UNLV Theses, Dissertations, Professional Papers, and Capstones

5-1-2022

Some Graph Laplacians and Variational Methods Applied to

Partial Differential Equations on Graphs
Daniel Anthony Corral

Follow this and additional works at: https://digitalscholarship.unlv.edu/thesesdissertations

Part of the Applied Mathematics Commons, and the Mathematics Commons

Repository Citation
Corral, Daniel Anthony, "Some Graph Laplacians and Variational Methods Applied to Partial Differential
Equations on Graphs" (2022). UNLV Theses, Dissertations, Professional Papers, and Capstones. 4498.
http://dx.doi.org/10.34917/33690268

This Dissertation is protected by copyright and/or related rights. It has been brought to you by Digital
Scholarship@UNLV with permission from the rights-holder(s). You are free to use this Dissertation in any way that
is permitted by the copyright and related rights legislation that applies to your use. For other uses you need to
obtain permission from the rights-holder(s) directly, unless additional rights are indicated by a Creative Commons
license in the record and/or on the work itself.

This Dissertation has been accepted for inclusion in UNLV Theses, Dissertations, Professional Papers, and
Capstones by an authorized administrator of Digital Scholarship@UNLV. For more information, please contact
[email protected].
SOME GRAPH LAPLACIANS AND VARIATIONAL METHODS APPLIED TO PARTIAL

DIFFERENTIAL EQUATIONS ON GRAPHS

By

Daniel Anthony Corral

Bachelor of Science - Mathematics

University of Nevada, Las Vegas
2006

Master of Science - Mathematics

University of Nevada, Las Vegas
2007

Master of Arts - Mathematics

University of Kentucky
2011

A dissertation submitted in partial fulfillment

of the requirements for the

Doctor of Philosophy - Mathematical Sciences

Department of Mathematical Sciences

College of Sciences
The Graduate College

University of Nevada, Las Vegas

May 2022
Copyright © 2022 by Daniel Anthony Corral
All Rights Reserved
 Dissertation Approval
The Graduate College
The University of Nevada, Las Vegas

April 8, 2022

This dissertation prepared by

Daniel Anthony Corral

entitled

Some Graph Laplacians and Variational Methods Applied to Partial Differential

Equations on Graphs

is approved in partial fulfillment of the requirements for the degree of

Doctor of Philosophy - Mathematical Sciences

Department of Mathematical Sciences

David Costa, Ph.D. Kathryn Hausbeck Korgan, Ph.D.

Examination Committee Chair Vice Provost for Graduate Education &
Dean of the Graduate College
Ebrahim Salehi, Ph.D.
Examination Committee Member

Hossein Tahrani, Ph.D.

Examination Committee Member

Pushkin Kachroo, Ph.D.

Graduate College Faculty Representative

ii
 ABSTRACT

SOME GRAPH LAPLACIANS AND VARIATIONAL METHODS APPLIED

TO PARTIAL DIFFERENTIAL EQUATIONS ON GRAPHS
by

Daniel Anthony Corral

Dr. David Costa, Examination Committee Chair

Professor of Mathematical Sciences
University of Nevada, Las Vegas, USA

In this dissertation we will be examining partial differential equations on graphs. We start by presenting

some basic graph theory topics and graph Laplacians with some minor original results. We move on to

computing original Jost graph Laplacians of friendly labelings of various finite graphs. We then continue on

to a host of original variational problems on a finite graph. The first variational problem is an original basic

minimization problem. Next, we use the Lagrange multiplier approach to the Kazdan-Warner equation on

a finite graph, our original results generalize those of Dr. Grigor’yan, Dr. Yang, and Dr. Lin. Then we

do an original saddle point approach to the Ahmad, Lazer, and Paul resonant problem on a finite graph.

Finally, we tackle an original Schrödinger operator variational problem on a locally finite graph inspired by

some papers written by Dr. Zhang and Dr. Pankov. The main keys to handling this difficult breakthrough

Schrödinger problem on a locally finite graph are Dr. Costa’s definition of uniformly locally finite graph and

the locally finite graph analog Dr. Zhang and Dr. Pankov’s compact embedding theorem when a coercive

potential function is used in the energy functional. It should also be noted that Dr. Zhang and Dr. Pankov’s

deeply insightful Palais-Smale and linking arguments are used to inspire the bulk of our original linking

proof.

iii
 ACKNOWLEDGEMENTS

I would like to start by thanking my PhD advisor Dr. David Costa for being so inspirational and

supportive during the entire process of producing this PhD dissertation. Dr. Costa has a keen level of

insight into difficult mathematical problems that he shares with all those around him. I appreciate his

saint-like patience with me as I take a long time to understand mathematical concepts.

I also am deeply thankful for Dr. Costa for sending me to the 1141st AMS Meeting at University of

Delaware in September of 2018 to present my talk “Remarks on Lagrange Multiplier Approch to Kadzan-Warner

Equations on a Finite Graph” which is the result of joint work between him and I. I would also like to thank

Dr. Alexander Pankov (Morgan State University, USA), Dr. Junping Shi (College of William and Merry,

USA) and Dr. Jun Wang (Jiangsu University, China) for inviting me to the AMS sectional in Delaware.

I also want to thank my PhD committee for their help in this process. My committee members are

chair Dr. David Costa, Dr. Ebrahim Salehi, Dr. Hossein Tehrani, outside committee member Dr. Pushkin

Kachroo. Serving on a PhD committee takes a lot of time from their already busy schedules: and I truly

appreciate their service and help.

I would like to extend a special thank you to Dr. Hossein Tehrani for his extremely helpful questions and

edit suggestions prior to my defense. These edit suggestions have substantially improved this dissertation

to its final form. Dr. Tehrani’s questions both before and during the defense helped to make the defense

stronger. I am very grateful for the time it took him to prepare these helpful comments.

I would like to give a special thank you to Dr. Ebrahim Salehi for his introducing me to the field of

graph labeling while he was chair of my master’s thesis and for his generous working with me on our research

project on fully cordial trees in our talk “Fully Cordial Trees” presented at the 28th Midwest Conference on

Combinatorics and Combinatorial Computing at University of Nevada, Las Vegas in October of 2014 and

our paper “Fully Cordial Trees” published in 2016. I would also like to thank Dr. Salehi for providing the

figures used in this dissertation.

Dr. Costa and I both wish to thank Dr. Alexander Pankov, who has sadly passed away, for his suggesting

working on a Schrödinger operator variational problem on a locally finite graph inspired by Dr. Zhang and

Dr. Pankov papers. We are deeply saddened that he did not get to see the completion of this project. I

iv
would personally like to thank both Dr. Costa and and Dr. Pankov for working on this problem with me.

I need to give a big thank you to my Dad for being the best Dad in the world and being so supportive

of me in all my endeavors. I appreciate him doing all the household chores in order to give me the free time

to devote to working on my graduate school work and PhD training activities. Also, my Dad drives me

everywhere as I cannot drive.

I give a giant shout out to my close friend Sean Trendell for all of his help with the typing of some this

dissertation in LATEX and formatting of this dissertation to comply with the graduate college regulations.

Without Sean, typing of this dissertation would never have been completed in time.

I have a lot of friends I wish to thank by name, however you know who you are. Thank you for putting

up with me throughout this PhD experience.

v
 TABLE OF CONTENTS

ABSTRACT iii

ACKNOWLEDGEMENTS iv

LIST OF FIGURES vii

1 INTRODUCTION 1

2 GRAPH LAPLACIANS APPLIED TO FRIENDLY LABELINGS ON

FINITE GRAPHS 14
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2 Friendly Labelings and Friendly Index Set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.3 Fully Cordial Trees . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.4 Jost Graph Laplacian Applied to Friendly Labelings on Various Finite Graphs . . . . . . . . 25
2.5 Weighted Laplacian-Like Operator Applied to Friendly Labelings on Various Finite Graphs . 37
2.6 Future Research Objectives Dealing with Graph Laplacians of Friendly Labelings . . . . . . . 39

3 VARIATIONAL METHODS ON FINITE GRAPHS 43

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.2 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.3 A Minimization Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.4 Lagrange Multiplier Approach to Kazdan-Warner Equation on a Finite Graph . . . . . . . . 50
3.5 Saddle Point Approach to Ahmad, Lazer, and Paul Resonant Problem on a Finite Graph . . 58

4 VARIATIONAL METHODS ON LOCALLY FINITE GRAPHS 62

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.2 Definitions and Energy Norm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.3 Spectrum and Compact Embedding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.4 Schrödinger Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.5 Energy Functional . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.6 A Nonlinear Schrödinger Problem on a Locally Finite Graph . . . . . . . . . . . . . . . . . . 70

BIBLIOGRAPHY 90

CURRICULUM VITAE 94

vi
 LIST OF FIGURES

2.1 A caterpillar of diameter n + 1 (Pn -spine)[52][20]. . . . . . . . . . . . . . . . . . . . . . . . . . 20

2.2 A caterpillar of diameter 4, CR(8, 9, 8)[52][20]. . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.3 The complete graphs K1 , K2 ,K3 , K4 [20]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.4 Two complete bipartite graphs K(2, 3) is on the left and K(3, 3) is on the right [20]. . . . . . 29
2.5 The star with 7 edges [20]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.6 A caterpillar of diameter 4, CR(8, 9, 8)[52][20]. . . . . . . . . . . . . . . . . . . . . . . . . . . 34

vii
 CHAPTER 1

INTRODUCTION

In this section we introduce some very basic definitions and concepts from both graph theory and partial

differential equation theory on graphs. The definitions and theorems in this chapter are standard graph

theory definitions and theorems. They are taken from [27] and [55] unless stated otherwise.

Definition 1.1. A graph G is a triple consisting of a vertex set V (G), an edge set E(G), and a relation that

associates with each edge two vertices (not necessarily distinct) called endpoints.

In general, we will simply say a graph G has a vertex set V and an edge set E.

Definition 1.2. Let G be a graph. Two vertices x, y ∈ G are said to be adjacent if there is an edge connecting

x to y.

In a graph two vertices being adjacent is denoted by x ∼ y.

Definition 1.3. A weighted graph is a couple (Γ, µ) where Γ = (V, E) is a graph with vertex set V and edge

set E and µxy is a non-negative function on V × V such that

1. µxy = µyx ,

2. µxy > 0 if and only if x ∼ y,

P
3. µ(x) = µxy ,
y∼x

where, µ is called the weight (or measure) of the graph Γ.

It should be noted that this definition implies that µxy = 0 if and only if x and y are not adjacent.

Note that µxy can be considered as a positive function on the set E of edges that is extended to be 0 on

non-edge pairs (x, y). This means that µ : E → R≥0 can be thought of as what graph theorists usually refer

to as a graph labeling. Now we are ready to state some more useful definitions.

1
Definition 1.4. Given a weighted graph (Γ, µ) then µ is said to be simple if and only if x ∼ y implies that

µxy = 1.

A word of caution should be noted at this point. Notice that Alexander Grigoryan is talking about a

simple measure here. Where the typical notion of simple in graph theory usually refers to a simple graph.

A simple graph is a graph where there are no loops.

There are numerous different graph laplacians. We now give a definition of the kind of Laplace operator

studied by Jost, Banerjee, Bauer, and Hao Chen. The following definition is from Alexander Grigoryan text

Analysis on Graphs. We begin by defining what it means for a graph to be locally finite. However, before

we can define what it means for a graph to be locally finite we define the notion of the degree of a vertex.

We adapt the definition of degree which Douglas B. West gives in his text ”Introduction to Graph Theory”

(not a direct citation). The adaptation is more suited to notation of this manuscript.

Definition 1.5. Given a graph G = (V, E) and x ∈ V . The degree of x, deg(x), is the number of edges

incident to x, except that each loop at x counts twice. Keep in mind when graph theorists refer to a simple

graph that G has no loops.

Definition 1.6. A graph (V, E) is said to be a locally finite graph if and only if 0 < deg(x) < ∞ ∀x ∈ V .

Definition 1.7. [55] If G has a x, y-path, then the distance form x to y, written dG (x, y) or simply d(x, y),

is the least length of all x, y-paths.

Note if there are no x, y-paths, then d(x, y) = ∞ [55].

Definition 1.8. [55] The diameter of a graph G, (diam G), is the max d(x, y).
x,y∈V

Now we may proceed in defining the Laplace operators studied by Jost and Grigoryan.

Definition 1.9. (Degree Laplace Operator) Let (V, E) be a locally finite connected graph. For any function

f : V → R, define the Laplace operator, ∆f , by

1 X
∆f (x) = f (y) − f (x). (1.1)
deg(x) y∼x

The Laplace operator studied by Jost is L = −∆f .

Unless it is said otherwise when we mention Laplace operator we are referring to the Jost Laplace operator.

2
Definition 1.10. (Weighted Laplace Operator) Let G = (V, E) be a locally finite connected graph. Suppose

that G has a weight µ. For any function f : V → R, define the weighted µ-Laplace operator, ∆µ f , by

1 X
∆µ f (x) = f (y)µxy − f (x). (1.2)
µ(x) y∼x

The normalized graph Laplace operator studied by Grigoryan is Lµ = −∆µ f .

It is well known that the Laplace operators studied by Jost and Grigoryan are linear. We will just

demonstrate the linearity of ∆µ and a couple of other well known results.

Theorem 1.1. Let G = (V, E) be a locally finite connected graph. Suppose that G has a weight µ. Then ∆µ

is linear.

Proof. Let F be the set of all real-valued functions on V . Suppose that f, g ∈ F. Let λ ∈ R. Note that

1 X
∆µ (λf + g) (x) = (λf (y) + g (y)) µxy − (λf (x) + g (x))
µ (x) y∼x
!
1 X 1 X
=λ f (y) − f (x) + g (y) − g (x)
µ (x) y∼x µ (x) y∼x
= λ∆µ f (x) + ∆µ g (x)

Therefore ∆µ is linear.

Theorem 1.2. Let G = (V, E) be a locally finite connected graph. Suppose that G has a weight µ. Let c ∈ R.

Then ∆µ (c) = 0.

1 1
1 − c = c µ(x)
P P
Proof. Let c ∈ R. Observe that ∆µ c = µ(x) c − c = c µ(x) µ(x) − c = c − c = 0.
y∼x y∼x

As was mentioned earlier, there are numerous different graph laplacians. We will now define a few more

of them.

The first Laplace operator that we will discuss was studied by Manfredi, Obermann, and Sviridov in

their paper ”Nonlinear Elliptic Partial Differential Equations And P-Harmonic Functions On Graphs”[44].

We will call this Laplace operator the Manfredi operator and denote it by ∆man f (x).

Definition 1.11. (Laplace Operator Studied by Manfredi) Let G = (V, E) be a locally finite connected graph.

Suppose that G has a weight µ. For any function f : V → R, define the Manfredi Laplace operator, ∆man f ,

by
X
∆man f (x) = (f (y) − f (x))µxy . (1.3)
y∼x

3
 We show that the Manfredi operator is also linear. This result is fundamental enough that I suspect it

is proven elsewhere. However, I could not find a proof anywhere while doing research.

Theorem 1.3. Let G = (V, E) be a locally finite connected graph. Suppose that G has a weight µ. Then

∆man f is linear.

Proof. Let F be the set of all real-valued functions on V . Suppose that f, g ∈ F. Let λ ∈ R. Note that

X
∆man (λf + g) (x) = ((λf + g)(y) − (λf + g)(x)) µxy
y∼x
X
= ((λf (y) + g(y)) − (λf (x) + g(x))) µxy
y∼x
X
= ((λf (y) − λf (x)) + (g(y) − g(x))) µxy
y∼x
X
= ((λf (y) − λf (x))µxy + (g(y) − g(x))µxy )
y∼x
X X
=λ (f (y) − f (x))µxy + (g(y) − g(x))µxy
y∼x y∼x

= λ∆man f + ∆man g

Therefore ∆man f is linear.

Let x ∈ Rd . The gradient studied by Manfredi is

∇u(x) = (w(exy1 )(u(y1 ) − u(x)), ..., w(exyd )(u(yd ) − u(x))). (1.4)

→
−
Note that the Laplace operator studied by Manfredi may be written as ∆man f (x) = 1 · ∇u(x) instead

of (1.3).

The next we will state a number of Laplace operators that are defined by Manfredi, Obermann, and

Sviridov in [44]. The Laplace operators we star are positive and negative eikonal operators, the infinity

Laplacian, the homogeneous Dirichlet problem for the positive eikonal operator, and the 1-Laplacian.

Definition 1.12. Let G = (V, E) be a locally finite connected graph. For x ∈ Rd let min(x) = min{x1 , ..., xd }

and max(x) = max{x1 , ..., xd }. The positive and negative eikonal operators on a graph are

|∇u(x)|+ = max(∇u(x)) = max(f (y) − f (x)))µxy ;

y∼x
(1.5)
|∇u(x)|− = min(∇u(x)) = min(f (y) − f (x)))µxy .
y∼x

Definition 1.13. Let G = (V, E) be a locally finite connected graph and let x ∈ Rd . The Infinity Laplacian
|∇u(x)|+ +|∇u(x)|−
is defined by ∆∞ u(x) = 2 .

4
Definition 1.14. The homogeneous Dirichlet problem (D) for the positive eikonal operator is defined by
|∇u(x)|+ − 1 = 0, x ∈ V − ∂V ;
(1.6)
u(x) = 0, x ∈ ∂V.
Definition 1.15. The 1-Laplacian is defined by

∆1 u(x) = median(∇u(x)). (1.7)

We will now discuss a relationship between the graph Laplacians studied by Grigoryan and Manfredi. To

our knowledge the following theorem is original work.

Theorem 1.4. Let (V, E) be a locally finite connected graph. Suppose that G has a weight µ. The relationship

between the graph Laplacians studied by Grigoryan and Manfredi may be described by:
1
∆µ f (x) = ∆man f (x);
µ(x)
(1.8)
1
Lµ f (x) = − ∆man f (x).;
µ(x)
Proof. Note that,
1 X
∆µ f (x) = f (y)µxy − f (x)
µ(x) y∼x
1 X
= ( f (y)µxy − f (x)µ(x))
µ(x) y∼x
1 X X
= ( f (y)µxy − f (x) µxy )) (1.9)
µ(x) y∼x y∼x
1 X
= (f (y) − f (x))µxy
µ(x) y∼x
1
= ∆man f (x).
µ(x)
1
Hence, Lµ f (x) = − µ(x) ∆man f (x).

The following theorem shows that the sum of the graph Laplacian studied by Manfredi over every vertex

is zero. To our knowledge the statement of this theorem, the corollary that follows it and their proofs are

original. The theorem may have applications in network flow theory.

Theorem 1.5. Let (V, E) be a connected finite graph. Then

X
∆man f (x) = 0. (1.10)
x∈V

Furthermore,
XX
(∇xy f ) µxy = 0. (1.11)
x∈V y∼x

5
Proof. Consider an arbitrary edge ab ∈ E on the graph. Notice that when x = a then b ∼ a and
P P P
(f (b) − f (a)) µab is in the sum ∆man f (x) = (f (y) − f (x)) µxy . Also when x = b then a ∼ b
x∈V x∈V y∼x
P P P
and (f (a) − f (b)) µba is in the sum ∆man f (x) = (f (y) − f (x)) µxy . It is important to note that
x∈V x∈V y∼x
µab = µba because the adjacency operation is symmetric.

Hence,

X XX
∆man f (x) = (f (y) − f (x)) µxy
x∈V x∈V y∼x
X
= ((f (b) − f (a)) µab + (f (a) − f (b)) µba )
ab∈E
X
= (f (b)µab − f (a)µab + f (a)µba − f (b)µba )
ab∈E
X
= (f (b)µab − f (b)µba + f (a)µba − f (a)µab )
ab∈E
X
= (f (b)µab − f (b)µab + f (a)µab − f (a)µab )
ab∈E
X
= 0
ab∈E
= 0 · |E|

=0

P P
Furthermore since, ∇xy f = f (y) − f (x), then (∇xy f ) µxy = 0.
x∈V y∼x

The following corollary shows that the discrete integral of the Grigoryan graph laplacian over every

vertex is also zero. It will be very useful in our proof of the Lagrange multiplier approach to the Kadzan

Warner equation on a finite graph later in this dissertation. We will again arrive at this corollary later in

the dissertation by using a discrete version of Green’s Theorem on a graph (integration by parts).

Corollary 1.6. Let (V, E) be a connected finite graph. Then

X
∆µ f (x)µ(x) = 0 (1.12)
x∈V

and
X
Lµ f (x)µ(x) = 0 (1.13)
x∈V

Proof. Observe that,

X X 1 X
∆µ f (x)µ(x) = ∆man f (x)µ(x) = ∆man f (x) = 0 (1.14)
µ(x)
x∈V x∈V x∈V

6
and
X X 1 X
Lµ f (x)µ(x) = − ∆man f (x)µ(x) = −∆man f (x) = 0. (1.15)
µ(x)
x∈V x∈V x∈V

It should be noted that Grigoryan, in his text Analysis on Graphs [27], defines the Dirichlet problem for

the Grigoryan graph Laplacian and proves the existence of its solution. We start by giving the definitions

for boundary, interior, subharmonic and superharmonic.

Definition 1.16. Let (V, µ) be a connected locally finite weighted graph and let Ω be a nonempty subset of

V . Define Ωc = V − Ω (in other words V = Ω ∪ Ωc ). The boundary of Ω is defined to be ∂Ω = {y ∈ Ωc :

y ∼ x for some x ∈ Ω}. The interior of Ω is defined by Ωo = Ω − ∂Ω.

Definition 1.17. Let (V, µ) be a connected locally finite weighted graph and let Ω be a nonempty subset of

V . A function u : V → R is called subharmonic in Ω if ∆µ u(x) ≥ 0 ∀x ∈ Ω, and superharmonic in Ω if

∆µ u(x) ≤ 0 ∀x ∈ Ω. A function u is called harmonic in Ω if it is both subharmonic and superharmonic, that

is, if it satisfies the Laplace equation: ∆µ u(x) = 0.

Next, we prove a helpful result which leads to the maximum/minimum principle.

Theorem 1.7. Let G = (V, E) be a connected graph with vertex set V and edge set E. Suppose H is a

non-empty subset of V . Suppose further that H has the property that ∀x ∈ H, if y ∈ V and y ∼ x then

y ∈ H. Then, H = V .

Proof. Let x, y ∈ H be arbitrary. Since G = (V, E) is connected, there exists a path {xk }nk=0 ⊆ V from x to

y such that x = x0 ∼ x1 ∼ x2 ∼ · · · ∼ xn = y. One proceeds by the Principle of Mathematical Induction.

Notice that, x1 ∈ V and x1 ∼ x0 = x ∈ H implies that x1 ∈ H, and this serves as our basis case for the

induction. For the sake of induction suppose that xk ∈ H and xk ∼ xk+1 , then by the property xk+1 ∈ H.

By the Principal of Mathematical Induction: xm ∈ H ∀m such that 1 ≤ m ≤ n. In particular, y = xn ∈ H.

Since y ∈ V was arbitrary, then V ⊆ H ⊆ V . Hence, H = V .

Theorem 1.8. (A maximum/minimum principle from Grigoryan Analysis on Graphs) Let (V, µ) be a

connected locally finite weighted graph. Let Ω ⊆ V such that Ω is finite and Ωc ̸= ∅. Then, for any

function u : V → R, that is subharmonic in Ω, we have

max u ≤ sup u (1.16)

Ω Ωc

7
and for any function u : V → R, that is superharmonic in Ω, we have

min u ≥ infc u. (1.17)

Ω Ω

Proof. It suffices to prove the first claim. First note that if supΩc u = +∞ then the result already holds. So

let’s suppose that supΩc u < ∞. Note that u can be replaced by u − c where c is some constant. So without

loss of generality take sup u(x) = 0. Observe that, sup u(x) = 0 implies u(x) ≤ 0 ∀ x ∈ Ωc .
Ωc Ωc
Let M = maxΩ u.

Suppose to the contrary that, M > 0. Define X = {x ∈ V : u(x) = M }. By definition: X ⊆ V . Note

that, |Ω| < ∞ implies that maxΩ u is obtained for some w ∈ Ω. In particular, there is a w ∈ Ω such that

u(w) = M . Thus, w ∈ X. Hence, X ̸= ∅.

Let x ∈ X be arbitrary and y ∈ V such that y ∼ x is arbitrary. By definition: u(x) = M . Note that,

supΩc u = 0 implies that u(y) ≤ 0 < M ∀y ∈ Ωc . From M = maxΩ u it is seen that u(y) ≤ M ∀y ∈ Ω.

Combining the previous two sentences results in: u(y) ≤ M ∀ y ∈ V as V = Ω ∪ Ωc . Since u : V → R is

subharmonic in Ω, then ∆µ (x) ≥ 0. So then,

1 X
M = u(x) ≤ u(y)µxy
µ(x) y∼x
1 X
≤ u(y)µxy
µ(x) y∼x
1 X
≤ M µxy
µ(x) y∼x (1.18)
1 X
=M µxy
µ(x) y∼x
µ(x)
=M
µ(x)
= M.
1
P
Consequently, µ(x) u(y)µx,y = M with u(y) ≤ M ∀ y ∈ V such that y ∼ x. Hence, u(y) = M . Thus,
y∼x
X has the property that ∀x ∈ X, if y ∈ V is such that y ∼ x then y ∈ X. Now by the previous theorem,

X = V . Therefore, u(x) = M > 0 ∀ x ∈ V . However, this contradicts the fact that u(x) ≤ 0 ∀x ∈ Ωc .

Now the Dirichlet problem is stated. In the following theorem we show a uniqueness result using the

min/max principle. We will also prove the existence of a solution to this Dirichlet problem.

Theorem 1.9. [27] Let (V, µ) be a connected locally finite weighted graph and let Ω be a finite, connected,

8
proper subset of V . Consider the following Dirichlet problem

∆µ u(x) = f (x) ∀x ∈ Ω
(1.19)
c
u(x) = g(x) ∀x ∈ Ω ,

where u : V → R is an unknown function while the functions f : Ω → R and g : Ωc → R are given. If Ω is

finite and Ωc ̸= ∅ then, ∀f, g as above, the Dirichlet equation (1.19) has a unique solution.

Proof. The easy part is to prove the uniqueness of the solution if one exists. Suppose that there are two

solutions u1 and u2 of the Dirichlet problem above. Let u = u1 − u2 . Note that,

∆µ u(x) = 0 ∀x ∈ Ω,
(1.20)
u(x) = 0 ∀x ∈ Ωc .

This implies that, u is harmonic in Ω (i.e. u is both subharmonic and superharmonic in Ω). Now by the

previous maximum/minimum principle,

0 = infc u ≤ min u ≤ max u ≤ sup u = 0. (1.21)

Ω Ω Ω Ωc

Hence, u1 (x) − u2 (x) = u(x) = 0 ∀x ∈ V = Ω ∪ Ωc . Thus, u1 = u2 . Therefore, if the above Dirichlet problem

has a solution, then that solution is unique.

Now the existence of a solution to the Dirichlet problem above the equation is proven. By definition,

∆µ u(x) = f (x) reads

1 X
u(y)µxy − u(x) = f (x). (1.22)
µ(x)
y∼x,y∈V

Using V = Ω ∪ Ωc , the previous equation becomes

1 X 1 X
u(y)µxy + u(y)µxy − u(x) = f (x). (1.23)
µ(x) µ(x)
y∼x,y∈Ω y∼x,y∈Ωc

Hence,
1 X 1 X
u(y)µxy − u(x) = f (x) − u(y)µxy . (1.24)
µ(x) µ(x)
y∼x,y∈Ω y∼x,y∈Ωc

Note that u(y) = g(y) on Ωc and the previous equation implies that,

1 X 1 X
u(y)µxy − u(x) = f (x) − g(y)µxy . (1.25)
µ(x) µ(x)
y∼x,y∈Ω y∼x,y∈Ωc

Let X be the set of all real-valued functions u on Ω and define L : X → X by

1 X
Lu(x) = u(y)µxy − u(x). (1.26)
µ(x)
y∼x,y∈Ω

9
 1
P
Note that L : X → X is an operator. By setting z = f (x) − µ(x) g(y)µxy it is seen that Lu = z.
y∼x,y∈Ωc
Since X is a linear space with {1{x} }x∈Ω as a basis, then dim X = |ω| < ∞. This implies that L : X → X is

a linear operator in a finite dimensional space which is injective (L being injective comes from the uniqueness

part of the proof ). Since any injective operator mapping spaces of equal dimensions must be bijective, then

L : X → X is subjective. Therefore, ∀ z ∈ X, there is a solution u = L−1 z.

