A four-regular bipartite Cayley graph with point spectrum C∖{0}\mathbb C\setminus\{0\} on CG\mathbb C^G

15.89

Problem

For every infinite connected vertex-transitive graph of finite valency, is every complex number an adjacency eigenvalue on the space of all complex-valued functions?

∀λ∈C,∃f∈CV(Γ)∖{0}:∑u∼vf(u)=λf(v)(v∈V(Γ))  ?\forall\lambda\in\mathbb C,\quad \exists f\in\mathbb C^{V(\Gamma)}\setminus\{0\}:\quad \sum_{u\sim v}f(u)=\lambda f(v)\quad(v\in V(\Gamma))\;?

There is no finite-support, boundedness, or summability condition on ff.

The graph and its point spectrum

For a locally finite simple graph Γ\Gamma, its adjacency operator on the space of all complex-valued functions is

AΓ:CV(Γ)⟶CV(Γ),(AΓf)(v)=∑u∼vf(u).A_\Gamma:\mathbb C^{V(\Gamma)}\longrightarrow\mathbb C^{V(\Gamma)},\qquad (A_\Gamma f)(v)=\sum_{u\sim v}f(u).

Each sum is finite. We write

Eig⁡(AΓ)={λ∈C:∃f∈CV(Γ)∖{0}, AΓf=λf}.\operatorname{Eig}(A_\Gamma)= \{\lambda\in\mathbb C:\exists f\in\mathbb C^{V(\Gamma)}\setminus\{0\},\ A_\Gamma f=\lambda f\}.

There is no support, boundedness or summability condition on ff. Problem 15.89 of the Kourovka Notebook, proposed by V. I. Trofimov, asks whether Eig⁡(AΓ)=C\operatorname{Eig}(A_\Gamma)=\mathbb C for every infinite connected vertex-transitive graph of finite valency (1).

Theorem 1. Put

G=C3∗C2=⟨a,t∣a3=t2=1⟩,S={ata,ata2,a2ta,a2ta2}.G=C_3*C_2=\langle a,t\mid a^3=t^2=1\rangle, \qquad S=\{ata,ata^2,a^2ta,a^2ta^2\}.

The graph Γ=Cay⁡(G,S)\Gamma=\operatorname{Cay}(G,S) is infinite, connected, simple, undirected, four-regular, bipartite and vertex-transitive. On CG\mathbb C^G its adjacency operator AA is bijective and

Eig⁡(A)=C∖{0}.(4) \tag{4} \operatorname{Eig}(A)=\mathbb C\setminus\{0\}.

More precisely, for (Rgf)(x)=f(xg)(R_gf)(x)=f(xg),

A−1=14(Ra+Ra2−Id)Rt(Ra+Ra2−Id).(5) \tag{5} A^{-1}=\tfrac14(R_a+R_{a^2}-\mathrm{Id})R_t(R_a+R_{a^2}-\mathrm{Id}).

Trofimov gives a finite-support criterion for a missing eigenvalue (2, Theorem 5.3). His complex examples in §§8.2–8.4 have several vertex orbits; the vertex-transitive examples in §8.1 use positive characteristic. The graph in (3) has unbounded degrees. Theorem 1 gives a four-regular vertex-transitive complex example with explicit point spectrum.

The graph also has a description by the three-arc construction (4). Put H=⟨a⟩H=\langle a\rangle. The graph with vertices gHgH and edges {gH,gtH}\{gH,gtH\} is the cubic tree: reduced normal forms in C3∗C2C_3*C_2 give connectedness and exclude non-backtracking closed paths. The map g↦(gH,gtH)g\mapsto(gH,gtH) is a bijection onto its oriented edges. For an oriented edge (u,v)(u,v), three-arc adjacency joins it to (x,y)(x,y) when (v,u,x,y)(v,u,x,y) is a non-backtracking path of length three. Taking (u,v)=(gH,gtH)(u,v)=(gH,gtH) gives exactly x=gaitHx=ga^itH and y=gaitajtHy=ga^ita^jtH, where i,j∈{1,2}i,j\in\{1,2\}. Thus Γ\Gamma is the three-arc graph of the cubic tree; the spectral assertion concerns this established construction.

Verification of the graph properties

We use composition as multiplication of affine maps. Fix a primitive cube root of unity ω∈C\omega\in\mathbb C, so that

ω3=1,ω≠1,1+ω+ω2=0.(6) \tag{6} \omega^3=1,\qquad \omega\ne1,\qquad 1+\omega+\omega^2=0.

