Normalized root words and reflective completion G^=lim←⁡G/Pn(G)\widehat G=\varprojlim G/P_n(G) with exact power filtration

10.52

Problem

Find a categorical setting, containing nilpotent groups, in which completion preserves the intrinsic power-subgroup topology and is reflective onto the complete Hausdorff objects.

For a prime pp, put

Pn(G)=⟨gpn:g∈G⟩,G^=lim←⁡nG/Pn(G).P_n(G)=\langle g^{p^n}:g\in G\rangle,\qquad \widehat G=\varprojlim_nG/P_n(G).

The comparison sought is between Pn(G^)P_n(\widehat G) and ker⁡(G^→G/Pn(G))\ker(\widehat G\to G/P_n(G)).

The category and the comparison theorem

Let pp be a prime. For a group GG, put

Pn(G)=⟨gpn:g∈G⟩,Qn(G)=G/Pn(G),n≥0.(1) \tag{1} P_n(G)=\langle g^{p^n}:g\in G\rangle, \qquad Q_n(G)=G/P_n(G), \qquad n\geq0.

The subgroups Pn(G)P_n(G) form a descending sequence of fully invariant normal subgroups. The power-subgroup uniformity has basic entourages

Un(G)={(a,b)∈G2:a−1b∈Pn(G)}.(2) \tag{2} U_n(G)=\{(a,b)\in G^2:a^{-1}b\in P_n(G)\}.

Its neighbourhoods of the identity are the subgroups Pn(G)P_n(G). All group homomorphisms are uniformly continuous for these uniformities.

The quotient maps ρn,m:Qm(G)→Qn(G)\rho_{n,m}:Q_m(G)\to Q_n(G), for n≤mn\leq m, give the group

G^={(zn)∈∏n≥0Qn(G):ρn,m(zm)=zn whenever n≤m}.(3) \tag{3} \widehat G= \left\{(z_n)\in\prod_{n\geq0}Q_n(G): \rho_{n,m}(z_m)=z_n\text{ whenever }n\leq m\right\}.

Each quotient is given the discrete uniformity. Write πn:G^→Qn(G)\pi_n:\widehat G\to Q_n(G) for the coordinate projection,

ηG(g)=(gPn(G))n≥0,Kn(G)=ker⁡πn.\eta_G(g)=(gP_n(G))_{n\geq0}, \qquad K_n(G)=\ker\pi_n.

The inverse-limit uniformity has basic entourages defined by equality of the nnth coordinates, or equivalently by x−1y∈Kn(G)x^{-1}y\in K_n(G).

Problem 10.52 of the Kourovka Notebook, attributed to R. Mines and communicated by S. V. Rychkov, asks for a categorical setting for the comparison of the intrinsic and induced topologies on completion, including nilpotent groups (3). Power-subgroup completions of nilpotent groups are treated in Warfield’s monograph (4, §7). Fay and Walls (1) study reflection functors associated with nilpotent-group completions, including Mal’cev and pp-cotorsion completions and a restricted pp-profinite construction. General relations between reflective completion and closure are considered in (2).

Here the object condition is a pair of group-word identities. The proof separates two questions: constructing a normalized root word in each finite nilpotency class, and using that word to make roots in the quotient tower compatible. The latter step gives equality of the indexed filtrations, before any appeal to completeness.

Let Cp\mathcal C_p be the full category of groups for which there are an integer r≥1r\geq1 and a word w∈F(x,y)w\in F(x,y) satisfying

w(a,b)p=apbpr,w(a,1)=a(a,b∈G).(5) \tag{5} w(a,b)^p=a^p b^{p^r}, \qquad w(a,1)=a \qquad(a,b\in G).

The witnesses r,wr,w are fixed for each object. Let Dp\mathcal D_p be the full subcategory of Cp\mathcal C_p consisting of the complete Hausdorff groups for (2).

Theorem 1. The category Cp\mathcal C_p contains every nilpotent group. For each G∈CpG\in\mathcal C_p, the group G^\widehat G belongs to Dp\mathcal D_p and

Pn(G^)=Kn(G)(n≥0).(6) \tag{6} P_n(\widehat G)=K_n(G)\qquad(n\geq0).