Given a graph G = (V, E), [27] defines F to be the set of real-valued functions on V . In [27] it is stated

that F is a vector space and provides an inner product for this space defined for any two functions f, g ∈ F

by:
X
(f, g) = f (x)g(x)µ(x). (1.27)
x∈V

Note that the above equation is the integration of f g against the measure µ on V . As usual the inner product

induces a norm on F defined for f ∈ F by:

sX
p
∥f ∥G = (f, f ) = (f (x))2 µ(x). (1.28)
x∈V

With this in mind it is seen that F equipped with the norm ∥·∥G gives the Hilbert space l2 (F) with measure

of integration µ(x).

Grigoryan defines the difference operator for x, y ∈ V , by ∇xy f = f (y) − f (x). Note that
1 X
∆µ f (x) = f (y)µxy − f (x)
µ(x) y∼x
1 X 1 X X
= f (y)µxy − f (x) µx,y (as µ(x) = µx,y )
µ(x) y∼x µ(x) y∼x y∼x
1 X 1 X
= f (y)µxy − f (x)µx,y (1.29)
µ(x) y∼x µ(x) y∼x
1 X
= (f (y) − f (x))µxy
µ(x) y∼x
1 X
= (∇xy f )µxy .
µ(x) y∼x

Now we states a discrete analogue of Green’s formula on connected locally finite weighted graph.

Theorem 1.10. [27] Let G = (V, E) be a locally finite connected graph. Suppose that G has a weight µ. Let

Ω ⊆ V be such that Ω ̸= ∅ and Ω is finite. Then for any two functions f, g on V:

X 1 X X X
∆µ f (x)g(x)µ(x) = − (∇xy f )(∇xy g)µxy + (∇xy f )g(x)µxy . (1.30)
2 c
x∈Ω x,y∈Ω x∈Ω y∈Ω

10
If V is finite and Ω = V then Ωc is empty so that the last ”boundary” term in the equation above vanishes

yielding the following equation:

X 1 X
∆µ f (x)g(x)µ(x) = − (∇xy f )(∇xy g)µxy . (1.31)
2
x∈Ω x,y∈Ω

We now provide a proof of Corollary 1.6 using Green’s Theorem (integration by parts).

Corollary 1.11. Let (V, E) be a finite connected graph. Then

X
∆µ f (x)µ(x) = 0 (1.32)
x∈V

and
X
Lµ f (x)µ(x) = 0. (1.33)
x∈V

Proof. Apply Green’s Formula with g(x) = 1 for every vertex x ∈ V .

Next we state the result that ∆µ is symmetric with respect to the inner product discussed at the beginning

of this section.

Theorem 1.12. [27] Let G = (V, E) be a finite connected graph. Suppose that G has a weight µ. The

operator ∆µ is symmetric with respect to (·, ·), that is,

(∆µ f, g) = (f, ∆µ g). (1.34)

∀f, g ∈ F.

Proof. Using the discrete Green’s formula (1.31) yields:

X
(∆µ f, g) = ∆µ f (x)g(x)µ(x)
x∈V
1 X
=− (∇xy f )(∇xy g)µxy
2
x,y∈V (1.35)
1 X
=− (∇xy g)(∇xy f )µxy
2
x,y∈V

= (f, ∆µ g).

Let G = (V, E) be a locally finite connected graph. Suppose that G has a weight µ. We earlier noted

that Grigoryan has found it useful to study the positive definite Laplacian: Lµ = −∆µ f . Using the Green’s

formula, It is noted that the Rayleigh quotient of L is:

P
(Lµ f, f ) (∆µ f, f ) 1 x,y∈V (∇xy f )(∇xy f )µxy
R= =− = P . (1.36)
(f, f ) (f, f ) 2 x∈V (f (x))2 µ(x)

11
Theorem 1.13. [27]

For any finite, connected, weighted graph (V, µ) with |V | > 1, the following is true:

(a) Zero is a simple eigenvalue of L.

(b) All the eigenvalues of L are contained in [0, 2].

(c) If (V, µ) is not bipartite then all the eigenvalues of L are in [0, 2).

Proof. (a) First recall that Lc = 0 ∀c ∈ R. Consequently, the constant function is an eigenfunction with the

eigenvalue 0. On the other-hand, suppose that Lf = 0. Hence,

P
(Lf, f ) (∆µ f, f ) 1 x,y∈V (∇xy f )(∇xy f )µxy
R= =− = P 2
= 0. (1.37)
(f, f ) (f, f ) 2 x∈V (f (x)) µ(x)

Thus, {x,y∈V :x∼y} (f (y) − f (x))2 µxy = 0. So then, f (x) = f (y) if and only if x ∼ y. Let x, y ∈ V be
P

arbitrary. Since G = (V, E) is connected, then there exists a path {xk }m

k=0 where x = x0 ∼ x1 ∼ · · · ∼ xN =

y. Hence, f (x) = f (x0 ) = f (x1 ) = · · · = f (xN ) = f (y). Therefore, f is constant.

It should be noted that you can prove that zero is a simple eigenvalue by appealing to the max/min

principle discussed earlier. Pick a single vertex a ∈ V and define Ω = V − {a}. Since Ω is finite and f is

harmonic, then the max/min principle implies:

f (a) = infc f ≤ min f ≤ max f ≤ sup f = f (a). (1.38)

Ω Ω Ω Ωc

Therefore, f is constant.

(b) Suppose that λ is an eigenvalue of L with an eigenfunction f . Then by definition, Lf = λf . Hence,

X X
λ (f (x))2 µ(x) = (Lf )(x)f (x)µ(x)
x∈V x∈V
1 X
= (∇xy f )(∇xy f )µxy
2 (1.39)
x,y∈V
1 X
= (f (y) − f (x))2 µxy .
2
{x,y∈V :x∼y}

2 2 2
Using the inequality (a + b) ≤ 2(a + b ), the previous equation yields the following inequality:
X X
λ (f (x))2 µ(x) ≤ ((f (x))2 + (f (y))2 )µxy
x∈V {x,y∈V :x∼y}
X X
= (f (y))2 µxy + (f (x))2 µxy
x,y∈V x,y∈V
X X (1.40)
2
= (f (y)) µ(x) + (f (x))2 µ(x)
y∈V x∈V
X
2
=2 (f (x)) µ(x).
x∈V

12
 (f (x))2 µ(x) yields: λ ≤ 2.
P
Dividing the previous equation by
x∈V
(c) We proceed by way of the contrapositive. Suppose λ = 2 is an eigenvalue of L with an eigenfunction
1
f . Replacing λ = 2 in the previous inequalities yields: 2 (f (y) − f (x))2 = ((f (x))2 + (f (y))2 ). Thus,

(f (y)−f (x))2 = 2((f (x))2 +(f (y))2 ). This in turn implies, (f (x)+f (y))2 = (f (x))2 +2f (x)f (y)+(f (y))2 = 0

. So then, f (x) + f (y) = 0. If f (v) = 0 for some v ∈ V then, f (w) = 0 for any w ∈ V such that w ∼ v.

By the connectedness of G = (V, E), f would be identically zero contradicting that f is an eigenfunction.

This implies that, f (x) ̸= 0 ∀x ∈ V . It is seen now that V = A ∪ B where A = {x ∈ V : f (x) > 0} and

B = {x ∈ V : f (x) < 0}. Therefore, V is bipartite.

Note that the eigenvalues of L can be placed in increasing the ordering: 0 = λ0 < λ1 ≤ λ2 · · · ≤ λN −1 .

Next we state the following additional theorem about eigenvalues of L. The proof is not provided here.

Theorem 1.14. [27] For any finite, connected, weighted graph (V, µ) with |V | > 1 and suppose the notation

in the previous sentence, the following is true:

N
(a) λ1 +λ2 +· · ·+λN −1 ≤ N . Furthermore, λ1 ≤ N −1 . If (V, µ) has no loops then, λ1 +λ2 +· · ·+λN −1 = N .
N
and λN −1 ≥ N −1 .

N
(b) If (V, µ) is a complete graph with a simple weight µ, then λ1 = λ2 = · · · = λN −1 = N −1 .

(c) If (V, µ) is non-complete then λ1 ≤ 1.

(d) If (V, µ) is a bipartite and λ is an eigenvalue of L, then 2 − λ is also an eigenvalue of L with the same

multiplicity as λ. In particular, 2 is a simple eigenvalue of L.

The original Laplacian of a graph from traditional graph theory obeys similar properties concerning its

eigenvalues. A great reference for this Laplacian is ”Spectral Graph Theory” by Fan R. K. Chung [15].

13
 CHAPTER 2

GRAPH LAPLACIANS APPLIED TO FRIENDLY LABELINGS ON FINITE

GRAPHS

2.1

Introduction

In this chapter we introduce potential applications of the concept of the graph Laplacian to friendly

labelings on weighted graphs. Hopefully, we may use insight gained from studying such concepts and problems

and apply this insight to social network theory, partial differential equations on graphs, or graph theory in

general.

Section 2.2 gives a brief history about friendly labelings and friendly index sets.

Section 2.3 is original work between Dr. Ebrahim Salehi and me. We give the definition of fully cordial

graph and show some trees are fully cordial. The main original result of this section is that trees with near

perfect matching are fully cordial. We also determine if and only if conditions for caterpillars of dimeter 4

to be fully cordial and show ST (n, k) is fully cordial if and only if k is odd.

In Section 2.4 we provide original computations of the Jost graph Laplacian of friendly labelings on

various finite connected simple graphs. These computations are done for the complete graph Kn , the

complete bipartite graph K(m, n), the star ST (n), and a particular caterpillar of diameter 4. The most

interesting result in this section is when Kn has an even number of vertices, then the Jost graph Laplacian

of any friendly labeling is a constant multiple of the eigenvalue of the Jost graph Laplacian operator on Kn .

In Section 2.5 we provide an original definition for a wieghted Laplacian like operator. We provide an

original computation of the weighted Laplacian like operator of friendly labelings on the complete graph Kn .

The goal of this section is to provide an operator that may take into consideration edge labelings that may

occur in the study of friendly index set theory.

Section 2.6 deals with future research objectives dealing with graph Laplacians of friendly labelings. The

14
main original result of this section shows, if q ∈ F I(G), then q − 2 ∈ F I(G), when you have an odd number

of vertices and a leaf. In this section we suggest potential ideas that may help in the search for if and only

if conditions for a tree to be fully cordial and other potential research ideas.

2.2

Friendly Labelings and Friendly Index Set

We begin this section with a discussion of the concept of friendly labelings of finite simple connected

graphs.

If we have a graph G with set of vertices V (G), set of edges E(G), and an abelian group A, then a labeling

f : V (G) → A creates an edge labeling f ∗ defined by f ∗ (xy) = f (x) + f (y). For i ∈ A, let vf (i) = |f −1 (i)|,

ef (i) = |f ∗−1 (i)|, and let X = {|ef (i) − ef (j)| : i, j ∈ A}. The graph G is called A-cordial if it has a graph

labeling f : V (G) → A such that for every i, j ∈ A,

|vf (i) − vf (j)| ≤ 1 and |ef (i) − ef (j)| ≤ 1.

Graph theorists have examined cordial graphs quite a bit. The concept of cordial graphs was introduced

by Cahit in [10][11][12]. Hovey studied A-cordial labelings where A is a Abelian group [35]. The labeling f

is called A-friendly if |vf (i) − vf (j)| ≤ 1 for every i, j ∈ A.

Given a graph G = (V, E) with a coloring f : V (G) → Z2 , let vf (i) = |f −1 (i)|. When |vf (1) − vf (0)| ≤ 1,

the coloring f is said to be friendly. This coloring produces an edge labeling f ∗ : E(G) → Z2 defined by

f ∗ (xy) ≡ f (x) + f (y) (mod 2) for every x, y ∈ E(G). Also let ef (i) = |f ∗−1 (i)|. For a graph G, the friendly

index set of the graph G is denoted by F I(G) = {|ef (1) − ef (0)| : f is a friendly labeling}. The concept of

a friendly index set was first introduced by Chartrand, Lee, and Zhang in [16].

It should be noted that a graph G is cordial if 0 or 1 is in the friendly index set F I(G) [41]. Thus, the

study of friendly index sets is just a generalization of cordiality. Our goal will be to investigate A-friendly

labelings where A = Z2 and F I(G) = {|ef (1) − ef (0)| : f is a friendly labeling of G} will denote the friendly

index set. Note we will often drop the subscript f in the case where the context is clear.

Sin-Min Lee and Ho Kuen Ng made the first calculations of friendly index sets of finite graphs in [41].

Cairne and Edwards proved that computational complexity in determining whether a graph has a cordial

labeling is NP-complete in [13]. In fact the question on if a connected graph of diameter 2 admits a cordial

15
labeling, which is a smaller scale question is also NP-complete. Thus, Cairne and Edwards concluded that

determining the friendly index sets of graphs is difficult.

We now state a few known results in this area from the literature of friendly indices of finite graphs.

From now on we assume G has p vertices and q edges, i.e. |V (G)| = p and |E(G)| = q.

Theorem 2.1. [41] For any finite graph G with q edges: F I(G) ⊆ q − 2i : i = 0, 1, 2, ..., 2q .
  

Theorem 2.2. [41] Suppose that, 1 ≤ m ≤ n. For the complete bipartite graph Km,n :

(m − 2i)2 : 0 ≤ i ≤ m
   
2 , if m + n is even
F I(Km,n ) = (2.1)
{i(i + 1) : 0 ≤ i ≤ m } , if m + n is odd.

Note that for n ≥ 2 that the complete bipartite graph K(1, n) is called a star and in the literature is

usually denoted by ST (n). It should also be noted that stars are the trees of diameter 2. Now we state the

friendly index set of a star from the literature.

Theorem 2.3. [41]


{0, 2}, if n is even
F I(ST (n)) = (2.2)
{1}, if n is odd.

Theorem 2.4. [41] The friendly index set of a full binary tree with depth d > 1 is {0, 2, 4, ..., 2d+1 − 4}.

Ebrahim Salehi and Sin Min Li in [54], found the friendly index sets of trees with perfect matching,

Fibonacci trees, and Lucas Trees. We state a few of these results below. Also the definition of the concepts

of matchings and perfect matchings are given.

Definition 2.1. [54] A matching in a graph is a set of edges with no shared endpoints. A matching M in a

graph G is said to be a perfect matching if every vertex of G is incident with an edge in M.

Definition 2.2. [54][55] A matching M in a graph G is said to be a near perfect matching if it covers all the

vertices of G but one. G is called a near perfect matching graph if any maximal matching is a near perfect

matching.

Theorem 2.5. [54] If T = (p , q) is a tree with perfect matching, then F I(T ) = {1, 3, 5, · · · , q}.

Corollary 2.6. [54] Given a tree T with q edges, the coronation T ©K 1 is a tree with perfect matching and

its friendly index set is {1, 3, 5, · · · , 2q + 1}.

16
 n j q ko
n
Theorem 2.7. [54] For n ≥ 3, the friendly index set of F Tn is |En | − 2i : i = 0, 1, 2, · · · , . Moreover,
2
every element of the index set can be obtained by a friendly coloring f : Vn → ZZ 2 with N (f ) = ef (1) − ef (0),

and the color of root is 1.

Theorem 2.8. [54] For n ≥ 3, the friendly index set of the Lucas tree LTn = (Vn , En ) is { 0, 2, 4, · · · , |En |}.

Moreover, every element b of this index set can be obtained by a friendly coloring f : Vn → ZZ 2 with the

following properties:

(a) f (rn ) = 1;

(b) b = N (f ) = ef (1) − ef (0);

(c) If b ≡ 0 (mod 4), then vf (1) = vf (0) + 1; and

(d) If b ≡ 2 (mod 4), then vf (0) = vf (1) + 1.

2.3

Fully Cordial Trees

Notice from the last 4 theorems in the previous section that every theoretical possible friendly index from

minimal index to the number of edges was obtained, i.e. F I(G) = q − 2i : i = 0, 1, 2, ..., 2q . This led to
  

a definition of the concept of a fully cordial graph. The work in this section is original work I did with Dr.

Ebrahim Salehi and was published as the paper ”Fully Cordial Trees” [52].

Definition 2.3. A graph G is said to be fully cordial if

F I(G) = {q − 2i : i = 0, 1, 2, . . . , ⌊q/2⌋}.

The next leema we present shows that ef (1) ̸= 0.

Lemma 2.9. Let G be a non trivial connected graph and f : V (G) → Z2 any friendly coloring of G. Then

ef (1) ≥ 1.

Proof. The two sets A = {u ∈ V (G) : f (u) = 0} and B = {v ∈ V (G) : f (v) = 1} partition V (G). Since G

is connected, there are vertices u ∈ A and v ∈ B that are adjacent. The label of edge uv is 1. Therefore,

ef (1) ≥ 1.

17
Corollary 2.10. For any graph G, q ∈ F I(G) if and only if ef (0) = 0 for some friendly coloring f : V (G) →

Z2 .

Lemma 2.11. A tree T with near perfect matching M contains at least a P3 pendent u1 ∼ u2 ∼ u3 such

that deg u1 = 1, deg u2 = 2 and u1 u2 ∈ M.

Proof. Let P : u1 ∼ u2 ∼ u3 ∼ · · · ∼ uk−2 ∼ uk−1 ∼ uk be the longest path in T. Clearly, deg u1 = deg uk =

1. Also, deg u2 = 2 or deg uk−1 = 2. Otherwise, any maximum matching of T would miss at least two vertices.

If deg u2 = deg uk−1 = 2, then u1 u2 ∈ M or uk uk−1 ∈ M. Suppose (WLOG) deg u2 = 2 and deg uk−1 > 2.

Then u1 u2 ∈ M. Otherwise, any maximum matching of T would miss at least two vertices.

Next we state the main result of [52].

Theorem 2.12. Any near perfect matching tree is fully cordial.

Proof. Note that T is a tree of odd order, |T | = 2n + 1. We proceed by induction on n. Clearly, the statement

of theorem is true for n = 1. Suppose the statement is true for any tree of order 2n + 1 and let T be a tree

of order 2n + 3 with near-perfect matching M . By Lemma 2.11, T contains vertices u ∼ v ∼ w such that

deg u = 1, deg v = 2 and the edge uv is in M. Now consider the tree S = T − {u, v} which has order 2n + 1
′
and has near perfect matching M = M − {uv}. Therefore, by the induction hypothesis

F I(S) = {0, 2, 4, · · · , 2n}.

We need to show that F I(T ) = {0, 2, · · · , 2n, 2n + 2}. Consider a friendly coloring f : V (S) → Z2 of S

and extend it to g : V (T ) → Z2 by defining g(v) = f (w), g(u) = 1 − f (w). Then g is a friendly coloring of

T with eg (1) = ef (1) + 1, eg (0) = ef (0) + 1. Therefore, N (g) = N (f ). This implies that

F I(S) = {0, 2, 4, · · · , 2n} ⊆ F I(T ).

It only remains to show that 2n + 2 ∈ F I(T ). Let ϕ : V (S) → Z2 be a friendly coloring of S with

index 2n. We may assume that e(1) = 2n, e(0) = 0, and extend ϕ to ψ : V (T ) → Z2 by defining ψ(v) =

1 − ϕ(w), ψ(u) = ϕ(w). Then ψ is a friendly coloring of T with eψ (1) = eϕ (1) + 2, eψ (0) = eϕ (0) = 0.

Therefore, N (ψ) = N (ϕ) + 2 = 2n + 2.

Corollary 2.13. For n ≥ 2, the path of order n is fully cordial.

18
Proof. This is an immediate consequence of Theorems 2.5 and 2.12. Because, any path Pn is either near

perfect matching or is a perfect matching tree. Therefore, it is fully cordial.

By Theorem 2.7 it is shown that any Fibonacci tree is fully cordial [54]. We now present a different proof

of Theorem 2.7 using the concepts of perfect matching and near perfect matching.

Theorem 2.14. For n ≥ 1, every Fibonacci tree F Tn is fully cordial.

Proof. Note that every Fibonacci tree has either a perfect matching or is a near perfect matching tree. In

fact, if n ≡ 1 (mod 3), then F Tn is a near perfect matching tree; otherwise, it has a perfect matching.

We prove this statement by induction on n. Clearly the statement is true for n = 1, 2, 3. Now suppose the

statement is true for all positive integers less than n (3 < n) and let F Tn be the Fibonacci tree of order n.

We consider the following cases:

(A) n ≡ 1 (mod 3). In this case, by the induction hypothesis, both the left and right children have perfect

matchings. Let M1 and M2 be perfect matchings of F Tn−1 and F Tn−2 , respectively. Then M1 ∪ M2 is

a maximum matching of F Tn that covers all the vertices but its root. Therefore, F Tn is a near perfect

matching tree.

(B) n ≡ 2 (mod 3). In this case, by the induction hypothesis, the left child F Tn−1 is near perfect matching

while the right child F Tn−2 has a perfect matching. Let M1 be a maximum matching of F Tn−1

(we may assume that M1 leaves the root rn−1 out) and M2 be a perfect matching of F Tn−2 . Then

M1 ∪ M2 ∪ {rn rn−1 } will form a perfect matching of F Tn .

(C) n ≡ 0 (mod 3). The argument is similar to the previous case.

Definition 2.4. Star-like tree, denoted by ST (n, k), is a graph formed by n copies of Pk when all of them

share exactly an end-vertex. This common end-vertex is clearly the center of the graph.

We observe that ST (1, k) ≃ Pk , ST (2, k) ≃ P2k−1 , ST (n, 2) ≃ K1,n and ST (n, 1) the trivial graph

having just one vertex. The friendly index sets of these graphs have been determined. Therefore, from now

on we assume that n, k ≥ 3.

Corollary 2.15. For any n, k ≥ 3 the star-like tree ST (n, k) is fully cordial if and only if k is odd.

19
Proof. If k is odd, then ST (n, k) is a near perfect matching tree and by Theorem 2.12 it is fully cordial.

When k is even, then e(0) ̸= 0 holds for any friendly coloring of the graph. Hence, by Corollary 2.10, the

maximum possible friendly index cannot be achieved. In fact, when k is even, then e(0) ≥ ⌊(n − 1)/2⌋.

Next we give two definitions a particular type of graph called a caterpillar. We will then give some

notation we use to describe caterpillar. Then we give visual examples of caterpillars.

Definition 2.5. [52][20] A caterpillar is a tree having the property that the removal of its end-vertices results

in a path called the spine.

Definition 2.6. [55] A caterpillar is a tree in which a single path called the spine contains every edge.

We use the notation CR(a1 , a2 , a3 , . . . , an ) to represent a caterpillar with a Pn -spine. We label the ith

vertex of Pn as ui and say deg(ui ) = ai . We will assume that ai ≥ 2. Note that a caterpillar with a Pn -spine

has a diameter of n + 1.

u1 u2 u3 un
...

Figure 2.1: A caterpillar of diameter n + 1 (Pn -spine)[52][20].

Definition 2.7. A double star is a tree of diameter 3. Double stars have two central vertices u and v and

are denoted by DS(a, b), where deg u = a and deg v = b.

Double star DS(a, b) has a + b vertices and its friendly index set is known and stated below.

Theorem 2.16. [53][20] Let a ≤ b. Then


{1, 3, . . . , 2a − 1} if a + b is even;
F I(DS(a, b)) =
{0, 2, . . . , 2a} if a + b is odd.

Corollary 2.17. Double star DS(a, b) is fully cordial if and only if |a − b| ≤ 1.

For a caterpillar of diameter 4, CR(a, b, c) we call the vertices on the spine u, v, and w. We calculate the

number of vertices by |V | = a + b + c − 1 and the number of edges by |E| = a + b + c − 2.

The friendly index set of G = CR(a, b, c), when a + b + c is odd, is determined and stated below

20
 u v w

v1
u2 u1 v2 w1 w2

Figure 2.2: A caterpillar of diameter 4, CR(8, 9, 8)[52][20].

Theorem 2.18. [53][20] Let a, b, c ≥ 2 and a + b + c be odd. Then F I(CR(a, b, c)) = A ∪ B ∪ C, where

A = {| 2a − 4i − 1 | : mA ≤ i ≤ MA } ;
B = {| 2b − 4j − 1 | : mB ≤ j ≤ MB } ;
C = {| 2c − 4k − 1 | : mC ≤ k ≤ MC } ;

and
mA = max 0, a−b−c+3 MA = min a − 1, a+b+c−3
 
2 ; 2 ;
mB = max 0, −a+b−c+1
2 ; M B = min b − 2, a+b+c−3
2 ;
mC = max 0, −a−b+c+3 ; MC = min c − 1, a+b+c−3

2 2 .

Lemma 2.19. Let a + b + c be odd. The caterpillar G = CR(a, b, c) has the maximum possible friendly index

if and only if |b − a − c + 1| ≤ 1.

Proof. By the Corollary 2.10, such a friendly labeling f exists if and only if all edges are labeled 1. Let

f (u) = f (w) = 1 and f (v) = 0. Then all the end-vertices adjacent to u and w are labeled 0, and all

end-vertices adjacent to v are labeled 1. That is, a + c − 1 vertices are labeled 0 and b vertices are labeled

1. But for this labeling to be friendly we require |b − a − c + 1| ≤ 1.

Theorem 2.20. Let a + b + c be odd. Then G = CR(a, b, c) is fully cordial if and only if b = a + c − 1 and

a = 2 or c = 2.

Proof. Suppose G is fully cordial. Then by Lemma 2.19, b = a + c − 1. Also, a = 2 or c = 2. Otherwise,

using the notation of Theorem 2.18, the sets A and C are subsets of B and

F I(G) = B = {| 2b − 4j − 1 | : 0 ≤ j ≤ b − 2}.

However, this set has b − 1 odd numbers; the smallest is 1 and the largest element is 2b − 1. Therefore, one

odd number between 1 and 2b − 1 is missing. In fact, 2b − 3 is not in F I(G).

Conversely, let b = a + c − 1 and a = 2 or c = 2. Without loss of generality, we may assume a = 2. In

this case G = CR(2, c + 1, c). Using Theorem 2.18, one can easily see that F I(G) = {1, 3, . . . , 2c + 1} which

shows that G is fully cordial.

21
 In what follows, we consider the caterpillar G = CR(a, b, c), when a + b + c is even. First we determine

its friendly index set, then we completely identify those that are fully cordial. As mentioned before, G has

a + b + c − 1 vertices and a + b + c − 2 edges.

We observe that any friendly coloring f : G → Z2 that labels the central vertices the same will result

in either index N (f ) = 0 or N (f ) = 2, which are not very interesting. Therefore, we consider the cases in

which the central vertices are labeled differently.

Case 1. Suppose we label the central vertices by f (u) = 0, and f (v) = f (w) = 1 and label all other

vertices by 1 except
f (u1 ) = f (u2 ) = · · · = f (ui ) = 0;
f (v1 ) = f (v2 ) = · · · = f (vj ) = 0; (2.3)
f (w1 ) = f (w2 ) = · · · = f (wk ) = 0.
Then v(0) = i + j + k + 1 and e(1) = a − i + j + k. For this labeling to be friendly we require either

a+b+c
i+j+k+1= , (2.4)
2

or
a+b+c−2
i+j+k+1= , (2.5)
2

Equation (2.4) yields N (f ) = |e(1) − e(0)| = |2a − 4i|. In this situation,

a+b+c a+b+c−2
i+1≤ and a − i + 1 ≤ ,
2 2

which provide the inequalities

a−b−c+4 a+b+c−2
≤i≤ .
2 2

Therefore, the friendly indices obtained in this case would be

A = {|2a − 4i| : mA ≤ i ≤ MA } ,

where
   
a−b−c+4 a+b+c−2
mA = max 0, and MA = min a − 1, .
2 2

Equation (2.5) gives us |e(1) − e(0)| = |2a − 4i − 2|. In this situation,

a+b+c−2 a+b+c
i+1≤ and a − i + 1 ≤ ,
2 2

which provide the inequalities

a−b−c+2 a+b+c−4
≤i≤ .
2 2

22
Therefore, the friendly indices obtained in this situation would be

D = {|2a − 4i − 2| : mD ≤ i ≤ MD } ,

where
   
a−b−c+2 a+b+c−4
mD = max 0, and MD = min a − 1, .
2 2

Case 2. Let f (v) = 0, and f (u) = f (w) = 1 be the labeling of the central vertices and all other vertices

be labeled 1 except for those specified in equation (2.3). In this case, v(0) = i+j +k+1 and e(1) = b−j +i+k.

