Papers
Topics
Authors
Recent
Search
2000 character limit reached

Paley Graphs: Algebraic and Spectral Insights

Updated 31 December 2025
  • Paley graphs are finite graphs constructed over finite fields where vertices represent field elements and edges exist if their difference is a quadratic residue.
  • They are strongly regular and quasi-random, with explicit spectral properties that bridge arithmetic and combinatorial graph theory.
  • Their unique structure supports applications in coding theory, extremal combinatorics, and enables advanced SDP relaxations for bounding clique numbers.

A Paley graph is a fundamental object in algebraic combinatorics, defined over a finite field of odd characteristic, in which adjacency encodes quadratic residue structure. These graphs integrate arithmetic, algebraic, and spectral properties, serving as archetypes for strongly regular and quasi-random graphs, and they catalyze connections to coding theory, extremal combinatorics, spectral graph theory, and arithmetic Ramsey theory.

1. Definition and Construction

Let q=peq=p^e be a prime power with q1(mod4)q\equiv1\pmod4, and let Fq\mathbb{F}_q denote the finite field of order qq. Define the subgroup of nonzero quadratic residues,

Q={xFq:x is a square}.Q = \{\,x\in\mathbb{F}_q^* : x \text{ is a square} \,\}.

The Paley graph P(q)P(q) has vertex set V(P(q))=FqV(P(q)) = \mathbb{F}_q, with two distinct vertices x,yx, y adjacent if and only if xyQx-y\in Q. The adjacency relation is symmetric, because 1-1 is a square in Fq\mathbb{F}_q when q1(mod4)q\equiv1\pmod4.

  • P(q)P(q) is an undirected, loopless, (q1)/2(q-1)/2-regular graph.
  • P(q)P(q) is self-complementary, as for any quadratic nonresidue rFqr\in\mathbb{F}_q^*, the map xrxx\mapsto r x is an automorphism exchanging edges and non-edges (Elsawy, 2012, Jones, 2017).
  • P(q)P(q) is a strongly regular graph with parameters:

v=q,k=q12,λ=q54,μ=q14v=q, \qquad k=\frac{q-1}{2},\qquad \lambda=\frac{q-5}{4},\qquad \mu=\frac{q-1}{4}

where λ\lambda and μ\mu count common neighbors for adjacent and non-adjacent pairs, respectively (Elsawy, 2012, Kim et al., 2024).

2. Spectral and Quasi-Random Properties

Let AA be the adjacency matrix of P(q)P(q). The spectrum is explicit:

  • Eigenvalues:

λ1=q12 (simple),λ2=1+q2, λ3=1q2\lambda_1 = \frac{q-1}{2} \ \text{(simple)}, \qquad \lambda_2 = \frac{-1+\sqrt{q}}{2},\ \lambda_3 = \frac{-1-\sqrt{q}}{2}

both with multiplicity (q1)/2(q-1)/2 (Mináč et al., 2022, Jones, 2017, Kim et al., 2024).

  • λ2=λ3=12q+O(1)=o(q)|\lambda_2| = |\lambda_3| = \frac{1}{2}\sqrt{q} + O(1) = o(q).

By Chung–Graham–Wilson (Kim et al., 2024), a sequence of graphs with edge-density $1/2$ is quasi-random if the edge distribution in each induced subgraph matches that of a random (density-$1/2$) graph to within o(n2)o(n^2). The expander-mixing lemma shows that Paley graphs P(q)P(q) meet these criteria:

e(S)=14S2+o(q2)e(S) = \frac{1}{4}|S|^2 + o(q^2)

for all SV(P(q))S\subset V(P(q)), and the second-largest eigenvalue is o(q)o(q), so P(q)P(q) is a standard example of a quasi-random graph.

3. Automorphism Group and Symmetries

The full automorphism group of P(q)P(q) is

Aut(P(q))={xaxγ+b:a(Fq)2,bFq,γGal(Fq/Fp)},\mathrm{Aut}(P(q)) = \{\, x\mapsto a x^\gamma + b : a\in(\mathbb{F}_q^*)^2,\, b\in\mathbb{F}_q,\, \gamma\in\mathrm{Gal}(\mathbb{F}_q/\mathbb{F}_p)\,\},

a semidirect product AΔL1(q)(Fq)2FqGal(Fq/Fp)A\Delta L_1(q) \cong (\mathbb{F}_q^*)^2 \ltimes \mathbb{F}_q \rtimes \mathrm{Gal}(\mathbb{F}_q/\mathbb{F}_p) (Jones, 2017). The action is vertex- and edge-transitive; every affine map with square multiplier is a graph automorphism.

P(q)P(q) is self-complementary via multiplication by any quadratic nonresidue.

4. Extremal Subgraph Structure and SDP Bounds

Clique and independence numbers: By classic Fourier methods and subsequent quasi-random analysis, for n=V(P(q))n=|V(P(q))|,

ω(P(q)), α(P(q))(1o(1))log3.008q\omega(P(q)),\ \alpha(P(q)) \geq (1-o(1))\log_{3.008} q

where ω\omega is the clique number and α\alpha the independence number (Kim et al., 2024).