In particular, its intrinsic power-subgroup uniformity equals the inverse-limit uniformity. The functor G↦G^G\mapsto\widehat G is left adjoint to the full inclusion Dp↪Cp\mathcal D_p\hookrightarrow\mathcal C_p, with unit ηG\eta_G. Thus restriction induces natural bijections

Hom⁡(G^,H)⟶Hom⁡(G,H),f⟼fηG(G∈Cp, H∈Dp).(7) \tag{7} \operatorname{Hom}(\widehat G,H) \longrightarrow \operatorname{Hom}(G,H), \qquad f\longmapsto f\eta_G \qquad(G\in\mathcal C_p,\ H\in\mathcal D_p).

No injectivity of ηG\eta_G is assumed. The nilpotent inclusion has no finite-generation, torsion or cardinality restriction. We prove it in the following uniform form.

Theorem 2. For every integer q≥2q\geq2 and every positive integer cc, there is a word Wc,q∈F(x,y)W_{c,q}\in F(x,y) such that, in every group of nilpotency class at most cc,

Wc,q(a,b)q=aqbqc,Wc,q(a,1)=a.(8) \tag{8} W_{c,q}(a,b)^q=a^q b^{q^c}, \qquad W_{c,q}(a,1)=a.

We use [x,y]=xyx−1y−1[x,y]=xyx^{-1}y^{-1} and the usual indexing γ1(G)=G\gamma_1(G)=G, γj+1(G)=[γj(G),G]\gamma_{j+1}(G)=[\gamma_j(G),G].

Words in the last lower-central term

A simple commutator of weight cc is a left-normed expression [s1,s2,…,sc][s_1,s_2,\ldots,s_c], with a weight-one expression interpreted as s1s_1.

Lemma 3. If G=⟨S⟩G=\langle S\rangle, then γc(G)\gamma_c(G) is normally generated by the simple commutators of weight cc in SS. If γc+1(G)=1\gamma_{c+1}(G)=1, these commutators generate γc(G)\gamma_c(G) as an ordinary subgroup.

Proof. For a subset T⊆GT\subseteq G, let NN be the normal closure of {[t,s]:t∈T, s∈S}\{[t,s]:t\in T,\ s\in S\}. In G/NG/N, the image of each t∈Tt\in T commutes with the images of all elements of SS, and hence is central. The normal closure of the image of TT is therefore central. It follows that

[⟨ ⁣⟨T⟩ ⁣⟩,G]=⟨ ⁣⟨[t,s]:t∈T, s∈S⟩ ⁣⟩.[\langle\!\langle T\rangle\!\rangle,G] =\langle\!\langle[t,s]:t\in T,\ s\in S\rangle\!\rangle.

The reverse inclusion used here holds because each displayed commutator belongs to the subgroup on the left, which is normal. Starting with the normal closure of SS, induction gives the first assertion. If γc+1(G)=1\gamma_{c+1}(G)=1, then γc(G)\gamma_c(G) is central. The subgroup generated by its displayed generators is consequently normal, so normal and ordinary generation agree. ◻

Lemma 4. Suppose that γc+1(G)=1\gamma_{c+1}(G)=1. Let t(x,y)t(x,y) be a simple commutator of weight cc, and let dd be the number of occurrences of yy in it. Then

t(a,bq)=t(a,b)qd(a,b∈G).(10) \tag{10} t(a,b^q)=t(a,b)^{q^d}\qquad(a,b\in G).

If c≥2c\geq2 and d=0d=0, then t(a,b)=1t(a,b)=1.

Proof. For weight one the assertion is immediate. Suppose c≥2c\geq2 and write t=[u,s]t=[u,s], where uu has weight c−1c-1 and ss is either xx or yy. The group G/Z(G)G/Z(G) has class at most c−1c-1. Induction shows that u(a,bq)u(a,b^q) and u(a,b)qd′u(a,b)^{q^{d'}} have the same image in G/Z(G)G/Z(G), where d′d' counts the occurrences of yy in uu. Elements differing by a central factor have equal commutators with every fixed element.

The commutator [u(a,b),s(a,b)][u(a,b),s(a,b)] belongs to the central subgroup γc(G)\gamma_c(G). For a central commutator one has

[vm,zn]=[v,z]mn(m,n≥0).(11) \tag{11} [v^m,z^n]=[v,z]^{mn}\qquad(m,n\geq0).