Again, for this labeling to be friendly we require either (2.4) or (2.5).

The equation (2.4) gives us N (f ) = |e(1) − e(0)| = |2b − 4j|. In this instance,

a+b+c a+b+c−2
j+1≤ and b − j ≤ ,
2 2

which provide the inequalities

b−a−c+2 a+b+c−2
≤j≤ .
2 2

Therefore, the friendly indices obtained in this case would be

B = {|2b − 4j| : mB ≤ j ≤ MB } ,

where
   
b−a−c+2 a+b+c−2
mB = max 0, and MB = min b − 2, .
2 2

The equation (2.5) yields N (f ) = |e(1) − e(0)| = |2b − 4j − 2|. In this instance,

a+b+c−2 a+b+c
j+1≤ and b − j ≤ ,
2 2

which provide the inequalities

b−a−c a+b+c−4
≤j≤ .
2 2

Therefore, the friendly indices obtained in this subcase would be

E = {|2b − 4j − 2| : mE ≤ j ≤ ME } ,

where
   
b−a−c a+b+c−4
mE = max 0, and ME = min b − 2, .
2 2

23
 Case 3. Suppose we label the central vertices by f (w) = 0, f (u) = f (v) = 1 and label all other vertices

by 1 except for those specified in equation (2.3). Then v(0) = i + j + k + 1 and e(1) = b − k + i + j. Again,

for this labeling to be friendly we require either (2.4) or (2.5).

The equation (2.4) gives us N (f ) = |e(1) − e(0)| = |2c − 4k|. In this situation,

a+b+c a+b+c−2
k+1≤ and c − k + 1 ≤ ,
2 2

which provide the inequalities

c−a−b+4 a+b+c−2
≤k≤ .
2 2

Therefore, the friendly indices obtained in this subcase would be

C = {|2c − 4k| : mC ≤ k ≤ MC } ,

where
   
c−a−b+4 a+b+c−2
mC = max 0, and MC = min c − 1, .
2 2

The equation (2.5) gives us N (f ) = |e(1) − e(0)| = |2c − 4k − 2|. In this situation,

a+b+c−2 a+b+c
k+1≤ and c − k + 1 ≤ ,
2 2

which provide the inequalities

c−a−b+2 a+b+c−4
≤k≤ .
2 2

Therefore, the friendly indices obtained in this subcase would be

F = {|2c − 4k| : mF ≤ k ≤ MF } ,

where
   
c−a−b+2 a+b+c−4
mF = max 0, and MF = min c − 1, .
2 2

We summarize the above discussion in the following theorem.

Theorem 2.21. Suppose a + b + c is even and a, b, c ≥ 2. Then

F I(CR(a, b, c)) = A ∪ B ∪ C ∪ D ∪ E ∪ F,

where
A = {|2a − 4i| : mA ≤ i ≤ MA } ; D = {|2a − 4i − 2| : mD ≤ i ≤ MD } ;
B = {|2b − 4j| : mB ≤ j ≤ MB } ; E = {|2b − 4j − 2| : mE ≤ j ≤ ME } ;
C = {|2c − 4k| : mC ≤ k ≤ MC } ; F = {|2c − 4k − 2| : mF ≤ k ≤ MF } ;

24
Lemma 2.22. Let a + b + c be even. Then the caterpillar G = CR(a, b, c) has a maximum friendly index if

and only if |b − a − c + 1| = 1.

Proof. By Corollary 2.10, such a friendly labeling f exists if and only if all edges are labeled 1. Without loss

of generality we may assume that f (u) = f (w) = 0 and f (v) = 1. Then all the end-vertices adjacent to u

and w are labeled 1, and all end-vertices adjacent to v are labeled 0. That is, v(0) = b. However this labeling

is friendly if and only if either 2b = a + b + c or 2b = a + b + c − 2 which proves the lemma.

Lemma 2.23. Let |b − a − c + 1| = 1. Then F I(CR(a, b, c)) = ∆ ∪ Ω where

∆ = |2b − 4j| : 1 ≤ j ≤ ⌊ 2b ⌋ ; Ω = |2b − 4j − 2| : 0 ≤ j ≤ ⌊ b−1

 
2 ⌋ .

Proof. We utilize Theorem 2.21 and note that

A = {|2a − 4i| : 0 ≤ i ≤ a − 1} ; D = {|2a − 4i − 2| : 0 ≤ i ≤ a − 1} ;

B = {|2b − 4j| : 1 ≤ j ≤ b − 2} ; E = {|2b − 4j − 2| : 0 ≤ j ≤ b − 2} ;
C = {|2c − 4k| : 0 ≤ k ≤ c − 1} ; F = {|2c − 4k − 2| : 0 ≤ k ≤ c − 1} .

Since A ∪ D, C ∪ F ⊂ B ∪ E then by 2.21, F I(G) = B ∪ E. We also observe that |2b − 4j| produces the same

number for j and b − j. Similarly, |2b − 4k − 2| produces the same number for k and b − k + 1. Therefore,

B = ∆ and E = Ω.

Theorem 2.24. Let a + b + c be even. Then G = CR(a, b, c) is fully cordial if and only if |b − a − c + 1| = 1.

Proof. Suppose G is fully cordial. Then G achieves its maximum friendly index and by Lemma 2.22, |b −

a − c + 1| = 1.

Conversely, let |b − a − c + 1| = 1. Then by Lemma 2.23, F I(G) = ∆ ∪ Ω. Also, we observe that G has

either 2b − 2 or 2b edges and the set ∆ ∪ Ω generates exactly either {0, 2, 4, . . . , 2b − 2} or {0, 2, 4, . . . , 2b}.

These numbers are the full spectrum of friendly indices of G.

2.4

Jost Graph Laplacian Applied to Friendly Labelings on Various Finite Graphs

I was once asked by Dr. Ebrahim Salehi what are if and only if conditions for a tree to be fully cordial. It

was also recently brought to my attention by Dr. Pushkin Kachroo that there are potential applications of the

concepts of friendly index sets of finite graphs and the concept of fully cordial graphs to network theory and

the study of social media. In light of these ideas, a research goal is to compute graph Laplacians of friendly

25
labelings of certain graphs with various types of edge labelings. In particular we focus on the Jost Graph

Laplacian of friendly labelings on various finite graphs. The goal will be to discover what combinatorial

properties continually arise and see if these new properties can be applied to partial differential equation

theory on graphs, the study of network theory, the study of fully cordial graphs, or graph theory in general.

We will recall the definition of the weighted graph Laplacian operators given earlier in this manuscript.

Definition 2.8. [27](Degree Laplace Operator) Let (V, E) be a locally finite connected graph. For any

function f : V → R, define the Laplace operator, ∆f , by

1 X
∆deg(x) f (x) = f (y) − f (x). (2.6)
deg(x) y∼x

The Laplace operator studied by Jost is L = −∆f .

Definition 2.9. [27](Weighted Laplace Operator) Let G = (V, E) be a connected, locally finite graph. Suppose

that G has a weight µ. For any function f : V → R, define the weighted µ-Laplace operator, ∆µ f , by

1 X
∆µ f (x) = f (y)µxy − f (x). (2.7)
µ(x) y∼x

The normalized graph Laplace operator studied by Grigoryan is Lµ = −∆µ f .

We first compute the graph Laplacian studied by Jost for various friendly labelings for various finite

graphs. This means we are setting µxy = 1 ∀{x, y} ∈ E(G). We first do such a computation using the

complete graph G = Kn .

Definition 2.10. A graph G is complete if every two distinct vertices are joined by an edge (adjacent).

n(n−1)
Note that a complete graph of order n is denoted by Kn and has 2 edges.

K1 K2 K3 K4

Figure 2.3: The complete graphs K1 , K2 ,K3 , K4 [20].

Theorem 2.25. Let f : Kn → Z2 be a friendly labeling. Then

 n o
n n
 − , , if n is even
 n 2(n−1) 2(n−1)

 o
1
∆f (Kn ) = , − n+1 , if n is odd, vf (0) = n+1
and vf (1) = n−1

 n 2 2(n−1) o 2 2
 − , 1 n+1 n−1 n+1
2 2(n−1) , if n is odd, vf (0) = and vf (1) = 2 .

2

26
 Consequently,
 n o
n n
 − , , if n is even
 n 2(n−1) 2(n−1)

 o
1 n+1 n+1 n−1
Lf (Kn ) = − , , if n is odd, vf (0) = and vf (1) =

 n 2 2(n−1) o 2 2
 1 , − n+1 , if n is odd, vf (0) = n−1
and vf (1) = n+1
2 .

2 2(n−1) 2

Proof. Suppose that f : Kn → Z2 be a friendly labeling. Let x ∈ V (Kn ). We have a three cases to consider

each with two subcases depending on if f (x) = 0 or f (x) = 1.

n n
Case 1: Suppose that n is even. This means vf (0) = 2 and vf (1) = 2.

Subcase 1: Let f (x) = 0. Then,

1 X 1  n  n n
∆f (x) = f (y) − f (x) = 0· −1 +1· −0= . (2.8)
deg(x) y∼x n−1 2 2 2(n − 1)
n
Thus, Lf (x) = − 2(n−1) .

Subcase 2: Let f (x) = 1. Then,

1 X 1  n n  n
∆f (x) = f (y) − f (x) = 0· +1· −1 −1=− . (2.9)
deg(x) y∼x n−1 2 2 2(n − 1)
n
Thus, Lf (x) = 2(n−1) .

n+1 n−1
Case 2: Suppose that n is odd, vf (0) = 2 and vf (1) = 2 .

Subcase 1: Let f (x) = 0. Then,

   
1 X 1 n+1 n−1 1
∆f (x) = f (y) − f (x) = 0· −1 +1· −0= . (2.10)
deg(x) y∼x n−1 2 2 2

Thus, Lf (x) = − 21 .

Subcase 2: Let f (x) = 1. Then,

    
1 X 1 n+1 n−1 n+1
∆f (x) = f (y) − f (x) = 0· +1· −1 −1=− . (2.11)
deg(x) y∼x n−1 2 2 2(n − 1)

n+1
Thus, Lf (x) = 2(n−1) .

n−1 n+1
Case 3: Suppose that n is odd, vf (0) = 2 and vf (1) = 2 .

Subcase 1: Let f (x) = 0. Then,

   
1 X 1 n−1 n+1 n+1
∆f (x) = f (y) − f (x) = 0· −1 +1· −0= . (2.12)
deg(x) y∼x n−1 2 2 2(n − 1)

n+1
Thus, Lf (x) = − 2(n−1) .

27
Subcase 2: Let f (x) = 1. Then,
  
1 X 1 n−1 n+1 1
∆f (x) = f (y) − f (x) = 0· +1· −1 −1=− . (2.13)
deg(x) y∼x n−1 2 2 2

Thus, Lf (x) = 12 .

In view of Theorem 1.14 part (b) concerning the eigenvalues of a complete graph with simple weight µ

the previous theorem is stated as the following corollary.

Corollary 2.26. Let f : Kn → Z2 be a friendly labeling. Suppose that λ is an eigenvalue of Laplace operator

∆ over Kn . Then
  1 1

 − 2 λ, 2 λ , if n is even
 n (n−1) o
∆f (Kn ) = 2n λ, − (n+1)
2n λ , if n is odd, vf (0) = n+1
2 and vf (1) = n−1
2
n o
 − (n−1) λ, (n+1) λ ,
 n−1 n+1
if n is odd, vf (0) = and vf (1) = 2 .

2n 2n 2

Consequently,
 1
λ, − 12 λ , if n is even
 n 2 (n−1)

 o
Lf (Kn ) = − 2n λ, (n+1) 2n λ , if n is odd, vf (0) = n+1
2 and vf (1) = n−1
2
n o
 (n−1)
 (n+1) n−1 n+1

2n λ, − 2n λ , if n is odd, vf (0) = 2 and vf (1) = 2 .

Proof. From Theorem 1.14 part (b) we recall that for the complete graph Kn with a simple weight µ, that
n
the Laplace operator ∆ has eigenvalues λ1 = λ2 = · · · = λn−1 = n−1 . So for any given eigenvalue λ of the
n
Laplace operator ∆: λ = n−1 . Comparing this with the previous theorem yields the corollary.

Note Corollary 2.26 shows when Kn has an even number of vertices, then the Jost graph Laplacian of

any friendly labeling is a constant multiple of the eigenvalue of the Jost graph Laplacian operator on Kn .

Now we recall the definitions of bipartite and complete bipartite finite graphs in order to compute the

graph laplacian. Then we give an example complete bipartite graph.

Definition 2.11. [55] A graph G is called bipartite if V (G) can be partitioned into two disjoint subsets S

and T , called partite sets, such that every edge of G joins a vertex of S and a vertex of T . A graph that is

both complete and bipartite is called a complete bipartite graph. A complete bipartite graph such that |S| = m

and |T | = n is denoted by K(m, n).

In the following theorem we present sets that have the range of the Jost graph Laplace operator on a

friendly labeling of a complete bipartite graph as a subset.

28
 Figure 2.4: Two complete bipartite graphs K(2, 3) is on the left and K(3, 3) is on the right [20].

Theorem 2.27. Let m ≤ n. Let f : K(m, n) → Z2 be a friendly labeling. Then

i m−j j
  n−i n−m
n , −n, m , −m : 2 ≤ i ≤ m+n
2 , 0 ≤ j ≤ m, if m + n is even



 m+n
and i + j = ,




 2

i m−j j
 n−i n−m+1
n , −n, m , −m : ≤ i ≤ m+n+1 , 0 ≤ j ≤ m, if m + n is odd and


 2 2
∆f (K(m, n)) ⊆ m+n+1

 and i + j = 2 , vf (0) = vf (1) + 1

 
n−i i m−j j n−m−1
n , −n, m , −m : ≤ i ≤ m+n−1 , 0 ≤ j ≤ m, if m + n is odd and


2 2




m+n−1
and i + j = , vf (1) = vf (0) + 1.


 2

Consequently,
− n , n , − m−j j
  n−i i n−m
m , m : 2 ≤ i ≤ m+n
2 , 0 ≤ j ≤ m, if m + n is even



 m+n
and i + j = 2 ,






 − n , n , − m−j j
 n−i i n−m+1
m , m : ≤ i ≤ m+n+1 , 0 ≤ j ≤ m, if m + n is odd and


2 2
Lf (K(m, n)) ⊆ m+n+1

 and i + j = ,
2 vf (0) = vf (1) + 1

 m−j j
− n−i i n−m−1
≤ i ≤ m+n−1

n , n, − m , m : , 0 ≤ j ≤ m, if m + n is odd and


2 2




m+n−1
and i + j = , vf (1) = vf (0) + 1.


 2

Proof. Let m ≤ n. Suppose that f : K(m, n) → Z2 be a friendly labeling. Let i be the number of vertices in

K(m, n) of degree m such that f (x) = 0 and j be the number of vertices in K(m, n) of degree n such that

f (x) = 0. Let x ∈ V (K(m, n)).

Regardless of the parity of m + n we have the following 4 cases for computing the Laplace operator at a

vertex x:

Case 1: Let deg(x) = n and f (x) = 0. Then

1 X 1 n−i
∆f (x) = f (y) − f (x) = (0 · i + 1 · (n − i)) − 0 = . (2.14)
deg(x) y∼x n n

Thus, Lf (x) = − n−i

n .

Case 2: Let deg(x) = n and f (x) = 1. Then

1 X 1 i
∆f (x) = f (y) − f (x) = (0 · i + 1 · (n − i)) − 1 = − . (2.15)
deg(x) y∼x n n
i
Thus, Lf (x) = n.

29
Case 3: Let deg(x) = m and f (x) = 0. Then

1 X 1 m−j
∆f (x) = f (y) − f (x) = (0 · j + 1 · (m − j)) − 0 = . (2.16)
deg(x) y∼x m m

Thus, Lf (x) = − m−j

m .

Case 4: Let deg(x) = m and f (x) = 1. Then

1 X 1 j
∆f (x) = f (y) − f (x) = (0 · i + 1 · (m − j)) − 1 = − . (2.17)
deg(x) y∼x m m
j
Thus, Lf (x) = m.

Now we need to determine which values of i, j are allowed under the friendly labeling f on K(m, n). We

have a three cases to consider.

m+n m+n
Case 1: Suppose that m + n is even. Here vf (0) = 2 and vf (1) = 2 . Note that, 0 ≤ i ≤ n and
n−m m+n n−m
2 ≤i≤ 2 . Hence, 2 = max{0, n−m m+n
2 } ≤ i ≤ min{n, 2 } =
m+n
2 . Observe that, 0 ≤ j ≤ m and
m−n m+n
2 ≤j≤ 2 . Thus, 0 = max{0, m−n m+n
2 } ≤ j ≤ min{m, 2 } = m. Also since for a friendly labeling we

m+n
have a given max number of zeros overall we see i + j = vf (0) = 2 .

m+n+1 m+n−1
Case 2: Suppose that m + n is odd, vf (0) = 2 and vf (1) = 2 . Note that, 0 ≤ i ≤ n and
n−m+1 m+n+1 n−m+1
2 ≤i≤ 2 . Hence, 2 = max{0, n−m+1
2 } ≤ i ≤ min{n, m+n+1
2 }= m+n+1
2 (as m < n implies
m−n+1 m+n+1
n − m ≥ 1 and this in turn implies n ≥ m + 1). Observe that, 0 ≤ j ≤ m and 2 ≤j≤ 2 . Thus,

0 = max{0, m−n+1
2 } ≤ j ≤ min{m, m+n+1
2 } = m. Also since for a friendly labeling we have a given max
m+n+1
number of zeros overall we see i + j = vf (0) = 2 .

m+n−1 m+n+1
Case 3: Suppose that m + n is odd, vf (0) = 2 and vf (1) = 2 . Note that, 0 ≤ i ≤ n and
n−m−1 m+n−1 n−m−1
2 ≤i≤ 2 . Hence, 2 = max{0, n−m−1
2 } ≤ i ≤ min{n, m+n−1
2 }= m+n+1
2 (as m < n implies
m−n−1 m+n−1
n − m ≥ 1 and this in turn implies n ≥ m + 1). Observe that, 0 ≤ j ≤ m and 2 ≤ j ≤ 2 .

Thus, 0 = max{0, m−n−1

2 } ≤ j ≤ min{m, m+n−1
2 } = m (as n − m ≥ 1 implies m − n ≤ −1 which in turn

implies m ≤ n − 1). Also since for a friendly labeling we have a given max number of zeros overall we see
m+n−1
i + j = vf (0) = 2 .

30
 It is important to note in Theorem 2.27 we get ”∆f (K(m, n)) ⊆” because depending on the labeling of

the graph some of the cases in the proof do not occur. To be more explicit you could have a graph with

no vertices x such that deg(x) = n and f (x) = 0 or you could have a graph with no vertices x such that

deg(x) = n and f (x) = 1. For example, given K(2, 4) there is a friendly labeling where both vertices of

degree 4 are labeled 1. This means for this graph there are no vertices x such that deg(x) = 4 and f (x) = 0.

Next we have way to calculate the Jost Graph Laplace operator for a vertex of a given friendly labeling

on a complete bipartite graph. Note this is just a way to shortening the calculation so we do not have to
1
P
find deg(x) f (y) − f (x) for each vertex.
y∼x

Corollary 2.28. Let m ≤ n. Let x ∈ V (K(m, n)). Let f : K(m, n) → Z2 be a friendly labeling with i

vertices, xi , of degree n such that f (xi ) = 0 and j, xj , vertices of degree m such that f (xi ) = 0.Then
 n−i
, when deg(x) = n and f (x) = 0
 −ni , when deg(x) = n and f (x) = 1


∆f (x) = n
m−j

 m , when deg(x) = m and f (x) = 0
j
−m , when deg(x) = m and f (x) = 1.


Proof. These cases follow directly from the proof of Theorem 2.27.

Here we introduce a special type of complete bipartite graph called a star and then display an example

of a star. We denote a star with n edges by ST (n) and has n + 1 vertices. It is the complete bipartite graph

K(1, n).

Definition 2.12. A star is a graph where every exterior vertex is connected by an edge to a single common

point called the center.

Figure 2.5: The star with 7 edges [20].

Theorem 2.29. Let f : ST (n) → Z2 be a friendly labeling for star with n edges. Let x, c ∈ V (ST (n)),

31
where c is the center vertex. Then

if x is an exterior vertex, f (x) = 0, and f (c) = 0 or

 0,


 if x is an exterior vertex, f (x) = 1, and f (c) = 1


1, if x is an exterior vertex, f (x) = 0, and f (c) = 1






−1, if x is an exterior vertex, f (x) = 1, and f (c) = 0





 1
2, if x = c, f (c) = 0, n is even, and vf (0) = vf (1) + 1




∆f (x) = n+2

 2n , if x = c, f (c) = 0, n is even, and vf (1) = vf (0) + 1
n+1

2n , if x = c, f (c) = 0, and n is odd





− n+2

2n , if x = c, f (c) = 1, n is even, and vf (0) = vf (1) + 1





− 21 ,




 if x = c, f (c) = 1, n is even, and vf (1) = vf (0) + 1

− n+1

2n , if x = c, f (c) = 1, and n is odd.



Proof. Let f : ST (n) → Z2 be a friendly labeling for star with n edges. Let x, c ∈ V (ST (n)), where c is

the center vertex. We break this proof into to cases. The first case is when f (c) = 0 ad the second is when

f (c) = 1.

Case 1: Let f (c) = 0.

Subcase 1: Let x be an exterior vertex such that f (x) = 0. Note that for this calculation it does not matter

if n is odd or even. Then

1 X 1
∆f (x) = f (y) − f (x) = (0) − 0 = 0.
deg x y∼x 1

Subcase 2: Let x be an exterior vertex such that f (x) = 1. Note that for this calculation it does not matter

if n is odd or even. Then

1 X 1
∆f (x) = f (y) − f (x) = (0) − 1 = −1.
deg x y∼x 1

Subcase 3: Let f (c) = 0, n be even, and vf (0) = vf (1) + 1. Since n is even and vf (0) = vf (1) + 1, then
n n
vf (0) = 2 + 1 and vf (1) = 2 because f is a friendly labeling. Then

1 X 1  n  n  1
∆f (c) = f (y) − f (c) = 0 +1 −0= .
deg c y∼c n 2 2 2

Subcase 4: Let f (c) = 0, n be even, and vf (1) = vf (0) + 1. Since n is even and vf (1) = vf (0) + 1, then
n n
vf (0) = 2 and vf (1) = 2 + 1 because f is a friendly labeling. Then

1 X 1  n  n  n+2
∆f (c) = f (y) − f (c) = 0 −1 +1 +1 −0= .
deg c y∼c n 2 2 2n

32
 n+1
Subcase 5: Let f (c) = 0 and n be odd. Since n is odd, then vf (0) = vf (1) = 2 because f is a friendly

labeling. Then
    
1 X 1 n+1 n+1 n+1
∆f (c) = f (y) − f (c) = 0 −1 +1 −0= .
deg c y∼c n 2 2 2n

Case 2: Let f (c) = 1.

Subcase 1: Let x be an exterior vertex such that f (x) = 0. Note that for this calculation it does not matter

if n is odd or even. Then

1 X 1
∆f (x) = f (y) − f (x) = (1) − 0 = 1.
deg x y∼x 1

Subcase 2: Let x be an exterior vertex such that f (x) = 1. Note that for this calculation it does not matter

if n is odd or even. Then

1 X 1
∆f (x) = f (y) − f (x) = (1) − 1 = 0.
deg x y∼x 1

Subcase 3: Let f (c) = 1, n be even, and vf (0) = vf (1) + 1. Since n is even and vf (0) = vf (1) + 1, then
n n
vf (0) = 2 + 1 and vf (1) = 2 because f is a friendly labeling. Then

1 X 1  n  n  n+2
∆f (c) = f (y) − f (c) = 0 +1 +1 −1 −1=− .
deg c y∼c n 2 2 2n

Subcase 4: Let f (c) = 1, n be even, and vf (1) = vf (0) + 1. Since n is even and vf (1) = vf (0) + 1, then
n n
vf (0) = 2 and vf (1) = 2 + 1 because f is a friendly labeling. Then

1 X 1  n  n  1
∆f (c) = f (y) − f (c) = 0 +1 −1=− .
deg c y∼c n 2 2 2

n+1
Subcase 5: Let f (c) = 1 and n be odd. Since n is odd, then vf (0) = vf (1) = 2 because f is a friendly

labeling. Then
    
1 X 1 n+1 n+1 n+1
∆f (c) = f (y) − f (c) = 0 +1 −1 −1=− .
deg c y∼c n 2 2 2n

Corollary 2.30. Let f : ST (n) → Z2 be a friendly labeling for star with n edges. Let x, c ∈ V (ST (n)),

33
where c is the center vertex. Then

if x is an exterior vertex, f (x) = 0, and f (c) = 0 or

 0,


 if x is an exterior vertex, f (x) = 1, and f (c) = 1


−1, if x is an exterior vertex, f (x) = 0, and f (c) = 1






1, if x is an exterior vertex, f (x) = 1, and f (c) = 0






− 21 , if x = c, f (c) = 0, n is even, and vf (0) = vf (1) + 1




Lf (x) = − n+2

 2n , if x = c, f (c) = 0, n is even, and vf (1) = vf (0) + 1
− n+1

2n , if x = c, f (c) = 0, and n is odd





n+2

2n , if x = c, f (c) = 0, n is even, and vf (0) = vf (1) + 1





1

2, if x = c, f (c) = 1, n is even, and vf (1) = vf (0) + 1





n+1

2n , if x = c, f (c) = 1, and n is odd.



Proof. Since Lf (x) = −∆f (x) we get the desired result.

Theorem 2.31. Let f : ST (n) → Z2 be a friendly labeling for star with n edges. Let i be the number of

zeros of degree m as in the proof of Theorem 2.27. Then i = 0 if and only if n = 2, f (c) = 0 and both the

exterior vertices are all labeled by a 1 i.e. it is the graph ST (2) = K(1, 2).

Proof. ⇐= Consider the graph ST (2) = K(1, 2) with the center labeled by a zero and both the exterior

vertices are all labeled by a 1. In this case observe that i = 0.

=⇒ Conversely let i = 0, vf (1) = n and vf (0) = n + 1 − n = 1. Then |vf (1) − vf (0)| = n − 1. Since

f : ST (n) → Z2 is a friendly labeling, n − 1 ≤ 1 =⇒ n ≤ 2. Note if n = 1 we have the path P1 = K2 and

for f to be friendly means i = 1 ̸= 0. So n = 2.

Reall for a caterpillar of diameter 4, CR(a, b, c) we call the vertices on the spine u, v, and w. See figure

2.6 for an example. We calculate the number of vertices by |V | = a + b + c − 1 and the number of edges by

|E| = a + b + c − 2.

u v w

v1
u2 u1 v2 w1 w2

Figure 2.6: A caterpillar of diameter 4, CR(8, 9, 8)[52][20].

We now compute the Jost graph Laplacian for all the vertices of a caterpillar of diameter 4, CR(a, b, c)

with a specific labeling applied to the vertices on the spine.

34
Theorem 2.32. Let f : CR(a, b, c) → Z2 be a friendly labeling for a caterpillar of diameter 4. Let x, u, v, w ∈

V (CR(a, b, c)) where u is the vertex of degree a, v is the vertex of degree b, and w is the vertex of degree c.

