Conjecture on the largest noncyclic abelian group with a 2-element spanning set

From papers

Let s3s\geq 3 be an integer. An ss-spanning set of size 22 in a finite abelian group is a set of two elements whose signed sums of at most ss terms cover the group. Largest noncyclic group conjecture. If 2s+12s+1 is prime, then the largest noncyclic group with an ss-spanning set of size 22 has order 2s22s^2. If 2s+12s+1 is composite and pp is its smallest prime divisor, then the largest such noncyclic group has order

2s2+2sp212.2s^2+2s-\frac{p^2-1}{2}.

In particular, every noncyclic group GG with an ss-spanning set of size 22 satisfies

G2s2+2s4.|G|\leq 2s^2+2s-4.

The cyclic case is known, while the noncyclic case remains open when s≢1(mod3)s\not\equiv 1\pmod 3.

Progress summary

Open

The conjecture has matching constructions and broad upper bounds, but its exact answer remains unproved outside one congruence class.

A 2024 preprint formulates the largest-noncyclic-group conjecture for two-element ss-spanning sets: the proposed maximum is 2s22s^2 when 2s+12s+1 is prime, and 2s2+2s(p21)/22s^2+2s-(p^2-1)/2 otherwise. It records substantial partial results but no proof of the full statement.

Known results

  • The conjectured values are attained by explicit groups and spanning sets, so they are established as lower bounds (Theorem 1.7, 2024).
  • Every noncyclic group with such a spanning set satisfies G2s2+2s4|G|\leq 2s^2+2s-4.
  • For ss-regular rank-22 groups, the same upper bound holds; equality is characterized by Z3×Z(2s2+2s4)/3\mathbb{Z}_3\times\mathbb{Z}_{(2s^2+2s-4)/3} when s1(mod3)s\equiv1\pmod3 (Theorem 3.3, 2024).
  • Groups of the form Z2×Z2k\mathbb{Z}_2\times\mathbb{Z}_{2k} satisfy the upper bound 2s22s^2.

Current status (as of August 2026): the conjectured lower bounds and general upper bound are known, but the exact noncyclic maximum remains open for s≢1(mod3)s\not\equiv1\pmod3; no proof or counterexample was found.

Sources
Sources & referencesView supporting material

Primary source

Bela Bajnok and W. Kyle Beatty, “On the Diameter of Undirected Cayley Graphs of Finite Abelian Groups”, arXiv:2406.04045 (2024).

Solutions 1

Proof

Put

Bs={(x,y)Z2:x+ys},L=2s+1.B_s=\{(x,y)\in\mathbb Z^2:|x|+|y|\leq s\}, \qquad L=2s+1.

Let GG be a noncyclic finite abelian group with an ss-spanning pair a,ba,b. The surjection

π:Z2G,(x,y)xa+yb,\pi:\mathbb Z^2\longrightarrow G,\qquad (x,y)\longmapsto xa+yb,

has kernel Λ\Lambda. Since GG is noncyclic and two-generated,

GZc×ZckG\cong\mathbb Z_c\times\mathbb Z_{ck}

for some c2c\geq2. Choose a prime pcp\mid c. Smith normal form implies

ΛpZ2.\Lambda\subseteq p\mathbb Z^2.

Therefore reduction modulo pp induces a surjection

G(Z/pZ)2,G\twoheadrightarrow(\mathbb Z/p\mathbb Z)^2,

each fiber having size G/p2|G|/p^2. Since π(Bs)=G\pi(B_s)=G, every residue class contains at least G/p2|G|/p^2 points of BsB_s. Hence

Gp2μp(s),μp(s)=minρ(Z/pZ)2Bs(ρ+pZ2).(1)|G|\leq p^2\mu_p(s), \qquad \mu_p(s)= \min_{\rho\in(\mathbb Z/p\mathbb Z)^2} |B_s\cap(\rho+p\mathbb Z^2)|. \tag{1}

For odd pp, write

L=pm+r,0r<p.L=pm+r,\qquad 0\leq r<p.

The invertible change of coordinates modulo pp

u=x+y,v=xyu=x+y,\qquad v=x-y

identifies BsB_s with pairs

(u,v)[s,s]2,uv(mod2).(u,v)\in[-s,s]^2,\qquad u\equiv v\pmod2.

For a residue α(modp)\alpha\pmod p, let Eα,OαE_\alpha,O_\alpha count respectively the even and odd integers in [s,s][-s,s] congruent to α\alpha. Set

nα=Eα+Oα,δα=EαOα.n_\alpha=E_\alpha+O_\alpha,\qquad \delta_\alpha=E_\alpha-O_\alpha.

