Central-quotient eigengap conjecture for nilpotent Cayley graphs

From papers

Let GG be a finite nilpotent group with upper central series

1=Z0(G)Z1(G)Zc(G)=G,1=Z_0(G)\leq Z_1(G)\leq \cdots \leq Z_c(G)=G,

and let Γ=Cay(G,S)\Gamma=\operatorname{Cay}(G,S) be the underlying undirected Cayley graph associated to a chosen generating set SS. Let λi\lambda_i denote the eigenvalues of the normalised Laplacian spectrum, and define

k>1=min{i:λi+1λi>1}.k_{>1}=\min\{i:\lambda_{i+1}-\lambda_i>1\}.

Central-quotient eigengap conjecture. Either k>1=G1k_{>1}=|G|-1, or

k>1=G/Zj(G)k_{>1}=|G/Z_j(G)|

for some 1jc1\leq j\leq c. The conjecture proposes that the first dominant eigengap in these Cayley graphs occurs at an index determined by a quotient by a term of the upper central series, reflecting the layered structure of finite nilpotent groups. Its validity beyond the empirical computations described in the source remains open.

Progress summary

Open

A July 2026 preprint appears to study the conjecture, but no independently verified proof or counterexample was found.

The conjecture predicts that the first normalized-Laplacian eigengap exceeding one in an undirected Cayley graph of a finite nilpotent group occurs either at the final possible index or at an index determined by a quotient by the upper central series. The conjecture is publicly associated with a July 2026 preprint by Rashid Barket, Enrico Grimaldi, Yacoub Hendi, Edward Hirst, Adam Onus, and Harmeet Singh.

July 2026 preprint

The preprint is publicly listed and was announced with its title and author list, but the retrieved material does not independently establish whether it proves or refutes this conjecture. No verified proof, counterexample, exposition, or response was found.

Current status (as of August 2026): the conjecture remains open on the verified record, with no confirmed proof or counterexample.

Sources
Sources & referencesView supporting material

Primary source

Rashid Barket, Enrico Grimaldi, Yacoub Hendi, Edward Hirst, Adam Onus and Harmeet Singh, “Learning the Graphical Nature of Symmetries”, arXiv:2607.12026 (2026).

Solutions 1

Counterexample

A minimal counterexample to the central-quotient eigengap conjecture

Source. Rashid Barket, Enrico Grimaldi, Yacoub Hendi, Edward Hirst, Adam Onus, and Harmeet Singh, Learning the Graphical Nature of Symmetries, arXiv:2607.12026v1, Section 4.4, Conjecture 4.4. The conjecture concerns the normalized Laplacian of the underlying undirected Cayley graph and uses eigenvalues indexed starting at one.

We disprove the conjecture using the cyclic group of order five. Moreover, this is the smallest possible counterexample by group order, and it is the only counterexample among Cayley graphs that are simple cycles.

1. The precise conjecture

Let GG be a finite nilpotent group with upper central series

1=Z0(G)Z1(G)Zc(G)=G.1=Z_0(G) \le Z_1(G) \le\cdots\le Z_c(G)=G.

Given a generating set SS, let Γ\Gamma be the underlying undirected graph of Cay(G,S)\operatorname{Cay}(G,S). Write AA and DD for its adjacency and degree matrices. The source defines the normalized Laplacian by

L=ID1/2AD1/2,(1)\mathcal L = I-D^{-1/2}AD^{-1/2}, \tag{1}

and orders its eigenvalues with multiplicity as

0=λ1λ2λG2.0=\lambda_1 \le\lambda_2 \le\cdots\le\lambda_{|G|} \le2.

Whenever a consecutive normalized eigengap exceeds one, define

k>1=min{i:λi+1λi>1}.(2)k_{>1} = \min\left\{ i:\lambda_{i+1}-\lambda_i>1 \right\}. \tag{2}

Conjecture 4.4 asserts that either

k>1=G1,(3)k_{>1}=|G|-1, \tag{3}

or

k>1=G/Zj(G)for some 1jc.(4)k_{>1} = |G/Z_j(G)| \quad\text{for some }1\le j\le c. \tag{4}

2. A five-vertex cyclic Cayley graph

Take the finite cyclic group

G=Z/5ZG=\mathbb Z/5\mathbb Z

and its single-element generating set

S={1}.S=\{1\}.

The directed Cayley graph has the arcs

xx+1(mod5).x\longmapsto x+1\pmod 5.

Its underlying undirected Cayley graph is therefore the five-cycle

Γ=C5.(5)\Gamma=C_5. \tag{5}

Equivalently, one may start directly with the inverse-closed generating set S={1,1}S=\{1,-1\} and obtain exactly the same undirected graph.