Let f (u) = 1, f (v) = 0, and f (w) = 1. Let i be the number of vertices adjacent to u such that f (x) = 0, j be

the number of vertices adjacent to v such that f (x) = 0, and k be the number of vertices adjacent to w such

that f (x) = 0. Then


if x is an exterior vertex of u or w and f (x) = 1 or
0,


if x is an exterior vertex of v and f (x) = 0








 1, if x is an exterior vertex of u or w and f (x) = 0

∆f (x) = −1, if x is an exterior vertex of v and f (x) = 1
a−i
a − 1, if x = u




b−j

b , if x = v




 c−k
c − 1, if x = w.



Proof. Let f : CR(a, b, c) → Z2 be a friendly labeling for a caterpillar of diameter 4. Let x, u, v, w ∈

V (CR(a, b, c)) where u is the vertex of degree a, v is the vertex of degree b, and w is the vertex of degree c.

Let f (u) = 1, f (v) = 0, and f (w) = 1. Let i be the number of vertices adjacent to u such that f (x) = 0, j

be the number of vertices adjacent to v such that f (x) = 0, and k be the number of vertices adjacent to w

such that f (x) = 0.

Case 1: Let x be an exterior vertex of u or w such that f (x) = 0. Then

1 X 1
∆f (x) = f (y) − f (x) = (1) − 0 = 1.
deg x y∼x 1

Case 2: Let x be an exterior vertex of u or w such that f (x) = 1. Then

1 X 1
∆f (x) = f (y) − f (x) = (1) − 1 = 0.
deg x y∼x 1

Case 3: Let x be an exterior vertex of v such that f (x) = 0. Then

1 X 1
∆f (x) = f (y) − f (x) = (0) − 0 = 0.
deg x y∼x 1

Case 4: Let x be an exterior vertex of v such that f (x) = 1. Then

1 X 1
∆f (x) = f (y) − f (x) = (0) − 1 = −1.
deg x y∼x 1

Case 5: Let x be the vertex u. Then

1 X 1 a−i
∆f (u) = f (y) − f (u) = (i · 0 + (a − i)1) − 1 = − 1.
deg u y∼u a a

35
Case 6: Let x be the vertex v. Then

1 X 1 b−j
∆f (v) = f (y) − f (v) = (j · 0 + (b − j)1) − 0 = .
deg v y∼v b b

Case 7: Let x be the vertex w. Then

1 X 1 c−k
∆f (w) = f (y) − f (w) = (k · 0 + (c − k)1) − 1 = − 1.
deg w y∼w c c

Corollary 2.33. Let f : CR(a, b, c) → Z2 be a friendly labeling for a caterpillar of diameter 4. Let

x, u, v, w ∈ V (CR(a, b, c)) where u is the vertex of degree a, v is the vertex of degree b, and w is the vertex

of degree c. Let f (u) = 1, f (v) = 0, and f (w) = 1. Let i be the number of vertices adjacent to u such that

f (x) = 0, j be the number of vertices adjacent to v such that f (x) = 0, and k be the number of vertices

adjacent to w such that f (x) = 0. Then


if x is an exterior vertex of u or w and f (x) = 1 or

 0,



 if x is an exterior vertex of v and f (x) = 0





 −1, if x is an exterior vertex of u or w and f (x) = 0


 1, if x is an exterior vertex of v and f (x) = 1
Lf (x) =
a−i


 − a −1 , if x = u



  

b−j
− , if x = v


b




 − c−k


c −1 ,

 if x = w.

Proof. Since Lf (x) = −∆f (x) we get the desired result.

In Theorem 2.32 we only calculated the graph Laplacian for a friendly labeling when spine had the

labeling f (u) = f (w) = 1 and f (v) = 0. Note that there are 7 more possible spine labelings for a caterpillar

of diameter 4. Theses cases are

(a) f (u) = f (v) = f (w) = 1,

(b) f (u) = f (v) = f (w) = 0,

(c) f (u) = f (v) = 0 and f (w) = 1,

(d) f (u) = f (w) = 0 and f (v) = 1,

(e) f (v) = f (w) = 0 and f (w) = 1,

(f) f (u) = f (v) = 1 and f (w) = 0, and

36
 (g) f (v) = f (w) = 1 and f (u) = 0.

We can preform similar calculations for each case. We leave these calculations for later work outside of this

dissertation.

2.5

Weighted Laplacian-Like Operator Applied to Friendly Labelings on Various

Finite Graphs

Now we consider a graph operator on friendly labelings of finite graphs that is graph Laplacian like. The

following original result I proved below supposed the weighted edge labeling arises from adding the vertex

labelings in Z2 , and allowing for µxy = 0 for some edges. Of course in such problems one loses the classical

notion of the graph Laplacian identifying specific edges between given vertices. It should be noted that the

only type of finite graph that I have encountered where this operator is well defined at every vertex is the

complete graph Kn . Indeed it is easy for one to generate examples of finite graphs G (complete bipartite

graphs, caterpillars of diameter 3, and caterpillars diameter 4) where µ(x) = 0 at some of the vertices

x ∈ V (G). Let us now give a more formal definition for this operator.

Definition 2.13. (Graph Weight Allowing For µxy = 0 For Some Edges) Here we consider a couple (Γ, µ)

where Γ = (V, E) is a graph with vertex set V and edge set E and µxy is a non-negative function on V × V

such that

1. µxy = µyx ,

2. µxy ≥ 0,

P
3. µ(x) = y∼x µxy ,

4. µ(x) ̸= 0 (i.e. there exists adjacent vertices u, v in the graph such that uuv ̸= 0)

where, µ is called the weight (or measure) of the graph Γ.

Definition 2.14. (Weighted Laplacian-Like Operator) Let G = (V, E) be a connected, locally finite graph.

Suppose that G has a generalized weight µ. Let A be an Abelian group. For any function f : V → A, define

the weighted Laplace-Like (LL) operator, ∆LL:µ f , by

37
 1 X
∆LL:µ f (x) = f (y)µxy − f (x). (2.18)
µ(x) y∼x

The normalized graph Laplace-Like (LL) operator is LLL:µ = −∆LL:µ f .

It should be noted that one loses the classical notion of the graph Laplacian identifying specific edges

between given vertices when one chooses to study weighted Laplacian-Like Operators.

Now we compute the weighted graph Laplace-Like (LL) of a friendly labeling of Kn where the weighted

measure is determined by an induced edge labeling is achieved by addition of the vertex labelings of a given

edge in Z2 and we allow for µxy = 0 for some edges.

Theorem 2.34. Let f : Kn → Z2 be a friendly labeling. Then ∆LL:µ f (Kn ) = {−1, 1}. Consequently,

LLL:µ f (Kn ) = {−1, 1}.

Proof. Suppose that f : Kn → Z2 be a friendly labeling. Let x ∈ V (Kn ). We have a couple of cases to

consider.
n n
Case A: Suppose that n is even. Here vf (0) = 2 and vf (1) = 2.

Subcase A1: Let f (x) = 0.

n n n
Note that x connects to 2 − 1 other ”zero” vertices and”one” vertices yielding
2 2 − 1 induced labeled
n
µxy = n2 . Hence,
P
”zero” edges and 2 induced labeled ”one” edges. Then µ(x) =
y∼x

1 X 2n
∆LL:µ f (x) = f (y)µxy − f (x) = − 0 = 1. (2.19)
µ(x) y∼x n2

Thus, LLL:µ f (x) = −1.

Subcase A2: Let f (x) = 1.

n n n
Note that x connects to 2 − 1 other ”one” vertices and”zero” vertices yielding
2 2 − 1 induced labeled
n
µxy = n2 . Hence,
P
”zero” edges and 2 induced labeled ”one” edges. Then µ(x) =
y∼x

1 X 2
∆LL:µ f (x) = f (y)µxy − f (x) = (0) − 1 = −1. (2.20)
µ(x) y∼x n

Thus, LLL:µ f (x) = 1.

Case B: Suppose that n is odd.

n n+2 n
Subcase B1: Let f (x) = 0. Without loss of generality, vf (0) = 2 +1= 2 and vf (1) = 2.

38
 n n n
Note that x connects to 2 other ”zero” vertices and
”one” vertices yielding
2 2 induced labeled ”zero”
n
µxy = n2 . Hence,
P
edges and 2 induced labeled ”one” edges. Then µ(x) =
y∼x

1 X 2n
∆LL:µ f (x) = f (y)µxy − f (x) = − 0 = 1. (2.21)
µ(x) y∼x n2

Thus, LLL:µ f (x) = −1.

Subcase B2: Let f (x) = 1.

n n−2 n+2
Note that x connects to 2 −1 = 2 other ”one” vertices and ”zero” vertices yielding n−2
2 2 induced

n+2
µxy = n+2
P
labeled ”zero” edges and 2 induced labeled ”one” edges. Then µ(x) = 2 . Hence,
y∼x

1 X 2
∆LL:µ f (x) = f (y)µxy − f (x) = (0) − 1 = −1. (2.22)
µ(x) y∼x n+2

Thus, LLL:µ f (x) = 1.

2.6

Future Research Objectives Dealing with Graph Laplacians of Friendly

Labelings

To my knowledge it remains an open problem to determine what are if and only if conditions for a tree

to be fully cordial. We now state a known theorem from the literature that is useful in the study of fully

cordial trees.

Theorem 2.35. [45] A graph G has q as a friendly index if and only if G is isomorphic to a spanning

subgraph of K(m, m) or K(m, m + 1).

Perhaps Theorem 2.27 and Corollary 2.28 can be used to study the possible Jost graph Laplacians of

friendly labelings of fully cordial trees.

Another tool that may be of aid in the search for this fully cordial tree condition may be the weighted

laplacian-like operator considered in section 2.5 of this chapter. The reason we were studying this operator

is that it would allow one to have a ”Laplaicain-like” operator that would allow for µ(x) created by induced

edge labelings caused by adding vertex labelings modulo 2 when studying friendly index sets for some graphs.

In the course of searching for fully cordial caterpillars of diameter 4 in [52], finding an if and only if

condition for a fully cordial caterpillar of diameter 4 to achieve its highest theoretical index q was needed.

We state both of these lemmas now.

39
Lemma 2.36. [52] Let a+b+c be odd. Then the caterpillar G=CR(a,b,c) has the maximum possible friendly

index if and only if |b − a − c + 1| ≤ 1.

Lemma 2.37. [52] Let a + b + c be even. Then the caterpillar G=CR(a,b,c) has the maximum possible

friendly index if and only if |b − a − c + 1| = 1.

These two lemmas were proven in [52] by forming a friendly labeling which produced the maximum

possible index q in these two cases. These lemmas and their proofs in [52] were presented at the 28th

Midwest Conference on Combinatorics and Combinatorial Computing at University of Nevada, Las Vegas

on October 22-24, 2014 in the talk “Fully Cordial Trees.” It is joint work between Dr. Sahlehi and I. It

should be noted that these two lemmas can now be more efficiently achieved by using [45] and Theorem 2.35.

Recall earlier we mentioned Theorems 2.20 and 2.24 from [52]. We restate these two theorems next for

the convenience of the reader.

Theorem 2.38. [52] Let a + b + c be odd. The caterpillar G = CR(a, b, c) is fully cordial if and only if

b = a + c − 1 and a = 2 or c = 2.

Theorem 2.39. [52] Let a + b + c be even. The caterpillar G = CR(a, b, c) is fully cordial if and only if

|b − a − c + 1| = 1.

The interesting thing in the situation of the caterpillar G = CR(a, b, c) of diameter 4 is that there are

way more instances of being fully cordial in the a + b + c even case than in the a + b + c odd case. What

happens in the a + b + c even case is that when we get the highest possible friendly index q we also get the

second highest possible friendly index q − 2. On the other hand, in the a + b + c odd case, there are lots

of situations where you get the highest possible friendly index q but fail to get the second highest possible

friendly index q − 2. This makes one wonder what caused the second highest possible friendly index q − 2

to be missed in the a + b + c, |b − a − c + 1| ≤ 1 case. Or in other words, what is so special in the a + b + c

even,|b − a − c + 1| = 1 case that allowed us to get the second highest possible friendly index q − 2. A partial

answer to this question is that in the a + b + c even,|b − a − c + 1| = 1 case the caterpillar G = CR(a, b, c)

has an odd number of vertices. The next theorem, which we believe to be original, takes a first step at this

partial answer.

Theorem 2.40. Suppose that G(V, E) is a connected finite graph having at least one leaf that has terminal

vertex w. Let |V (G)| = n be odd and |E(G)| = q ≥ 2. If q ∈ F I(G), then q − 2 ∈ F I(G).

40
Proof. Suppose that G(V, E) is a connected finite graph having at least one leaf that has terminal vertex w.

Let |V (G)| = n be odd and |E(G)| = q ≥ 2. Suppose further that q ∈ F I(G).

Since n is odd then n = 2k + 1 for some positive integer k.

Note that q ∈ F I(G) and Corollary 2.10 implies that there exists a friendly labeling f : V (G) → Z2 such

that ef (1) = q and ef (0) = 0. Let vf (i) = max{vf (0), vf (1)}. Either f (w) = i or f (w) ̸= i. If f (w) ̸= i,

then relabel V (G) by the inverse friendly labeling g = 1 − f to get g(w) = i. So without loss of generality

we will take f (w) = i.

Also without loss of generality let vf (i) = vf (0) (i.e. vf (0) ≥ vf (1)). So the vf (0) = k + 1 and vf (1) = k.

Now consider the relabeling of V (G) by h : V (G) → Z2 defined by


f (v), if v ̸= w
h(v) = (2.23)
1 − f (w) = 1 − i = 1 − 0 = 1, if v = w.

Note that vh (0) = k and vh (1) = k + 1. Thus, |vh (1) − vh (0)| = k + 1 − k = 1 ≤ 1. Hence, h : V (G) → Z2

is friendly. Observe that, eh (0) = 1 and eh (1) = q − 1. Consequently, |eh (1) − eh (1)| = q − 1 − 1 = q − 2.

Therefore, q − 2 ∈ F I(G).

Note that Theorem 2.40 gives a partial answer or reason why the caterpillar G = CR(a, b, c) in the

a + b + c even,|b − a − c + 1| = 1 case can have the second highest possible friendly index q − 2. What matters

is this leaf with terminal vertex w. It is also worth to note that the fully cordial caterpillars in Theorem 2.20

are merely formed by adding a leaf to the fully cordial caterpillars of diameter 3, DS(a, b), in Corollary 4.2

of [52] by adding the terminal vertex w and one edge. Jost and Banerjee refer to the terminal vertex w as a

pending vertex [5]. In [5] Jost and Banerjee use pending vertices in processes such as Motif doubling, graph

splitting, and joining. In fact they state in [5] that they start by studying Lemma 1.1 from their paper.

We paraphrase this Lemma concerning the eigenfunctions for the eigenvalue 1 of the Jost graph laplacian

operator below.

Lemma 2.41. [5] Let w be a pending vertex of G. Then any eigenfunction f1 for the eigenvalue 1 vanishes

at the unique neighbor of w.

The importance of the pending vertex of Jost and Banerjee from [5] in our study of fully cordial

Caterpillars of diameter 4 lead me to believe that processes such as Motif doubling, graph splitting, and

joining may be of use in finding an if and only if condition for fully cordial trees. In our case, this pending

vertex actually showed in some fashion when G = CR(a, b, c) is fully cordial. Perhaps in future research

41
we can use eigenfunctions for the eigenvalue 1 of the Jost graph laplacian operator and these techniques to

further study this problem. More support for this line of thinking is the paper [14], where Chen and Jost

study the minimum vertex cover for these pending vertices. We state a result from [14] concerning these

pending vertices below.

Theorem 2.42. [14] For a tree, let L be the set of its leaves (vertices of degree 1). Then L is not a subset

of any minimum vertex cover.

Chen and Jost also list numerous theorems concerning minimum vertex covers and matchings from graph

theory that they use in the study of such pending vertices. In section 5 of [14], Chen and Jost study the

minimum vertex cover and the eigenvectors associated with the eigenvalue 1 of the Jost graph laplacian

operator.

Note that in [54] Salehi and Lee demonstrated that trees with perfect matching are fully cordial. Recall

we discussed this in Theorem 2.5 and Corollary 2.6. In [52] Salehi and Corral demonstrated that trees with

near perfect matching are fully cordial. We discussed this in Theorem 2.12. The proofs of these come down

to having a pendent P3 path with specific conditions. These pendants have a pending vertex.

This concludes our discussion of fully cordial trees. It is still a research objective to find an application

of friendly labelings and fully cordial trees to social media problems. At this point, I am uncertain how I

would proceed in such a study.

42
 CHAPTER 3

VARIATIONAL METHODS ON FINITE GRAPHS

3.1

Introduction

In this chapter we study variational methods on finite graphs.

Section 3.2 provide preliminary definitions and concepts.

Section 3.3 provides an original minimization result. The main result of this section is Theorem 3.5 which

shows there exists a weak solution to a particular nonlinear Dirichlet problem. Lemma 3.4 is a minor original

technical lemma.

Section 3.4 is original joint work between Dr. David Costa and me. It was presented at the 1141st

AMS Meeting at the University of Delaware on September 29-30, 2018 as the talk “Remarks on Lagrange

Multiplier Approach to Kadzan-Warner Equations on a Finite Graph”. The main result of this section is

Theorem 3.12 which gives an if an only if condition for the Kazdan-Warner equation, ∆µ u = h(x) − k(x)eu ,

to have a weak solution for a non-constant function h on a finite graph. Lemmas 3.9, 3.10, and 3.11 are

original work used to prove the main result. Lemma 3.11 in particular applies the Lagrange multipliers to

the corresponding energy functional.

Section 3.5 is original joint work between Dr. David Costa and me. This section is an original application

of the Saddle Point theorem of Rabinowitz to the analogue of Ahmad, Lazer, and Paul resonant problem on

a finite graph. The main result is Theorem 3.18 which shows that the analogue of Ahmad, Lazer, and Paul

resonant problem on a finite graph has weak solution and thus a strong solution. Theorems 3.16 and 3.17

are minor original results used to prove the main result.

43
 3.2

Definitions

The classical approach of calculus of variations was to find critical points of Φ by solving some partial

differential equation D(x) = 0 (i.e. arriving at the Euler-Lagrange Equations and solving them). Now

the modern approach is to find critical points of some functional Φ and classify them in order to solve

some partial differential equation D(x) = 0. These modern variational techniques consider an associated

real-valued function Φ : X → R, whose derivative is equal to D(x), and by looking for points of minimum,

maximum, or minimax of Φ. In these situations, Φ′ (x) = 0.

We begin with a few basic definitions. Recall from earlier, definition 1.10 of the weighted Laplace operator

from Alexander Grigoryan’s text Analysis on Graphs [27].

Next we state some definitions by Grigoryan, Lin, and Yang in their papers ”Yamabe type equations on

finite graphs” [30] and ”Kazdan-Warner equations on graphs” [29].

Definition 3.1. [29] [30] (Corresponding Gradient Form For Weighted Laplace Operator) Let G = (V, E)

be a connected finite graph. Suppose that G has a weight µ. For any functions f, g : V → R, define the

corresponding gradient form for the weighted Laplace operator by

1 X
∇f · ∇g = Γµ (f, g)(x) = (f (y) − f (x))(g(y) − g(x))µxy . (3.1)
2µ(x) y∼x

In the literature: Γµ (f ) = Γµ (f, f ).

Definition 3.2. [29] [30] (Corresponding Length Of Gradient) Let G = (V, E) be a locally finite connected

graph. Suppose that G has a weight µ. For any function f : V → R, define the corresponding norm for
s
p 1 X
|∇f |(x) = Γ(f )(x) = (f (y) − f (x))2 µxy . (3.2)
2µ(x) y∼x

We need to define the relevant Rayleigh Quotient and first eigenvalue to be involved with our variational

problems we wish to study. Again consider the graph G = (V, E) which is finite or locally finite. Let Ω be a

finite subgraph of V . Recall that for any function f : V → R,

Z X
f dµ = f (x)µ(x). (3.3)
Ω x∈Ω

The first eigenvalue of the Laplacian with respect to the Dirichlet boundary condition reads
P
|∇f |2 dµ
R
ΩR 1 x,y∈Ω (∇xy f )(∇xy f )µxy
λ1 (Ω) = inf = P 2
. (3.4)
u̸=0,u|∂Ω=0
Ω
f2 2 x∈Ω f (x)µ(x)

44
Definition 3.3. [29] [30] (Norm For Sobolev Space) Let G = (V, E) be a locally finite connected graph.

Suppose that G has a weight µ, Ω ⊆ V such that Ω ̸= ∅, ∂Ω is the boundary of Ω and Ωo is the interior of

Ω. For any function f : V → R, the norm for W 1,2 (Ω) = H 1 (Ω) is

sZ Z sX X
∥f ∥W 1,2 (Ω) = 2
|f | dµ + 2
|∇f | dµ = |f |2 µ(x) + |∇f |2 µ(x). (3.5)
Ω Ω Ω Ω

Recall ∂Ω and Ωo are defined in Definition 1.16. The space W 1,2 (Ω) = H 1 (Ω) is defined to be the

functions f such that ∥f ∥W 1,2 (Ω) < ∞.

Following the classical tradition of the field of PDE’s one defines H01 (Ω) = W01,2 (Ω).

Definition 3.4. [29] [30] (H01 (Ω) = W01,2 (Ω)) Let G = (V, E) be a locally finite connected graph. Suppose

that G has a weight µ, Ω ⊆ V such that Ω ̸= ∅, ∂Ω is the boundary of Ω and Ωo is the interior of Ω. C01 (Ω)

is defined to be the set of all functions f : Ω → R with f = |∇f | = 0 on ∂Ω. Now H01 (Ω) = W01,2 (Ω) is

defined to be the completion of C01 (Ω) by the norm in the previous definition.

3.3

A Minimization Problem

Before we may proceed in our discussion of minimization type critical point theory for graphs we must

state a couple of classical results. The first is an abstract minimization property for compact topological

spaces.

Theorem 3.1. [21] Suppose X is a compact topological space. Let Φ : X → R be a lower-semicontinuous

functional. Then Φ is bounded from below and there exists a ∈ X such that Φ(a) = inf X Φ(x).

Proof. (Classical proof from literature) The lower semi-continuity of Φ implies that Φ−1 (−n, ∞) is open.

Now note that compactness of X and X = ∪∞

n=1 Φ
−1
(−n, ∞) yields that X = ∪m
n=1 Φ
−1
(−n, ∞) for some

m ∈ N. Consequently, Φ(x) > −m for every x ∈ X, this means that Φ is bounded from below.

Let b = inf X Φ(x). Suppose to the contrary that Φ(x) > b for every x ∈ X. Observe that, X =

∪∞
n=1 Φ
−1
(b + n1 , ∞). Since X is compact, there is a t ∈ N such that Φ(x) > b + 1
t for every x ∈ X. Thus,

b = inf X Φ(x) ≥ b + 1t , which implies the contradiction 0 ≥ 1

t. Therefore, there exists a ∈ X such that

Φ(a) = inf X Φ(x).

Another well-known theorem is as follows.

45
Theorem 3.2. [21] Suppose X is a reflexive Banach space. Let Φ : X → R be

1. weakly lower-semi-continuous,

2. coercive.

Then Φ is bounded from below and there exists a ∈ X such that Φ(a) = inf X Φ(x).

Proof. (Classical proof from literature) Since Φ is coercive, then one may pick a radius R > 0 where

Φ(x) > Φ(0) for every x ∈ X satisfying ∥x∥ ≥ R. A consequence of Alaoglu’s Theorem gives us that,

B, the closed ball of radius R centered at 0 is compact in the weak topology. Now since Φ : B → R is

lower-semi-continuous and B is compact in the weak topology, then the previous theorem says there exists

a ∈ B such that Φ(a) = inf B Φ(x) = inf X Φ where last equality comes from how R was picked.

Now we present some more recent results in the literature that will help us in our study of non-linear

operators on graphs. Alexander Grigoryan, Yong Lin, and Yunyan Yang in [30] [29] ”Yamabe type equations

on finite graphs” gives the following version of the Sobolev embedding theorem.

Theorem 3.3. [29] [30] Let G = (V, E) be a finite graph, Ω be a nonempty subset of V such that Ωo ̸= ∅. Let

m by any positive integer and p > 1. Then W0m,p is embedded in Lq (Ω) for all 1 ≤ q ≤ +∞. In particular,

there exists a constant C depending only on m, p, and Ω such that

Z  q1 Z  p1
q m p
|u| dµ ≤C |∇ u| dµ . (3.6)
Ω Ω

for all 1 ≤ q ≤ +∞ and for all u ∈ W0m,p (Ω). Moreover, W0m,p (Ω) is pre-compact, namely if uk is bounded

in W0m,p (Ω), then up to a subsequence, there exists some u ∈ W0m,p (Ω) such that uk → u in W0m,p (Ω).

Applying this theorem to the Sobolev space H01 (Ω) = W01,2 (Ω) tells us that H01 is embedded in ls for

every 1 ≤ s ≤ ∞ and provides that there exists a cs such that ∥h∥ls ≤ cs ∥h∥.

Now we are prepared to present a lemma which will be useful in proving the first non-linear Dirichlet

problem we discuss has weak solutions.

Lemma 3.4. Let G = (V, E) be a finite graph with weight µ. Suppose that Ω ⊆ V . Let f : V × R → R be
Ry
continuous in the second variable and 1r + 1s = 1. Suppose that F (x, y) = 0 f (x, z)dz. Then the functional
Z  
1
Φ(u) = |∇u|2 − F (x, u) dx
Ω 2
X 1 1 X
! (3.7)
= (u(y) − u(x)) µxy − F (x, u) µ(x), u ∈ H01 (Ω)
2
4 µ(x) y∼x
x∈Ω

46


is well defined. Furthermore, Φ ∈ C 1 H01 (Ω), R and
Z
′
Φ (u) · h = (∇u · ∇h − f (x, u)h)dx
Ω
X 1  (3.8)
= (u(y) − u(x))(h(y) − h(x)) − f (x, u)h µ(x), ∀u, h ∈ H01 (Ω).
2µ(x)
x∈Ω

Proof. Note that Φ(u) = q(u) − Ψ(u) where,

1
q(u) = ∥u∥2
2

and !
X X Z u(x)
Ψ(u) = F (x, u)µ(x) = f (x, z)dz µ(x).
x∈Ω x∈Ω 0

Note that,
|q(u + h) − q(u) − ⟨u, h⟩| |q(u + h) − q(u) − ⟨u, h⟩|
lim = lim
∥h∥→0 ∥h∥ ∥h∥→0 ∥h∥
| 12 ∥u∥2 + ⟨u, h⟩ + 12 ∥h∥2 − 12 ∥u∥2 − ⟨u, h⟩|
= lim
∥h∥→0 ∥h∥
2
(3.9)
1 |∥h∥
= lim
2 ∥h∥→0 ∥h∥
1
= lim ∥h∥ = 0.
2 ∥h∥→0

Hence, q ∈ C ∞ (H01 (Ω), R). Thus, q ∈ C 1 (H01 (Ω), R). Observe that,
X
δ(h) = Ψ(u + h) − Ψ(u) − f (x, u)hµ(x)
x∈Ω
X X
= (F (x, u + h) − F (x, u))µ(x) − f (x, u)hµ(x)
x∈Ω x∈Ω
X Z 1  Z 1X
d
= F (x, u + th)dt µ(x) − f (x, u)hµ(x)dt
x∈Ω 0 dt 0 x∈Ω

X Z 1  Z 1X (3.10)
= f (x, u + th)hdt µ(x) − f (x, u)hµ(x)dt)
x∈Ω 0 0 x∈Ω
Z 1X Z 1 X
= f (x, u + th)hµ(x)dt − f (x, u)hµ(x)dt
0 x∈Ω 0 x∈Ω
Z 1X
= (f (x, u + th) − f (x, u))hµ(x)dt.
0 x∈Ω

By Holder’s inequality,

Z 1
|δ(h)| ≤ ∥f (x, u + th) − f (x, u)∥Lr ∥h∥ls dt (3.11)
0

By the continuous Sobolev Embedding H01 (Ω) is embedded in ls , so u + th → u in ls as h → 0 in H01 (Ω).

