This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

On the Paley graph of a quadratic character

Ján Mináč, Lyle Muller, Tung T. Nguyen, NguyÊ~\tilde{\text{\^{E}}}n Duy Tân Department of Mathematics, Western University, London, Ontario, Canada N6A 5B7 [email protected] Department of Mathematics, Western University, London, Ontario, Canada N6A 5B7 [email protected] Department of Mathematics, Western University, London, Ontario, Canada N6A 5B7 [email protected] School of Applied Mathematics and Informatics, Hanoi University of Science and Technology, 1 Dai Co Viet Road, Hanoi, Vietnam [email protected]
Abstract.

Paley graphs form a nice link between the distribution of quadratic residues and graph theory. These graphs possess remarkable properties which make them useful in several branches of mathematics. Classically, for each prime number pp we can construct the corresponding Paley graph using quadratic and non-quadratic residues modulo pp. Therefore, Paley graphs are naturally associated with the Legendre symbol at pp which is a quadratic Dirichlet character of conductor pp. In this article, we introduce the generalized Paley graphs. These are graphs that are associated with a general quadratic Dirichlet character. We will then provide some of their basic properties. In particular, we describe their spectrum explicitly. We then use those generalized Paley graphs to construct some new families of Ramanujan graphs. Finally, using special values of LL-functions, we provide an effective upper bound for their Cheeger number. As a by-product of our approach, we settle a question raised in [8] about the size of this upper bound.

Key words and phrases:
Paley graphs, Ramanujan graphs, Cheeger number, Gauss sums, Special values of LL-functions.
2020 Mathematics Subject Classification:
Primary 05C25, 05C50, 11M06
Jan Minac is partially supported by the Natural Sciences and Engineering Research Council of Canada (NSERC) grant R0370A01. Jan Minac also gratefully acknowledges Faculty of Sciences Distinguished Research Professorship award for 2020/21. Jan Minac, Lyle Muller and Tung T Nguyen acknowledge the support of the Western Academy for Advanced Research. Nguyê~\tilde{\text{\^{e}}}n Duy Tân is is partially supported by Vietnam National University, Hanoi (VNU) under the project number QG.23.48

1. Introduction

Let q=pnq=p^{n} be a prime power. The Paley graph PqP_{q} is the graph with the following data

  1. (1)

    The vertex set of PqP_{q} is 𝔽q{\mathbb{F}}_{q}; here 𝔽q{\mathbb{F}}_{q} is the finite field with qq elements;

  2. (2)

    Two vertices u,vu,v are connected if and only uvu-v a nonzero square in 𝔽q{\mathbb{F}}_{q}; i.e., vuv-u is an element of the set

    S={x2x𝔽q,x0}.S=\{x^{2}\mid x\in{\mathbb{F}}_{q},x\neq 0\}.

Let us recall the definition of the Caley graph Γ(G,S)\Gamma(G,S) where GG is a group and a set SS is a subset of GG which has the property that 1S1\not\in S. The vertices of Γ(G,S)\Gamma(G,S) are elements of GG and the edges of Γ(G,S)\Gamma(G,S) are given by

E={(g,gs)gG,sS}G×G.E=\{(g,gs)\mid g\in G,s\in S\}\subset G\times G.

(See [6, 19]). We note that in [18, Definition 2.3.1], the author requires that SS is symmetric in the sense that sSs\in S if and only if s1Ss^{-1}\in S (this symmetric condition guarantees that Γ(G,S)\Gamma(G,S) is an undirected graph). In the case G=𝔽qG={\mathbb{F}}_{q} as an additive group, we see that PqP_{q} is exactly the Caley graph Γ(G,S)\Gamma(G,S) with S=(𝔽q×)2:={x2x𝔽q×}S=({\mathbb{F}}_{q}^{\times})^{2}:=\{x^{2}\mid x\in{\mathbb{F}}_{q}^{\times}\}. (Here we use the standard notation 𝔽q×=𝔽q{0}{\mathbb{F}}_{q}^{\times}={\mathbb{F}}_{q}\setminus\{0\}.) We remark that 1S-1\in S if and only if q1(mod4).q\equiv 1\pmod{4}. Therefore, PqP_{q} is an undirected graph for q1(mod4).q\equiv 1\pmod{4}.

The origin of Paley graphs can be traced back to Paley’s 1933 paper [27] in which he implicitly used them to construct Hadamard matrices. These graphs were later rediscovered by Carlitz in [5], though this time in a different context. Gareth A. Jones had the following remark in [16] about Paley graphs

Anyone who seriously studies algebraic graph theory or finite permutation groups will, sooner or later, come across the Paley graphs and their automorphism groups.

Due to its arithmetic and representation theoretic nature, we have various tools to study Paley graphs. Consequently, they have interesting properties that make them useful for graph theory and related fields. For example, Paley graphs have found applications in coding and cryptography theory (see [12, 14]).

In this article, we define and study generalized Paley graphs which are associated with general quadratic characters (we refer the reader to Section 2 for the precise definition of these graphs). The terminology ”generalized Paley graphs” was used earlier in some papers of C.E. Praeger and her collaborators. Their generalization of Paley graphs, however, went in another direction from our generalization (see [21]). Therefore a possible confusion is unlikely. Using the theory of Gauss sums and circulant matrices, we describe explicitly the spectrum of these generalized Paley graphs. We then give a complete answer to the following question: which generalized Paley graphs are Ramanujan? In the final section, we provide an effective upper bound for the Cheeger number of these generalized Paley graphs using special values of LL-functions. As a by-product of our approach, we settle a question raised in [8, Section 2.2] about the size of this upper bound.

In closing the introduction, we remark that our interest in this topic arises naturally from our recent works on generalized Fekete polynomials (see [24, 23]), on the join of circulant matrices (see [7, 10]), and on some new constructions of Ramanujan graphs (see [11]).

2. Quadratic characters and the associated generalized Paley graphs

2.1. Kronecker symbol

In this section, we recall the definition of the Kronecker symbol.

Let a,na,n be integers. Suppose that n0n\not=0 and nn has the following factorization into the product of distinct prime numbers

n=sgn(n)p1e1p2e2prer.n=\text{sgn}(n)p_{1}^{e_{1}}p_{2}^{e_{2}}\cdots p_{r}^{e_{r}}.

Here sgn(n)\text{sgn}(n) is the sign of nn, which is 11 if n>0n>0 and 1-1 otherwise. Then, the Kronecker symbol (an)\left(\frac{a}{n}\right) is defined as

(an)=(asgn(n))(ap1)e1(ap2)e2(apr)er.\left(\frac{a}{n}\right)=\left(\frac{a}{\text{sgn}(n)}\right)\left(\frac{a}{p_{1}}\right)^{e_{1}}\left(\frac{a}{p_{2}}\right)^{e_{2}}\cdots\left(\frac{a}{p_{r}}\right)^{e_{r}}.

Here

(a1)={1if a01if a<0,(a2)={0if 2|a 1if a±1(mod8)1if a±3(mod8),\left(\frac{a}{-1}\right)=\begin{cases}1&\text{if }a\geq 0\\ -1&\text{if }a<0\end{cases},\quad\left(\frac{a}{2}\right)=\begin{cases}0&\text{if $2|a$ }\\ 1&\text{if }a\equiv\pm{1}\pmod{8}\\ -1&\text{if }a\equiv\pm{3}\pmod{8},\end{cases}

and for an odd prime pp, (ap)\left(\frac{a}{p}\right) is the Legendre symbol which is defined as follows

(ap)={0if p|a 1if a is a square modulo p 1else. \left(\frac{a}{p}\right)=\begin{cases}0&\text{if $p|a$ }\\ 1&\text{if $a$ is a square modulo p }\\ -1&\text{else. }\end{cases}

Also, the Kronecker symbol (a0)\left(\frac{a}{0}\right) is defined as

(a0)={1if a=±10otherwise. \left(\frac{a}{0}\right)=\begin{cases}1&\text{if }a=\pm{1}\\ 0&\text{otherwise. }\end{cases}

We note that, the Kronecker symbol is defined for all integers aa and nn.

