A closed Y⊆GL⁡3(F5(u)‾)Y\subseteq\operatorname{GL}_3(\overline{\mathbb F_5(u)}) with P∈Ym‾∖YmP\in\overline{Y^m}\setminus Y^m for every m≥2m\ge2

16.28(a)

Problem

Let XX be a Zariski-closed subset of a connected reductive linear algebraic group in positive characteristic. Must some power XcX^c, c>1c>1, be closed?

Xc={x1⋯xc:xi∈X},∃c>1: Xc‾=Xc  ?X^c=\{x_1\cdots x_c:x_i\in X\},\qquad \exists c>1:\ \overline{X^c}=X^c\;?

The power consists of products of exactly cc factors.

The statement and the counterexample

Problem 16.28(a) of the Kourovka Notebook, proposed by E. P. Vdovin, asks whether every closed subset XX of a connected linear reductive algebraic group in positive characteristic has a closed power XcX^c for some integer c>1c>1 (1). Here

Xc={x1⋯xc:x1,…,xc∈X}X^c=\{x_1\cdots x_c:x_1,\ldots,x_c\in X\}

denotes the set of products of exactly cc factors.

Linear algebraic groups are Zariski-closed matrix subgroups over an algebraically closed field. Reductivity means triviality of the connected normal unipotent radical (2). Closedness is taken in the matrix Zariski topology: closed subsets of GL3(k)\mathrm{GL}_3(k) are intersections with polynomial zero sets in the nine entries.

Let uu be an indeterminate and put k=F5(u)‾k=\overline{\mathbb F_5(u)}. We also write uu for its image in kk. For x∈k×x\in k^\times, define

B(x)=(1xx−101000u),P=diag⁡(1,1,u2),B(x)=\begin{pmatrix}1&x&x^{-1}\\0&1&0\\0&0&u\end{pmatrix}, \qquad P=\operatorname{diag}(1,1,u^2),

and set

Z={λI3:λ∈k×},D={λB(x):λ,x∈k×},Y=Z∪D.(3) \tag{3} Z=\{\lambda I_3:\lambda\in k^\times\},\qquad D=\{\lambda B(x):\lambda,x\in k^\times\},\qquad Y=Z\cup D.

Theorem 1. The group GL3(k)\mathrm{GL}_3(k) is connected and reductive. The subset YY in (3) is closed and contains I3I_3. For every integer m≥2m\geq2,

P∈Ym‾∖Ym.P\in\overline{Y^m}\setminus Y^m.

The field kk is algebraically closed of characteristic five. For a>b≥0a>b\geq0, the equality ua=ubu^a=u^b would imply the false rational-function identity ua−b=1u^{a-b}=1. The embedding F5(u)↪k\mathbb F_5(u)\hookrightarrow k therefore gives an element of infinite multiplicative order.

The ambient group

Affine nine-space over kk is irreducible. Indeed, two proper closed subsets admit nonzero vanishing polynomials; if they covered the space, the product would be a nonzero polynomial vanishing everywhere over the infinite field kk, a contradiction. The nonempty open subset det⁡≠0\det\ne0 is consequently irreducible. Thus GL3(k)\mathrm{GL}_3(k) is connected. The following lemma proves its reductivity.

Lemma 2. Let VV be a finite-dimensional vector space over a field. If H⊴GL(V)H\trianglelefteq\mathrm{GL}(V) and every element of HH is unipotent, then H={1}H=\{1\}.

Proof. Suppose A∈HA\in H and A≠1A\ne1, and put N=A−1N=A-1. The map NN is nonzero and nilpotent. A nilpotent chain gives a vector ww such that v=Nw≠0v=Nw\ne0 and Nv=0Nv=0: if Nq=0N^q=0 with qq minimal, choose zz with Nq−1z≠0N^{q-1}z\ne0 and take w=Nq−2zw=N^{q-2}z. The vectors w,vw,v are independent, since applying NN to a relation between them first kills its coefficient of ww. Extend them to a basis and choose a linear functional ff satisfying

f(w)=0,f(v)=1.f(w)=0,\qquad f(v)=1.

Define S(x)=f(x)wS(x)=f(x)w and T=1+ST=1+S. Then S2=0S^2=0 and T−1=1−ST^{-1}=1-S. Put R=ASA−1R=ASA^{-1}. Since Aw=w+vAw=w+v and A−1w=w−vA^{-1}w=w-v, one has

RS(x)=−f(x)Aw.RS(x)=-f(x)Aw.

For a rank-one map x↦f(x)ax\mapsto f(x)a, its trace is f(a)f(a). Hence

tr(S)=0,tr(R)=tr(S)=0,tr(RS)=−f(Aw)=−1.\mathop{\mathrm{tr}}(S)=0,\qquad \mathop{\mathrm{tr}}(R)=\mathop{\mathrm{tr}}(S)=0,\qquad \mathop{\mathrm{tr}}(RS)=-f(Aw)=-1.