Every vertex of C5C_5 has degree two. Consequently, in the cyclic vertex order 0,1,2,3,40,1,2,3,4, its normalized Laplacian is

L(C5)=12(2100112100012100012110012).(6)\mathcal L(C_5) = \frac12 \begin{pmatrix} 2&-1&0&0&-1 \\ -1&2&-1&0&0 \\ 0&-1&2&-1&0 \\ 0&0&-1&2&-1 \\ -1&0&0&-1&2 \end{pmatrix}. \tag{6}

Its characteristic polynomial factors exactly as

det(xIL(C5))=x16(4x210x+5)2.(7)\det\left(xI-\mathcal L(C_5)\right) = \frac{x}{16} \left(4x^2-10x+5\right)^2. \tag{7}

Thus the ordered normalized Laplacian eigenvalues, including multiplicity, are

λ1=0,λ2=λ3=554,λ4=λ5=5+54.(8)\begin{aligned} \lambda_1&=0, \\ \lambda_2=\lambda_3 &=\frac{5-\sqrt5}{4}, \\ \lambda_4=\lambda_5 &=\frac{5+\sqrt5}{4}. \end{aligned} \tag{8}

The consecutive gaps are therefore

λ2λ1=554<1,λ3λ2=0,λ4λ3=52>1,λ5λ4=0.(9)\begin{aligned} \lambda_2-\lambda_1 &=\frac{5-\sqrt5}{4}<1, \\ \lambda_3-\lambda_2 &=0, \\ \lambda_4-\lambda_3 &=\frac{\sqrt5}{2}>1, \\ \lambda_5-\lambda_4 &=0. \end{aligned} \tag{9}

The strict middle inequality follows from 5>45>4. Hence the first gap greater than one occurs at the one-based index

k>1=3.(10)\boxed{k_{>1}=3.} \tag{10}

Since GG is abelian, it is nilpotent of class one and

Z1(G)=G.Z_1(G)=G.

Its only quotient permitted in (4) therefore has order

G/Z1(G)=1,|G/Z_1(G)|=1,

whereas the exceptional index permitted in (3) is

G1=4.|G|-1=4.

Consequently

k>1=3{G/Z1(G), G1}={1,4}.(11)\boxed{ k_{>1}=3 \notin \left\{ |G/Z_1(G)|,\ |G|-1 \right\} =\{1,4\}. } \tag{11}

This contradicts Conjecture 4.4 for a finite nilpotent group, a valid minimal generating set, and precisely the normalized undirected spectrum stipulated by the source.

3. Minimality and the complete cycle-family behavior

No smaller finite group supplies a counterexample. Every group of order at most four is abelian. A connected undirected Cayley graph of order two or three is respectively K2K_2 or K3K_3. A connected undirected Cayley graph of order four, being regular, is either C4C_4 or K4K_4.

For a complete graph, the normalized Laplacian spectrum is

spec(L(Kn))={0,(nn1)(n1)}.\operatorname{spec}\left(\mathcal L(K_n)\right) = \left\{ 0, \left(\frac{n}{n-1}\right)^{(n-1)} \right\}.

The only gap greater than one is at index 11, exactly the permitted central-quotient order. For C4C_4, the spectrum is

spec(L(C4))={0,1,1,2},\operatorname{spec}\left(\mathcal L(C_4)\right) =\{0,1,1,2\},

so there is no gap strictly greater than one and the conjecture's conditional index is undefined. Therefore order five is the sharp minimum.

More generally, consider the standard cyclic Cayley graph

Cay(Z/nZ,{1,1})=Cn,n3.\operatorname{Cay} \left(\mathbb Z/n\mathbb Z,\{1,-1\}\right) =C_n, \qquad n\ge3.

Its normalized eigenvalues are

1cos(2πjn),0j<n.(12)1-\cos\left(\frac{2\pi j}{n}\right), \qquad 0\le j<n. \tag{12}

After sorting and retaining multiplicities, every nonzero consecutive gap is of the form

cos(2πjn)cos(2π(j+1)n)=2sin((2j+1)πn)sin(πn)2sin(πn).(13)\begin{aligned} \cos\left(\frac{2\pi j}{n}\right) -\cos\left(\frac{2\pi(j+1)}{n}\right) &= 2\sin\left(\frac{(2j+1)\pi}{n}\right) \sin\left(\frac{\pi}{n}\right) \\ &\le 2\sin\left(\frac{\pi}{n}\right). \end{aligned} \tag{13}

For every n6n\ge6,

2sin(πn)2sin(π6)=1.2\sin\left(\frac{\pi}{n}\right) \le 2\sin\left(\frac{\pi}{6}\right) =1.

Thus k>1k_{>1} is undefined for every cycle CnC_n with n6n\ge6. The triangle C3=K3C_3=K_3 has k>1=1k_{>1}=1, the square C4C_4 has no qualifying gap, and C5C_5 has the forbidden index 33. Hence C5C_5 is the unique counterexample within the complete family of cyclic Cayley graphs generated by a single element.

0 endorsements
Shivam Patel ·