Then the occupancy of the residue pair (α,β)(\alpha,\beta) equals

EαEβ+OαOβ=nαnβ+δαδβ2.(2)E_\alpha E_\beta+O_\alpha O_\beta =\frac{n_\alpha n_\beta+\delta_\alpha\delta_\beta}{2}. \tag{2}

Exactly rr residue classes have size m+1m+1, and prp-r have size mm. Furthermore,

δα=0 if nα is even,δα{1,1} if nα is odd.\delta_\alpha=0\ \text{if }n_\alpha\text{ is even}, \qquad \delta_\alpha\in\{-1,1\}\ \text{if }n_\alpha\text{ is odd}.

Ordering residue classes by first appearance in the interval, the prp-r minimum-size classes occur consecutively, and their δ\delta-signs alternate whenever mm is odd. It follows from (2) that

μp(s)={m2/2,m even,(m21)/2,m odd, rp2,(m2+1)/2,m odd, r=p1.(3)\mu_p(s)= \begin{cases} m^2/2,&m\text{ even},\\[2mm] (m^2-1)/2,&m\text{ odd},\ r\leq p-2,\\[2mm] (m^2+1)/2,&m\text{ odd},\ r=p-1. \end{cases} \tag{3}

Indeed, the middle case has two minimum-size classes with opposite signs. In the last case there is exactly one minimum-size class, and its self-pair gives (m2+1)/2(m^2+1)/2; every pair involving a size-(m+1)(m+1) class has occupancy at least m(m+1)/2m(m+1)/2.

If pLp\mid L, then r=0r=0 and mm is odd, so

Gp2(m21)2=L2p22=2s2+2sp212.(4)|G| \leq\frac{p^2(m^2-1)}2 =\frac{L^2-p^2}{2} = 2s^2+2s-\frac{p^2-1}{2}. \tag{4}

If pLp\nmid L, then r1r\geq1. The first two cases of (3) give

p2μp(s)(L1)22=2s2.p^2\mu_p(s)\leq\frac{(L-1)^2}{2}=2s^2.

In the third case, r=p1r=p-1, the same conclusion reduces to

p2(m2+1)(pm+p2)2.p^2(m^2+1)\leq(pm+p-2)^2.

The difference between right and left is

2pm(p2)4p+4,2pm(p-2)-4p+4,

which is positive for p5,m1p\geq5,m\geq1, and for p=3p=3 whenever the odd integer m3m\geq3. The sole exceptional case p=3,m=1p=3,m=1 gives s=2s=2, outside the hypothesis. Thus

pL, s3G2s2.(5)p\nmid L,\ s\geq3 \quad\Longrightarrow\quad |G|\leq2s^2. \tag{5}

For p=2p=2, exactly s2s^2 points of BsB_s satisfy

x+y≢s(mod2).x+y\not\equiv s\pmod2.

If ss is even, these split equally between the residue classes (0,1)(0,1) and (1,0)(1,0), giving

μ2(s)s2/2.\mu_2(s)\leq s^2/2.

If ss is odd, they occupy (0,0)(0,0) and (1,1)(1,1), giving

μ2(s)(s21)/2.\mu_2(s)\leq(s^2-1)/2.

Equation (1) again yields G2s2|G|\leq2s^2.

If L=2s+1L=2s+1 is prime, a prime pcp\mid c cannot equal LL: formula (3) would give μL(s)=0\mu_L(s)=0, contradicting spanning. Therefore pLp\nmid L, and

G2s2.|G|\leq2s^2.

If LL is composite and p0p_0 is its least prime divisor, either pLp\nmid L, yielding (5), or pLp\mid L, in which case pp0p\geq p_0 and (4) gives

G2s2+2sp0212.|G| \leq 2s^2+2s-\frac{p_0^2-1}{2}.

Since p02Lp_0^2\leq L, this bound exceeds 2s22s^2 and therefore covers both cases.

Finally, both upper bounds are attained by the constructions already established in Theorem 1.7. The group

Zs×Z2s\mathbb Z_s\times\mathbb Z_{2s}

has the ss-spanning pair

{(0,1),(1,1)}\{(0,1),(1,1)\}

and order 2s22s^2. If L=p0tL=p_0t, the group

Zp0×Zp0(t21)/2\mathbb Z_{p_0} \times \mathbb Z_{p_0(t^2-1)/2}

has the ss-spanning pair

{(1,t12),(1,t+12)}\left\{ \left(1,\frac{t-1}{2}\right), \left(1,\frac{t+1}{2}\right) \right\}

and order

L2p022=2s2+2sp0212.\frac{L^2-p_0^2}{2} = 2s^2+2s-\frac{p_0^2-1}{2}.

These matching upper and lower bounds prove both cases of the conjecture for every s3s\geq3.

0 endorsements
Shivam Patel ·