From the continuity of f in the second variable, f (·, u + th) → f (·, u) in Lr as all norms are equivalent on a

47
finite dimensional vector space. By Lebesgue Dominated Convergence Theorem and the Sobolev inequality,

∥h∥ls ≤ cs ∥h∥, we get

1
|δ(h)| |δ(h)|
Z
≤ cs ≤ ∥f (·, u + th) − f (·, u)∥Lr dt → 0. (3.12)
∥h∥ ∥h∥ls 0

Consequently, Ψ : H01 → R is Frechet differentiable at any u ∈ H01 with Frechet derivative

X
Ψ′ (u) · h = f (x, u)hµ(x) (3.13)
x∈Ω

for every u, h ∈ H01 (Ω).

Now the continuity of Ψ′ : H01 → (H01 )∗ will be demonstrated. By again appealing to Holder’s inequality

and Sobolev embedding (∥h∥ls ≤ c∥h∥) and Lr continuity of f , we observe the following computation:

∥Ψ′ (u + v) − Ψ′ (u)∥H −1 = sup |(Ψ′ (u + v) − Ψ′ (u)) · h|

∥h∥≤1

X
= sup (f (x, u + v) − f (x, u))hµ(x)
∥h∥≤1 x∈V
X
≤ sup |(f (x, u + v) − f (x, u))| ∥h∥µ(x)
∥h∥≤1 x∈V
X (3.14)
≤ sup ∥(f (x, u + v) − f (x, u))∥Lr ∥h∥ls
∥h∥≤1 x∈V
X
≤ cs sup ∥(f (x, u + v) − f (x, u))∥Lr ∥h∥
∥h∥≤1 x∈V
X
≤ cs sup ∥(f (x, u + v) − f (x, u))∥Lr → 0 as v → 0 in H01 .
∥h∥≤1 x∈V

Therefore, Φ ∈ C 1 (H01 (Ω), R) with Frechet derivative

Z
′
Φ (u) · h = (∇u · ∇h − f (x, u)h)dx
Ω
X 1 
= (u(y) − u(x))(h(y) − h(x)) − f (x, u)h µ(x), ∀u, h ∈ H01 (Ω).
2µ(x)
x∈Ω

We now study a nonlinear Dirichlet problem on a graph which will be solved using the technique of

minimization.

Theorem 3.5. Let G = (V, E) be a finite graph with weight µ. Suppose that Ω ⊆ V . Let f : Ω × R → R be
1 1
continuous in the second variable and r + s = 1. Suppose further that there exists α < λ1 , where λ1 (Ω) is
f (x,y)
the first eigenvalue of −∆u = λu in Ω such that u = 0 on ∂Ω, such that lim sup|y|→∞ y ≤ α uniformly

48
for x ∈ Ω. Suppose also that, there exists c, d > 0 and 0 ≤ σ < ∞ such that |f (x, y)| ≤ c|y|σ + d. Then there

exists a weak solution to the following nonlinear Dirichlet problem:

Lµ u(x) = f (x) ∀x ∈ Ω
(3.15)
u(x) = 0 ∀x ∈ ∂Ω.

Proof. Note that to find solutions of the weak form of this Dirichlet problem is to look for functions u ∈ H01 (Ω)
R P 1
1
such that Ω (∇u · ∇h − f (x, u)h)dx = 2 (u(y) − u(x))(h(y) − h(x)) − f (x, u)h µ(x) = 0 ∀h ∈ H0 (Ω).
x∈Ω
By the previous theorem finding such a weak solution is the same as finding a critical point of the functional

Φ ∈ C 1 H01 (Ω), R given by Φ(u) = q(u) − Ψ(u) where q(u) = 21 ∥u∥2 and Ψ(u) =

P
F (x, u)µ(x).
x∈Ω
∞
First notice that q is weakly lower semi continuous (as q ∈ C (H01 (Ω), R)
and so q is continuous) and
P
Ψ(u) = F (x, u)µ(x) is weakly continuous. Hence, Φ is weakly lower semi continuous.
x∈Ω
f (x,y)
Since there exists α < λ1 such that lim sup|y|→∞ y ≤ α uniformly for x ∈ Ω, then by l’Hopital’s Rule

2F (x, y) f (x, y)
lim sup = lim sup ≤ α uniformly ∀x ∈ Ω. (3.16)
|y|→∞ y2 |y|→∞ y

This implies that there exists α1 where α < α1 < λ1 (Ω) and R1 > 0 such that F (x, y) ≤ 12 α1 y 2 ∀x ∈ Ω and

|y| ≥ R1 . Now, in view of the fact that, there exists c, d > 0 and 0 ≤ σ < ∞ such that |f (x, y)| ≤ c|y|σ + d

then, there is a γ such that F (x, y) ≤ γ ∀x ∈ Ω and |y| ≤ R1 . Thus F (x, y) ≤ 21 α1 y 2 + γ ∀x ∈ Ω, ∀y ∈ R.

Hence,
1X 1 X 2
Φ(u) ≥ (u(y) − u(x))2 µ(x) − α1 u µ(x) − γ|Ω| as Ω is bounded. (3.17)
2 2
Ω Ω

An application of the Poincaré inequality λ1 Ω u2 µ(x) ≤ Ω (u(y) − u(x))2 µ(x) results in:
P P

 
1 α1 X 1
Φ(u) ≥ 1− (u(y) − u(x))2 µ(x) − C = a∥u∥2 − C, (3.18)
2 λ1 2
Ω

α1
where C = γ|Ω| and a = 1 − λ1 > 0. The previous sentence gives us that Φ is coercive on H01 .

Now since Φ is both weakly lower semi continuous and coercive on H01 and H01 is a reflexive Banach

space, then by Theorem 3.2, Φ is bounded from below and there exists v ∈ X such that Φ(v) = inf X Φ(u).

Therefore, v is a critical point of Φ. Now from the first paragraph of this proof we see that v is a weak

solution to equation (3.15).

49
 3.4

Lagrange Multiplier Approach to Kazdan-Warner Equation on a Finite Graph

In 2016 [29] Grigoryan, Lin, and Yang studied the finite graph analogue of the Kazdan-Warner equation.

Given a finite graph V and c a constant, they studied the equations:

∆µ u = c − k(t)eu in V, (3.19)

and

∆µ u = −k(t)eu in V. (3.20)

They determined if and only if conditions for these equations to have a solution. They followed closely

the technique used by J.Kazdan and F. Warner in their 1974 paper ”Curvature functions for compact

2-manifolds” [38] for proving existence of solution in the continuous analogue of this equation. The results

that Grigoryan, Yang, and Yang found in the paper include:

Theorem 3.6. [29] Let G = (V, E) be a finite graph, and k (≡ / 0) be a function on V . Then the equation
R
(3.20) above has a solution u if and only if k changes sign and V kdµ < 0.

Theorem 3.7. [29] Let G = (V, E) be a finite graph, and k (≡

/ 0) be a function on V and c be a positive

constant. Then the equation (3.19) above has a solution u if and only if k is positive somewhere.

We now provide an extension of the aforementioned results of the Kazdan-Warner equation on a finite

graph by replacing the constant c with a function h : V → R. Our technique is based upon the concept of

Lagrange Multipliers. The work in this section was presented at the 1141st AMS Meeting at the University of

Delaware on September 29-30, 2018 as the talk “Remarks on Lagrange Multiplier Approch to Kadzan-Warner

Equations on a Finite Graph”. The work is joint work between myself and Dr. Costa. We will start by

stating the variational principle of Lagrange Multipliers.

Theorem 3.8. [21] Let Φ ∈ C 1 (X, R) and suppose M ⊆ X is a C 1 -sub-manifold of codimension k, say

M = {u ∈ U : Ψj (u) = 0, j = 1, ..., k} where ψj ∈ C 1 (U, R), j = 1, .., k, U ⊆ X is an open set and

Ψ′1 (u), ..., Ψ′k (u) are linearly independent for each u ∈ U . If u ∈ M is a critical point of Φ|M , then there

exists λ = (λ1 , ..., λk ) ∈ Rk such that:

k
X
′ ′
Φ (u) = λ · Ψ (u) = λj Ψ′j (u). (3.21)
j=1

The λj ’s are called Lagrange multipliers.

50
 Again let V be a finite graph. The version of Kazdan-Warner equation for a graph that we study is

∆µ u = h(x) − k(x)eu in V (3.22)

P P
where both h, k : V → R and h(x)µ(x) = 0 and k(x)µ(x) < 0. The weak form of this graph PDE is:
x∈V x∈V
P1 P
(∇xy u)(∇xy v)µxy µ(x) + hvµ(x) = k(x)eu vµ(x). The corresponding functional Φ : H 1 (V ) → R
P P
2
Ω y∼x Ω Ω
needed is:
Z  
1
Φ(u) = |∇u|2 + hu dx
V 2
!
X 1 1 X
2
= (u(y) − u(x)) µxy + hu µ(x) (3.23)
4 µ(x) y∼x
V
X1X X
= (∇xy u)2 µxy + huµ(x) ∀ u, h ∈ H 1 (V ) ,
4 y∼x
V V

and it’s Fréchet derivative is

Z
′
Φ (u) · v = (∇u · ∇v + hv)dx
V
!
X 1 1 X
= (u(y) − u(x))(v(y) − v(x))µxy + hv µ(x) (3.24)
2 µ(x) y∼x
V
X1X X
= (∇xy u)(∇xy v)µxy + hvµ(x) ∀ u, h ∈ H 1 (V ) .
2 y∼x
V V

We begin by providing the existence of a non-trivial ”basic solution”, u0 , for the equation ∆µ u =

h in H1 (V ). The fact that ∆µ u : V → Rn is symmetric with respect to (·, ·) on H 1 (V ) by spectral

decomposition assures there exists an orthonormal basis of eigenvectors {e1 , e2 , e3 , · · · , en } for ∆µ over

H 1 (V ) with eigenvalues {λ1 , λ2 , λ3 , · · · , λn } respectively. Our proof of a non-trivial solution is essentially a

version of the Fredholm alternative for our finite-dimensional situation.

Lemma 3.9. Let G = (V, E) be a finite graph such that |V | = n. Then the equation ∆µ u = h has a solution

u0 ̸= 0 if and only if h ∈ Span{ei : λi ̸= 0}.

P P  αi 
Proof. For the first direction suppose that h ∈ Span{ei : λi ̸= 0}. So then h = αi ei = λi (λi ei ).
λi ̸=0 λi ̸=0
P  αi  1
Note that u0 = λi ei ∈ H (V ) and
λi ̸=0
 
X  αi  X  αi  X  αi  X
∆µ u0 = ∆µ  ei =
 ∆µ ei = λ i ei = αi ei = h.
λi λi λi
λi ̸=0 λi ̸=0 λi ̸=0 λi ̸=0

Hence, u0 ̸= 0 is a classical solution to ∆µ u = h. Conversely, suppose that ∆µ u = h has a solution u0 ̸= 0

in H 1 (V ). Since {e1 , e2 , · · · , en } is an orthonormal basis of basis of eigenvectors for ∆µ over H 1 (V ) with

51
eigenvalues {λ1 , λ2 , λ3 , · · · , λn } (we only need the fact that there is a basis for H 1 (V )), then u0 =
P
λi ̸=0 βi e i .

Therefore,  
X X X
h = ∆µ u0 = ∆µ  βi e i  = βi ∆ µ e i = βi λi ei ∈ Span{ei : λi ̸= 0}.
λi ̸=0 λi ̸=0 λi ̸=0

The next lemma shows that every solution to ∆µ u = h is a constant off of u0 ̸= 0 defined in the proof of

the previous lemma. This is why we refer to this u0 ̸= 0 as a basic solution to ∆µ u = h.

Lemma 3.10. Let G = (V, E) be a finite graph. Then any solution of the equation ∆µ u = h has the form

u = u0 + c where c ∈ R is a constant.

Proof. Suppose u ∈ V is an arbitrary solution to ∆µ u = h. By the previous lemma, ∆µ u0 (x) = h(x). Now

∆µ u = h becomes ∆µ u = ∆µ u0 . By the linearity of ∆µ shown in Theorem 1.1 we see ∆µ (u − u0 ) = 0. Since

zero is a simple eigenvalue of ∆µ by the proof of Theorem 1.13, then u − u0 = c for some constant c ∈ R.

Therefore, u = u0 + c.

Now define the following set

( )
X X
1 u(x)
M= u ∈ H (V ) : k(x)e µ(x) = 0, u(x)µ(x) = 0 . (3.25)
x∈V x∈V

In the following lemma, we show that M is a C 1 sub-manifold of codimension 2.

P
Lemma 3.11. Let G = (V, E) be a finite graph. Suppose that h, k : V → R satisfy h(x)µ(x) = 0 and
x∈V
k(x)eu0 (x) µ(x) < 0. Then we have the following:
P
x∈V

1. M ⊆ H 1 (V ) is a C 1 sub-manifold of codimension 2;

2. Furthermore, any critical point u of Φ|M is (up to a constant) a weak solution to ∆µ u = h(x) − k(x)eu

in H 1 (V ), and this in turn implies that u is a solution of ∆µ u = h(x) − k(x)eu .

Proof. We begin by proving M ⊆ H 1 (V ) is a C 1 sub-manifold of codimension 2. Note that, Grigoryan, Lin,

and Yang showed M ̸= ∅ [29]. Observe that the functionals ψ1 , ψ2 : H 1 (V ) → R defined by

X X
Ψ1 (u) = k(x)eu(x) µ(x) , Ψ2 (u) = u(x)µ(x),
x∈V x∈V

52
are of class C ∞ , hence, C 1 . Furthermore, we now show that Ψ′1 (u) and Ψ′2 (u) are linearly independent

∀u ∈ M . Indeed, observe that,

X X
(αΨ′1 (u) + βΨ′2 (u)) · v = α k(x)eu(x) v(x)µ(x) + β v(x)µ(x) = 0 ∀v ∈ H 1 (V ),
x∈V x∈V

implies that β = 0 (by the choice of v(x) = 1) and then α = 0 (by the choice of v(x) = eu0 (x)−u(x) ).

Hence, Ψ′1 (u) and Ψ′2 (u) are linearly independent ∀u ∈ M . Therefore, M ⊆ H 1 (V ) is a C 1 sub-manifold of

codimension 2.

Now we prove for any critical point of Φ|M is (up to a constant) a solution to ∆µ u = h(x) − k(x)eu .

Suppose that u ∈ M is a critical point of Φ|M . Since M ⊆ H 1 (V ) is a C 1 sub-manifold of codimension 2

then by Theorem 3.8 (Lagrange Multipliers), there exists λ1 , λ2 ∈ R such that

Z
Φ′ (u) · v = (∇u · ∇v + hv)dx
V
X1 X X
= (∇xy u)(∇xy v)µxy + hvµ(x) (3.26)
2µ(x) y∼x
V V
X X
= λ1 k(x)eu(x) v(x)µ(x) + λ2 v(x)µ(x) ∀v ∈ H 1 (V ).
x∈V x∈V

Therefore, u ∈ H 1 (V ) is a weak solution to −∆µ u + h(x) = λ1 k(x)eu(x) + λ2 . In view of Theorem 1.10

(Integration By Parts) and V = Ω ∪ Ωc and Ωc = ∅ (i.e. Ω = V ) then

X 1 X
∆µ u(x)v(x)µ(x) = − (∇xy u)(∇xy v)µx,y ∀v ∈ H 1 (V ).
2
x∈V x,y∈V

This means
X
((−∆µ u) + h − λ1 keu − λ2 )vµ(x) = 0 ∀v ∈ H 1 (V ).
x∈V

Consequently, u satisfies (−∆µ u) + h − λ1 keu − λ2 = 0 in H 1 (V ) (by choosing v = sgn((−∆µ u) + h −

λ1 keu − λ2 ))). Hence, u satisfies

−∆µ u + h(x) = λ1 k(x)eu(x) + λ2

in H 1 (V ) (in other words u is a solution in H 1 (V )). The previous equation becomes:

X X
(−∆µ u + h(x))µ(x) = (λ1 k(x)eu(x) + λ2 )µ(x). (3.27)
x∈V x∈V

k(x)eu(x) µ(x) = 0 by equation (3.25) and

P P P
Since h(x)µ(x) = 0, (−∆µ u)µ(x) = 0 by Corollaries
x∈V x∈V x∈V
1.6 and 1.11 then, equation (3.27) implies that λ2 = 0 and hence

−∆µ u + h(x) = λ1 k(x)eu(x) . (3.28)

53
Note that, λ1 ̸= 0, for if λ1 = 0 then ∆µ u = h(x) would imply that u = u0 + c for some constant c ∈ R by

k(x)eu(x) µ(x) = 0 would become

P
Lemma 3.10. Now
x∈V
X X
0= k(x)eu(x) µ(x) = ec k(x)eu0 (x) µ(x) < 0,
x∈V x∈V

which is a contradiction. Consequently, λ1 ̸= 0. Now let u = u0 + w (i.e. w = u − u0 ). Then equation (3.28)

becomes:

−∆µ (u0 + w) + h = λ1 keu0 +w .

The previous equation becomes:

−∆µ w = λ1 keu0 +w

as ∆µ u0 = h. Multiplying the previous equation by e−w and integrating over V yields:

X X
− (∆µ w)e−w µ(x) = λ1 keu0 µ(x).
x∈V x∈V

Applying Theorem 1.10 (Integration By Parts) to the left hand side of the previous equation yields:

1 X X
(∇xy w)(∇xy e−w )µxy = λ1 keu0 µ(x).
2
x,y∈V x∈V

The previous equation is equivalent to:

1 X X
(w(y) − w(x))(e−w(y) − e−w(x) )µxy = λ1 keu0 µ(x).
2
x,y∈V x∈V

The previous equation directly implies that λ1 > 0 as

1 X
(w(y) − w(x))(e−w(y) − e−w(x) )µxy < 0
2
x,y∈V

and
X
keu0 µ(x) < 0.
x∈V

Since λ1 > 0 then there exists a constant d ∈ R such that λ1 = ed . Letting û = u + d we observe that:

∆µ û + keû = ∆µ (u + d) + keu+d

= ∆µ u + ∆µ d + keu ed

= ∆µ u + 0 + λ1 keu

= ∆µ u + λ1 keu

= h,

54
in view of equation (3.28). Therefore, û is a solution to ∆µ u + keu = h.

Now we present a theorem concerning a necessary and sufficient condition for the existence of solutions

to ∆µ u = h(x) − k(x)eu in H 1 (V ).

Theorem 3.12. Let G = (V, E) be a finite graph. Suppose that h : V → R where h ∈ span{ei : λi ̸= 0} and

k(x) is not the zero function. Then ∆µ u = h(x) − k(x)eu has a solution if and only if k(x)eu0 (x) µ(x) < 0
P
x∈V
and k(x) changes sign.

Proof. First suppose that u is a solution to ∆µ u + keu = h in V . Integrating both sides of the previous

equation over V yields:

X X X
(∆µ u)µ(x) + keu µ(x) = hµ(x). (3.29)
x∈V x∈V x∈V

ei (x)µ(x) = λ1i
P P P P
Note that h(x)µ(x) = 0 as h = αi ei and (∆µ ei )(x) = 0. Now since
x∈V λi ̸=0 x∈V x∈V
P P
h(x)µ(x) = 0 and (∆µ u)µ(x) = 0 then equation (3.29) becomes:
x∈V x∈V
X
keu µ(x) = 0.
x∈V

The previous equation implies that k(x) must change sign. Letting u = u0 + w makes ∆µ u + keu = h become

∆µ (u0 + w) + keu0 +w = h in V . Thus, ∆µ u0 + ∆µ w + keu0 ew = h in V . The previous equation becomes

h + ∆µ w + keu0 ew = h as ∆µ u0 = h by Lemma 3.10. Hence, ∆µ w + keu0 ew = 0. Multiplying both sides by

e−w and integrating both sides of the previous equation over V yields:
X X
(∆µ w)e−w µ(x) + keu0 µ(x) = 0.
x∈V x∈V

Applying Theorem 1.24 (Integration By Parts) to the left hand side of the previous equation yields:

1 X X
− (∇xy w)(∇xy e−w )µx,y + keu0 µ(x) = 0.
2
x,y∈Ω x∈V

The previous equation along with (k(x) ̸= 0 implying ∇xy w ̸= 0) implies that
X
keu0 µ(x) < 0.
x∈V
u0
P
Conversely, suppose that ke µ(x) < 0. Then, in view of Lemma 3.13, in order to demonstrate that
x∈V
∆µ u = h(x) − k(x)eu has a solution, it suffices to find a critical point of Φ|M . Note that, M ⊆ X0 =
 
u ∈ H 1 (V ) :
P
uµ(x) = 0 and on X0 we have the following Wirtinger inequality:
x∈V
X X
|u|2 µ(x) ≤ C (u(y) − u(x))2 µ(x).
x∈V x∈V

55
Hence,
X 1 
Φ(u) = (u(y) − u(x))2 + hu µ(x) ≥ Ĉ∥u∥2 −∥h∥l2 ∥u∥ → +∞
2
x∈V

as ∥u∥ → +∞ for u ∈ X0 . Consequently, since Φ is weakly lower semi continuous and M is weakly closed in

X, Theorem 3.2 shows the existence of v ∈ M such that Φ(v) = inf M Φ. Hence, v ∈ M is a critical point of

Φ|M . Therefore, by Lemma 3.11, v is a solution to ∆µ u = h(x) − k(x)eu in V.

Our non-constant h(x) style of Lagrange Multiplier approach also agrees with the Grigoryan, Yang, Lin

result when h(x) is the zero function. Here, we take the fundamental solution to the PDE ∆µ u = h to be

the zero-vector (or constant zero graph labeling over the vertex set) denoted by u0 = 0. A further note is

that using this notation gives us eu0 (x) = 1 is the graph labeling each vertex by 1. Note in this situation
P
that h(x)µ(x) = 0 as h(x) is the constant zero graph labeling. Using the same manifold of codimension
x∈V
2 in the previous argument and replacing the non-zero u0 with u0 = 0 in all the previous arguments results

in the result of Grigoryan, Yang, Lin:

Theorem 3.13. [29] Let G = (V, E) be a finite graph, and k (≡ / 0) be a function on V . Then the equation
R
(3.20) has a solution u if and only if k changes sign and V kdµ < 0.

To arrive at the result of Grigoryan, Yang, Lin regarding h(x) = c where |c| =
̸ 0 by a Lagrange Multiplier

approach requires use of a manifold of codimension 1. This is what we discuss in the next proof. First we

paraphrase the result of Grigoryan, Yang, Lin that we wish to prove via Lagrange Multipliers:

Theorem 3.14. [29] Let G = (V, E) be a finite graph, and k (≡

/ 0) be a function on V . Suppose that h(x) = c

for every vertex x ∈ V and |c| =

̸ 0. Then the equation (3.20) has a solution.

Proof. We start by considering the manifold of codimension 1 given by

( )
X
1 u(x)
M = u ∈ H (V ) : k(x)e µ(x) = cvol(V ) . (3.30)
x∈V

Note that, Grigoryan, Lin, and Yang showed M ̸= ∅ [29].

We now prove M ⊆ H 1 (V ) is a C 1 sub-manifold of codimension 1. Note that, M ̸= ∅. Observe that, the

functional ψ1 : H 1 (V ) → R defined by

X
Ψ1 (u) = k(x)eu(x) µ(x),
x∈V

56
is of class C ∞ , hence, C 1 . Furthermore, we now show that {Ψ′1 (u)} is linearly independent ∀u ∈ M . Now

we observe that,
X
(αΨ′1 (u)) · v = α k(x)eu(x) v(x)µ(x) = 0 ∀v ∈ H 1 (V ),
x∈V

implies that α = 0. Hence, {Ψ′1 (u)} is linearly independent ∀u ∈ M . Therefore, M ⊆ H 1 (V ) is a C 1

sub-manifold of codimension 1. Now we prove for any critical point of Φ|M is (up to a constant) a solution

to ∆µ u = h(x) − k(x)eu . Suppose that u ∈ M is a critical point of Φ|M . Since M ⊆ H 1 (V ) is a C 1

sub-manifold of codimension 1 then by Theorem 3.8 (Lagrange Multipliers): there exists λ ∈ R such that

X 1 X X X
(∇xy u)(∇xy v)µxy + cvµ(x) = λ k(x)eu(x) v(x)µ(x).
2µ(x) y∼x
V V x∈V

Using integration by parts the previous equation yields:

−∆µ u + c = λk(x)eu(x) .

We sum both sides of previous equation over V to get:

X X X
(−∆µ u)µ(x) + cµ(x) = λ k(x)eu(x) µ(x).
x∈V x∈V x∈V

The previous equation gives us

X X
c µ(x) = λ k(x)eu(x) µ(x).
x∈V x∈V
P
as (−∆µ u)µ(x) = 0. From the previous equation we get:
x∈V

cvol(V ) = λcvol(V ).

k(x)eu(x) µ(x) = cvol(V ). Hence, (λ − 1)cvol(V ) = 0. Now since (λ − 1)cvol(V ) = 0 and cvol(V ) ̸= 0,
P
as
x∈V
then λ − 1 = 0. Thus, λ = 1. Consequently,

−∆µ u + c = k(x)eu(x) .

Note that Grigoryan, Yang, and Lin successfully proved that the functional

X 1
Φ(u) = ( |∇u|2 + cu)µ(x)
2
x∈V

is coercive on M [29]. Consequently, since Φ is weakly lower semi continuous and M is weakly closed in X,

Theorem 3.7 shows the existence of v ∈ M such that Φ(v) = inf M Φ. Hence, v ∈ M is a critical point of

Φ|M . Therefore, by Lemma 3.16, v is a solution to ∆µ u = h(x) − k(x)eu in V.

57
 3.5

Saddle Point Approach to Ahmad, Lazer, and Paul Resonant Problem on a

Finite Graph

Now we are about to embark on an original application of the Saddle Point theorem of Rabinowitz to

the analogue of Ahmad, Lazer, and Paul resonant problem on a finite graph. We start with the definition of

the Palais-Smale condition.

Definition 3.5. (Palais-Smale condition) [21] A functional Φ ∈ C 1 (X, R) is said to satisfy the Palais-Smale

condition if and only if any sequence {un }∞ ′

0 such that Φ(un ) is bounded and Φ (un ) → 0 possesses a

convergent sub-sequence.

Now we state the famous Saddle Point Theorem of Rabinowitz.

Theorem 3.15. [21][50] Let X = V ⊕ W be a Banach space, where dim V < ∞, and let Φ ∈ C 1 (X, R) be a

functional satisfying the Palais-Smale condition. If D is a bounded neighborhood of 0 on V such that

a = max Φ < inf Φ = b,

∂D W

then

c = inf max Φ(h(u))

h∈Γ u∈D

is a critical value of Φ with c ≥ b. (Here Γ is the class of deformations of D in X which fix ∂D point-wise,

that is, Γ = {h ∈ C(D, X) : h(u) = u ∀u ∈ ∂D}.)

We now consider the finite graph analogue of Ahmad, Lazer, and Paul resonant problem [1]:

−∆µ u = λk u + g(x, u) ∀x ∈ Ω ⊆ V
(3.31)
u(x) = 0 ∀x ∈ Ωc ,

where V is a finite graph, u : V → R is an unknown function while the function g : V × R → R is continuous

and uniformly bounded and given, say |g(x, s)| ≤ M ∀x ∈ Ω,∀s ∈ R. Also we impose the following condition:

XZ v(x)
g(x, t)dtµ(x) → ±∞ as ∥v∥ → ∞, v ∈ Nk .
x∈Ω 0

where Nk the null-space of the operator L : H01 → H01 defined by < Lu, u >= x∈Ω 21 (u(y) − u(x))2 − λk u2 .
P

R v(x)
The functional we study is Φ(u) = 21 < Lu, u > − x∈Ω 0 g(x, t)dtµ(x). Observe that H01 (Ω) = X− ⊕
P

X0 ⊕ X+ such that X0 = Nk , X+ is the subspace where the operator L : H01 (Ω) → H01 (Ω) is positive definite,

58
X− is the subspace where the operator L : H01 (Ω) → H01 (Ω) is negative definite. The respective orthogonal

projections are denoted by P0 , P+ , and P− .

