104.jpg)
Enough to takeA andB as unions of intervals of theform [ j
N, j+1
N]
For eachA× B it is only the ordinary LLN forindependent random variables.
Error Boundse−cN2
Large DeviationsandRandom Graphs – p.30/39
![Page 105: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/105.jpg)
Enough to takeA andB as unions of intervals of theform [ j
N, j+1
N]
For eachA× B it is only the ordinary LLN forindependent random variables.
Error Boundse−cN2
Number of rectangles2n × 2n. That is good. LLNHolds.
Large DeviationsandRandom Graphs – p.30/39
![Page 106: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/106.jpg)
Enough to takeA andB as unions of intervals of theform [ j
N, j+1
N]
For eachA× B it is only the ordinary LLN forindependent random variables.
Error Boundse−cN2
Number of rectangles2n × 2n. That is good. LLNHolds.
Three fourths of the battle!
Large DeviationsandRandom Graphs – p.30/39
![Page 107: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/107.jpg)
If K were compact in the cut topology we would bedone.
Large DeviationsandRandom Graphs – p.31/39
![Page 108: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/108.jpg)
If K were compact in the cut topology we would bedone.
But it is not. The projectionf →∫
f(x, y)dy iscontinuous. The image isL1 and not compact.
Large DeviationsandRandom Graphs – p.31/39
![Page 109: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/109.jpg)
If K were compact in the cut topology we would bedone.
But it is not. The projectionf →∫
f(x, y)dy iscontinuous. The image isL1 and not compact.
The problem is invariant under a huge group. Thepermutation groupΠN .
Large DeviationsandRandom Graphs – p.31/39
![Page 110: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/110.jpg)
If K were compact in the cut topology we would bedone.
But it is not. The projectionf →∫
f(x, y)dy iscontinuous. The image isL1 and not compact.
The problem is invariant under a huge group. Thepermutation groupΠN .
The functionHp(f), rΓ(f) are invariant under thegroupσ ∈ Σ of measure preserving transformationsof [0, 1].
Large DeviationsandRandom Graphs – p.31/39
![Page 111: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/111.jpg)
Go toK/Σ.
Large DeviationsandRandom Graphs – p.32/39
![Page 112: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/112.jpg)
Go toK/Σ.
It is compact! ( Lovász-Szegedy)
Large DeviationsandRandom Graphs – p.32/39
![Page 113: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/113.jpg)
Go toK/Σ.
It is compact! ( Lovász-Szegedy)
But what is it?
Large DeviationsandRandom Graphs – p.32/39
![Page 114: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/114.jpg)
Go toK/Σ.
It is compact! ( Lovász-Szegedy)
But what is it?
It is the space of "graphons"
Large DeviationsandRandom Graphs – p.32/39
![Page 115: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/115.jpg)
Go toK/Σ.
It is compact! ( Lovász-Szegedy)
But what is it?
It is the space of "graphons"
What is a graphon?
Large DeviationsandRandom Graphs – p.32/39
![Page 116: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/116.jpg)
Gn a sequence of graphs. Becoming infinite.
Large DeviationsandRandom Graphs – p.33/39
![Page 117: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/117.jpg)
Gn a sequence of graphs. Becoming infinite.
We sayGn has limit if
limn→∞
rGn(Γ) = r(Γ)
exists for every finite graphΓ
Large DeviationsandRandom Graphs – p.33/39
![Page 118: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/118.jpg)
Gn a sequence of graphs. Becoming infinite.
We sayGn has limit if
limn→∞
rGn(Γ) = r(Γ)
exists for every finite graphΓ
Graphon is the mapΓ → r(Γ)
Large DeviationsandRandom Graphs – p.33/39
![Page 119: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/119.jpg)
Gn a sequence of graphs. Becoming infinite.
We sayGn has limit if
limn→∞
rGn(Γ) = r(Γ)
exists for every finite graphΓ
Graphon is the mapΓ → r(Γ)
It has a representation asrΓ(f) for somef in K
Large DeviationsandRandom Graphs – p.33/39
![Page 120: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/120.jpg)
f is not unique. ButrΓ(f) ≡ rΓ(g) if and only iff(x, y) = g(σx, σy) for someσ ∈ Σ
Large DeviationsandRandom Graphs – p.34/39
![Page 121: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/121.jpg)
f is not unique. ButrΓ(f) ≡ rΓ(g) if and only iff(x, y) = g(σx, σy) for someσ ∈ Σ
In other wordsr(·) ∈ K/Σ
Large DeviationsandRandom Graphs – p.34/39
![Page 122: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/122.jpg)
SinceK/Σ is compact it is enough to prove theupper bound for ballsB(f̃ , ǫ) in K/Σ
Large DeviationsandRandom Graphs – p.35/39
![Page 123: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/123.jpg)
SinceK/Σ is compact it is enough to prove theupper bound for ballsB(f̃ , ǫ) in K/Σ
This means estimating
QN,p[∪σ∈ΣB(σf, ǫ)]
Large DeviationsandRandom Graphs – p.35/39
![Page 124: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/124.jpg)
SinceK/Σ is compact it is enough to prove theupper bound for ballsB(f̃ , ǫ) in K/Σ
This means estimating
QN,p[∪σ∈ΣB(σf, ǫ)]
Szemerédi’s regularity lemma.
Large DeviationsandRandom Graphs – p.35/39
![Page 125: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/125.jpg)
SinceK/Σ is compact it is enough to prove theupper bound for ballsB(f̃ , ǫ) in K/Σ
This means estimating
QN,p[∪σ∈ΣB(σf, ǫ)]
Szemerédi’s regularity lemma.
The permutation groupΠN ⊂ Σ by permutingintervals of length1
N.