2.2. Quadratic characters

We first recall the theory of Dirichlet characters (we refer the reader to [1, Chapter 1] for some further discussions).

Definition 2.1.

Let mm be a positive integer. A Dirichlet character mod mm is a group homomorphism

χ:(/m)××.\chi:({\mathbb{Z}}/m)^{\times}\to{\mathbb{C}}^{\times}.

We say that χ\chi is primitive if it does not factor through (/n)×({\mathbb{Z}}/n)^{\times} for some proper divisor nn of m.m.

We remark that we can extend a Dirichlet character to an arithmetic function χ:\chi:{\mathbb{Z}}\to{\mathbb{C}} by the following convention

χ(a)={χ(amodm)if gcd(a,m)=10else.\chi(a)=\begin{cases}\chi(a\mod{m})&\text{if }\gcd(a,m)=1\\ 0&\text{else.}\end{cases}

In this article, we will focus on quadratic characters. These characters are closely related to the theory of quadratic fields and Kronecker symbols which we know recall. Let mm be a squarefree integer. Let Δ\Delta be the discriminant of the quadratic extension (m)/{\mathbb{Q}}(\sqrt{m})/{\mathbb{Q}}, which is given by

Δ={mif m1(mod4)4mif m2,3(mod4).\Delta=\begin{cases}m&\text{if }m\equiv 1\pmod{4}\\ 4m&\text{if }m\equiv 2,3\pmod{4}.\end{cases}

From this definition, we can see that Δ\Delta necessarily determines the quadratic extension (m)/.{\mathbb{Q}}(\sqrt{m})/{\mathbb{Q}}. We have the following theorem.

Theorem 2.2.

([25, Theorem 9.13]) Let χΔ:×\chi_{\Delta}\colon{\mathbb{Z}}\to\mathbb{C}^{\times} be the function given by

χΔ(a)=(Δa),\chi_{\Delta}(a)=\left(\frac{\Delta}{a}\right),

where (Δa)\left(\frac{\Delta}{a}\right) is the Kronecker symbol. Then χΔ\chi_{\Delta} is a primitive quadratic character of conductor D=|Δ|.D=|\Delta|. Furthermore, every primitive quadratic character is given uniquely this way.

2.3. Generalized Paley graphs and their spectra

Let χ=χΔ\chi=\chi_{\Delta} be the primitive quadratic character of conductor D=|Δ|D=|\Delta| as explained in the previous section. We introduce the following definition (see also [4] for a similar approach).

Definition 2.3.

The generalized Paley graph PΔP_{\Delta} is the graph with the following data

  1. (1)

    The vertices of PΔP_{\Delta} are {0,1,,D1}.\{0,1,\ldots,D-1\}.

  2. (2)

    Two vertices (u,v)(u,v) are connected iff χΔ(vu)=1.\chi_{\Delta}(v-u)=1.

Example 2.4.

We refer to Figure 2 and Figure 2 for two concrete examples of generalized Paley graphs.

Refer to caption
Figure 1. The generalized Paley graph P12P_{12}
Refer to caption
Figure 2. The generalized Paley graph P21P_{21}.

We remark that if Δ<0\Delta<0, then χΔ(1)=1.\chi_{\Delta}(-1)=-1. In this case, PΔP_{\Delta} is a directed graph. Otherwise, if Δ>0\Delta>0 then χΔ(1)=1\chi_{\Delta}(-1)=1 and therefore, PΔP_{\Delta} is an undirected graph.

Since the connection in PΔP_{\Delta} is determined by (vu)modD(v-u)\mod{D}, we conclude that PΔP_{\Delta} is a circulant graph with respect to the cyclic group /D{\mathbb{Z}}/D (see [7, 9]). In fact, its adjacency matrix is generated by the following vector

v=[12χ(a)(χ(a)+1)]0aD1.v=\left[\frac{1}{2}\chi(a)(\chi(a)+1)\right]_{0\leq a\leq D-1}.

This follows from the fact that

12χ(a)(1+χ(a))={1 if χ(a)=10 else.\frac{1}{2}\chi(a)(1+\chi(a))=\begin{cases}1&\text{ if }\chi(a)=1\\ 0&\text{ else.}\end{cases}

We first observe the following.

Proposition 2.5.

PΔP_{\Delta} is a regular graph of degree 12φ(D).\frac{1}{2}\varphi(D).

Proof.

We have

2deg(PΔ)\displaystyle 2\deg(P_{\Delta}) =a=0D1χ(a)[1+χ(a)]=a=0D1χ(a)+a=0D1χ2(a)\displaystyle=\sum_{a=0}^{D-1}\chi(a)[1+\chi(a)]=\sum_{a=0}^{D-1}\chi(a)+\sum_{a=0}^{D-1}\chi^{2}(a)
=0+0aD1,gcd(a,D)=11=φ(D).\displaystyle=0+\sum_{0\leq a\leq D-1,\gcd(a,D)=1}1=\varphi(D).

Here we use the fact that

a=0D1χ(a)=0.\sum_{a=0}^{D-1}\chi(a)=0.

We conclude that deg(PΔ)=12φ(D).\deg(P_{\Delta})=\frac{1}{2}\varphi(D).

Corollary 2.6.

Suppose that Δ>0\Delta>0. Then PΔP_{\Delta} is a cycle graph if and only if Δ=5\Delta=5 or Δ=8\Delta=8 or Δ=12.\Delta=12.

Proof.

Suppose that PΔP_{\Delta} is a cycle graph. Then the degree of PΔP_{\Delta} is 22. By Proposition 2.5, we must have φ(D)=4.\varphi(D)=4. Therefore Δ=D{5,8,12}.\Delta=D\in\{5,8,12\}. Conversely, if Δ{5,8,12}\Delta\in\{5,8,12\}, then χ(a)=1\chi(a)=1 if and if a=±1.a=\pm{1}. Consequently, an edge in PΔP_{\Delta} must have the form (u,u+1)(u,u+1) or (u,u1)(u,u-1) for u/D.u\in{\mathbb{Z}}/D. In other words, PΔP_{\Delta} is a cycle graph. ∎

3. The spectrum of PΔP_{\Delta}

In this section, we compute the spectrum of PΔP_{\Delta}. By the Circulant Diagonalization Theorem (see [9]), the spectrum of PΔP_{\Delta} is given by

{λ(ω):=12a=0D1χ(a)(1+χ(a))ωa},\left\{\lambda(\omega):=\frac{1}{2}\sum_{a=0}^{D-1}\chi(a)(1+\chi(a))\omega^{a}\right\},

where ω\omega runs over the set of all DDth roots of unity. Note that for each ω\omega, there exists a unique positive integer d|Dd|D such that ω\omega is a primitive ddth root of unity. We first recall the definition of the Möbius function

Definition 3.1.