The maps

α(z)=ωz,τ(z)=1−z\alpha(z)=\omega z,\qquad \tau(z)=1-z

satisfy α3=τ2=Id\alpha^3=\tau^2=\mathrm{Id}. They therefore determine a homomorphism from GG to the affine group of C\mathbb C, taking aa to α\alpha and tt to τ\tau. On the invariant lattice Z+Zω\mathbb Z+\mathbb Z\omega, their coordinate maps are α(u,v)=(−v,u−v)\alpha(u,v)=(-v,u-v) and τ(u,v)=(1−u,−v)\tau(u,v)=(1-u,-v).

Lemma 2. The four elements of SS are distinct, nonidentity, inverse-closed and generate GG. The group GG is infinite.

Proof. For i,j∈{1,2}i,j\in\{1,2\} the affine image of aitaja^ita^j is

z⟼ωi−ωi+jz.(8) \tag{8} z\longmapsto\omega^i-\omega^{i+j}z.

Its constant term is nonzero. Equality of two such maps implies equality of their constant terms, hence of ii; equality of their linear terms then implies equality of jj. Thus the elements are distinct and nonidentity. Moreover,

(aitaj)−1=a3−jta3−i∈S.(a^ita^j)^{-1}=a^{3-j}ta^{3-i}\in S.

The identities

(ata)(a2ta)=a2,(a2)2=a,a2(ata)a2=t(ata)(a^2ta)=a^2,\qquad (a^2)^2=a, \qquad a^2(ata)a^2=t

show that SS generates GG.

The affine image of tata−1tat a^{-1} is the translation

z⟼z+1−ω.z\longmapsto z+1-\omega.

Its kkth power translates by k(1−ω)k(1-\omega), which is nonzero for every positive integer kk. Thus GG contains an element of infinite order. ◻

Define adjacency by g∼hg\sim h if h=gsh=gs for some s∈Ss\in S. Lemma 2 gives symmetry and absence of loops. Cancellation in GG identifies the neighbors of gg with the four elements of SS. Generation by SS gives connectedness. For every b∈Gb\in G, left multiplication g↦bgg\mapsto bg preserves adjacency, and these automorphisms act transitively on vertices.

Finally, the homomorphism

ε:G⟶Z/2Z,ε(a)=0,ε(t)=1\varepsilon:G\longrightarrow\mathbb Z/2\mathbb Z,\qquad \varepsilon(a)=0,\quad\varepsilon(t)=1

is well-defined by the presentation. Every s∈Ss\in S has ε(s)=1\varepsilon(s)=1. Hence every edge joins the two fibers of ε\varepsilon, proving bipartiteness.

The finite-range inverse

Translations on CG\mathbb C^G satisfy RgRh=RghR_gR_h=R_{gh}. Put

U=Ra+Ra2,T=Rt.U=R_a+R_{a^2},\qquad T=R_t.

On each left coset of ⟨a⟩\langle a\rangle, UU is the matrix J−IJ-I with JJ the 3×33\times3 all-ones matrix, as in the local decomposition of the non-backtracking operator (5). The adjacency sum factors as

A=UTU.(14) \tag{14} A=UTU.

The defining group relations give

U2=Ra2+2Ra3+Ra4=U+2Id,T2=Id.\begin{align*} U^2&=R_{a^2}+2R_{a^3}+R_{a^4}=U+2\mathrm{Id}, \tag{15}\\ T^2&=\mathrm{Id}.\nonumber \end{align*}

Consequently the operator V=(U−Id)/2V=(U-\mathrm{Id})/2 satisfies

UV=VU=Id.UV=VU=\mathrm{Id}.

Using (14),

(VTV)A=VT(VU)TU=VU=Id,A(VTV)=UT(UV)TV=UV=Id.(VTV)A=VT(VU)TU=VU=\mathrm{Id}, \qquad A(VTV)=UT(UV)TV=UV=\mathrm{Id}.

This proves (5). Expanding it yields

4A−1=Rata+Rata2+Ra2ta+Ra2ta2−Rat−Ra2t−Rta−Rta2+Rt.\begin{align*} 4A^{-1}={}&R_{ata}+R_{ata^2}+R_{a^2ta}+R_{a^2ta^2} \tag{18}\\ &-R_{at}-R_{a^2t}-R_{ta}-R_{ta^2}+R_t.\nonumber \end{align*}

These finite operator identities imply

Af=0⟹f=A−1Af=0,Af=0\quad\Longrightarrow\quad f=A^{-1}Af=0,

so 0∉Eig⁡(A)0\notin\operatorname{Eig}(A).

