0% found this document useful (0 votes)
37 views

Computer Network Homework Help

The document contains 3 computer network homework problems. Problem 1 asks about clustering in random graph models and finds that the expected number of triangles is constant, the clustering coefficient decays to 0 for one model but is a constant for another, and the ratio of diameters approaches 0 for one model pair but remains undefined for another. Problem 2 demonstrates how eigenvector centrality can change drastically by adding one edge to an infinite graph. Problem 3 analyzes a network of oscillators and derives conditions for stability of small perturbations around the origin state.
Copyright
© © All Rights Reserved
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
37 views

Computer Network Homework Help

The document contains 3 computer network homework problems. Problem 1 asks about clustering in random graph models and finds that the expected number of triangles is constant, the clustering coefficient decays to 0 for one model but is a constant for another, and the ratio of diameters approaches 0 for one model pair but remains undefined for another. Problem 2 demonstrates how eigenvector centrality can change drastically by adding one edge to an infinite graph. Problem 3 analyzes a network of oscillators and derives conditions for stability of small perturbations around the origin state.
Copyright
© © All Rights Reserved
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 12

For any Homework related queries, Call us at : - 

+1 678 648 4277


You can mail us at : - [email protected] or
reach us at : - https://www.computernetworkassignmenthelp.com/

Computer Network Homework


Problem 1
Clustering
Help
Consider the Erdos-Renyi random graph G1(n, p) with mean degree
a.
(a) Show that in the limit of large n, the expected number of
triangles in the network is a constant.
(b) Calculate the clustering coefficient C in the limit of large n. Note:
The clustering coefficient is defined as three times the number of
triangles divided by the number of connected triplets. A
“connected triplet” means three vertices uvw with edges (u, v)
and (v,w). The edge (u,w) can be present or not.
(c) Calculate the clustering coefficient C for the Erdos-Renyi random
graph G2(n, p) with p(n) = a log(n)/n. Compare your answer with
part (b) in the limit of large n.
(d) Compare the ratio of the diameters of G1 and G2 in the limit of
large n.
(e) Now, consider a different construction for a random graph
model. We take n vertices and go through each distinct trio of
three vertices and with independent probability
connect the trio using three edges to form a
triangle. Compute the mean vertex degree and clustering coefficient
for this network model.

Problem 2.
Centrality in Infinite Graphs.

In this problem, you will demonstrate an example that shows that


eigenvector centrality can be very sensitive to minimal changes in a
network. The problem is broken into different components that
finally lead to the conclusion.

Part I
First, consider the infinite ring network as in Figure 1.

Assume that xi is the eigenvector centrality measure of node i.

(a) Show that the xi’s are computed by finding the largest λ for
which there exists a set of xi for i = 0, 1, 2, . . . such that:
Note that we can always normalize the eigenvector centrality by
dividing xi by x0 for all i, so that x0 = 1.
(b) Show that all the nodes have equal ranking. (Hint: Show that
the is xi = 1 for all i. This should be a straightforward conclusion
and can be proved by inspection.).

Part II
Next, as shown in Figure 2, we add an edge between two nodes
so that the infinite ring is divided into two symmetric halves. We
will examine the eigenvector centrality of this new network. By
symmetry, we only need to find the eigenvector centrality

measures indexed by x0, x1, x2, . . . . As before, we always


normalize it so that x0 = 1.

(c) Write down the system of equations which characterize the


eigenvalue centrality.

(d) Show that the eigenvector centrality must satisfy:

(Hint: start with an initial guess xi = 1 for all i, and try to iteratively
compute the eigenvector. You can prove the inequalities by
induction.)
(e) Show that the largest eigenvalue λ must satisfy:

2≤λ≤3

(Hint: observe the system of equations you wrote down for (c).)

(f) We have shown in Part (e) that xn is positive and decreasing in n.


Please prove that limn→∞ xn = 0.

(Hint: consider writing the system of equations in Q2 into the form of


a linear dynamical system with state y[n] given by:

write down the recursive equation y[n + 1] = Ay[n] that describes the
evolution of the linear dynamics, and think about the equilibrium.)
Now you have demonstrated that by adding a single edge one can
change the relative centrality measure x x n 0 drastically.

Problem 3.
Synchronization.

An oscillator is a simple dynamical system that can be modeled by a


first order differential equation. A network of n oscillators can be
modeled by a system of differential equations of the form:

where θi represents the phase angle and is the state of the oscillator
on vertex i, ω is a constant, and the function g(x) has g(0) = 0 and
respects the rotational symmetry of the phases, meaning that g(x +
2π) = g(x) for all x.
(a) Characterize all solutions of the form θi(t) = ait+bi to the
set of dynamical equations, i.e., find ai, bi, i = 1, . . . , n.