The Möbius function μ(n)\mu(n) is defined as follow

μ(n)={0 if n has one or more repeated prime factors1 if n=1(1)k if n is a product of k distinct primes..\mu(n)=\begin{cases}0\text{ if $n$ has one or more repeated prime factors}\\ 1\text{ if $n=1$}\\ (-1)^{k}\text{ if $n$ is a product of $k$ distinct primes}.\end{cases}.

We also recall Ramanujan’s sum (see [13, Section 5.6]). Let m,nm,n be two positive integers. The Ramanujan sum cn(m)c_{n}(m) is defined as follow

cn(m)=1angcd(a,n)=1exp(2πiman).c_{n}(m)=\sum_{\begin{subarray}{c}1\leq a\leq n\\ \gcd(a,n)=1\end{subarray}}{\rm exp}(\frac{2\pi ima}{n}).

If gcd(n,m)=a\gcd(n,m)=a and n=aNn=aN then by [13, Theorem 272], we have

cn(m)=μ(N)φ(n)φ(N).c_{n}(m)=\frac{\mu(N)\varphi(n)}{\varphi(N)}.

In particular, when m=1m=1 and n=dn=d, we have (see [13, Equation 16.4.4])

Lemma 3.2.

Let dd be a positive integer. Let ω\omega be a primitive ddth root of unity. Then

1idgcd(i,d)=1ωi=μ(d).\sum_{\begin{subarray}{c}1\leq i\leq d\\ \gcd(i,d)=1\end{subarray}}\omega^{i}=\mu(d).

In other words, the sum of all primitive ddth roots of unity is equal to μ(d).\mu(d).

We recall the theory of Gauss sum. Let ζD=exp(2πiD)\zeta_{D}=\exp\left(\frac{2\pi i}{D}\right) be a primitive DDth root of unity.

Definition 3.3.

([1, Chapter V]) The Gauss sum G(b,χ)G(b,\chi) is defined as follow

G(b,χ)=a=1D1χ(a)ζDab.G(b,\chi)=\sum_{a=1}^{D-1}\chi(a)\zeta_{D}^{ab}.

We recall the following fundamental property [1, Theorem 4.12, page 312]

G(b,χ)=χ(b)G(1,χ).G(b,\chi)=\chi(b)G(1,\chi).

Furthermore, by [1, Theorem 4.17], we have G(1,χ)=ΔG(1,\chi)=\sqrt{\Delta}. Consequently G(b,χ)=χ(b)ΔG(b,\chi)=\chi(b)\sqrt{\Delta}. In particular, if ω\omega is not a primitive DDth root of unity then

a=1D1χ(a)ωa=0.\sum_{a=1}^{D-1}\chi(a)\omega^{a}=0.

On the other hand, if ω\omega is a primitive DDth root of unity, namely ω=ζDb\omega=\zeta_{D}^{b} with gcd(b,D)=1\gcd(b,D)=1 then

a=0D1χ(a)ωa=χ(b)Δ.\sum_{a=0}^{D-1}\chi(a)\omega^{a}=\chi(b)\sqrt{\Delta}.

In summary, we have the following lemma.

Lemma 3.4.

Let ω\omega be a DDth root of unity.

  1. (1)

    If ω\omega is not a primitive DDth root of unity then

    a=1D1χΔ(a)ωa=0.\sum_{a=1}^{D-1}\chi_{\Delta}(a)\omega^{a}=0.
  2. (2)

    If ω=ζDb\omega=\zeta_{D}^{b} with gcd(b,D)=1\gcd(b,D)=1 then

    a=1D1χΔ(a)ωa=χ(b)Δ.\sum_{a=1}^{D-1}\chi_{\Delta}(a)\omega^{a}=\chi(b)\sqrt{\Delta}.

We can now simplify λ(ω).\lambda(\omega). In fact

2λ(ω)\displaystyle 2\lambda(\omega) =a=0D1χ(a)ωa+a=0D1χ2(a)ωa\displaystyle=\sum_{a=0}^{D-1}\chi(a)\omega^{a}+\sum_{a=0}^{D-1}\chi^{2}(a)\omega^{a}
=a=0D1χ(a)ωa+0aD1gcd(a,D)=1ωa.\displaystyle=\sum_{a=0}^{D-1}\chi(a)\omega^{a}+\sum_{\begin{subarray}{c}0\leq a\leq D-1\\ \gcd(a,D)=1\end{subarray}}\omega^{a}.

We consider two cases.
Case 1. ω\omega is a primitive ddth root of unity with d<D.d<D. In this case

a=0D1χ(a)ωa=0,\sum_{a=0}^{D-1}\chi(a)\omega^{a}=0,

and

0aD1gcd(a,D)=1ωa=φ(D)φ(d)μ(d).\sum_{\begin{subarray}{c}0\leq a\leq D-1\\ \gcd(a,D)=1\end{subarray}}\omega^{a}=\frac{\varphi(D)}{\varphi(d)}\mu(d).

Therefore

λ(ω)=12φ(D)φ(d)μ(d).\lambda(\omega)=\frac{1}{2}\frac{\varphi(D)}{\varphi(d)}\mu(d).

Case 2. ω=ζDb\omega=\zeta_{D}^{b} is a primitive DDth root of unity; namely ω=ζDb\omega=\zeta_{D}^{b} with gcd(b,D)=1\gcd(b,D)=1. In this case, we have

a=0D1χ(a)ωa=χ(b)D,\sum_{a=0}^{D-1}\chi(a)\omega^{a}=\chi(b)\sqrt{D},

and

0aD1gcd(a,D)=1ωa=μ(D).\sum_{\begin{subarray}{c}0\leq a\leq D-1\\ \gcd(a,D)=1\end{subarray}}\omega^{a}=\mu(D).

Consequently

λ(ω)=12[χ(b)Δ+μ(D)].\lambda(\omega)=\frac{1}{2}\left[\chi(b)\sqrt{\Delta}+\mu(D)\right].

Let us write [a]b[a]_{b} for the multiset {a,,ab times}\{\underbrace{a,\ldots,a}_{\text{$b$ times}}\}. By the above calculations, we have the following conclusion.

Theorem 3.5.

The spectrum of the generalized Paley graph PΔP_{\Delta} is the union of the following multisets

[12φ(D)φ(d)μ(d)]φ(d)for d|Dand d<D,\left[\frac{1}{2}\frac{\varphi(D)}{\varphi(d)}\mu(d)\right]_{\varphi(d)}\quad\text{for }d|D\quad\text{and }d<D,

and

[12(Δ+μ(D))]φ(D)2,\left[\frac{1}{2}(\sqrt{\Delta}+\mu(D))\right]_{\frac{\varphi(D)}{2}},

and

[12(Δ+μ(D))]φ(D)2.\left[\frac{1}{2}(-\sqrt{\Delta}+\mu(D))\right]_{\frac{\varphi(D)}{2}}.

We discuss a corollary of this theorem. First, we recall that an undirected graph G=(V,E)G=(V,E) is called a bipartite graph if V(G)V(G) can be decomposed into two disjoint sets such that no two vertices within the same set are adjacent. As explained in [26], if GG is regular of degree rr then GG is a bipartite graph if and only if r-r is an eigenvalue of G.G.

Corollary 3.6.

Suppose that Δ>0\Delta>0. Then, the generalized Paley graph PΔP_{\Delta} is a bipartite graph if and only if Δ\Delta is even.

Proof.