Eigenfunctions for every nonzero complex number

For s∈Cs\in\mathbb C, consider the matrices

D=(100ω),Bs=(s11−s2−s).D=\begin{pmatrix}1&0\\0&\omega\end{pmatrix},\qquad B_s=\begin{pmatrix}s&1\\1-s^2&-s\end{pmatrix}.

They satisfy

D3=I,Bs2=I.D^3=I,\qquad B_s^2=I.

The presentation of GG therefore defines a representation

ρs:G⟶GL⁡2(C),ρs(a)=D,ρs(t)=Bs.\rho_s:G\longrightarrow\operatorname{GL}_2(\mathbb C), \qquad \rho_s(a)=D,\quad \rho_s(t)=B_s.

By (6), D+D2=diag⁡(2,−1)D+D^2=\operatorname{diag}(2,-1). Thus

∑u∈Sρs(u)=(D+D2)Bs(D+D2)=Ms:=(4s−2−2(1−s2)−s).(23) \tag{23} \sum_{u\in S}\rho_s(u) =(D+D^2)B_s(D+D^2) =M_s:=\begin{pmatrix}4s&-2\\-2(1-s^2)&-s\end{pmatrix}.

Its characteristic polynomial is

det⁡(λI−Ms)=λ2−3sλ−4.(24) \tag{24} \det(\lambda I-M_s)=\lambda^2-3s\lambda-4.

Fix λ≠0\lambda\ne0 and set

sλ=λ2−43λ,vλ=(24sλ−λ).(25) \tag{25} s_\lambda=\frac{\lambda^2-4}{3\lambda},\qquad v_\lambda=\begin{pmatrix}2\\4s_\lambda-\lambda\end{pmatrix}.

Then

Msλvλ=λvλ.M_{s_\lambda}v_\lambda=\lambda v_\lambda.

Indeed, the first coordinate is 8sλ−2(4sλ−λ)=2λ8s_\lambda-2(4s_\lambda-\lambda) =2\lambda; the second is −4+sλλ-4+s_\lambda\lambda, equal to λ(4sλ−λ)\lambda(4s_\lambda-\lambda) by λ2−3sλλ−4=0\lambda^2-3s_\lambda\lambda-4=0.

Define the matrix coefficient

fλ(g)=(1  0)ρsλ(g)vλ(g∈G).(27) \tag{27} f_\lambda(g)=(1\ \ 0)\rho_{s_\lambda}(g)v_\lambda \qquad(g\in G).

It is nonzero since fλ(1)=2f_\lambda(1)=2. Equations (23)–(27) give, for every g∈Gg\in G,

(Afλ)(g)=∑u∈S(1  0)ρsλ(gu)vλ=(1  0)ρsλ(g)(∑u∈Sρsλ(u))vλ=(1  0)ρsλ(g)Msλvλ=λfλ(g).\begin{align*} (Af_\lambda)(g) &=\sum_{u\in S}(1\ \ 0)\rho_{s_\lambda}(gu)v_\lambda\\ &=(1\ \ 0)\rho_{s_\lambda}(g) \left(\sum_{u\in S}\rho_{s_\lambda}(u)\right)v_\lambda\\ &=(1\ \ 0)\rho_{s_\lambda}(g)M_{s_\lambda}v_\lambda =\lambda f_\lambda(g). \end{align*}

Thus every λ≠0\lambda\ne0 belongs to Eig⁡(A)\operatorname{Eig}(A). Together with (5), this proves Theorem 1.

References

Preprint · Lean (GitHub)

  1. E. I. Khukhro and V. D. Mazurov (eds.), The Kourovka Notebook: Unsolved Problems in Group Theory, 21st ed., version of 1 September 2026. arXiv:1401.0300v46.
  1. V. I. Trofimov, On adjacency operators of locally finite graphs, Izv. Math. 88 (2024), no. 3, 542–589. MathNet: im9408. Author version arXiv:2208.01926v3.
  1. V. I. Trofimov, Infinite locally finite connected graphs with countable complements in C\mathbb C of the sets of eigenvalues, Proc. Steklov Inst. Math. 329 (2025), suppl. 1, S250–S256. doi:10.1134/S0081543825600942.
  1. G. Xu and S. Zhou, Three-arc graphs: characterization and domination, Discrete Applied Mathematics 193 (2015), 200–207. doi:10.1016/j.dam.2015.04.019.
  1. O. Angel, J. Friedman and S. Hoory, The non-backtracking spectrum of the universal cover of a graph, Trans. Amer. Math. Soc. 367 (2015), no. 6, 4287–4318. Author version arXiv:0712.0192.