Theorem 3.16. Let G = V be a finite graph and let Ω ⊆ V . Suppose |g(x, s)| ≤ M ∀x ∈ Ω,∀s ∈ R and

XZ v(x)
g(x, t)dtµ(x) → −∞ as ∥v∥ → ∞, v ∈ Nk ,
x∈Ω 0

then

(a). Φ(u) → −∞ as ∥u∥ → ∞ and u ∈ X− ,

(b). Φ(u) → ∞ as ∥u∥ → ∞ and u ∈ X0 ⊕ X+ .

Proof. (a). Set u = P− u ∈ X− and observe

1 X Z P− u(x)
Φ(u) = ⟨LP− u, P− u⟩ − g(x, t)dtµ(x)
2
x∈Ω 0
1 X
≤ − α∥P− u∥2 − M P− uµ(x)
2
x∈Ω
1 X
= − α∥P− u∥2 −M P− uµ(x)
2
x∈Ω
1 X
≤ − α∥P− u∥2 +M |P− u|µ(x)
2
x∈Ω
1
= − α∥P− u∥2 +M ∥P− u∥L1
2
1
≤ − α∥P− u∥2 +M ∥P− u∥→ −∞
2

as ∥u∥ =∥P− u∥ → ∞.

(b). Set u = P0 u + P+ u ∈ X0 ⊕ X+ and observe

1 X Z u(x)
Φ(u) = ⟨LP+ u, P+ u⟩ − g(x, t)dtµ(x)
2
x∈Ω 0

1 X Z u(x)
≥ α∥P+ u∥2 − g(x, t)dtµ(x)
2
x∈Ω 0
!
1 X Z u(x) Z P0 u(x) X Z P0 u(x)
2
≥ α∥P+ u∥ − g(x, t)dt − g(x, t)dt µ(x) − g(x, t)dt
2 0 0
x∈Ω x∈Ω 0

1 X X Z P0 u(x)
2
≥ α∥P+ u∥ −M |P+ u|µ(x) − g(x, t)dt
2
x∈Ω x∈Ω 0

1 X Z P0 u(x)
≥ α∥P+ u∥2 −M ∥P+ u∥L1 − g(x, t)dt
2
x∈Ω 0

1 X Z P0 u(x)
≥ α∥P+ u∥2 −A∥P+ u∥− g(x, t)dt → ∞
2 0
x∈Ω

59
as ∥u∥ → ∞ where u ∈ X0 ⊕ X+ .

Note that in the previous theorem if one replaces the assumption

XZ v(x)
g(x, t)dtµ(x) → −∞ as ∥v∥ → ∞, v ∈ Nk
x∈Ω 0

with
XZ v(x)
g(x, t)dtµ(x) → ∞ as ∥v∥ → ∞, v ∈ Nk
x∈Ω 0

one clearly gets the following theorem.

Theorem 3.17. Let G = V be a finite graph and let Ω ⊆ V . Suppose |g(x, s)| ≤ M ∀x ∈ Ω,∀s ∈ R and
XZ v(x)
g(x, t)dtµ(x) → ∞ as ∥v∥ → ∞, v ∈ Nk ,
x∈Ω 0

then

(a). Φ(u) → −∞ as ∥u∥ → ∞ and u ∈ X− ⊕ X0 ,

(b). Φ(u) → ∞ as ∥u∥ → ∞ and u ∈ X+ .

We now state and prove the discrete analog of the Ahmad, Lazer, and Paul [1] resonant problem.

Theorem 3.18. Suppose that V is a finite graph, u : V → R is an unknown function while the function

g : V × R → R is continuous and uniformly bounded and given, say |g(x, s)| ≤ M ∀x ∈ Ω,∀s ∈ R. Also we

impose one of the following conditions:

XZ v(x)
g(x, t)dtµ(x) → ∞ as ∥v∥ → ∞, v ∈ Nk
x∈Ω 0

or
XZ v(x)
g(x, t)dtµ(x) → −∞ as ∥v∥ → ∞, v ∈ Nk .
x∈Ω 0
1
where Nk the null-space of the operator L : H01 → H01 defined by ⟨Lu, u⟩ = − u(x))2 − λk u2 .
P
2 (u(y)
x∈Ω
Then our Ahmad, Lazer, and Paul resonant problem equation (3.31) has a weak solution and thus a strong

solution.

Proof. Let us show that the functional Φ satisfies the Palais-Smale condition. Suppose that |Φ(un )| ≤ C
P R v(x)
and Φ′ (un ) → 0 while 0
g(x, t)dtµ(x) → −∞ as ∥v∥ → ∞, v ∈ Nk . Then for every n large enough:
x∈Ω

XZ un (x)
|Φ′ (un ) · h| = |⟨Lun , h⟩ − g(x, t)hdtµ(x)| ≤∥h∥ ∀h ∈ H01 (Ω).
x∈Ω 0

60
This implies that

∥P+ un ∥≥ |Φ′ (un ) · P+ un | ≥ α∥P+ un ∥2 −M ∥P+ un ∥L1 ≥ α∥P+ un ∥2 −A∥P+ un ∥.

Hence, ∥P+ un ∥ is bounded. Note that for ∥un ∥ sufficiently large we have

|Φ′ (un ) · h| ≤∥h∥ ∀h ∈ H01 (Ω)

also implies that

−∥P− un ∥≤ |Φ′ (un ) · P− un | ≤ −α∥P− un ∥2 +M ∥P− un ∥L1 ≤ −α∥P− un ∥2 +A∥P− un ∥.

Thus, ∥P− un ∥ is bounded. Therefore, ∥(P+ + P− )un ∥=∥un − P0 un ∥ is bounded. Note that

1
Φ(un ) = ⟨L(un − P0 un ), un − P0 un ⟩
2 !
X Z un (x) Z P0 un (x) XZ P0 un (x)
− g(x, t)dt − g(x, t)dt µ(x) − g(x, t)dt.
x∈Ω 0 0 x∈Ω 0

Since Φ(un ) is bounded together with the first two terms on the right side of the previous equation, we
P R P0 un (x) P R P0 un (x)
conclude that 0
g(x, t)dt is bounded. Now since 0
g(x, t)dt is bounded and
x∈Ω x∈Ω

XZ un (x)
g(x, t)dtµ(x) → −∞ as ∥un ∥ → ∞, un ∈ Nk ,
x∈Ω 0

then ∥P0 un ∥ is necessarily bounded. Since ∥(P+ + P− )un ∥ and ∥P0 un ∥ are bounded, then ∥un ∥ is bounded.

Now since H01 (Ω) is a reflexive Banach space, and un is bounded, then un has a weakly convergent subsequence

which we may pass to, so we may write that uk converges weakly to u. Applying the compact operator

T (u) = u − ∇Φ(u) to this weakly convergence subsequence uk we get that uk = ∇Φ(uk ) + T (uk ) → 0 + T (u)

strongly. Consequently, uk → u strongly in H01 (Ω). Hence, uk is a H01 (Ω) convergent subsequence of un .

Therefore, Φ satisfies the Palais-Smale condition.

Now by Theorem 3.17, we may apply the saddle-point theorem with V = X− , W = X+ ⊕ X0 to conclude

the existence of a critical point of Φ. Therefore, our Ahmad, Lazer, and Paul resonant problem equation

(3.31) has a weak solution. Since on a finite graph a solution is weak if and only if it is a strong solution,

our Ahmad, Lazer, and Paul resonant problem equation (3.31) has a strong solution.

61
 CHAPTER 4

VARIATIONAL METHODS ON LOCALLY FINITE GRAPHS

4.1

Introduction

In this chapter we study a nonlinear Schrödinger partial differential equations on uniformly locally finite

graphs. The particular equation we study is equation (4.16).

Section 4.2 gives definitions and defines the energy norm we will be working with.

Section 4.3 is original joint work between Dr. David Costa and me. In this section we give an original

compact embedding from E into lp (V ) for 2 ≤ p ≤ ∞ inspired by Theorem 2.1 in [47] by Zhang and Pankov.

This breakthrough compact embedding is necessary for our proofs in Section 4.6.

Section 4.4 goes over some Schrödinger applications and defines the nonlinear Schrödinger partial differential

equation that we are studying in this chapter.

In Section 4.5 we present the energy functional that corresponds to our Schrödinger partial differential

equation. Then we present its Fréchet derivative. This is an original application.

Section 4.6 is original joint work between Dr. David Costa and me. In this section we study an original

nonlinear Schrödinger problem on a locally finite graph using a equation (4.16). We do this by using and

original linking agument. The main original result in this section is Theorem 4.15. Theorem 4.8 along with

Lemmas 4.10, 4.11, 4.12, and 4.14 are original results used to help prove this main result. In particular

Lemmas 4.10 shows our functional satisfies the Palais-Smale condition and Lemma 4.12 shows our functional

satisfies the linking geometry.

62
 4.2

Definitions and Energy Norm

Recall that we defined locally finite graphs in definition 1.6. Now we will define what it means for a

graph to be uniformly locally finite.

Definition 4.1. A graph G(V,E) is uniformly locally finite if there exist a positive integer, U , such that

0 < deg(x) ≤ U ∀x ∈ V .

For example in the graph of Z × Z, which can be visualized as the Cartesian coordinate plane, each vertex

has degree 4 and thus U = 4 in this case. Also, in the graph of Z × Z × Z, which can be visualized as the

3D Cartesian coordinate plane, each vertex has degree 6 and thus U = 6 in this case. The vertex set V may

be infinite for both locally finite graphs and uniformly locally finite graphs. For a locally finite graph the

vertex set V may be uncountably infinite. However, for a uniformly locally finite graph the vertex set v is

at most countably infinite. Note this is drastically different from the graphs discussed in Chapter 3 where

all graphs have a finite vertex set.

In the next problem we consider some nonlinear Schrödinger partial differential equations on locally finite

graphs. So we will first define Schrödinger operators on locally finite graphs. Let Γ = (V, E) be a graph

with vertex set V and edge set E and suppose that there are no loops and multiple edges. Suppose further

that the graph is infinite, locally finite and connected. The distance function d(·, ·) on V is defined in the

standard way. Now we fix a function m : V → (0, ∞). Note that m generates a measure on V that is a

more general measure compared to µ(x) that we used in previous chapters. We suppose further that m is

bounded below by a positive constant m0 . Also we fix a function µxy : E → (0, ∞) where the value µxy

can be considered as the weight of the edge xy as shown in definition 1.3. Now we consider a more general

Laplacian (more precisely, the m-Laplacian) defined formally by

1 X
∆m f (x) = (f (y) − f (x))µxy .
m(x) y∼x

where f is a function on V . Let V : V → R be a function that is bounded below. In particular we take

V ≥ 1. We consider the Schrödinger operator formally defined by

Lf (x) = −∆m f (x) + V(x)f (x). (4.1)

Recall, by definition 1.3, µxy > 0 if and only if x ∼ y and µxy = 0 if and only if x and y are not adjacent.

63
 Let lp (V ) stand for the Banach space of all u : V → R such that ∥u∥plp (V ) = |u(x)|p m(x) < ∞, where
P
x∈V
p ∈ [1, ∞). l2 (V ) is a Hilbert space with the inner product (u, v) = u(x)v(x)m(x). l∞ (V ) stands for
P
x∈V
the space of bounded functions on V endowed with the sup-norm. We also denote by c0 the vector space

of finitely supported functions on V (note we use a nonstandard notation). Obviously, lq (V ) ⊂ lp (V ) if

1 ≤ q ≤ p ≤ ∞ , and this embedding is continuous. The inequality from this embeding is

∥u∥lp (V ) ≤ ∥u∥lq (V ) . (4.2)

The operator L is well-defined, symmetric operator on c0 with respect to the inner product (·, ·).

Furthermore, under our assumptions this operator is essentially self-adjoint in l2 (V ). Its unique self-adjoint

extension is still denoted by L. The (self-adjoint) operator L can be defined in terms of quadratic forms as

follows. The quadratic form

X X
q(u) = (u(x) − u(y))2 µxy + V(x)u2 (x)m(x), (4.3)
x,y∈V,x∼y x∈V

with the domain

 
 X X 
u ∈ l2 (V ) : (u(x) − u(y))2 µxy + V(x)|u2 (x)m(x) < ∞
 
x,y∈V,x∼y x∈V

is closed, and the domain is a Hilbert space E with respect to the norm

X X
∥u∥2E = (u(x) − u(y))2 µxy + V(x)u2 (x)m(x). (4.4)
x,y∈V,x∼y x∈V

The inner product associated with ∥u∥2E is

X X
(u, v)E = (u(x) − u(y))(v(x) − v(y))µxy + V(x)(x)u(x)v(x)m(x). (4.5)
x,y∈V,x∼y x∈V

It is important to point out that [40] uses 21 q(u) instead of q(u) in equation (4.3) as do many other papers

using this kind of quadratic form including but not limited to [28], [29] and [30]. We denote the vector space

of finitely supported functions on V by c0 . Furthermore, c0 is dense in E. Hence, q generates a self-adjoint

operator which is exactly the operator L.

If q is a quadratic form, then its associated bilinear form is denoted by q(u, v).

Our energy space can be defined as

 
 X 
E = u ∈ l2 (V ) : (u(x) − u(y))2 µxy < ∞ and V 1/2 u ∈ l2 (V ) (4.6)
 
x,y∈V,x∼y

64
where V ≥ 1. Note that V 1/2 u ∈ l2 (V ) means V 1/2 u l2 (V )
< ∞. In terms of the inner product of l2 (V )

this is equivalent to saying

2  
V 1/2 u = V 1/2 u, V 1/2 u < ∞.
l2 (V ) l2 (V )

Since V 1/2 is the square root of V and V 1/2 is the adjoint of V 1/2 , we have

 
V 1/2 u, V 1/2 u 2 = (Vu, u)l2 (V ) .
l (V )

Thus V 1/2 u ∈ l2 (V ) is equivalent to

X X
(Vu)(x)u(x)m(x) = (V)(x)u(x)u(x)m(x) < ∞.
x∈V x∈V

Therefore, V 1/2 u ∈ l2 (V ) is exactly

X
V(x)(u(x))2 m(x) < ∞. (4.7)
x∈V

So we can also define the energy space as

( )
X
E= u ∈ l2 (V ) : u ∈ dom(∆m ) and V(x)(u(x))2 m(x) < ∞ (4.8)
x∈V

where V ≥ 1. Recall that E is a Hilbert space. We introduce an equivalent version of the norm

1/2
∥u∥E = (−∆ + V) u = L1/2 u . (4.9)
l2 (V ) l2 (V )

We will be using this version of the energy norms in some later proofs of the Palais-Smale condition and

linking geometry. The majority of the theorems and proofs in this chapter are motivated by theorems and

proofs in [48] by Zhang and Pankov.

4.3

Spectrum and Compact Embedding

Throughout the rest of the text we make the following assumption: ”The spectrum σ(L) of the operator

L is purely discrete, i.e., consists of isolated eigenvalues of finite multiplicity.” Let λ1 < λ2 < · · · < λk <

· · · → ∞ be distinct eigenvalues of L. Due to our assumption on V, 1 ≤ λ1 .

A necessary and sufficient condition (not that efficient) for the discreteness of the spectrum is given in

[40]. Note also that the discreteness of the spectrum is equivalent to the compactness of the embedding
2 p
E ⊂ lm (hence, E ⊂ lm for 2 ≤ p). Therefore, Lemma 2.2 of [28] provides a sufficient condition for our

65
assumption given above. We should also mention that Lemma 2.1 of [28] can be useful to the study of

bifurcation of solutions.

Now we define what it means for an function to be coercive on a graph. We will need this definition to

prove some theorems on the compactness of the embedding.

Definition 4.2. An function h : V → R is said to be coercive on a graph if for all real numbers M > 0,

there exists a real number RM > 0 such that for all x where d(x0 , x) > RM , where x0 ∈ V is a fixed vertex,

normally called the origin, then h(x) > M .

A similar definition holds for a self-adjoint operator A : H → H, where H is a real Hilbert Space, is called

coercive if there exists a constant c > 0 such that (Ax, x) ≥ c∥x∥2 for all x ∈ H. A mapping f : X → Y

between two normed vector space is coercive if and only if ∥f (x)∥ → ∞ as ∥x∥ → ∞. Note that for the rest

of this dissertation when we say V is coercive we mean all three of the definitions hold.

Lemma 4.1. If A is coercive, then A1/2 is coercive.

Proof. Let A be coercive. Since A is coercive, A1/2 is the square root of A, and A1/2 is self-adjoint, then

2  
A1/2 u = A1/2 u, A1/2 u = (Au, u) ≥ c∥u∥2 .

Thus

A1/2 u → ∞ as ∥u∥ → ∞.

Therefore, A1/2 is coercive.

Lemma 4.2. If A is coercive, then A−1 is compact.

Note we stated Lemma 4.2 without proof as it is a well-known property.

Next we prove an original compact embedding from E into lp (V ) for 2 ≤ p ≤ ∞ inspired by Theorem

2.1 in [47] by Zhang and Pankov.

Theorem 4.3. If V is coercive, then

(1) for any 2 ≤ p ≤ ∞, the embedding map form E into lp (V ) is compact. Denote the best embedding
1
constant Cp = max .
∥u∥lp (v) =1 ∥u∥E

(2) the spectrum σ(L) is discrete, i.e., it consists of eigenvalues of finite multiplicity.

66
Proof. Note that L−1/2 is a bounded operator form l2 (V ) to E and V 1/2 is a operator for E to l2 (V ). Hence

L−1/2 V 1/2 is an operator from E to E. For any u ∈ E we have

   
L−1/2 V 1/2 u, u = V 1/2 u, L−1/2 u ≤ V 1/2 u L−1/2 u ≤ C∥u∥2E .
l2 (V ) l2 (V )

The first inequality comes form the Cauchy–Schwarz inequality. The second inequality uses the definition

of E and the boundedness of L−1/2 . Thus L−1/2 V 1/2 is a bounded operator from E to E. Since V is

coercive, then V 1/2 is coercive by Lemma 4.1. Since V 1/2 is coercive, then V −1/2 is compact by Lemma 4.2.

Therefore, L−1/2 = L−1/2 V 1/2 V −1/2 is a compact from l2 (V ) to E. The compact embedding from E into

l2 (V ) follows from this fact. The embedding from l2 (V ) into lP (V ) is continuous. Thus part (1) holds.

Note that L−1 = (I + L−1 ∆)V −1 or L−1 = L−1/2 L−1/2 . Then by part (1) we have L−1 is compact. Note

that part (2) is a consequence of the usual spectral theory of a compact operator.

4.4

Schrödinger Applications

In [48] Pankov and Zhang study the one-dimensional discrete nonlinear Schrödinger equation

iψ˙n + ∆ψn − vn ψn + σγn f (ψn ) = 0, n ∈ Z (4.10)

where σ = ±1 and

∆ψn = ψn+1 − 2ψn + ψn−1 (4.11)

is the discrete Laplacian operator. In equation (4.10) σ characterizes the focusing property of the equation.

When σ = 1, the equation is self-focusing and when σ = −1 the equation is defocusing [48]. Also in equation

(4.10) the nonlinearity f is to be a gauge invariant complex valued function of complex variables [48]. Note

this means f (eiω u) = eiω f (u) for any ω ∈ R [48].

Then Pankov looked at special solutions, called standing wave solutions or breather solutions, to equation

(4.10) where ψn = e−itω un , where un is a sequence of complex numbers such that

lim un = 0.
n→±∞

Inserting this into equation (4.10) we get the infinite nonlinear system of equations

−(∆un ) + vn un − ωun − σγn f (un ) = 0, n ∈ Z. (4.12)

67
By letting H = −∆ + v equations (4.12). Becomes

Hun − ωun − σγn f (un ) = 0. (4.13)

Next we prove the standing wave equation ψn = e−itω un is also a solution to equation (4.12) using

equation (4.10). Pankov stated this in [48] but did not prove it. However, we prove it here for educational

purposes and for the sake of completion. By substituting ψn = e−itω un into equation (4.10) we see it becomes

˙ un + ∆e−itω un − vn e−itω un + σγn f (e−itω un ) = 0.

ie−itω

Taking the derivative of the first term we get

i(−iω)e−itω un + e−itω u˙n + ∆e−itω un − vn e−itω un + σγn f (e−itω un ) = 0.

Since un is a sequence of real number u˙n = 0 ∀n ∈ Z and −i2 = 1. So we get

ωe−itω un + ∆e−itω un − vn e−itω un + σγn f (e−itω un ) = 0.

Note that the nonlinearity of f is supposed to be a gauge invariant complex valued function of complex

variables. This means f (e−itω un ) = e−itω f (un ). So we have

ωe−itω un + ∆e−itω un − vn e−itω un + σγn e−itω f (un ) = 0.

Finally by factoring out −e−itω and rearranging the terms we get.

−e−itω (−(∆un ) + vn un − ωun − σγn f (un )) = 0, n ∈ Z.

Therefore, ψn = e−itω un is also a solution to equation (4.12).

In [46] Pankov studies the discrete nonlinear Schrödinger equation

iψ˙n + ∆ψn − vn ψn + γn |ψn |p−2 ψn = 0, n ∈ Z (4.14)

where

∆ψn = ψn+1 + ψn−1 − 2ψn (4.15)

is the discrete one-dimensional Laplacian operator. In equation (4.14) vn and γn are real valued sequences,

vn is the potential parameter and γn is the anharmonic parameter.

Equation (4.10) is a very important discrete linear model that can be used for a number of topics ranging

from solid-state and condensed-matter physics to biology [3][22][23][24][25][31].

68
 Now we will state the nonlinear Schrödinger equation that we will be studying which is a generalization

of equation (4.12). The equation we will be studying is

−∆m u(x) + V(x)u(x) − λu(x) = κf (x, u(x)), x ∈ V. (4.16)

using equation (4.1) we get

Lu(x) − λu(x) = κf (x, u(x)), x ∈ V. (4.17)

Here, κ = ±1 is responsible for self-focusing when κ = 1 or defocusing when κ = −1 character of the

nonlinearity, and λ is a real parameter. Note that the κ in equation (4.16) corresponds to σ in equation

(4.12) and λ in equation (4.16) corresponds to ω in equation (4.12). Due to the presence of λ, the previously

imposed assumption 1 ≤ V(x) is not restrictive. Also note that L in equation (4.17) corresponds to H in

equation (4.13).

4.5

Energy Functional

We assume that, for every x ∈ V , f (x, u) is a continuous function of u ∈ R and f (x, 0) = 0. Let

Z u
F (x, u) = f (x, s)ds. (4.18)
0

Observe that the functional we will use is Φ ∈ C 1 (E, R) defined by

1 X
Φ(u) = ((L − λ)u, u)l2 (V ) − κ F (x, u(x))m(x). (4.19)
2
x∈V

Hence,

1 X
Φ′ (u) · v = ((L − λ)u, v)l2 (V ) − κ f (x, u(x))v(x)m(x). (4.20)
2
x∈V

One can take the inner product with respect to ∥u∥E of both sides of equation (4.16) with u(x) we get

(−∆m u(x), u(x))E + (V(x)u(x), u(x))E − (λu(x), u(x))E = κ(f (x, u(x)), u(x))E . (4.21)

Hence,

(Lu, u)E − λ∥u∥2E = κ(f (x, u(x)), u(x))E . (4.22)

Instead we take the inner product with respect to ∥u∥l2 (V ) of both sides of equation (4.16) with u(x) we get

(−∆m u(x), u(x))l2 (V ) + (V(x)u(x), u(x))l2 (V ) − (λu(x), u(x))l2 (V ) = κ(f (x, u(x)), u(x))l2 (V ) . (4.23)

69
Motivated by the idea of weak solutions we use this inner product in equation (4.23) for computations in

the remainder of this dissertation. Equation (4.23) defines what it means to be a weak solution in our space.

For the remainder of this dissertation we will only be concerned with the self-focusing situation when κ = 1.

Now consider,
  X 1 
1 1 1
Φ(u) − Φ′ (u) · u = − ((L − λI)u, u)l2 (V ) + κ f (x, u)u − F (x, u) m(x). (4.24)
q 2 q q
x∈V

This is equivalent to,

  X 1 
1 1 1 
Φ(u) − Φ′ (u) · u = − (Lu, u)l2 (V ) − λ∥u∥2l2 (V ) + κ f (x, u)u − F (x, u) m(x). (4.25)
q 2 q q
x∈V

Letting κ = 1 this becomes

   X 1 
1 1 1
Φ(u) − Φ′ (u) · u = − (Lu, u)l2 (V ) − λ∥u∥2l2 (V ) + f (x, u)u − F (x, u) m(x). (4.26)
q 2 q q
x∈V

Using the (AR) condition that there exists a constant 2 < q such that 0 < qF (x, u) ≤ f (x, u)u, s ̸= 0 for all

x ∈ V , yields,
 
1 1 1 
Φ(u) − Φ′ (u) · u ≥ − (Lu, u)l2 (V ) − λ∥u∥2l2 (V ) . (4.27)
q 2 q

Since (Lu, u)l2 (V ) ≥ λ1 ∥u∥2l2 (V ) with λ1 > 0 then

   1 1
1 1 1
Φ(u) − Φ′ (u) · u ≥ − λ1 ∥u∥2l2 (V ) −λ∥u∥2l2 (V ) = − (λ1 − λ)∥u∥2l2 (V ) . (4.28)
q 2 q 2 q

4.6

A Nonlinear Schrödinger Problem on a Locally Finite Graph

Here we list some basic assumptions on the nonlinearity of f (x, u).

Assumption 4.4. Assume that f (x, s) ∈ C(R).

Assumption 4.5. There exists positive constant C1 and 2 < p < ∞ such that

|f (x, s)| ≤ C1 |s| + |s|p−1 .

(4.29)

Assumption 4.6. Assume that f is superlinear near zero, this means

f (s)
lim = 0. (4.30)
u→0 |s|

70
Assumption 4.7. There is a constant 2 < q < ∞ such that

0 ≤ qF (x, s) ≤ f (x, s)s, ∀s ̸= 0

where
Z u(x)
F (x, u(x)) = f (x, s)ds.
0

Note that Assumption 4.7 is the (AR) condition form Theorem 4.8.

From equations (4.29) and (4.30) it is easy to show that for any ϵ > 0, There exists A = A(ϵ) > 0 such

that for any u(x) ∈ R

f (x, u(x))u(x) ≤ ϵ|u(x)|2 + A|u(x)|p (4.31)

and
ϵ A
F (x, u(x)) ≤ |u(x)|2 + |u(x)|p . (4.32)
2 p

An elementary differential inequality argument argument shows for every ϵ > 0 there exists C2 > 0 such that

F (x, u(x))u(x) ≥ C2 |u(x)|q − ϵ|u(x)|2 , ∀u(x) ∈ R. (4.33)

Combining equations (4.29) and (4.30), we can conclude that 2 < q ≤ p.

Theorem 4.8. Suppose that V is a locally finite graph, u : V → R is an unknown function f (x, u) is a

continuous function of u ∈ R and f (x, 0) = 0. Assume Assumptions 4.4, 4.5, 4.6, and 4.7 are satisfied. Also

suppose that κ = 1 and λ ≤ 0.

Then Φ(u) satisfies the Palais-Smale condition.