If Δ\Delta is even then we can decompose V(PΔ)V(P_{\Delta}) into the following two sets

Vodd={aV(G)|a1(mod2)},V_{\text{odd}}=\{a\in V(G)|a\equiv 1\pmod{2}\},

and

Veven={aV(G)|a0(mod2)}.V_{\text{even}}=\{a\in V(G)|a\equiv 0\pmod{2}\}.

If u,vu,v both belong to either VoddV_{\text{odd}} or VevenV_{\text{even}} then uvu-v is even. Therefore, χ(uv)=0.\chi(u-v)=0. By definition, uu and vv are not adjacent. We conclude that PΔP_{\Delta} is a bipartite graph.

Conversely, let us assume that PΔP_{\Delta} is a bipartite graph. By the above remark and the fact that PΔP_{\Delta} is regular of degree 12φ(D)\frac{1}{2}\varphi(D), we conclude that one of its eigenvalues must be 12φ(D).-\frac{1}{2}\varphi(D). Clearly this eigenvalue cannot be either 12(Δ+μ(D))\frac{1}{2}(\sqrt{\Delta}+\mu(D)) or 12(Δ+μ(D))\frac{1}{2}(-\sqrt{\Delta}+\mu(D)) because these values are not integers. Consequently, by Theorem 4.3, there must exists d|Dd|D and d<Dd<D such that

12φ(D)φ(d)μ(d)=12φ(D).\frac{1}{2}\frac{\varphi(D)}{\varphi(d)}\mu(d)=-\frac{1}{2}\varphi(D).

Equivalently, we must have φ(d)=μ(d){1,1}.\varphi(d)=-\mu(d)\in\{-1,1\}. We conclude that d=2d=2 and hence DD is even. ∎

4. Ramanujan Paley graphs

In this section, we investigate the following question: which PΔP_{\Delta} is a Ramanujan graph? For this question to make sense, we need to assume that PΔP_{\Delta} is an undirected graph. This requires χΔ\chi_{\Delta} is an even character; or equivalently Δ>0.\Delta>0. We first recall the definition of a Ramanujan graph (see [22, 26]).

Definition 4.1.

Let GG be a connected rr-regular graph with NN vertices, and let r=λ1λ2λNr=\displaystyle\lambda_{1}\geq\lambda_{2}\geq\cdots\geq\lambda_{N} be the eigenvalues of the adjacency matrix of GG. Since GG is connected and rr-regular, its eigenvalues satisfy |λi|r,1iN.|\lambda_{i}|\leq r,1\leq i\leq N. Let

λ(G)=max|λi|<r|λi|.\lambda(G)=\max_{|\lambda_{i}|<r}|\lambda_{i}|.

The graph GG is a Ramanujan graph if

λ(G)2r1.\lambda(G)\leq 2{\sqrt{r-1}}.

It is well-known that PΔP_{\Delta} is a Ramanujan graph if Δ=p\Delta=p, where pp is a prime number, which is congruent to 1 modulo 4. In fact, by Proposition 2.5, PΔP_{\Delta} is a regular graph of degree r:=p12r:=\dfrac{p-1}{2}. By Theorem 3.5, the eigenvalues of the adjacency matrix of PΔP_{\Delta} which are smaller than rr are ±p12\dfrac{\pm\sqrt{p}-1}{2}. Because p1(mod4)p\equiv 1\pmod{4} we see that p5p\geq 5 and λ(PΔ)=p+12<2r1=2p32\lambda(P_{\Delta})=\dfrac{\sqrt{p}+1}{2}<2\sqrt{r-1}=2\sqrt{\dfrac{p-3}{2}}. When DD is not a prime number, we have the following.

Lemma 4.2.

Suppose DD is not a prime number. The graph PΔP_{\Delta} is a Ramanujan graph if and only if

φ(D)(p1)2+16φ(D)8,\dfrac{\varphi(D)}{(p-1)^{2}}+\dfrac{16}{\varphi(D)}\leq 8,

where pp is the smallest odd prime divisor of DD, and

{φ(D)D18+14D+178D if D is oddφ(D)D18+2D if D is even.\begin{cases}\dfrac{\varphi(D)}{D}\geq\dfrac{1}{8}+\dfrac{1}{4\sqrt{D}}+\dfrac{17}{8D}&\text{ if $D$ is odd}\\ \dfrac{\varphi(D)}{D}\geq\dfrac{1}{8}+\dfrac{2}{D}&\text{ if $D$ is even}\end{cases}.
Proof.

Let dd be an arbitrary divisor of DD such that 1<d<D1<d<D and d2d\not=2 if 4D4\mid D. One has

φ(D)2φ(d)2φ(D)21φ(D)2φ(d)28φ(D)16φ(D)φ(d)2+16φ(D)8.\dfrac{\varphi(D)}{2\varphi(d)}\leq 2\sqrt{\dfrac{\varphi(D)}{2}-1}\Leftrightarrow\dfrac{\varphi(D)^{2}}{\varphi(d)^{2}}\leq 8\varphi(D)-16\Leftrightarrow\dfrac{\varphi(D)}{\varphi(d)^{2}}+\dfrac{16}{\varphi(D)}\leq 8.

Clearly if dd has an odd prime divisor then φ(D)φ(d)2+16φ(D)φ(D)(p1)2+16φ(D)\dfrac{\varphi(D)}{\varphi(d)^{2}}+\dfrac{16}{\varphi(D)}\leq\dfrac{\varphi(D)}{(p-1)^{2}}+\dfrac{16}{\varphi(D)}, where pp is the smallest odd prime divisor of DD.

One also has

D+122φ(D)21(1+D)28φ(D)16φ(D)D18+14D+178D,\dfrac{\sqrt{D}+1}{2}\leq 2\sqrt{\dfrac{\varphi(D)}{2}-1}\Leftrightarrow(1+\sqrt{D})^{2}\leq 8\varphi(D)-16\Leftrightarrow\dfrac{\varphi(D)}{D}\geq\dfrac{1}{8}+\dfrac{1}{4\sqrt{D}}+\dfrac{17}{8D},

and

D22φ(D)21D8φ(D)16φ(D)D18+2D.\dfrac{\sqrt{D}}{2}\leq 2\sqrt{\dfrac{\varphi(D)}{2}-1}\Leftrightarrow D\leq 8\varphi(D)-16\Leftrightarrow\dfrac{\varphi(D)}{D}\geq\dfrac{1}{8}+\dfrac{2}{D}.

The lemma now follows from Theorem 3.5. ∎

Theorem 4.3.

The graph PΔP_{\Delta} is a Ramanujan graph if and only if

  1. (1)

    either D=8D=8,

  2. (2)

    or D=4pD=4p, where pp is a prime number, p3(mod4)p\equiv 3\pmod{4},

  3. (3)

    or D=8pD=8p, where pp is an odd prime number,

  4. (4)

    or D=4p1p2D=4p_{1}p_{2} where p1p_{1} and p2p_{2} are primes, p1p23(mod4)p_{1}p_{2}\equiv 3\pmod{4} and p1<p24p15p_{1}<p_{2}\leq 4p_{1}-5.

  5. (5)

    or D=8p1p2D=8p_{1}p_{2} where p1p_{1} and p2p_{2} are primes and 2<p1<p22p132<p_{1}<p_{2}\leq 2p_{1}-3,

  6. (6)

    or DD is a prime number pp with p1(mod4)p\equiv 1\pmod{4},

  7. (7)

    or D=p1p2D=p_{1}p_{2} where p1p_{1} and p2p_{2} are primes, p1p21(mod4)p_{1}p_{2}\equiv 1\pmod{4} and p1<p28p19p_{1}<p_{2}\leq 8p_{1}-9.

