Problem
For a fixed prime p, does sufficiently large nilpotency class force the mixed variety generated by a finite p-group to be an ordinary variety?
∃cp∀G finite p-group,cl(G)≥cp⟹mvar(G) ordinary?
The prime must remain fixed while the nilpotency classes grow.
Mixed identities and the class threshold
Let X={x0,x1,…} and write G[X]=G∗F(X). For a homomorphism ϕ:G→H and a map ξ:X→H, denote by
evϕ,ξ:G[X]⟶H
the homomorphism extending ϕ and ξ. Put
IG[X]={w∈G[X]:ev1G,ξ(w)=1 for every ξ:X→G}.
Following Anashin (1), the mixed variety generated by G is defined by
H∈mvar(G)⟺∃ϕ:G→Hevϕ,ξ(w)=1(w∈IG[X], ξ:X→H).(3)
The same coefficient homomorphism must work for all identities, and it need not be injective. An ordinary variety is a class defined by a set of words in F(X), without coefficients.
Problem 10.4 of the Kourovka Notebook (2) asks whether, for a fixed prime p, the mixed variety generated by an arbitrary finite p-group of sufficiently large nilpotency class is an ordinary variety. Anashin (1, p. 176) already gives the class-two example
⟨a,b∣ap=bp2=1, a−1ba=b1+p⟩(p odd),
which satisfies xp[x,a−1]=1 but generates a nonordinary mixed variety. We use a principal unit congruent to 1 modulo p2 to obtain unbounded nilpotency class for each fixed prime, including p=2. Throughout,
[x,y]=x−1y−1xy,γ1(G)=G,γj+1(G)=[γj(G),G].
Theorem 1. For every prime p and every integer k≥0, there is a finite p-group Gp,k such that
∣Gp,k∣=p4k+4,cl(Gp,k)≥k+2,mvar(Gp,k) is not ordinary.
The group is a split extension of a cyclic group of order p2k+3 by a cyclic group of order p2k+1.
Taking k at least any proposed threshold gives a negative answer to Problem 10.4. The prime remains fixed throughout the construction.
Lemma 2. Suppose that G contains an abelian subgroup A, and that r≥1 and d∈G satisfy
xr[x,d]=1(x∈G).
If some a∈A satisfies ar=1, then mvar(G) is not ordinary.
Proof. For any homomorphism ϕ:G→A, the transported mixed word has value
ar[a,ϕ(d)]=ar=1.Consequently A∈/mvar(G). The identity homomorphism gives G∈mvar(G). Ordinary identities restrict to subgroups, since their evaluations commute with the inclusion. Thus no set of ordinary identities defines mvar(G). ◻
Principal units and geometric sums
Write
Sm(u)=i=0∑m−1ui,S0(u)=0.
We will use
Sm+n(u)=Sm(u)+umSn(u),Smn(u)=Sm(u)Sn(um).(10)
These identities follow by splitting the sum into consecutive intervals of equal length.
Lemma 3. For every natural number p and every h≥0, there is an integer ah such that
(1+p2)ph=1+ph+2(1+pah).(11)
Moreover, if R is a commutative ring and u=1+p2z∈R, then for every h≥0 there is wh∈R such that
Sph(u)=ph(1+pwh).(12)
Here integer coefficients in the second identity are interpreted in R.
Proof. For (11), take a0=0. Suppose the formula holds at h, and put v=ph+2(1+pah). The binomial formula gives an integer zh with
(1+v)p=1+pv+v2zh.Substitution yields
(1+p2)ph+1=1+ph+3(1+p(ah+ph(1+pah)2zh)),which proves the induction.
For (12), every power of u has the form 1+p2z′: the property is preserved by multiplication. Hence
Sp(uph)=p+p2vhfor some vh∈R. The case h=0 is S1(u)=1. If Sph(u)=ph(1+pwh), then (10) gives
Sph+1(u)=ph(1+pwh)(p+p2vh)=ph+1(1+p(wh+vh+pwhvh)).This proves the assertion without division in R. ◻
Fix a prime p and k≥0, and put
N=p2k+3,M=p2k+1,r=p2k+2,t=p2k,R=Z/NZ.
Let q=1+p2∈R. Since gcd(1+p2,p)=1, the element q is a unit. Equation (11), with h=2k+1, gives qM=1. Therefore the additive cyclic group C=Z/MZ acts on the additive group of R by
j⋅z=q−jz.
Define G=Gp,k=R⋊C. Its multiplication and inverse are
(z,j)(w,ℓ)=(z+q−jw,j+ℓ),(z,j)−1=(−qjz,−j).(19)
In particular,
∣G∣=NM=p4k+4.
Thus G is a finite p-group and is nilpotent. Put
a=(0,1),b=(1,0),d=a−t=(0,−t).
All subsequent calculations involving q and r take place in R. We have
rp=0,r2=0,qt=1+r,q−t=1−r.(22)
The first two equalities follow from the modulus N. For the third, use (11) at h=2k and rp=0. The last follows from (1+r)(1−r)=1.
Every power of q or q−1 has the form 1+p2z. For the inverse, this follows directly from
q−1=1−p2q−1;
the set of elements of the form 1+p2z is closed under multiplication. It follows that, for every j∈C,
rqj=r,Sr(q−j)=r.(24)
Indeed, the first assertion uses rp=0, and the second follows from (12) with h=2k+2, again using rp=0.
For x=(z,j), induction using (19) gives
xm=(Sm(q−j)z,mj).
Since M∣r, equations (24) yield
xr=(rz,0).(26)
Also,
d−1xd=(q−tz,j),
and hence
[x,d]=(qj(q−t−1)z,0)=(−rz,0).(28)
Combining (26) and (28) proves
xr[x,d]=1(x∈G).(29)
The subgroup A=R×{0} is abelian. Its element b=(1,0) satisfies
br=(r,0)=(0,0),
because 0<r<N. Lemma 2 now shows that mvar(G) is not ordinary.
The lower-central-series bound
For j≥0, put vj=(p2j,0)∈G. The coordinate multiplication gives
[(z,0),a]=(p2z,0)(z∈R),[vj,a]=vj+1.
Since v0∈G=γ1(G), induction gives vj∈γj+1(G). In particular,
vk+1=(p2k+2,0)=1,vk+1∈γk+2(G).
Thus cl(G)≥k+2. Together with the order calculation and (29), this proves Theorem 1.
References
Preprint · Lean (GitHub)
- V. S. Anashin, Mixed identities and mixed varieties of groups, Math. USSR-Sb. 57 (1987), no. 1, 171–182, doi:10.1070/SM1987v057n01ABEH003062.
- E. I. Khukhro and V. D. Mazurov (eds.), The Kourovka Notebook: Unsolved problems in group theory, 21st ed., September 2026 update, Problem 10.4, arXiv:1401.0300v46.