Proof. Suppose that un is a sequence in E such that |Φ(un )| ≤ C for some C > 0 and lim Φ′ (un ) = 0. So
n→∞
′
then for all n sufficiently large enough, we have |Φ (un ) · un | ≤∥un ∥E . Then
  X 1 
1 1 1 
Φ(un ) − Φ′ (un ) · un = − ∥un ∥2E −λ∥un ∥2l2 (V ) + f (x, u)u − F (x, u) m(x)
q 2 q q
x∈V
 
1 1 
≥ − ∥un ∥2E −λ∥un ∥2l2 (V )
2 q
 
1 1
≥ − ∥un ∥2E
2 q

Hence,
1 1
∥un ∥2E ≤ Φ(un ) − Φ′ (un ) · un ≤ C + ∥un ∥E . (4.34)
q q

71
Thus, un is bounded in E. Since un is bounded in E and E is a reflexive Banach space, then un has a

subsequence uk weakly converging to u in E (a theorem of Mazur). By compact embedding we have E is

compactly embedded in lp (V ). So for any 2 ≤ p ≤ ∞

uk → u in lp (V ). (4.35)

We compute

X
∥uk − u∥2E = Φ′ (uk − u) · (uk − u) + λ∥uk − u∥2l2 (V ) +κ f (x, uk (x) − u(x))(uk (x) − u(x))m(x).
x∈V

The Palais-Smale assumptions makes the first term on the right-hand side of the above equation converges

to zero. The compact embedding of E into l2 (V ) ensures that the second term on the right-hand side of the

above equation converges to zero. All that remains is to show that the third term on the right-hand side

of the above equation converges to zero. By equation (4.31) and Hölder’s inequality we know there exists

C˜5 > 0 and C˜6 > 0 such that

X X
ϵ|uk | + A|un |p−1 |(un − u)| m(x)


f (un )(un − u)m(x) ≤
x∈V x∈V

≤ C˜5 ∥uk ∥l2 (V ) ∥uk − u∥l2 (V ) + C˜6 ∥uk ∥lp−1

p (V ) ∥uk − u∥lp (V )

So by the compact embedding of E in both l2 (v) and lp (V ) we have ∥un ∥2l2 (V ) ≤ C2E ∥un ∥2E and ∥un ∥2lp (V ) ≤

CpE ∥un ∥2E . Thus uk is bounded in both l2 (V ) and lp (V ). Using this and equation (4.44) the third term on

the right hand side converges to zero. Hence uk → u in E. Thus uk is a convergent subsequence of un in E.

Therefore, Φ(u) satisfies the Palis-Smale condition.

Theorem 4.8 is as much as we can prove using a Palais-Smale argument without using the fact that the

function V : V → R used in equation (4.1) and equation (4.16) is coercive.

Definition 4.3. Fix x0 ∈ V , normally the origin. Let R ∈ R. The ball

B(x0 ) = {x ∈ V : d(x, x0 ) ≤ R}.

In the following Palais-Smale and linking arguments we will assume V : V → R is coercive. Recall that

λ1 < λ2 < · · · < λk < · · · → ∞ be distinct eigenvalues of the function L defined in equation (4.1), due to

our assumption on V, 1 ≤ λ1 . Let ϕk be the associated normalized eigenfunction with λk for each k, That is

Lϕk = λk ϕk , ∥ϕk ∥l2 (V ) = 1. (4.36)

72
Further note that, {ϕk : k = 1, 2, 3, . . .} is an orthonormal basis of l2 (V ).

For any λ ≥ λ1 there exists a unique k such that λ ∈ [λk , λk+1 ). Let

Y = Span{ϕ1 , ϕ2 , ϕ3 , . . . , ϕk }, dim Y = k < ∞,

for λ < λ1 , we set Y = {0}. Then the Hilbert space E can be decomposed into the direct sum

∥·∥E
E = Y ⊕ Z, Z = Y ⊥ = Span{ϕj : j ≥ k + 1} .

Observe that the linking geometry becomes the mountain pass geometry when Y = {0}. This means the

Mountain Pass Theorem is one of the cases of the Linking Theorem.

Next we follow the typical Linking geometry setup used in [36] and [57]. Let ρ > r > 0 and z ∈ Z such

that ∥z∥E = 1. Then we define the following sets,

M = {u = y + λz : ∥u∥E ≤ ρ, λ ≥ 0, y ∈ Y }

∂M = M0 = {u = y + λz : y ∈ Y, ∥u∥E = ρ and λ ≥ 0 or ∥u∥E ≤ ρ and λ = 0}

N = {u ∈ Z : ∥u∥E = r}

Next we state the relevant linking theorem for the problem at hand.

Theorem 4.9. [48][50] Let Φ(u) ∈ C 1 (E, R) and assume that Φ satisfies the Palais-Smale condition. Assume

also that Φ possesses the following linking geometry

β = inf Φ(u) > sup Φ(u) = α. (4.37)

u∈N u∈∂M

Let Γ = {γ ∈ C(M, E) : γ = id on ∂M }. Then

c = inf sup Φ(γ(u)) (4.38)

γ∈Γ u∈M

is a critical value of Φ and

β ≤ c ≤ sup Φ(u). (4.39)

u∈M

In order to show Φ(u) satisfies the hypothesis of Theorem 4.9 we need to prove two lemmas. The first

lemma will show that Φ(u) satisfies the Palais-Smale condition when κ = 1. The second lemma, Lemma

4.12, will show that Φ(u) possesses the linking geometry when κ = 1.

Lemma 4.10. Assume Assumptions 4.4, 4.5, 4.6, and 4.7 are satisfied and the function V is coercive. For

κ = 1 and λ ∈ R, Φ(u) satisfies the Palis-Smale condition.

73
Proof. Suppose that un is a sequence in E such that |Φ(un )| ≤ B for some B > 0 and lim Φ′ (un ) = 0.
n→∞

Then, for all n sufficiently large enough, we have

Φ(un ) − βΦ′ (un ) · un ≤ B + β∥un ∥. (4.40)

The coercivity of V(x) implies there exists and R > 0 such that V(x) ≥ 2λ whenever d(x, x0 ) > R. Thus

X X
∥u∥2l2 (v) = (u(x))2 m(x) + (u(x))2 m(x). (4.41)
x∈B(x0 ) x∈V \B(x0 )

Recall equation (4.4)

X X
∥u∥2E = (u(x) − u(y))2 µxy + V(x)u2 (x)m(x).
x,y∈V,x∼y x∈V

(u(x) − u(y))2 µxy is positive we can see

P
Since
x,y∈V,x∼y
X X
∥u∥2E = (u(x) − u(y))2 µxy + V(x)u2 (x)m(x)
x,y∈V,x∼y x∈V
X
≥ V(x)u2 (x)m(x).
x∈V

Then using equation (4.41) we get

X X
∥u∥2E ≥ V(x)(u(x))2 m(x) + V(x)(u(x))2 m(x).
x∈B(x0 ) x∈V \B(x0 )

V(x)(u(x))2 m(x) is positive we get

P
Since
x∈B(x0 )
X
∥u∥2E ≥ V(x)(u(x))2 m(x).
x∈V \B(x0 )

Then since V(x) ≥ 2λ whenever d(x, x0 ) > R we get

X 1 X
∥u∥2E ≥ 2λ (u(x))2 m(x) =⇒ ∥u∥2E ≥ (u(x))2 m(x).
2λ
x∈V \B(x0 ) x∈V \B(x0 )

Hence equation (4.41) can be written as

X 1
∥u∥2l2 (v) ≤ (u(x))2 m(x) + ∥u∥2E . (4.42)
2λ
x∈B(x0 )

We start by showing ∥un ∥E is bounded.

Let κ = 1. Note that the case for λ ≤ 0 was handled in Theorem 4.8. So this proof will only talk about
1
the case when λ > 0. Take q < β < 21 .

For any ϵ > 0

0 < K(ϵ) = (u(x))2 − ϵ|(u(x))|q l∞ (B(x0 ))

< ∞. (4.43)

74
Thus

X X
(u(x))2 m(x) = (u(x))2 − ϵ|(u(x))|q + ϵ|(u(x))|q m(x)

x∈B(x0 ) x∈B(x0 )
X X
(u(x))2 − ϵ|(u(x))|q m(x) + (ϵ|(u(x))|q ) m(x)


=
x∈B(x0 ) x∈B(x0 )
X X
2 q
≤ (u(x)) − ϵ|(u(x))| l∞ (B(x0 ))
m(x) + (ϵ|(u(x))|q ) m(x)
x∈B(x0 ) x∈B(x0 )
X X
2 q
≤ (u(x)) − ϵ|(u(x))| l∞ (B(x0 ))
m(x) + (ϵ|(u(x))|q ) m(x)
x∈B(x0 ) x∈B(x0 )
X
q
= CK(ϵ) + (ϵ|(u(x))| ) m(x)
x∈B(x0 )

≤ CK(ϵ) + ϵ∥u(x)∥qlq (V ) ,

P
where 0 < C = m(x) < ∞ because B(x0 ) has finitely many vertices.
x∈B(x0 )
1


Let Tϵ = 2 − β λ + (βq − 1) ϵ. Then
 
1  X
Φ(u) − βΦ′ (u) · u = −β ∥u∥2E −λ∥u∥2l2 (V ) + (βf (x, u)u − F (x, u)) m(x)
2
x∈V
 
1  X
≥ −β ∥u∥2E −λ∥u∥2l2 (V ) + (βq − 1) F (x, u)m(x)
2
x∈V
 
1  X
∥u∥2E −λ∥u∥2l2 (V ) + (βq − 1) C2 |u|q − ϵ|u|2 m(x)


≥ −β
2
x∈V
 
1   
≥ −β ∥u∥2E −λ∥u∥2l2 (V ) + (βq − 1) C2 ∥u∥qlq (V ) − ϵ∥u∥2l2 (V )
2
    
1 1
= 2
− β ∥u∥E − − β λ + (βq − 1) ϵ ∥u∥2l2 (V ) + (βq − 1)C2 ∥u∥qlq (V )
2 2
 
1
= − β ∥u∥2E −Tϵ ∥u∥2l2 (V ) + (βq − 1)C2 ∥u∥qlq (V )
2
 
 
1 X 1
≥ − β ∥u∥2E −Tϵ  (u(x))2 m(x) + ∥u∥2E  + (βq − 1)C2 ∥u∥qlq (V )
2 2λ
x∈B(x0 )
1


2 −β λ
   
1 2 1 X
2
= − β ∥u∥E − −β λ (u(x)) m(x) − ∥u∥2E
2 2 2λ
x∈B(x0 )
X (βq − 1) ϵ
− (βq − 1) ϵ ∥u∥2E +(βq − 1)C2 ∥u∥qlq (V )
(u(x))2 m(x) −
2λ
x∈B(x0 )
     X
1 − 2β (βq − 1)ϵ 2 1
= − ∥u∥E − − β λ + (βq − 1) ϵ (u(x))2 m(x)
4 2λ 2
x∈B(x0 )

+ (βq − 1)C2 ∥u∥qlq (V )

 
1 − 2β (βq − 1)ϵ X
= − ∥u∥2E −Tϵ (u(x))2 m(x) + (βq − 1)C2 ∥u∥qlq (V )
4 2λ
x∈B(x0 )
 
1 − 2β (βq − 1)ϵ  
≥ − ∥u∥2E −Tϵ CK(ϵ) + ϵ∥u(x)∥qlq (V ) + (βq − 1)C2 ∥u∥qlq (V )
4 2λ

75
  
1 − 2β (βq − 1)ϵ
= − ∥u∥2E −Tϵ CK(ϵ) + ((βq − 1)C2 − Tϵ ϵ) ∥u∥qlq (V ) .
4 2λ

For small enough ϵ > 0 there exists C˜1 > 0 and C˜2 > 0 such that
 
1 − 2β
Φ(u) − βΦ′ (u) · u ≥ ∥u∥2E +C˜1 ∥u∥qlq (V ) − C˜2
4
 
1 − 2β
≥ ∥u∥2E −C˜2 .
4

Using this and equation (4.40) when n is sufficiently large we get

 
1 − 2β
′
B + β∥un ∥ ≥ Φ(un ) − βΦ (un ) · un ≥ ∥un ∥2E −C˜2 .
4

Therefore ∥un ∥E is bounded.

Next we will show ∥un ∥E has a convergent subsequences. Since un is bounded in E and E is a reflexive

Banach space, then un has a subsequence uk weakly converging to u in E (a theorem of Mazur). By compact

embedding we have E is compactly embedded in lp (V ). So for any 2 ≤ p ≤ ∞

uk → u in lp (V ). (4.44)

We compute

X
∥uk − u∥2E = Φ′ (uk − u) · (uk − u) + λ∥uk − u∥2l2 (V ) +κ f (x, uk (x) − u(x))(uk (x) − u(x))m(x).
x∈V

The Palais-Smale assumptions makes the first term on the right-hand side of the above equation converges

to zero. The compact embedding of E into l2 (V ) ensures that the second term on the right-hand side of the

above equation converges to zero. All that remains is to show that the third term on the right-hand side

of the above equation converges to zero. By equation (4.31) and Hölder’s inequality we know there exists

C˜5 > 0 and C˜6 > 0 such that

X X
ϵ|uk | + A|un |p−1 |(un − u)| m(x)


f (un )(un − u)m(x) ≤
x∈V x∈V

≤ C˜5 ∥uk ∥l2 (V ) ∥uk − u∥l2 (V ) + C˜6 ∥uk ∥lp−1

p (V ) ∥uk − u∥lp (V )

So by the compact embedding of E in both l2 (v) and lp (V ) we have ∥un ∥2l2 (V ) ≤ C2E ∥un ∥2E and ∥un ∥2lp (V ) ≤

CpE ∥un ∥2E . Thus uk is bounded in both l2 (V ) and lp (V ). Using this and equation (4.44) the third term on

the right hand side converges to zero. Hence uk → u in E. Thus uk is a convergent subsequence of un in E.

Therefore, Φ(u) satisfies the Palais-Smale condition.

76
 Next we state the definition of the Nehari manifold as given in [57] and [47]. We need the Nehari manifold

to help prove our next lemma which handles the λ < λ1 case needed for the proof of Lemma 4.12. Take Φ

as defined by equation (4.19) with Φ′ (0) = 0. A necessary condition for u ∈ E to be a critical point of Φ is

that Φ′ (u) · u = 0. We use this condition to define the Nehari manifold

N = {u ∈ E : Φ′ (u) · u = 0, u ̸= 0} . (4.45)

Note that a critical point u ̸= 0 of Φ is called a ground state critical point or least energy critical point if

Φ(u) = inf Φ.
N

To see a problem using Nehari Manifold look at [47] by Zhang and Pankov. The locally finite graph analog

of this problem remains open and is a future research objective.

Lemma 4.11. Suppose κ = 1 and λ < λ1 . Then there are constants C, D > 0 so that for any u ∈ N

∥u∥E ≥ C and Φ(u) ≥ D.

Proof. Let κ = 1. For any u ∈ N with Φ′ (u) · u = 0 observe that

1 X
0 = Φ′ (u) · u = ((L − λ)u, u)l2 (V ) − f (x, u)um(x)
2
x∈V
1 λ X
= (Lu, u)l2 (V ) − ∥u∥2l2 (V ) − f (x, u)um(x)
2 2
x∈V
1 λ X
= ∥u∥2E − ∥u∥2l2 (V ) − f (x, u)um(x)
2 2
x∈V
1  X
= ∥u∥2E − λ∥u∥2l2 (V ) − f (x, u)um(x).
2
x∈V

Thus
X 1 
f (x, u)um(x) = ∥u∥2E − λ∥u∥2l2 (V ) .
2
x∈V

By solving for ∥u∥2E we get

X
∥u∥2E = λ∥u∥2l2 (V ) + 2 f (x, u)um(x).
x∈V

So we have

1 X
Φ(u) = ((L − λ)u, u)l2 (V ) − F (x, u)m(x)
2
x∈V
1 λ X
= (Lu, u)l2 (V ) − ∥u∥2l2 (V ) − F (x, u)m(x)
2 2
x∈V

77
 1 λ X
= ∥u∥2E − ∥u∥2l2 (V ) − F (x, u)m(x)
2 2
x∈V
1  X
= ∥u∥2E − λ∥u∥2l2 (V ) − F (x, u)m(x)
2
x∈V
1  X1
≥ ∥u∥2E − λ∥u∥2l2 (V ) − f (x, u)um(x)
2 q
x∈V
1  1X
= ∥u∥2E − λ∥u∥2l2 (V ) − f (x, u)um(x)
2 q
x∈V
1  1  
= ∥u∥2E − λ∥u∥2l2 (V ) − ∥u∥2E − λ∥u∥2l2 (V )
2 2q
   
1 1 1 1
= − ∥u∥2E − − λ∥u∥2l2 (V ) .
2 2q 2 2q

Now consider two cases.

Case 1: Let λ ≤ 0. Since λ ≤ 0, then we have

X
∥u∥2E = λ∥u∥2l2 (V ) − 2 f (x, u)um(x)
x∈V
X
ϵ|u|2 + A|u|p m(x)


≤2
x∈V

= 2ϵ∥u∥2l2 (V ) + 2A∥u∥plp (V )

≤ 2ϵC22 ∥u∥2E + 2ACPP ∥u∥pE .

Thus we see

1 − 2ϵC22 ∥u∥2E ≤ 2ACPP ∥u∥pE .

1
Let ϵ = 4C22
. Hence
   
1
2ϵC22 ∥u∥2E 2ACPP ∥u∥pE C22 ∥u∥2E ≤ 2ACPP ∥u∥pE


1− ≤ =⇒ 1−2
4C22
1
=⇒ ∥u∥2E ≤ 2ACPP ∥u∥pE
2
1 p−2
=⇒ ≤ 2ACPP ∥u∥E
2
1
=⇒ ≤ ∥u∥p−2
E
4ACPP
 1
 p−2
1
=⇒ ≤ ∥u∥E
4ACPP

1
=⇒ 4ACPP 2−p ≤ ∥u∥E

Thus
1
∥u∥E ≥ 4ACPP

2−p
= C > 0.

78
Therefore,
   
1 1 1 1
Φ(u) ≥ − ∥u∥2E − − λ∥u∥2l2 (V )
2 2q 2 2q
 
1 1
≥ − ∥u∥2E
2 2q
 
1 1
≥ − C2
2 2q
= D > 0.

Case 2: Let 0 < λ < λ1 . Since λ1 is the smallest eigenvalue of L, then by the definition of Cp we get
1
λ1 = C22
. Then for any ϵ > 0 we have

X
∥u∥2E = λ∥u∥2l2 (V ) − 2 f (x, u)um(x)
x∈V
X
≤ λ∥u∥2l2 (V ) + 2 ϵ|u|2 + A|u|p m(x)

x∈V

= λ∥u∥2l2 (V ) + 2ϵ∥u∥2l2 (V ) + 2A∥u∥plp (V )

≤ λC22 ∥u∥2E + 2ϵC22 ∥u∥2E + 2ACpp ∥u∥pE .

1
By letting λ1 = C22
we see

∥u∥2E ≤ λC22 ∥u∥2E + 2ϵC22 ∥u∥2E + 2ACpp ∥u∥pE =⇒ ∥u∥2E ≤ λC22 ∥u∥2E + 2ACpp ∥u∥pE

=⇒ ∥u∥2E − λC22 ∥u∥2E ≤ 2ACpp ∥u∥pE

=⇒ 1 − λC22 ∥u∥2E ≤ 2ACpp ∥u∥pE

p−2
=⇒ 1 − λC22 ≤ 2ACpp ∥u∥E
1 − λC22 p−2
=⇒ ≤ ∥u∥E
2ACpp
 1
1 − λC22 p−2

=⇒ ≤ ∥u∥E
2ACpp
1
2ACpp
  2−p
=⇒ ≤ ∥u∥E
1 − λC22
! 1
2ACpp 2−p
=⇒ ≤ ∥u∥E
1 − λλ1

Thus 1
! 2−p
2ACpp
∥u∥E ≥ λ
= C > 0.
1− λ1

Therefore,
   
1 1 1 1
Φ(u) ≥ − ∥u∥2E − − λ∥u∥2l2 (V )
2 2q 2 2q

79
    
1 1 1 1
≥ − ∥u∥2E − − λC22 ∥u∥2E
2 2q 2 2q
 
1 1
1 − λC22 ∥u∥2E


= −
2 2q
 
1 1
1 − λC22 C 2


≥ −
2 2q
  
1 1 λ
≥ − 1− C2
2 2q λ1
= D > 0.

The next lemma will show that Φ(u) satisfies the linking geometry when κ = 1 needed for Theorem

4.9 . The proof of this lemma makes heavy use of the Pythagorean formula from functional analysis. The

Pythagorean formula states that if x1 , x2 , x3 , . . . , xn are orthogonal vectors in an inner product space, then
n 2 n
X X 2
xk = ∥xk ∥ . (4.46)
k=1 k=1

Lemma 4.12. There exists two positive constants ρ > r > 0 such that

inf Φ(u) > 0 ≥ sup Φ(u). (4.47)

u∈N u∈∂M

Proof. Let κ = 1. Since the case for λ < λ1 has been demonstrated by Lemma 4.11, so we will consider
k
P ∞
P
λk ≤ λ < λk+1 for some k ≥ 1. Let y = ai ϕi ∈ Y and z = bi ϕi ∈ Z with ∥z∥E = 1, that is
i=1 i=k+1

∞
X
L1/2 z = 1 ⇐⇒ λi b2i = 1.
l2 (V )
i=k+1

By using equation (4.46) we can see

k
X
∥y∥2E = λi a2i
i=1

and
k
X
∥y + ηz∥2E = λi a2i + η 2 .
i=1
∞
P
Let u = βi ϕi ∈ Z, and recall
i=k+1

1 X
Φ(u) = ((L − λI) u, u) − κ F (x, u)m(x). (4.48)
2
x∈V

Since we are ony considering the case when κ = 1, equation (4.48) becomes

1 X
Φ(u) = ((L − λI) u, u) − F (x, u)m(x). (4.49)
2
x∈V

80
 ∞
P
Since u = βi ϕi ∈ Z, we can see
i=k+1

∞ ∞
!
1 1 X X
((L − λI) u, u) = (L − λI) βi ϕ i , βi ϕ i
2 2
i=k+1 i=k+1
∞ ∞
!
1 X X
= (βi Lϕi − λβi ϕi ) , βi ϕ i
2
i=k+1 i=k+1
∞ ∞
!
1 X X
= (βi λi ϕi − λβi ϕi ) , βi ϕ i
2
i=k+1 i=k+1
∞
1 X
βi2 λi − λβi2


=
2
i=k+1
∞
1 X
= (λi − λ) βi2 .
2
i=k+1

Thus equation (4.49) becomes

∞
1 X X
Φ(u) = (λi − λ) βi2 − F (x, u)m(x). (4.50)
2
i=k+1 x∈V

For any ϵ > 0, there exists A = A(ϵ) > 0 such that for any u ∈ R

0 ≤ F (x, u) ≤ ϵ|u|2 + A|u|p =⇒ −F (x, u) ≥ − ϵ|u|2 + A|u|p

Thus equation (4.50) becomes

∞
1 X X
(λi − λ) βi2 − ϵ|u|2 + A|u|p m(x).


Φ(u) ≥ (4.51)
2
i=k+1 x∈V

Since, ∥u∥2l2 (V ) = |u|2 m(x) and ∥u∥plp (V ) = |u|p m(x), we can see
P P
x∈V x∈V
X
ϵ|u|2 + A|u|p m(x) = ϵ∥u∥2l2 (V ) + A∥u∥plp (V ) .

x∈V

∞
P
Then, becuase u = βi ϕi ∈ Z, we get
i=k+1

∞ 2 ∞ p
X X
ϵ∥u∥2l2 (V ) + A∥u∥plp (V ) =ϵ βi ϕ i +A βi ϕ i .
i=k+1 l2 (V ) i=k+1 lp (V )

Then, by equation (4.46),

∞ 2 ∞
X X
βi ϕ i = ∥βi ϕi ∥2l2 (V ) .
i=k+1 l2 (V ) i=k+1

Then we see
∞
X ∞
X
∥βi ϕi ∥2l2 (V ) = |βi |2 ∥ϕi ∥2l2 (V ) .
i=k+1 i=k+1

81
Given ∥ϕi ∥2l2 (V ) = 1 and inequality (4.2) we get
 p/2
∞ 2 ∞ p ∞ 2 ∞ 2
X X X X
ϵ βi ϕ i +A βi ϕ i =ϵ βi ϕ i + A βi ϕ i 
i=k+1 l2 (V ) i=k+1 lp (V ) i=k+1 l2 (V ) i=k+1 lp (V )
 p/2
∞ 2 ∞ 2
X X
≤ϵ βi ϕ i + A βi ϕ i 
i=k+1 l2 (V ) i=k+1 l2 (V )
∞ ∞
!p/2
X 2
X 2
=ϵ ∥βi ϕi ∥l2 (V ) +A ∥βi ϕi ∥l2 (V )
i=k+1 i=k+1
∞ ∞
!p/2
X X
=ϵ βi2 + A βi2 .
i=k+1 i=k+1

Therefore, equation (4.51) becomes

∞ ∞ ∞
!p/2
1 X 2
X X
Φ(u) ≥ (λi − λ) βi − ϵ βi2 − A βi2 . (4.52)
2
i=k+1 i=k+1 i=k+1

If u ∈ N , then
∞
X ∞
X
r2 = ∥u∥2E = λi βi2 ≥ λk+1 βi2 ,
i=k+1 i=k+1

which implies
∞
X r2
βi2 ≤ .
λk+1
i=k+1

Now we will do some algebraic manipulation to equation (4.51).

∞ ∞ ∞
!p/2
1 X X X
Φ(u) ≥ (λi − λ) βi2 − ϵ βi2 − A βi2
2
i=k+1 i=k+1 i=k+1
∞ ∞ ∞ ∞
!p/2
1 X λ X X X
= λi βi2 − βi2 −ϵ βi2 −A βi2
2 2
i=k+1 i=k+1 i=k+1 i=k+1
∞  ∞
 X ∞
!p/2
1 X λ X
= λi βi2 − +ϵ βi2 − A βi2
2 2
i=k+1 i=k+1 i=k+1
2
  2  2 p/2
r λ r r
≥ − +ϵ −A
2 2 λk+1 λk+1

Thus equation (4.52) becomes

!
rp
 
1 λ ϵ
Φ(u) ≥ − − r2 − A p/2
. (4.53)
2 2λk+1 λk+1 λk+1
λk+1 −λ
Choose ϵ = 4 . Thus

1 λ ϵ 1 λ λk+1 − λ
− − = − −
2 2λk+1 λk+1 2 2λk+1 4λk+1

82
 1 λ 1 λk+1 − λ
= − −
2 2λk+1 4 λk+1
 
1 λ 1 λk+1 λ
= − − −
2 2λk+1 4 λk+1 λk+1
   
1 λ 1 λ
= 1− − 1−
2 λk+1 4 λk+1
 
1 λ
= 1− .
4 λk+1

So equation (4.53) becomes !

rp
 
1 λ 2
Φ(u) ≥ 1− r −A p/2
. (4.54)
4 λk+1 λk+1
Next we create a function in terms of r of the left hand side of equation (4.54).
!
rp
 
1 λ 2
g(r) = 1− r −A p/2
. (4.55)
4 λk+1 λ k+1

Taking its derivative we get

!
Aprp−1
 
′ 1 λ
g (r) = 1− r− p/2
2 λk+1 λk+1
!!
Aprp−2
 
1 λ
=r 1− − p/2
2 λk+1 λk+1

By setting the derivative equal to zero we can find potential minimum or maximum values of g(r) in order

to find a lower bound for the function Φ(u).

!!
Aprp−2
 
′ 1 λ
g (r) = 0 =⇒ r 1− − p/2
=0
2 λk+1 λk+1
s
p/2 
λk+1

p−2 λ
=⇒ r = 0 or r = 1−
2Ap λk+1

Since we stated earlier that r > 0 we will not consider the case of r = 0 and show that the the function g(r)
r
p/2  
p−2 λk+1 λ
has a local maximum at the point r = 2Ap 1 − λk+1 .