Proof.

a) We first consider the case that DD is even. It is easy to check that P8P_{8} is a Ramanujan graph. So we suppose that D8D\not=8 and PΔP_{\Delta} is a Ramanujan graph. Let D=2ap1a1pkakD=2^{a}p_{1}^{a_{1}}\cdots p_{k}^{a_{k}} be the prime factorization of DD with 2<p1<<pk2<p_{1}<\cdots<p_{k}. One has a{2,3}a\in\{2,3\}, k1k\geq 1, and a2==ak=1a_{2}=\cdots=a_{k}=1. By Lemma 4.2,

8φ(D)(p11)2+16φ(D)>φ(D)(p11)2=2a1p21p11(pk1).8\geq\dfrac{\varphi(D)}{(p_{1}-1)^{2}}+\dfrac{16}{\varphi(D)}>\dfrac{\varphi(D)}{(p_{1}-1)^{2}}=2^{a-1}\dfrac{p_{2}-1}{p_{1}-1}\cdots(p_{k}-1).

This inequality forces k2k\leq 2. In fact, if k3k\geq 3 then

2a1p21p11(pk1)>2(p31)2(71)=12>8.2^{a-1}\dfrac{p_{2}-1}{p_{1}-1}\cdots(p_{k}-1)>2(p_{3}-1)\geq 2(7-1)=12>8.

Case 1: k=2k=2. In this case D=2ap1p2D=2^{a}p_{1}p_{2}, a{2,3}a\in\{2,3\}. One has

φ(D)D18+2D341p1+1p2+22a1p1p2.\dfrac{\varphi(D)}{D}\geq\dfrac{1}{8}+\dfrac{2}{D}\Leftrightarrow\dfrac{3}{4}\geq\dfrac{1}{p_{1}}+\dfrac{1}{p_{2}}+\dfrac{2^{2-a}-1}{p_{1}p_{2}}.

The last inequality is true because

1p1+1p2+22a1p1p213+15<34.\dfrac{1}{p_{1}}+\dfrac{1}{p_{2}}+\dfrac{2^{2-a}-1}{p_{1}p_{2}}\leq\dfrac{1}{3}+\dfrac{1}{5}<\dfrac{3}{4}.

Subcase 1.1: a=2a=2 (k=2k=2). In this case D=4p1p2D=4p_{1}p_{2}, where p1<p2p_{1}<p_{2} and p1p23(mod4)p_{1}p_{2}\equiv 3\pmod{4}. By Lemma 4.2, the graph PΔP_{\Delta} is a Ramanujan graph if and only if

φ(D)(p11)2+16φ(D)8\displaystyle\dfrac{\varphi(D)}{(p_{1}-1)^{2}}+\dfrac{16}{\varphi(D)}\leq 8 p21p11+4(p11)(p21)4\displaystyle\Leftrightarrow\dfrac{p_{2}-1}{p_{1}-1}+\dfrac{4}{(p_{1}-1)(p_{2}-1)}\leq 4
4(p11)(p21)4p21.\displaystyle\Leftrightarrow 4(p_{1}-1)-(p_{2}-1)\geq\dfrac{4}{p_{2}-1}.

In the case that p25p_{2}\geq 5, the latter condition is equivalent to

4(p11)(p21)1,4(p_{1}-1)-(p_{2}-1)\geq 1,

which is equivalent to

p24p15.p_{2}\leq 4p_{1}-5.

Subcase 1.2: a=3a=3 (k=2k=2). In this case D=8p1p2D=8p_{1}p_{2}, where 2<p1<p22<p_{1}<p_{2}. By Lemma 4.2, the graph PΔP_{\Delta} is a Ramanujan graph if and only if

φ(D)(p11)2+16φ(D)8\displaystyle\dfrac{\varphi(D)}{(p_{1}-1)^{2}}+\dfrac{16}{\varphi(D)}\leq 8 p21p11+1(p11)(p21)2\displaystyle\Leftrightarrow\dfrac{p_{2}-1}{p_{1}-1}+\dfrac{1}{(p_{1}-1)(p_{2}-1)}\leq 2
2(p11)(p21)1p21.\displaystyle\Leftrightarrow 2(p_{1}-1)-(p_{2}-1)\geq\dfrac{1}{p_{2}-1}.

The latter condition is equivalent to

2(p11)(p21)1,2(p_{1}-1)-(p_{2}-1)\geq 1,

which is equivalent to

p22p13.p_{2}\leq 2p_{1}-3.

Case 2: k=1k=1. In this case D=2ap1D=2^{a}p_{1}, a{2,3}a\in\{2,3\}. One has

φ(D)D18+2D341+22ap1.\dfrac{\varphi(D)}{D}\geq\dfrac{1}{8}+\dfrac{2}{D}\Leftrightarrow\dfrac{3}{4}\geq\dfrac{1+2^{2-a}}{p_{1}}.

The last inequality is true because

1+22ap123<34.\dfrac{1+2^{2-a}}{p_{1}}\leq\dfrac{2}{3}<\dfrac{3}{4}.

Note also that

φ(D)(p11)2+16φ(D)=2a1+25ap111032=5<8.\dfrac{\varphi(D)}{(p_{1}-1)^{2}}+\dfrac{16}{\varphi(D)}=\dfrac{2^{a-1}+2^{5-a}}{p_{1}-1}\leq\dfrac{10}{3-2}=5<8.

b) Now we consider the case DD is odd. It is well known that if Δ\Delta is a prime number then PΔP_{\Delta} is a Ramanujan graph.(For example, see the discussion after Definition 4.1.)

Suppose that DD is not a prime number and that PΔP_{\Delta} is a Ramanujan graph. Let D=p1a1p2a2pkakD=p_{1}^{a_{1}}p_{2}^{a_{2}}\cdots p_{k}^{a_{k}} be the prime factorization of DD with 2<p1<p2<<pk2<p_{1}<p_{2}<\cdots<p_{k}. Since dd is square-free, a1==ak=1a_{1}=\cdots=a_{k}=1, and k2k\geq 2. By Lemma 4.2,

8φ(D)(p11)2+16φ(D)>p21p11(pk1).8\geq\dfrac{\varphi(D)}{(p_{1}-1)^{2}}+\dfrac{16}{\varphi(D)}>\dfrac{p_{2}-1}{p_{1}-1}\cdots(p_{k}-1).

This inequality forces k3k\leq 3. In fact, if k4k\geq 4 then

p21p11(pk1)>(p31)(p41)(51)(71)=24>8.\dfrac{p_{2}-1}{p_{1}-1}\cdots(p_{k}-1)>(p_{3}-1)(p_{4}-1)\geq(5-1)(7-1)=24>8.

Case 1: k=3k=3. Since 8>p21p11(p31)8>\dfrac{p_{2}-1}{p_{1}-1}(p_{3}-1), we conclude that p3=7p_{3}=7. Hence (p1,p2)=(3,5)(p_{1},p_{2})=(3,5). However, in this case, one has p21p11(p31)>8\dfrac{p_{2}-1}{p_{1}-1}(p_{3}-1)>8, a contradiction.

Case 2: k=2k=2. In this case D=p1p2D=p_{1}p_{2}, where p1<p2p_{1}<p_{2} and p1p21(mod4)p_{1}p_{2}\equiv 1\pmod{4}. One has

φ(D)D18+14D+178D781p1+1p2+14p1p2+4p1p2.\dfrac{\varphi(D)}{D}\geq\dfrac{1}{8}+\dfrac{1}{4\sqrt{D}}+\dfrac{17}{8D}\Leftrightarrow\dfrac{7}{8}\geq\dfrac{1}{p_{1}}+\dfrac{1}{p_{2}}+\dfrac{1}{4\sqrt{p_{1}p_{2}}}+\dfrac{4}{p_{1}p_{2}}.

