Central-quotient eigengap conjecture for nilpotent Cayley graphs

Less than 1 year old · traced to

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

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

k>1=∣G/Zj(G)∣k_{>1}=|G/Z_j(G)|

for some 1≤j≤c1\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.

References

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).

Progress summary

Refreshed
Claimed solved

A reader-supplied calculation claims the conjecture is false already for a five-vertex cycle, but nobody has independently checked that claim.

The July 2026 preprint by Rashid Barket, Enrico Grimaldi, Yacoub Hendi, Edward Hirst, Adam Onus, and Harmeet Singh formulates this as a testable conjecture arising from empirical spectral regularities in nilpotent Cayley graphs.

Posted attempt

A reader-supplied argument claims a complete counterexample: for G=Z/5ZG=\mathbb{Z}/5\mathbb{Z} and the underlying cycle C5C_5, it computes k>1=3k_{>1}=3, whereas the conjecture permits only ∣G/Z1(G)∣=1|G/Z_1(G)|=1 or ∣G∣−1=4|G|-1=4. It also claims minimality and uniqueness among simple cycles, but the calculation has not been independently verified.

Current status (as of August 2026): the conjecture has no confirmed proof or counterexample; the claimed C5C_5 counterexample remains unverified.

Sources

Solutions 1

CounterexampleThis solution needs a summarySee full solutionHide full solution

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

and orders its eigenvalues with multiplicity as

0=λ1≤λ2≤⋯≤λ∣G∣≤2.0=\lambda_1 \le\lambda_2 \le\cdots\le\lambda_{|G|} \le2.

Whenever a consecutive normalized eigengap exceeds one, define

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

Conjecture 4.4 asserts that either

k>1=∣G∣−1,(3)k_{>1}=|G|-1, \tag{3}

or

k>1=∣G/Zj(G)∣for some 1≤j≤c.(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

x⟼x+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(2−100−1−12−1000−12−1000−12−1−100−12).(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⁡(xI−L(C5))=x16(4x2−10x+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=5−54,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

λ2−λ1=5−54<1,λ3−λ2=0,λ4−λ3=52>1,λ5−λ4=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

∣G∣−1=4.|G|-1=4.

Consequently

k>1=3∉{∣G/Z1(G)∣, ∣G∣−1}={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,(nn−1)(n−1)}.\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,n≥3.\operatorname{Cay} \left(\mathbb Z/n\mathbb Z,\{1,-1\}\right) =C_n, \qquad n\ge3.

Its normalized eigenvalues are

1−cos⁡(2πjn),0≤j<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 n≥6n\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 n≥6n\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.