Taking the second derivative of g(r) we get

!
Ap(p − 1)rp−2
 
′′ 1 λ
g (r) = 1− − p/2
.
2 λk+1 λk+1
r !
p/2  
′′ p−2 λk+1 λ
Calculating g 2Ap 1− λk+1 and simplifying we get

 p−2
 s  ! s 
p/2 p/2
λk+1 λk+1
    
p−2 λ = 1 λ Ap(p − 1)  p−2 λ
g ′′  1− 1− − p/2
1− 
2Ap λk+1 2 λk+1 λk+1 2Ap λk+1

83
 p/2  ! !
λk+1
 
1 λ Ap(p − 1)
λ
= 1− − p/2
1−
2 λk+1 λk+1 2Ap λk+1
    
1 λ p−1 λ
= 1− − 1−
2 λk+1 2 λk+1
  
1 p−1 λ
= − 1−
2 2 λk+1
  
2−p λ
= 1−
2 λk+1

Since p > 2, we see 2−p2 !

λ
< 0. Also since λ < λk+1 , we see λk+1 λ
< 1 and thus 1 − λk+1 > 0. Therefore,
r
p/2    
p−2 λk+1
g ′′ λ
= 2−p λ


2Ap 1 − λk+1 2 1 − λk+1 < 0. So the function g(r) obtains a maximum value at
r
p/2
 
p−2 λk+1 λ
r= 2Ap 1− λk+1 .
r
p/2  
p−2 λk+1 λ
Now since Φ(u) ≥ g(r) and g(r) has a maximum at r = 2Ap 1− λk+1 , then
 s 
p/2
λk+1
 
p−2 λ
inf Φ(u) ≥ g  1− .
u∈N 2Ap λk+1

r !
p/2  
p−2 λk+1 λ
Since we want to show inf Φ(u) > 0 we just need to show g 2Ap 1− λk+1 > 0.
u∈N
r ! r
p/2   p/2  
p−2 λk+1 λ p−2 λk+1 λ
To show g 2Ap 1 − λk+1 > 0 we start by writing 2Ap 1− λk+1 in another form.

s s
p/2 p/2
λk+1 λk+1
   
p−2 λ p−2 λk+1 − λ
1− =
2Ap λk+1 2Ap λk+1
v
p/2
u !
u λk+1 − λ
p−2
λk+1
= t
2Ap λk+1
1
1 p/2
! p−2
λk+1
  p−2
λk+1 − λ
=
2Ap λk+1
 1
 p−2
λk+1 − λ 
1/2

= λk+1 .
2Ap

Plugging this into g(r) we get

 1
 p−2 !     1 !2
λk+1 − λ 
1/2
 1 λk+1 − λ λk+1 − λ p−2  1/2 
g λk+1 = λk+1
2Ap 4 λk+1 2Ap
   1   p 
λk+1 −λ p−2 1/2
 2Ap λk+1 
− A p/2

λk+1
 

  2
 p−2
1 λk+1 − λ λk+1 − λ
= λk+1
4 λk+1 2Ap

84
  p
 p−2  
λk+1 −λ p/2
2Ap λk+1 
− A

p/2 
λk+1
 2
 p−2  p
 p−2
1 λk+1 − λ λk+1 − λ
= (λk+1 − λ) −A
4 2Ap 2Ap
  2    2
1 λk+1 − λ p−2 λk+1 − λ λk+1 − λ p−2
= (λk+1 − λ) −A
4 2Ap 2Ap 2Ap
 2
 p−2    2
λk+1 − λ λk+1 − λ λk+1 − λ λk+1 − λ p−2
= −
4 2Ap 2p 2Ap
 2
 p−2  
λk+1 − λ λk+1 − λ λk+1 − λ
= −
2Ap 4 2p
 2
 p−2  
λk+1 − λ (p − 2)(λk+1 − λ)
= .
2Ap 4p

Since λ < λk+1 , then λk+1 − λ > 0. Recall p > 2 so p − 2 > 0. Recall A > 0. Thus
 2
 p−2  
λk+1 − λ (p − 2)(λk+1 − λ)
> 0,
2Ap 4p

meaning
 1
 p−2 !
λk+1 − λ 
1/2

g λk+1 > 0.
2Ap

Therefore,

inf Φ(u) > 0.

u∈N

k
ϕk+1 P
Consider the special choice of z = 1/2 . Then z ∈ Z and ∥z∥E = 1. Let y = ai ϕi and
λk+1 i=1

u = y + ηz ∈ ∂M ⊂ Span{ϕ1 , ϕ2 , ϕ3 , . . . , ϕk+1 } = Y ⊕ {sϕk+1 :∈ R}.

We will consider two cases.

Case 1: Let η = 0. Note this means u = y. Then, by the definition of the set ∂M , ∥u∥E = ∥y∥E ≤ ρ. Then

by equation (4.46)
k
X
2 2
∥u∥l2 (V ) = ∥y∥l2 (V ) = λi a2i ≤ ρ2 .
i=1

Again we will start with equation (4.49)

1 X
Φ(y) = Φ(u) = ((L − λI) u, u) − F (x, u)m(x).
2
x∈V
P
Then since F (x, u)m(x) ≥ 0 equation (4.49) becomes
x∈V

1
Φ(u) ≤ ((L − λI) u, u) . (4.56)
2

85
 k
P
Since u = y = ai ϕi , we can see
i=1

k k
!
1 X X
Φ(u) ≤ (L − λI) ai ϕi , ai ϕi
2 i=1 i=1
k k
!
1 X X
= ai (Lϕi − λϕi ) , ai ϕi
2 i=1 i=1
k k
!
1 X X
= ai (λi ϕi − λϕi ) , ai ϕi
2 i=1 i=1
k k
!
1 X X
= (λi − λ) ai ϕi , ai ϕi
2 i=1 i=1

Then since ∥ϕi ∥2l2 (V ) = 1 equation (4.56) becomes

k
1X
Φ(u) ≤ (λi − λ) a2i . (4.57)
2 i=1

Since λi ≤ λk ≤ λ, then λi − λ ≤ 0. So we see equation (4.57) becomes

k
1X
Φ(u) ≤ (λi − λ) a2i ≤ 0. (4.58)
2 i=1

Case 2: Let η ≥ 0. Note this means u = y+ηz. Then, by the definition of the set ∂M , ∥u∥E = ∥y + ηz∥E = ρ.

Then by equation (4.46)

k
X
2 2
∥u∥l2 (V ) = ∥y + ηz∥l2 (V ) = λi a2i = ρ2 − η 2 .
i=1

Again we will start with equation (4.49)

1 X
Φ(y + ηz) = Φ(u) = ((L − λI) u, u) − F (x, u)m(x).
2
x∈V

1
Working out 2 ((L − λI) u, u), equation (4.49)

1 λ X
Φ(u) = ∥u∥2E − ∥u∥2l2 (v) − F (x, u)m(x). (4.59)
2 2
x∈V

For every ϵ > 0, there exists C2 > 0 such that for every u ∈ R

F (x, u) ≥ C2 |u|q − ϵ|u|2 .

So equation (4.59) becomes

1 λ X
∥u∥2E − ∥u∥2l2 (v) − C2 |u|q − ϵ|u|2 m(x).


Φ(u) ≤ (4.60)
2 2
x∈V

86
Reworking equation (4.60) we see

1 λ X
∥u∥2E − ∥u∥2l2 (v) − C2 |u|q − ϵ|u|2 m(x)


Φ(u) ≤
2 2
x∈V
1 λ
= ∥u∥2E − ∥u∥2l2 (v) − C2 ∥u∥qlq (v) + ϵ∥u∥2l2 (v) (4.61)
2 2
 
1 λ
2
= ∥u∥E − − ϵ ∥u∥2l2 (v) − C2 ∥u∥qlq (v)
2 2

Recall that in a finite dimensional space all norms are equivalent and y + ηz is an element of a finite

dimensional space. Consequently, there is a K > 0 which depends on k and q such that

∥y + ηz∥lq (V ) ≥ K ∥y + ηz∥l2 (V ) .

Hence, for 0 ≤ η ≤ ρ equation (4.60) becomes

 
1 λ
Φ(u) ≤ ∥u∥2E − − ϵ ∥u∥2l2 (v) − C2 K q ∥u∥ql2 (v) . (4.62)
2 2

Rewriting the last term we get

  q/2
1 λ 
Φ(u) ≤ ∥u∥2E − − ϵ ∥u∥2l2 (v) − C2 K q ∥u∥2l2 (v) . (4.63)
2 2

Using the fact that ∥u∥E = ρ we get

ρ2
  q/2
λ 
Φ(u) ≤ − − ϵ ∥u∥2l2 (v) − C2 K q ∥u∥2l2 (v) . (4.64)
2 2
k
λ2
Given the fact that ∥u∥2l2 (v) = a2i +
P
λk+1 we get
i=1

k
! k
!q/2
ρ2 η2 η2
  X
λ X
Φ(u) ≤ − −ϵ a2i + − C2 K q
a2i + . (4.65)
2 2 i=1
λk+1 i=1
λk+1

Reworking this we get

! q/2
λ
−ϵ ρ2
     
2 1 λ 1 1 1 1
Φ(u) ≤ ρ − 2 + −ϵ − η 2 − C2 K q − − η2 . (4.66)
2 λk 2 λk λk+1 λk λk λk+1

We will call the right hand side of the function g˜ϵ (η) to get the equation
!   q/2
λ  2 
2 −ϵ
 
2 1 λ 1 1 ρ 1 1
g˜ϵ (η) = ρ − + −ϵ − η 2 − C2 K q − − η2 . (4.67)
2 λk 2 λk λk+1 λk λk λk+1

Taking the derivative of this we get

 q−2
ρ2
     
′ 1 1 q 1 1 1 1 2
2

g˜ϵ (η) = (λ − 2ϵ) η − + C2 K qη − − − η . (4.68)

λk λk+1 λk λk+1 λk λk λk+1

87
For 0 ≤ η ≤ ρ we have g˜ϵ ′ (η) ≥ 0, when 0 < ϵ ≤ λ2 . Choosing ϵ = λ4 , gives

C2 K q
 
1 λ
Φ(u) ≤ max g˜ϵ (η) = − ρ2 − q/2
ρq = g˜ϵ (ρ). (4.69)
0≤λ≤ρ 2 4λk+1 λk+1

Therefore, there is ρ > r > 0 such that g˜ϵ (ρ) < 0. So we have

inf Φ(u) ≥ g(r) > 0 > g˜ϵ (ρ) ≥ sup Φ(u).

u∈N u∈∂M

Theorem 4.13. [48][50] Let E be an infinite dimensional Banach space and let Φ ∈ C 1 (E, R) be even,

satisfy the Palais-Smale condition, and Φ(0) = 0. If E = Y ⊕ Z where Y is finite dimensional and Φ satisfies

(1) there are constants r, α ≥ 0 such that Φ|∂Br ∩Z ≥ α, and

(2) for nested sequences E1 ⊂ E2 ⊂ · · · of increasing finite dimension, there exists ρi = ρ(Ei ) > 0 such

that Φ ≤ 0 on Bρci = {x ∈ Ei : ∥x∥ > ρi }, for i = 1, 2, . . .,

then Φ possesses an unbounded sequence of critical values.

We need to show that Φ(u) satisfies the hypothesis of Theorem 4.13. The fact that Φ(u) satisfies part

(1) of Theorem 4.13 has been verified by by the proof of Lemma 4.12. We need another lemma to show that

Φ(u) satisifies part (2) of Theorem 4.13. We define the nested sequence of finite dimensional space {Em } in

Theorem 4.13 as follows. For λ < λ1 we have Em = Span{ϕ1 , ϕ2 , . . . , ϕm }, and for λk ≤ λ < λk+1 we have

Em = Span{ϕ1 , ϕ2 , . . . , ϕm+k }, when m = 1, 2, . . ..

Lemma 4.14. There exist two positive constants c1 and c2 , depending on k and m, such that for any

u ∈ Em ,

Φ(u) ≤ c1 ∥u∥2E − c2 ∥u∥qE (4.70)

Proof. We break the proof into two case. The case where m = 1 and the case when m > 1.

Case 1: Let m = 1. This case be done by a similar calculation to the one done in Lemma 4.12. Note that

of any u ∈ E1 there exists unique y ∈ Y and η ∈ R such that u = y + ηz, where Y and z are defined in the

proof Lemma 4.12. So by a similar calculation to how we obtained equation (4.69) we get for u ∈ E1

C2 K q
 
1 λ
Φ(u) ≤ − ∥u∥2E − q/2
∥u∥qE . (4.71)
2 4λk+1 λk+1

This means when m = 1 equation (4.70) is true.

88
 λk+m−1 +λk+m
Case 2: Let m > 1. Take λm = 2 . We define the functional

1 X
Φm (u) = ((L − λm )u, u) − F (x, u)m(x),
2
x∈V

this is simply the functional Φ(u) with a different frequency. Note that λk+m−1 ≤ λm < λk+m . Thus by

replacing k + 1 with k + m in equation (4.71) we get for any u ∈ Em ,

C2 K q
 
1 λ
Φ(u) ≤ − ∥u∥2E − q/2 ∥u∥qE . (4.72)
2 4λk+m λ k+m

k+m
P
Express u = ai ϕi . Using
i=1
k+m
X k+m
X
∥u∥2E = λi a2i ≥ λ1 a2i ,
i=1 i=1

we get for any u ∈ Em ,

k+m
1 X
Φ(u) = Φm (u) + (λm − λ) a2i
2 i=1
C2 K q
 
1 λ λm − λ
≤ − ∥u∥2E − q/2 ∥u∥qE + ∥u∥2E
2 4λk+m λk+m 2λ 1
q
 
1 λ C 2K
≤ − ∥u∥2E − q/2 ∥u∥qE .
2 4λk+m λk+m
This menas when m > 1 equation (4.70) is true.

Therefore there exists positive constants c1 and c2 for any u ∈ Em such that equation (4.70) is true.

Theorem 4.15. Assume Assumptions 4.4, 4.5, 4.6, and 4.7 are satisfied and the function V is coercive. Let

κ = 1 and λ ∈ R. Then

(1) there is at least one nontrivial solution u ∈ E of equation (4.16);

(2) if in addition the nonlinearity f is odd, then there exists an unbounded sequence of critical values of

the function Φ(u). Consequently, there exists infinitely many distinct pairs of nontrivial solutions to

equation (4.16) in the space E.

Proof. Since Φ(u) satisfies the Palais-Smale condition by Lemma 4.10 and the linking geometry by Lemma

4.12, then the first part of Theorem 4.15 is a consequence of the Min/Max Linking Theorem 4.9. Therefore,

equation (4.16) has at least one nontrivial solution u ∈ E when κ = 1 and λ ∈ R.

Since Φ(u) satisfies part (1) of Theorem 4.13 by Lemma 4.12 and part (2) of Theorem 4.13 by Lemma

4.14, then by Theorem 4.13 there exists an unbounded sequence of critical values of Φ(u). Therefore, equation

(4.16) has infinitely many distinct pairs of nontrivial solutions in E when κ = 1 and λ ∈ R.

89
 BIBLIOGRAPHY

[1] S. Ahmad, A.C. Lazer and J.L. Paul, Elementary critical point theory and perturbations of
elliptic boundary value problems at resonance, Indiana Univ. Math. J. 25 (1976), 933-944, DOI
10.1512/iumj.1976.25.25074, MR427825

[2] A. Ambrosetti and P.R. Rabinowitz, Dual variational methods in critical point theory and applications,
Journal of Functional Analysis. 14 (1973), 349-381, DOI 10.1016/0022-1236(73)90051-7, MR0370183

[3] S. Aubry, Breathers in nonlinear lattices: Existence, linear stability and quantization, Physica D:
Nonlinear Phenomena 103 (1997), 201-250, DOI 10.1016/S0167-2789(96)00261-8, MR1464249

[4] A. Banerjee, J. Jost, Laplacian Spectrum and Protein-Protein Interaction Networks, arXiv:0705.3373,
2007.

[5] A. Banerjee, J. Jost, On the spectrum of the normalized graph Laplacian, Linear Algebra and its
Applications. 428 (2008), 3015-3022, DOI 10.1016/j.laa.2008.01.029, MR2416605

[6] F. Bauer, J. Jost, Bipartite and Neighborhood graphs and the spectrum of the normalized graph Laplace
operator, Comm. Anal. Geom 21 (2013), 787-845, DOI 10.4310/CAG.2013.v21.n4.a2, MR3078942

[7] M. Benson and S-M. Lee, On Cordialness of Regular Windmill Graphs, Congressus Numerantium 68
(1989), 49-58, MR995854

[8] Alain Bensoussan and Jose-Luis Menaldi, Difference Equations on Weighted Graphs, Journal of Convex
Analysis 12 (2003), 13-44, MR2135795

[9] H. Brezis, J.M. Caron and L. Nirenberg, Free vibrations for a nonlinear wave equation and a theorem of
P. Rabinowitz, Comm. Pure Appl. Math. 33 (1980), 667-684, DOI 10.1002/cpa.3160330507, MR586417

[10] I. Cahit, Cordial Graphs: a weaker version of graceful and harmonious graphs, Ars Combin. 23 (1987),
201-207, MR886954

[11] I. Cahit, On Cordial and 3-equitable Graphs, Utilitas Math. 37 (1990), 189-198, MR1068518

[12] I. Cahit, Recent Results and Open Problems on Cordial Graphs, Contemporary Methods in Graph Theory,
Bibligraphisches Inst. Mannhiem (1990), 209-230, MR1126229

[13] N. Cairnie and K.Edwards, The computational complexity of cordial and equitable labelling, Discrete
Math. 216 (2000), 29-34, DOI 10.1016/S0012-365X(99)00295-2, MR1750852

[14] H Chen and J Jost, Minimum vertex covers and the spectrum of the normalized Laplacian on trees, Linear
Algebra and its Applications, 437 (2012), 1089-1101, DOI 10.1016/j.laa.2012.04.005, MR2926157

90
[15] Fan R. K. Chung, Spectral Graph Theory, CBMS Regional Conference Series in Mathematics, vol92,
Published for the Conference Board of the Mathematical Sciences, Washington, DC; by the American
Mathematical Society, Providence, RI, 1997. MR1421568

[16] G. Chartrand, S-M Lee and P. Zhang, Uniformly Cordial Graphs, Discrete Mathematics, 306 (2006),
726-737, DOI 10.1016/j.disc.2005.11.025, MR2234980

[17] Fan R. K. Chung, A. Grigor’yan, and S.-T. Yau, Upper Bounds For Eigenvalues of The
Discrete and Continuous Laplace Operators, Advances in Mathematics, 117(1996), 165-178, DOI
10.1006/aima.1996.0006, MR1371647

[18] Fan R. K. Chung and Robert P. Langlands, A Combinatorial Laplacian with Vertex Weights, Journal of
Combinatorial Theory, A 75 (1996), 316-327, DOI 10.1006/jcta.1996.0080, MR1401006

[19] D.C. Clark, A variant of the Lusternik-Schnirelman theory, Ind. Univ. Math. J. 22 (1972), 65-74, DOI
10.1512/iumj.1972.22.22008, MR296777

[20] Daniel Anthony Corral, Friendly Index Sets of Starlike Graphs, Master’s Thesis, University of Nevada,
Las Vegas, 2007.

[21] David G. Costa, An Invitation to Variational Methods in Differential Equations, Birkhäuser Boston, Inc.,
Boston, MA, 2007, MR2321283

[22] J.C. Eilbeck and M. Johansson The discrete nonlinear Schrodinger equation - 20 years on, in Localization
and energy transfer in nonlinear systems (ed. L Vasquez, R.S. MacKay, and M.P. Zorzano), (2003), 44-67

[23] Jason W. Fleischer, Tal Carmon, Mordechai Segev, Nikos K. Efremidis, and Demetrios N.
Christodoulides, Observation of Discrete Solitons in Optically Induced Real Time Waveguide Arrays,
Phys. Rev. Lett., 90 (2003), 023902, DOI 10.1103/PhysRevLett.90.023902.

[24] Jason W. Fleischer, Mordechai Segev, Nikos K. Efremidis, and Demetrios N. Christodoulides, Observation
of two-dimensional discrete solitons in optically induced nonlinear photonic lattices, Nature , 422 (2003)
147–150, DOI 10.1038/nature01452.

[25] S. Flach and C.R. Willis, Discrete breathers, Physics Reports, 295 (1998), 181-264, DOI
10.1016/S0370-1573(97)00068-9, MR1607320

[26] R. Frucht and F. Harrary, On the Corona of Two Graphs, Aequationes Mathematicae 4 (1970), 322-325,
DOI 10.1007/BF01844162, MR281659

[27] Alexander Grigor’yan, Analysis on Graphs, University Lecture Series, vol 71, American Mathematical
Society, Providence, RI, 2009, MR3822363

[28] Alexander Grigor’yan, Young Lin, and Yunyan Yang, Existence of positive solutions to some nonlinear
equations on locally finite graphs, Sci. China Math. 60 (2017), DOI 10.1007/s11425-016-0422-y.
MR3665801

91
[29] Alexander Grigor’yan, Young Lin, and Yunyan Yang, Kazdan-Warner equations on graphs, Calc. Var.
Partial Differential Equations 55 (2016), DOI 10.1007/s00526-016-1042-3. MR3523107

[30] Alexander Grigor’yan, Young Lin, and Yunyan Yang, Yamabe type equations on finite graphs,
arXiv.1607.04521, 2016.

[31] D. Hennig and G. P. Tsironis, Wave transmission in nonlinear lattices, Physics Reports 307 (1999),
333-432, DOI 10.1016/S0370-1573(98)00025-8. MR1670429

[32] Sa’ar Hersonsky, Boundary Value Problems on Planar Graphs and Flat Surfaces with Integer
Cone Sigularites, I: The Dirichlet Problem, J. Reine Angew. Math. 670 (2012), 65-92, DOI
10.1515/crelle.2011.155. MR2982692

[33] Sa’ar Hersonsky,Boundary value problems on planar graphs and flat surfaces with integer cone
singularities, II: the mixed Dirichlet-Neumann problem, Differential Geom. Appl., 29 (2011) 329–347,
DOI 10.1016/j.difgeo.2011.03.003. MR2795842

[34] Y.S. Ho, S-M. Lee, and S.C. Shee, Cordial Labellings of the Cartesian Product and Composition of Graphs,
Ars Combin. 29 (1990), 169-180. MR1046105

[35] M. Hovay, A-cordial Graphs, Discrete Mathematics, 93 (1991), 183-194, DOI

10.1016/0012-365X(91)90254-Y.

[36] Y. Jabri, The mountain pass theorem, Encyclopedia of Mathematics and its Applications, vol.
95, Cambridge University Press, Cambridge, 2003. Variants, generalizations and some applications.
MR2012778.

[37] Jurgen Jost, Mathematical Methods in Biology and Neurobiology, Universitext, Springer, London, 2014.
MR3157168

[38] J. Kazdan, F. Warner, Curvature functions for compact 2-manifolds, Ann.of Math. (2) 99 (1974), 14-47,
DOI 10.2307/1971012. MR343205

[39] W.W. Kirchherr, On the cordiality of some specific graphs, Ars Combin. 31 (1991), 127-138. MR1110227

[40] M. Keller and D. Lenz, Unbounded Laplacians on graphs: basic spectral properties and the heat equation,
Math. Model. Nat. Phenom. 5 (2010), 198-224, DOI 10.1051/mmnp/20105409. MR2662456

[41] S-M. Lee and H.K. Ng, On Friendly Index Sets of Bipartite Graphs, Ars Combin. 86 (2008), 257-271,
MR2379169.

[42] S-M. Lee and H.K. Ng, On Friendly Index Sets of Spiders, Malays. J. Math. Sci. 8 (2014), 47-68.
MR3226031

[43] Y.H. Lee, H.M. Lee, and G.J. Chang, Cordial Labelings of Graphs, Chinese J. Math. 20 (1992), 263-273.
MR1183821

92
[44] J. J. Manfredi, A. M. Oberman, and A. P. Sviridov, Nonlinear Elliptic Partial Differential Equations
And P-Harmonic Functions On Graphs, Differential Integral Equations 28 (2013), 79-102. MR3299118

[45] D. McGinn, Friendly indices and fully cordial graphs, J. Combin. Math. Combin. Comput. 101 (2017),
83-99, MR3676192.

[46] A. Pankov Standing waves for discrete nonlinear Schrödinger equations: sign-changing nonlinearities,
Appl. Anal. 92 (2013), 308-317, DOI 10.1080/00036811.2011.609987. MR3022859

[47] A. Pankov and G. Zhang, Standing waves of the discrete nonlinear Schrödinger equations with growing
potentials, Commun. Math. Anal. 5 (2008), 38-49, MR2421490.

[48] A. Pankov and G. Zhang, Standing wave solutions of the discrete non-linear Schrödinger equations
with unbounded potentials, II, Appl. Anal. 89 (2010), 1541-1557, DOI 10.1080/00036810902942234,
MR2682118.

[49] P.H. Rabinowitz, Minimax methods in critical point theory with applications to differential equations,
CBMS Regional Conference Series in Mathematics, vol. 65, Published for the Conference Board of the
Mathematical Sciences, Washington, DC; by the American Mathematical Society, Providence, RI, 1986.
MR845785.

[50] P.H. Rabinowitz, Some Minimax Theorems and Applications to Nonlinear Partial Differential Equations,
Nonlinear analysis (collection of papers in honor of Erich H. Rothe) , Academic Press, New York, (1978),
pp.161-177. MR0501092

[51] E. Seah, On the construction of cordial graphs, Ars Combin. 31 (1991), 249-254. MR1110244

[52] Ebrahim Salehi and Daniel Corral, Fully cordial trees, J. Combin. Math. Combin. Comput. 98 (2016),
171-183. MR3560482

[53] E. Salehi and S. De, On a conjecture concerning the friendly index sets of trees, Ars Combin. 90 (2009),
371-381. MR2489539

[54] E. Salehi and S-M Lee, On friendly index sets of trees, Congr. Numer. 178 (2006), 173-183. MR2310230

[55] Douglas West, Introduction to graph theory, Prentice Hall, Inc., Upper Saddle River, NJ, 1996.
MR1367739

[56] M. Willem,Lectures on Critical Point Theory, Trabalhos de Matematica, University of Brasilia, Brasilia,
1983.

[57] M. Willem, Minimax Theorems,Progress in Nonlinear Differential Equations and their Applications, vol.
24, Birkhäuser Boston, Inc., Boston, MA, 1996. MR1400007.

93
 CURRICULUM VITAE
Graduate College
University of Nevada, Las Vegas

Daniel Anthony Corral

Degrees:
Bachelor of Science - Mathematics, 2006
University of Nevada, Las Vegas

Master of Science - Mathematics, 2007

University of Nevada, Las Vegas

Master of Arts - Mathematics, 2011

University of Kentucky

Special Honors and Awards:

Honorable mention NSF Graduate Research Fellowship Program (Spring 2005)
2004 Bhatnagar Award for the Top Major in Mathematical Sciences at UNLV
NSF EPSCoR Ring True Two Award, Award Number: EPS-0132556 (Summer 2004)
The National Dean’s List 2003-2004 edition (2004)
Summa Cum Laude
University Honors
Department Honors

Publications:
Ebrahim Salehi and Daniel Corral, ”Fully Cordial Trees”, Journal of Combinatorial
Mathematics and Combinatorial Computing 98 (2016), p. 171-183.

Conference and Poster Presentations:

Invited Speaker 1141st AMS Meeting at University of Delaware: “Remarks on Lagrange
Multiplier Approch to Kadzan-Warner Equations on a Finite Graph”, September 29-30
2018.
Presenter 28th Midwest Conference on Combinatorics and Combinatorial Computing
at University of Nevada, Las Vegas:“Fully Cordial Trees”, October 22-24, 2014.
Daniel Corral and Corran Webster: ”On a metric of the discrete Heisenberg group”
Southern Nevada Undergraduate Research Opportunities Poster Program, 2004,
Funded by National Science Foundation EPSCoR.
Daniel Corral: ”On the metric boundary of the discrete Heisenberg group”, Fall Honors
Thesis Poster Session, Fall 2004.

Dissertation Title:
Some Graph Laplacians and Variational Methods Applied to Partial Differential
Equations on Graphs

Dissertation Examination Committee:

Chairperson, David Costa, Ph.D.
Committee Member, Ebrahim Salehi, Ph.D.
Committee Member, Hossein Tehrani, Ph.D.
Graduate Faculty Representative, Pushkin Kachroo, Ph.D.

94