The last inequality is true because

1p1+1p2+14p1p2+4p1p213+15+1415+415<78.\dfrac{1}{p_{1}}+\dfrac{1}{p_{2}}+\dfrac{1}{4\sqrt{p_{1}p_{2}}}+\dfrac{4}{p_{1}p_{2}}\leq\dfrac{1}{3}+\dfrac{1}{5}+\dfrac{1}{4\sqrt{15}}+\dfrac{4}{15}<\dfrac{7}{8}.

By Lemma 4.2, the graph PΔP_{\Delta} is a Ramanujan graph if and only if

φ(D)(p11)2+16φ(D)8\displaystyle\dfrac{\varphi(D)}{(p_{1}-1)^{2}}+\dfrac{16}{\varphi(D)}\leq 8 p21p11+16(p11)(p21)8\displaystyle\Leftrightarrow\dfrac{p_{2}-1}{p_{1}-1}+\dfrac{16}{(p_{1}-1)(p_{2}-1)}\leq 8
8(p11)(p21)16p21.\displaystyle\Leftrightarrow 8(p_{1}-1)-(p_{2}-1)\geq\dfrac{16}{p_{2}-1}.

In the case that p217p_{2}\geq 17, the latter condition is clearly equivalent to

8(p11)(p21)1,8(p_{1}-1)-(p_{2}-1)\geq 1,

which is equivalent to

p28p19.p_{2}\leq 8p_{1}-9.

All pairs of primes (p1,p2)(p_{1},p_{2}) satisfying that p1<p2<17p_{1}<p_{2}<17 and p1p21(mod4)p_{1}p_{2}\equiv 1\pmod{4} are (3,7)(3,7), (3,11)(3,11), (7,11)(7,11) and (5,13)(5,13). One can check directly that for such a pair (p1,p2)(p_{1},p_{2}) we always have

8(p11)(p21)16p21 and p28p19.8(p_{1}-1)-(p_{2}-1)\geq\dfrac{16}{p_{2}-1}\text{ and }p_{2}\leq 8p_{1}-9.\qed
Remark 4.4.

There are infinitely pairs of primes (p1,p2)(p_{1},p_{2}) satisfying the conditions in part (4) of Theorem 4.3. In fact, let us first recall the prime number theorem for arithmetic progressions. Let aa, dd be two positive integers with gcd(a,d)=1\gcd(a,d)=1. Let

π(x;a,d)=|{p prime px,pa(modd)}|.\pi(x;a,d)=|\{p\text{ prime }\mid p\leq x,p\equiv a\pmod{d}\}|.

Then π(x;a,d)1φ(d)xlnx\pi(x;a,d)\sim\dfrac{1}{\varphi(d)}\dfrac{x}{\ln x} as xx\to\infty. (See for example [20, page 257].) Hence for any fixed number k>1k>1,

π(kx;a,d)π(x;a,d)=|{p prime x<pkx,pa(modd)}|k1φ(d)xlnx as x.\pi(kx;a,d)-\pi(x;a,d)=|\{p\text{ prime }\mid x<p\leq kx,\;p\equiv a\pmod{d}\}|\sim\dfrac{k-1}{\varphi(d)}\dfrac{x}{\ln x}\text{ as }x\to\infty.

In particular for xx big enough, there exists a prime pp such that

x<pkx,pa(modd).x<p\leq kx,\quad p\equiv a\pmod{d}.

Now let p1>5p_{1}>5 be a prime number congruent to 1 modulo 4 which is big enough so that there exists a prime p2p_{2} such that

p1<p23p1,p3(mod4).p_{1}<p_{2}\leq 3p_{1},\quad p\equiv 3\pmod{4}.

Then p1p23(mod4)p_{1}p_{2}\equiv 3\pmod{4} and p2<4p15p_{2}<4p_{1}-5. Hence the pair (p1,p2)(p_{1},p_{2}) satisfies the conditions in part (4) of Theorem 4.3.

Similarly, there are infinitely pairs of primes (p1,p2)(p_{1},p_{2}) satisfying the conditions in part (5), or in part (7), of Theorem 4.3.

5. Cheeger number of generalized Paley graphs

Let χ=χΔ\chi=\chi_{\Delta} be an even quadratic character (in other words, Δ>0\Delta>0) and PΔP_{\Delta} the corresponding generalized Paley graph. In this section, we provide an upper bound for the Cheeger number of PΔP_{\Delta} (also known by other names such as isoperimetric number or edge expansion ratio). Our estimation gives a natural generalization of the results in [8]. We also answer the authors’ suspicion about the relationship between the upper bounds that they used in this paper, namely the α\alpha-bound and the p14\dfrac{p-1}{4}-bound (see [8, Section 2.2].) While our result is more general and explicit, we remark that our approach is inspired by the method in [8].

First, let us recall the definition of the Cheeger number of a graph. Let G=(E,V)G=(E,V) be an undirected graph. Let FF be a subset of VV. For a subset FVF\subseteq V, the boundary of FF, denoted by F\partial{F}, is the set of all edges going from a vertex in FF to a vertex outside of FF. The Cheeger number of GG is defined as

h(G):=min{|F||F||FV(G),0<|F|12|V(G)|}.h(G):=\min\left\{\frac{|\partial{F}|}{|F|}\middle|\ F\subseteq V(G),0<|F|\leq{\tfrac{1}{2}}|V(G)|\right\}.

Cheeger number is of great importance in various scientific fields. For example, it has found applications in graph clustering, analysis of Markov chain, and image segmentation (see [15, 17, 28, 29].)

Let {α1,α2,,αk}\{\alpha_{1},\alpha_{2},\ldots,\alpha_{k}\} be the set of all elements on the interval [1,,D2]][1,\ldots,\lfloor\frac{D}{2}]] such that χ(αi)=1.\chi(\alpha_{i})=1. Note that since χ\chi is even, χ(Da)=χ(a)\chi(D-a)=\chi(a). Therefore, we can conclude that

{α1,Dα1,α2,Dα2,,αk,Dαk},\{\alpha_{1},D-\alpha_{1},\alpha_{2},D-\alpha_{2},\ldots,\alpha_{k},D-\alpha_{k}\},

is the set of a[0,D]a\in[0,D] such that χ(a)=1.\chi(a)=1. In particular, we see that k=φ(D)4.k=\dfrac{\varphi(D)}{4}.

The following proposition is a direct analog of [8, Proposition 6].

Proposition 5.1.

Let F={0,1,,D21}V(PΔ)F=\{0,1,\ldots,\lfloor\frac{D}{2}\rfloor-1\}\subset V(P_{\Delta}). Then |F|=D2|F|=\lfloor\frac{D}{2}\rfloor and

|F|=2i=1kαi.|\partial{F}|=2\sum_{i=1}^{k}\alpha_{i}.
Proof.

By definition, an element of F\partial{F} will have the form (i,i+a)(i,i+a) where iFi\in F, i+aFi+a\not\in F, and χ(a)=1.\chi(a)=1. In particular, we know that

a{α1,Dα1,α2,Dα2,,αk,Dαk}.a\in\{\alpha_{1},D-\alpha_{1},\alpha_{2},D-\alpha_{2},\ldots,\alpha_{k},D-\alpha_{k}\}.