(a) Consider a small perturbation away from the state θi =


ωt + i and show that the vector = (1, 2, . . . ,) satisfies

Your solution should specify L in terms of [Aij ], the adjacency


matrix of an undirected graph. (Hint: Using the Taylor series
approximation of g(·) around

(c) Show that L = MTM where M is the incidence matrix, i.e., the
rows correspond to the edges and columns correspond to the
vertices. Therefore, for every edge e = (i, j) between i, j where i < j
we have that

(d) Argue that L is a symmetric matrix and that for any vector x we
have that xTLx ≥ 0. Conclude from this that all the eigenvalues of L
are non–negative. (Hint: Use the fact that all eigenvalues, λ, of a
symmetric matrix, P, are of the form
where v is the corresponding eigenvector.)
(e) For what values of g 0 (0) is the system stable to small
perturbations around the origin?
(Hint: You can use a Lyapunov argument with quadratic Lyapunov
function V (x) = xT x to examine stability of the linearized system in
part (b))
Solution
Solution 1.
Clustering

Consider the Erdos-Renyi random graph G1(n, p) with mean degree


a.

(a)

(b) Call �i,j,k as the indicator random variable that denotes a


triple between i, j, k.

Clustering Coefficient = p

limn→∞ p = 0.
(c) As shown before, it is just p, and still 0. But since p is larger in
the latter case, there is a higher chance that a triple is actually a
triangle.

(d) For the value of p in G1 the graph is almost surely


disconnected. As a result it has infinite diameter. On the other
hand, G2 has a diameter that is Ω(log(n)) (and is almost surely
connected). Therefore, the ratio G2/G1 = 0
(e) Pick an arbitrary vertex i, let Δij be the indicator random variable
denoting a link between i, j. Then

Then the average degree is E[Δij ](n�1) = a. Observe that in this


case the number of triangles grow linearly as the number of nodes
Number of Triangle =

Also, the number of triples of uvw occurs when either uvw is a


triangle or there are triangles uvj and vwk and no triangle uvw
where j, k =6 u, v,w. Call the event of no triangle in uvw but there
exists a triple between them as �(uvw)

The ratio is roughly 3/(3a + 1) and points have been awarded to


show that the clustering coefficient is a constant (does not decay
to 0 as n → ∞).
Solution 2.
Centrality in Infinite Graphs.

Part I
(a) This follows by representing the ring network as Ax =
λx where λ is the largest eigenvalue and A is the
adjacency matrix.

(b) One can argue by symmetry or show that λ = 2. If λ = 2


all xi = 1 is a solution and hence the eigenvector. λ
cannot be larger than 2. If so, and xt was the largest
then
λ ≤ 2 indeed.

Part II

(c)

(d) Take initial guess as xi t = 1, where

Clearly for t = 0 this is true (after many iterations this will converge to
the true centrality). Assume this for t = k. Now observe for

The last inequality follows from inductive hypothesis.


(e) λ ≥ 2 because if we sum up the equation in (c) we get

Since x0 is the largest, we also have that

(e) Observe that

for all n ≥ 1. Then since xn ≤ xn�1 we only need to show that there
is an eigenvalue of A that is < 1. As it turns out that eigenvalue is

It can be shown that λA < 1 whenever λ > 2. This implies that xn →


0.

Solution 3.
Synchronization.
(a) Characterize all solutions of the form θi(t) = ait+bi to the set
of dynamical equations, i.e., find ai, bi, i = 1, . . . , n.

(b) Substitute θi = ωt + i. Then the Taylor approximation around


0 can be written as
Now stacking the i we get

(D - A) = L.

L is also called the graph Laplacian. We will prove that L has only non–
negative eigenvalues.

(c) Show that L = MTM where M is the incidence matrix, i.e., the
rows correspond to the edges and columns correspond to the
vertices. Therefore, for every edge e = (i, j) between i, j where i < j
we have that

The proof is here.

(d) Argue that L is a symmetric matrix and that for any vector x
we have that xTLx ≥ 0. Conclude from this that all the
eigenvalues of L are non–negative. (Hint: Use the fact that all
eigenvalues, λ, of a symmetric matrix, P, are of the form
where v is the corresponding eigenvector.)

(e) (1pt) For what values of g 0 (0) is the system stable to small
perturbations around the origin?

(Hint: You can use a Lyapunov argument with quadratic Lyapunov


function V (x) = xT x to examine stability of the linearized system in
part (b)).

g 0 (0) < 0 ensures system is stable.

You might also like