Large DeviationsandRandom Graphs – p.35/39
![Page 126: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/126.jpg)
SinceK/Σ is compact it is enough to prove theupper bound for ballsB(f̃ , ǫ) in K/Σ
This means estimating
QN,p[∪σ∈ΣB(σf, ǫ)]
Szemerédi’s regularity lemma.
The permutation groupΠN ⊂ Σ by permutingintervals of length1
N.
Givenǫ > 0, there is a finite set{gj} ⊂ K such thatfor sufficiently largeN ,
KN = ∪j ∪σ∈ΠNB(σgj, ǫ)
Large DeviationsandRandom Graphs – p.35/39
![Page 127: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/127.jpg)
It is therefore enough to estimate the probability
QN,p[∪j ∪σ∈ΠNB(σgj, ǫ)) ∩ [∪σ∈ΣB(σf, ǫ)]]
Large DeviationsandRandom Graphs – p.36/39
![Page 128: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/128.jpg)
It is therefore enough to estimate the probability
QN,p[∪j ∪σ∈ΠNB(σgj, ǫ)) ∩ [∪σ∈ΣB(σf, ǫ)]]
Sincej only varies over a finite set, it is enough toestimate for anyg
QN,p
[
[∪σ∈π(N)B(σg, ǫ)] ∩ [∪σ∈ΣB(σf, ǫ)]]
Large DeviationsandRandom Graphs – p.36/39
![Page 129: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/129.jpg)
It is therefore enough to estimate the probability
QN,p[∪j ∪σ∈ΠNB(σgj, ǫ)) ∩ [∪σ∈ΣB(σf, ǫ)]]
Sincej only varies over a finite set, it is enough toestimate for anyg
QN,p
[
[∪σ∈π(N)B(σg, ǫ)] ∩ [∪σ∈ΣB(σf, ǫ)]]
This is the same as
QN,p
[
∪σ∈π(N) [B(σg, ǫ) ∩ ∪σ∈ΣB(σf, ǫ)]]
Large DeviationsandRandom Graphs – p.36/39
![Page 130: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/130.jpg)
QN,p is ΠN invariant.N ! << ecN2
.
QN,p
[
B(g, ǫ) ∩ ∪σ∈ΣB(σf, ǫ)]
Large DeviationsandRandom Graphs – p.37/39
![Page 131: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/131.jpg)
QN,p is ΠN invariant.N ! << ecN2
.
QN,p
[
B(g, ǫ) ∩ ∪σ∈ΣB(σf, ǫ)]
If the intersection is nonempty, then it is containedin B(σf, 3ǫ) for someσ ∈ Σ.
Large DeviationsandRandom Graphs – p.37/39
![Page 132: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/132.jpg)
QN,p is ΠN invariant.N ! << ecN2
.
QN,p
[
B(g, ǫ) ∩ ∪σ∈ΣB(σf, ǫ)]
If the intersection is nonempty, then it is containedin B(σf, 3ǫ) for someσ ∈ Σ.
The choice ofσ does not depend onN . Only onǫ.
Large DeviationsandRandom Graphs – p.37/39
![Page 133: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/133.jpg)
QN,p is ΠN invariant.N ! << ecN2
.
QN,p
[
B(g, ǫ) ∩ ∪σ∈ΣB(σf, ǫ)]
If the intersection is nonempty, then it is containedin B(σf, 3ǫ) for someσ ∈ Σ.
The choice ofσ does not depend onN . Only onǫ.
Balls are weakly closed.
Large DeviationsandRandom Graphs – p.37/39
![Page 134: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/134.jpg)
QN,p is ΠN invariant.N ! << ecN2
.
QN,p
[
B(g, ǫ) ∩ ∪σ∈ΣB(σf, ǫ)]
If the intersection is nonempty, then it is containedin B(σf, 3ǫ) for someσ ∈ Σ.
The choice ofσ does not depend onN . Only onǫ.
Balls are weakly closed.
We have upper bounds.Hp(σf) = Hp(f)
Large DeviationsandRandom Graphs – p.37/39
![Page 135: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/135.jpg)
QN,p is ΠN invariant.N ! << ecN2
.
QN,p
[
B(g, ǫ) ∩ ∪σ∈ΣB(σf, ǫ)]
If the intersection is nonempty, then it is containedin B(σf, 3ǫ) for someσ ∈ Σ.
The choice ofσ does not depend onN . Only onǫ.
Balls are weakly closed.
We have upper bounds.Hp(σf) = Hp(f)
Done!
Large DeviationsandRandom Graphs – p.37/39
![Page 136: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/136.jpg)
With p = 12, we have done the counting.
Large DeviationsandRandom Graphs – p.38/39
![Page 137: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/137.jpg)
With p = 12, we have done the counting.
The quantity
D(N, ǫ) = #|{G : |rG(Γj)−rj| ≤ ǫ for j = 1, 2, . . . , k}|
Large DeviationsandRandom Graphs – p.38/39
![Page 138: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/138.jpg)
With p = 12, we have done the counting.
The quantity
D(N, ǫ) = #|{G : |rG(Γj)−rj| ≤ ǫ for j = 1, 2, . . . , k}|
Satisfies
limǫ→0
limN→∞
2
N2logD(N, ǫ) = log 2− ψ 1
2({Γj, rj})
Large DeviationsandRandom Graphs – p.38/39
![Page 139: Large Deviations Random Graphsdepts.washington.edu/ssproc/ChungLecturers/Varadhan_lecture.pdfWhat types of random graphs do we consider? A pair of vertices are connected by an edge](https://reader034.vdocument.in/reader034/viewer/2022042808/5f838ba06cd5f9676a77db2c/html5/thumbnails/139.jpg)
Thank You.
Large DeviationsandRandom Graphs – p.39/39