Let us fix an index ii where 1iφ(D)4.1\leq i\leq\dfrac{\varphi(D)}{4}. The number of edges of the form (j,j+αi)(j,j+\alpha_{i}) is equal to the number of jFj\in F such that

(1) (j+αi)(modD)D2.(j+\alpha_{i})\pmod{D}\geq\lfloor\frac{D}{2}\rfloor.

Because αD2\alpha\leq\lfloor\frac{D}{2}\rfloor, we conclude that the number of jFj\in F satisfying the inequality 1 is exactly αi.\alpha_{i}. Similarly, the number of edges of the form (j,j+Dαi)(j,j+D-\alpha_{i}) is equal to the number of jFj\in F such that

(2) (j+Dαi)(modD)=(jαi)(modD)D2(j+D-\alpha_{i})\pmod{D}=(j-\alpha_{i})\pmod{D}\geq\lfloor\frac{D}{2}\rfloor

Again, the number of jj satisfying the inequality 2 is exactly αi\alpha_{i}. Summing up over all ii, we have

|F|=2i=1kαi.|\partial{F}|=2\sum_{i=1}^{k}\alpha_{i}.\qed

By definition, we conclude that

h(PΔ)2D/2i=1kαi.h(P_{\Delta})\leq\dfrac{2}{\lfloor D/2\rfloor}\sum_{i=1}^{k}\alpha_{i}.

We can further simplify the right-hand side of the above estimate using zeta values. In order to do so, we first recall the definition of the LL-function L(χ,s)L(\chi,s) attached to χ\chi

L(χ,s)=n=1χ(n)ns.L(\chi,s)=\sum_{n=1}^{\infty}\frac{\chi(n)}{n^{s}}.

This LL-function is convergent for ss\in{\mathbb{C}} such that (s)>0\Re(s)>0 and it is absolutely convergent if (s)>1.\Re(s)>1. Furthermore, there is an Euler product formula

L(χ,s)=p11χ(p)ps,L(\chi,s)=\prod_{p}\dfrac{1}{1-\chi(p)p^{-s}},

where pp runs over the set of all prime numbers. This Euler product formula shows that L(χ,s)>0L(\chi,s)>0 if ss\in{\mathbb{R}} and s>1.s>1. We are now ready to state the following theorem.

Theorem 5.2.
h(PΔ)α:=18D/2(Dφ(D)μ(D)φ(D)8DDπ2(1χ(2)4)L(2,χ)).h(P_{\Delta})\leq\alpha:=\dfrac{1}{8\lfloor D/2\rfloor}\left(D\varphi(D)-\mu(D)\varphi(D)-\dfrac{8D\sqrt{D}}{\pi^{2}}\left(1-\dfrac{\chi(2)}{4}\right)L(2,\chi)\right).
Proof.

One has

2i=1kαi=a=1,gcd(a,D)=1D/2(1+χ(a))a=a=1,gcd(a,D)=1D/2a+a=1D/2χ(a)a.2\sum_{i=1}^{k}\alpha_{i}=\sum_{a=1,\gcd(a,D)=1}^{\lfloor D/2\rfloor}(1+\chi(a))a=\sum_{a=1,\gcd(a,D)=1}^{\lfloor D/2\rfloor}a+\sum_{a=1}^{\lfloor D/2\rfloor}\chi(a)a.

By [3, Theorem 13.1],

a=1D/2χ(a)a=DDπ2(1χ(2)4)L(2,χ).\sum_{a=1}^{\lfloor D/2\rfloor}\chi(a)a=-\dfrac{D\sqrt{D}}{\pi^{2}}\left(1-\dfrac{\chi(2)}{4}\right)L(2,\chi).

By [2, Theorem],

a=1,gcd(a,D)=1D/2a=18(Dφ(D)ϵψ(D)),\sum_{a=1,\gcd(a,D)=1}^{\lfloor D/2\rfloor}a=\dfrac{1}{8}(D\varphi(D)-\epsilon\psi(D)),

where ϵ=1\epsilon=1 if DD is odd and ϵ=0\epsilon=0 if DD is even; and ψ(D)=pD(1p)\psi(D)=\prod\limits_{p\mid D}(1-p). Note that in the case that DD is odd, DD is then square-free, hence ψ(D)=μ(D)φ(D)\psi(D)=\mu(D)\varphi(D). Also if DD is even then DD is divisible by 44, hence μ(D)=0\mu(D)=0. Thus

a=1,gcd(a,D)=1D/2a=18(Dφ(D)μ(D)φ(D)).\sum_{a=1,\gcd(a,D)=1}^{\lfloor D/2\rfloor}a=\dfrac{1}{8}(D\varphi(D)-\mu(D)\varphi(D)).

Therefore

α:=2i=1kαiD/2=18D/2(Dφ(D)μ(D)φ(D)8DDπ2(1χ(2)4)L(2,χ)).\alpha:=\dfrac{2\sum\limits_{i=1}^{k}\alpha_{i}}{\lfloor D/2\rfloor}=\dfrac{1}{8\lfloor D/2\rfloor}\left(D\varphi(D)-\mu(D)\varphi(D)-\dfrac{8D\sqrt{D}}{\pi^{2}}\left(1-\dfrac{\chi(2)}{4}\right)L(2,\chi)\right).

Remark 5.3.

Let the notation be as above. If DD is even, then

α=φ(D)42Dπ2(1χ(2)4)L(2,χ).\alpha=\dfrac{\varphi(D)}{4}-\dfrac{2\sqrt{D}}{\pi^{2}}\left(1-\dfrac{\chi(2)}{4}\right)L(2,\chi).

If DD is odd, then

α\displaystyle\alpha =14(D1)(Dφ(D)μ(D)φ(D)8DDπ2(1χ(2)4)L(2,χ))\displaystyle=\dfrac{1}{4(D-1)}\left(D\varphi(D)-\mu(D)\varphi(D)-\dfrac{8D\sqrt{D}}{\pi^{2}}\left(1-\dfrac{\chi(2)}{4}\right)L(2,\chi)\right)
=φ(D)414(D1)(8DDπ2(1χ(2)4)L(2,χ)(1μ(D))φ(D)).\displaystyle=\dfrac{\varphi(D)}{4}-\dfrac{1}{4(D-1)}\left(\dfrac{8D\sqrt{D}}{\pi^{2}}(1-\dfrac{\chi(2)}{4})L(2,\chi)-(1-\mu(D))\varphi(D)\right).
Corollary 5.4.

Let the notation be as above. Then α<φ(D)4\alpha<\dfrac{\varphi(D)}{4}.

Proof.

If DD is even then the statement follows from the previous corollary and the fact that L(2,χ)>0L(2,\chi)>0. So we suppose that DD is odd. One can check that for D=5D=5 or D=13D=13 then the statement is true. So we suppose further that D17D\geq 17.

Case 1: χ(2)=1\chi(2)=1. In this case,

L(2,χ)=n=1χ(n)n21+122n=31n2=52π26>56.L(2,\chi)=\sum_{n=1}^{\infty}\dfrac{\chi(n)}{n^{2}}\geq 1+\dfrac{1}{2^{2}}-\sum_{n=3}^{\infty}\dfrac{1}{n^{2}}=\dfrac{5}{2}-\dfrac{\pi^{2}}{6}>\dfrac{5}{6}.