SDP relaxations: The clique number satisfies the classical upper bound ω(P(q))q\omega(P(q))\leq\sqrt{q} (Kobzar et al., 2023), with recent computational evidence (block-diagonal SDP relaxations, such as L2L^2 and SOS-4) indicating actual growth may be sub-q\sqrt{q}: numerically, L2(P(q))=O(q0.456)L^2(P(q)) = O(q^{0.456}) (Kobzar et al., 2023).

  • The Lovász ϑ\vartheta function equals q\sqrt{q} and coincides with the value at the first level of the exact subgraph hierarchy (ESH).
  • The ESH remains at the Lovász bound up to level k0(q+3)/2k_0\sim(\sqrt{q}+3)/2; the local ESH, exploiting vertex-transitivity, gives strictly improved upper bounds already at low levels and is at least as tight as ESH (Gaar et al., 2024).

Table: Numerical Comparison for Small qq (Gaar et al., 2024) | qq | α(Pq)\alpha(P_q) | ϑ(Pq)\vartheta(P_q) | z2(Pq)z_2(P_q) | z2(Pq)z'_2(P_q) | |-----|--------------|------------------|------------|-------------| | 13 | 3 | 3.6056 | 3.6056 | 3.0000 | | 29 | 4 | 5.3852 | 5.3852 | 4.3177 |

5. Extremal and Combinatorial Properties

  • Pancyclicity: For q1(mod4)q\equiv 1\pmod4, q5q\ne5, P(q)P(q) is pancyclic: it contains cycles of every possible length 3kq3\leq k\leq q (Nishimura, 2023).
  • Subgraph enumeration: For the number of triangles and 4-cliques, explicit formulas in terms of Jacobi sums are known. For q1(mod4)q\equiv 1\pmod4,

K3(P(q))=q(q1)(q5)243,K4(P(q))=q(q1)((q9)24y2)293\mathcal{K}_3(P(q)) = \frac{q(q-1)(q-5)}{24\cdot3},\quad \mathcal{K}_4(P(q)) = \frac{q(q-1)((q-9)^2-4y^2)}{2^9\cdot3}

where q=x2+y2q=x^2+y^2 with x1(mod4)x\equiv 1\pmod 4, yy even (Dawsey et al., 2020).

  • Even induced subgraphs: The number of even induced subgraphs of Paley graphs matches that in random models for small sizes; the parity structure corresponds to the enumeration of MDS self-dual codes (Li et al., 22 Dec 2025).

6. Generalizations of Paley Graphs

The Paley construction motivates several generalizations:

  • Generalized Paley graphs: For kk dividing q1q-1 (and q1mod2kq\equiv 1\bmod 2k when qq odd), define adjacency via kk-th power residues:

V=Fq,  {x,y}E    xy(Fq)k.V = \mathbb{F}_q,\ \ \{x,y\} \in E \iff x-y \in (\mathbb{F}_q^*)^k.

Regular of degree (q1)/k(q-1)/k, often not strongly regular for k>2k>2 (Elsawy, 2012, Schneider et al., 2013, Bonini et al., 2024).

  • Automorphism group: For large qq relative to kk, Aut(GP(q,(q1)/k))AΓL(1,q)\mathrm{Aut}(\mathrm{GP}(q,(q-1)/k)) \leq \mathrm{A}\Gamma\mathrm{L}(1,q) (Ponomarenko, 23 Nov 2025).
  • Paley graphs in characteristic $2$: A distinct construction exists, using the trace map and Möbius transformations, resulting in a self-complementary, vertex-transitive pseudo-random graph on q+1q+1 points for q=2kq=2^k (Thomason, 2015).
  • Paley graphs over Zn\mathbb{Z}_n: With nn restricted to ensure 1-1 is a square in the unit group, a version exists for rings, where the underlying group is Zn\mathbb{Z}_n and adjacency is defined by units that are squares modulo nn (Bhowmik et al., 2020).

7. Applications and Structural Invariants

  • Random models: The multiplicative random-graph model more faithfully captures the clique-number fluctuations of Paley graphs than purely random Cayley graphs, matching the Graham–Ringrose phenomenon for cliques of size Ω(logplogloglogp)\Omega(\log p\log\log\log p) (Mrazović, 2016).
  • Coding theory: There is a tight connection between even/odd subgraph structure in Paley graphs and the existence and enumeration of MDS self-dual (extended) GRS codes (Li et al., 22 Dec 2025).
  • Critical group / Smith normal form: The critical (sandpile) group and Smith group of the adjacency matrix for P(q)P(q) are described explicitly in terms of the field order; the primary decomposition involves detailed number-theoretic and character-sum data (Chandler et al., 2014).

8. Infinite and Arithmetic Variants

  • Infinite Paley graphs: The direct limits of Paley graphs over towers of extensions yield, up to isomorphism, the universal Erdős–Rényi–Rado random graph RR for any (locally finite, infinite) field of odd characteristic. This is established via the extension property and Weil’s character sum estimates (Jones, 2019).
  • Ramanujan and energy properties: The spectrum of Paley and generalized Paley graphs provides explicit classes of Ramanujan graphs and infinite families of non-strongly regular, non-bipartite graphs which are equienergetic with their complements (Mináč et al., 2022, Podestá et al., 2022).

References:

Topic to Video (Beta)

Whiteboard

No one has generated a whiteboard explanation for this topic yet.

Follow Topic

Get notified by email when new papers are published related to Paley Graphs.