Normality implies Q=ATA−1T−1∈HQ=ATA^{-1}T^{-1}\in H, whereas

Q−1=(1+R)(1−S)−1=R−S−RS,tr(Q−1)=1.Q-1=(1+R)(1-S)-1=R-S-RS, \qquad \mathop{\mathrm{tr}}(Q-1)=1.

This contradicts unipotence of QQ, since a nilpotent endomorphism has trace zero. Thus no such AA exists. ◻

Applying Lemma 2 to the unipotent radical of GL3(k)\mathrm{GL}_3(k) proves that this radical is trivial. Thus GL3(k)\mathrm{GL}_3(k) is reductive.

Closedness and the obstruction

The scalar subgroup ZZ is closed, being defined by vanishing of the off-diagonal entries and equality of the diagonal entries. The set DD is the zero locus in GL3(k)\mathrm{GL}_3(k) of

a21=a31=a32=a23=0,a11=a22,a33=ua22,a12a13=a222.(9) \tag{9} \begin{gathered} a_{21}=a_{31}=a_{32}=a_{23}=0,\qquad a_{11}=a_{22},\qquad a_{33}=u a_{22},\\ a_{12}a_{13}=a_{22}^2. \end{gathered}

Indeed, these equations give det⁡(aij)=ua223\det(a_{ij})=u a_{22}^3. Invertibility forces λ=a22≠0\lambda=a_{22}\ne0. Dividing the matrix by λ\lambda gives B(x)B(x) with x=a12/λ≠0x=a_{12}/\lambda\ne0, because the last equation says (a12/λ)(a13/λ)=1(a_{12}/\lambda)(a_{13}/\lambda)=1. Conversely, every λB(x)\lambda B(x) satisfies (9). Thus DD, and consequently Y=Z∪DY=Z\cup D, is closed. Also I3∈ZI_3\in Z.

Fix m≥2m\geq2. Suppose a product of mm elements of YY equals PP, and let jj count the factors chosen from DD. Since scalar matrices are central, this product can be written

ΛB(x1)⋯B(xj),Λ∈k×.\Lambda B(x_1)\cdots B(x_j),\qquad \Lambda\in k^\times.

Its diagonal is (Λ,Λ,Λuj)(\Lambda,\Lambda,\Lambda u^j). Comparison with PP gives Λ=1\Lambda=1 and uj=u2u^j=u^2, hence j=2j=2. But direct multiplication yields

B(x)B(y)=(1x+yu/x+1/y01000u2).B(x)B(y)= \begin{pmatrix}1&x+y&u/x+1/y\\0&1&0\\0&0&u^2\end{pmatrix}.

For this to equal PP, one would need y=−xy=-x and (u−1)/x=0(u-1)/x=0, contradicting u≠1u\ne1. Therefore P∉YmP\notin Y^m.

To prove the opposite closure membership, consider the polynomial matrix curve

γ(t)=(10(u−1)t01000u2)(t∈k).\gamma(t)= \begin{pmatrix}1&0&(u-1)t\\0&1&0\\0&0&u^2\end{pmatrix} \qquad(t\in k).

Its determinant is the nonzero constant u2u^2, so it is defined in GL3(k)\mathrm{GL}_3(k) for every tt, and γ(0)=P\gamma(0)=P. For t≠0t\ne0,

γ(t)=B(t−1)B(−t−1)∈Y2⊆Ym,\gamma(t)=B(t^{-1})B(-t^{-1})\in Y^2\subseteq Y^m,

where the final inclusion inserts m−2m-2 identity factors. The inverse image γ−1(Ym‾)\gamma^{-1}(\overline{Y^m}) is closed, since substituting the polynomial entries of γ\gamma into any matrix polynomial gives a polynomial in tt. It contains k∖{0}k\setminus\{0\}, which is dense in the affine line over the infinite field kk. Hence it also contains zero. Thus P∈Ym‾P\in\overline{Y^m}, completing the proof of Theorem 1.

The counterexample, including the field and ambient-group hypotheses, is formalised in Lean 4 with Mathlib; the proof source accompanies this article.

References

Preprint · Lean (GitHub)

  1. E. I. Khukhro and V. D. Mazurov (eds.), Unsolved Problems in Group Theory: The Kourovka Notebook, 21st edition, revision of 1 September 2026, Problem 16.28, arXiv:1401.0300.
  1. J. S. Milne, Algebraic Groups: The Theory of Group Schemes of Finite Type over a Field, Cambridge Studies in Advanced Mathematics 170, Cambridge University Press, 2017.