Hence

8DDπ2(1χ(2)4)L(2,χ)(1μ(D))φ(D)>8DDπ234562D>0.\begin{aligned} \dfrac{8D\sqrt{D}}{\pi^{2}}(1-\dfrac{\chi(2)}{4})L(2,\chi)-(1-\mu(D))\varphi(D)>\dfrac{8D\sqrt{D}}{\pi^{2}}\end{aligned}\cdot\dfrac{3}{4}\cdot\dfrac{5}{6}-2D>0.

Thus α<φ(D)4\alpha<\dfrac{\varphi(D)}{4}.

Case 2: χ(2)=1\chi(2)=1. In this case,

L(2,χ)=n=1χ(n)n21n=21n2=2π26>13.L(2,\chi)=\sum_{n=1}^{\infty}\dfrac{\chi(n)}{n^{2}}\geq 1-\sum_{n=2}^{\infty}\dfrac{1}{n^{2}}=2-\dfrac{\pi^{2}}{6}>\dfrac{1}{3}.

Hence

8DDπ2(1χ(2)4)L(2,χ)(1μ(D))φ(D)>8DDπ254132D>0.\begin{aligned} \dfrac{8D\sqrt{D}}{\pi^{2}}(1-\dfrac{\chi(2)}{4})L(2,\chi)-(1-\mu(D))\varphi(D)>\dfrac{8D\sqrt{D}}{\pi^{2}}\end{aligned}\cdot\dfrac{5}{4}\cdot\dfrac{1}{3}-2D>0.

Thus α<φ(D)4\alpha<\dfrac{\varphi(D)}{4}. ∎

In the particular case when DD is a prime of the form 4k+14k+1, Corollary 5.4 shows that the α\alpha-bound discussed in [8] is better than the p14\dfrac{p-1}{4}-bound.

Remark 5.5.

It is interesting to see whether the α\alpha-bound provides the optimal value for h(PΔ)h(P_{\Delta}); namely h(PΔ)=α.h(P_{\Delta})=\alpha. This is true in the case PΔP_{\Delta} is a cycle graph (by Corollary 2.6, this happens if Δ{5,8,12}\Delta\in\{5,8,12\}). The reason is that in this case α=4D\alpha=\dfrac{4}{D} if D{8,12}D\in\{8,12\} and α=4D1\alpha=\dfrac{4}{D-1} if D=5D=5 which is precisely the Cheeger number of PΔP_{\Delta} (by the result in [19], it is known that for a cycle graph of order nn its Cheeger number is 4n\dfrac{4}{n} if nn is even and 4n1\dfrac{4}{n-1} if nn is odd.)

Data Availability Statement

Data sharing not applicable to this article as no datasets were generated or analysed during the current study.

Acknowledgements

The third named author is very grateful to Professor Moshe Rosenfeld who kindled his interest in using number theory to attack problems in graph theory and combinatorics.

References

  • [1] R. Ayoub. An introduction to the analytic theory of numbers. Mathematical Surveys. American Mathematical Society, Providence, R.I., 1963.
  • [2] J. Baum. A number-theoretic sum. Math. Mag., 55(2):111–113, 1982.
  • [3] B. C. Berndt. Classical theorems on quadratic residues. Enseign. Math.(2), 22(3–4):261–304, 1976.
  • [4] M. Budden, N. Calkins, W. Hack, J. Lambert, and K. Thompson. Dirichlet character difference graphs. Acta Math. Univ. Comen., 82(1):21–28, 2017.
  • [5] L. Carlitz. A theorem on permutations in a finite field. Proc. Amer. Math. Soc., 11(3):456–459, 1960.
  • [6] P. Cayley. Desiderata and suggestions: No. 2. the theory of groups: graphical representation. Amer. J. Math., 1(2):174–176, 1878.
  • [7] S. K. Chebolu, J. L. Merzel, J. Mináč, L. Muller, T. T. Nguyen, F. W. Pasini, and N. D. Tân. On the joins of group rings. arXiv preprint arXiv:2208.07413, 2022.
  • [8] K. Cramer, M. Krebs, N. Shabazi, A. Shaheen, and E. Voskanian. The isoperimetric and Kazhdan constants associated to a Paley graph. Involve, 9(2):293–306, 2016.
  • [9] P. J. Davis. Circulant matrices. American Mathematical Society, 2013.
  • [10] J. Doan, J. Mináč, L. Muller, T. T. Nguyen, and F. W. Pasini. Joins of circulant matrices. Linear Algebra Appl., 650:190–209, 2022.
  • [11] J. Doan, J. Mináč, L. Muller, T. T. Nguyen, and F. W. Pasini. On the spectrum of the joins of normal matrices and applications. arXiv preprint arXiv:2207.04181, 2022.
  • [12] D. Ghinelli and J. D. Key. Codes from incidence matrices and line graphs of Paley graphs. Adv. Math. Commun., 5(1):93–108, 2011.
  • [13] G. H. Hardy and E. M. Wright. An introduction to the theory of numbers. Oxford university press, 1979.
  • [14] J. Javelle. Cryptographie Quantique: Protocoles et Graphes. PhD thesis, Université de Grenoble, 2014.
  • [15] M. Jerrum, A. Sinclair, and E. Vigoda. A polynomial-time approximation algorithm for the permanent of a matrix with nonnegative entries. J.ACM, 51(4):671–697, 2004.
  • [16] G. A. Jones. Paley and the Paley graphs. In International workshop on Isomorphisms, Symmetry and Computations in Algebraic Graph Theory, pages 155–183. Springer, 2020.
  • [17] R. Kannan, S. Vempala, and A. Vetta. On clusterings: Good, bad and spectral. J. ACM, 51(3):497–515, 2004.
  • [18] E. Kowalski. An introduction to expander graphs. Société mathématique de France, 2019.
  • [19] M. Krebs and A. Shaheen. Expander families and Cayley graphs: a beginner’s guide. Oxford University Press, 2011.
  • [20] W. J. LeVeque. Topics in number theory. Vol. I, II. Dover Publications, Inc., Mineola, NY, 2002.
  • [21] T. K. Lim and C. E. Praeger. On generalized Paley graphs and their automorphism groups. Michigan Math. J., 58(1):293–308, 2009.
  • [22] A. Lubotzky, R. Phillips, and P. Sarnak. Ramanujan graphs. Combinatorica, 8(3):261–277, 1988.
  • [23] J. Mináč, T. T. Nguyen, and N. D. Tân. On the arithmetic of generalized Fekete polynomials. arXiv preprint arXiv:2206.11778, 2022.
  • [24] J. Mináč, T. T. Nguyen, and N. D. Tân. Fekete polynomials, quadratic residues, and arithmetic. J. Number Theory, 242:532–575, 2023.
  • [25] H. L. Montgomery and R. C. Vaughan. Multiplicative number theory I: Classical theory. Cambridge Studies in Advanced Mathematics, 97. Cambridge University Press, Cambridge, 2007.
  • [26] M. R. Murty. Ramanujan graphs. J. Ramanujan Math. Soc., 18(1):1–20, 2003.
  • [27] R. E. Paley. On orthogonal matrices. J. Math. Phys., 12(1-4):311–320, 1933.
  • [28] A. Sinclair and M. Jerrum. Approximate counting, uniform generation and rapidly mixing markov chains. Inform. and Comput., 82(1):93–133, 1989.
  • [29] D. A. Spielman and S.-H. Teng. Spectral partitioning works: Planar graphs and finite element meshes. In Proceedings of 37th conference on foundations of computer science, pages 96–105. IEEE, 1996.