Indeed, the identity [v1v2,z]=v1[v2,z]v1−1[v1,z][v_1v_2,z]=v_1[v_2,z]v_1^{-1}[v_1,z] gives [vm,z]=[v,z]m[v^m,z]=[v,z]^m by induction. Applying this with the variables interchanged and using [v,z]−1=[z,v][v,z]^{-1}=[z,v] gives the second exponent. If s=xs=x, formula (11) gives exponent qd′q^{d'}; if s=ys=y, it gives qd′q=qd′+1q^{d'}q=q^{d'+1}. This proves (10). A commutator of weight at least two involving only aa is trivial, proving the last assertion. ◻

Lemma 5. Let NN be generated by a,ba,b, with γc+1(N)=1\gamma_{c+1}(N)=1 and c≥2c\geq2. Suppose φ:N→N\varphi:N\to N is an endomorphism satisfying φ(a)=a\varphi(a)=a and φ(b)=bq\varphi(b)=b^q. For every e∈γc(N)e\in\gamma_c(N) there is z∈γc(N)z\in\gamma_c(N) such that

zq=φ(e).(12) \tag{12} z^q=\varphi(e).

Proof. By Lemma 3, the central subgroup T=γc(N)T=\gamma_c(N) is generated by the values of weight-cc simple commutators in a,ba,b. For a generator tt with d≥1d\geq1 occurrences of bb, Lemma 4 gives

φ(t)=tqd=(tqd−1)q.\varphi(t)=t^{q^d}=\bigl(t^{q^{d-1}}\bigr)^q.

Generators with d=0d=0 are trivial. The identity has the root 11. If ziq=φ(ei)z_i^q=\varphi(e_i) with zi∈Tz_i\in T, centrality gives

(z1z2)q=φ(e1e2),(z1−1)q=φ(e1−1).(z_1z_2)^q=\varphi(e_1e_2), \qquad (z_1^{-1})^q=\varphi(e_1^{-1}).

Induction under products and inverses now gives (12) for every e∈Te\in T. ◻

Proof of Theorem 2. Let F=F(x,y)F=F(x,y) and Nc=F/γc+1(F)N_c=F/\gamma_{c+1}(F). The images of x,yx,y will be denoted by the same letters. We construct a word whose power equation holds in NcN_c and whose normalization W(x,1)=xW(x,1)=x holds in FF itself.

For c=1c=1, take W1,q=xyW_{1,q}=xy. Suppose c≥2c\geq2 and the word v=Wc−1,qv=W_{c-1,q} has been constructed. The canonical map Nc→Nc−1N_c\to N_{c-1} has kernel γc(Nc)\gamma_c(N_c). Thus, writing vv also for its value in NcN_c,

e=(vq)−1xqyqc−1∈γc(Nc).e=(v^q)^{-1}x^q y^{q^{c-1}}\in\gamma_c(N_c).

The substitution x↦xx\mapsto x, y↦yqy\mapsto y^q induces an endomorphism φ\varphi of NcN_c, since homomorphisms preserve lower-central terms. Lemma 5 supplies z∈γc(Nc)z\in\gamma_c(N_c) with zq=φ(e)z^q=\varphi(e).

Surjective homomorphisms map each lower-central term onto the corresponding term. Choose t∈γc(F)t\in\gamma_c(F) mapping to zz, and put

Wc,q(x,y)=v(x,yq)t(x,y).(16) \tag{16} W_{c,q}(x,y)=v(x,y^q)t(x,y).

The element zz is central in NcN_c, so

Wc,q(x,y)q=(φ(v)z)q=φ(v)qφ(e)=φ(vqe)=xq(yq)qc−1=xqyqc.\begin{align*} W_{c,q}(x,y)^q &=(\varphi(v)z)^q =\varphi(v)^q\varphi(e) =\varphi(v^q e)\\ &=x^q(y^q)^{q^{c-1}} =x^q y^{q^c}. \end{align*}

Since c≥2c\geq2, substitution y=1y=1 kills γc(F)\gamma_c(F): its image lies in the corresponding lower-central term of the cyclic group generated by xx. Hence t(x,1)=1t(x,1)=1, and (16) gives Wc,q(x,1)=xW_{c,q}(x,1)=x in FF. Finally every pair a,ba,b in a group of class at most cc defines a homomorphism Nc→GN_c\to G. Evaluation transfers both identities to GG. ◻

Finite correction and compatible roots

Fix G∈CpG\in\mathcal C_p and witnesses r,wr,w for (5).

Lemma 6. For k≥0k\geq0, a∈Ga\in G and z∈Pr+k(G)z\in P_{r+k}(G), there is b∈Gb\in G such that

a−1b∈Pk(G),bp=apz.(18) \tag{18} a^{-1}b\in P_k(G),\qquad b^p=a^p z.

Proof. First let z=tpr+kz=t^{p^{r+k}}. Set b=w(a,tpk)b=w(a,t^{p^k}). The first identity in (5) gives bp=apzb^p=a^p z. In the quotient G/Pk(G)G/P_k(G), the second argument of ww becomes 11, so the normalization gives bPk(G)=aPk(G)bP_k(G)=aP_k(G).

The conclusion holds for z=1z=1 by taking b=ab=a. If it holds for zz, apply the first step to its chosen root bb to absorb a further factor tpr+kt^{p^{r+k}}. The coset condition is preserved because Pk(G)P_k(G) is a subgroup. Inverse factors are absorbed by replacing tt with t−1t^{-1}. This proves the assertion for the subgroup generated by all the indicated powers, namely Pr+k(G)P_{r+k}(G). ◻

For m≥nm\geq n and z∈Kn(G)z\in K_n(G), choose a representative g∈Gg\in G of πm(z)\pi_m(z). Compatibility of the coordinates shows that

g∈Pn(G),πm(ηG(g))=πm(z).(19) \tag{19} g\in P_n(G),\qquad \pi_m(\eta_G(g))=\pi_m(z).

This finite-coordinate approximation will be used repeatedly.

Lemma 7. For every k≥0k\geq0 and z∈Kr+k(G)z\in K_{r+k}(G) there is h∈Kk(G)h\in K_k(G) with hp=zh^p=z. Consequently, for every n≥0n\geq0,

z∈Krn(G)⟹z=hpn for some h∈G^.(20) \tag{20} z\in K_{rn}(G)\quad\Longrightarrow\quad z=h^{p^n}\text{ for some }h\in\widehat G.

Proof. Choose a0=1a_0=1. We construct ai∈Ga_i\in G such that

ei=ηG(aip)−1z∈Kr+k+i(G),ai−1ai+1∈Pk+i(G).\begin{align*} e_i&=\eta_G(a_i^p)^{-1}z\in K_{r+k+i}(G),\tag{21a}\\ a_i^{-1}a_{i+1}&\in P_{k+i}(G).\tag{21b} \end{align*}

The error condition holds at i=0i=0. Given aia_i, apply (19) to eie_i, with n=r+k+in=r+k+i and m=n+1m=n+1, to obtain ti∈Pr+k+i(G)t_i\in P_{r+k+i}(G) satisfying

ηG(ti)−1ei∈Kr+k+i+1(G).\eta_G(t_i)^{-1}e_i\in K_{r+k+i+1}(G).

Lemma 6, with k+ik+i in place of kk, gives ai+1a_{i+1} such that

ai+1p=aipti,ai−1ai+1∈Pk+i(G).a_{i+1}^p=a_i^p t_i, \qquad a_i^{-1}a_{i+1}\in P_{k+i}(G).

Then ei+1=ηG(ti)−1eie_{i+1}=\eta_G(t_i)^{-1}e_i, proving the induction.

Products of (21b) give

ai−1aj∈Pk+i(G)(i≤j),ai∈Pk(G).(24) \tag{24} a_i^{-1}a_j\in P_{k+i}(G)\quad(i\leq j), \qquad a_i\in P_k(G).

Define hn=anPn(G)h_n=a_nP_n(G). Equation (24) makes (hn)(h_n) a compatible sequence, hence an element h∈G^h\in\widehat G. Its kkth coordinate is trivial, so h∈Kk(G)h\in K_k(G). Since (21a) implies en∈Kn(G)e_n\in K_n(G), the nnth coordinate of hph^p is πn(z)\pi_n(z) for every nn. Therefore hp=zh^p=z.

Iterating this assertion proves (20). At the induction step, take a ppth root in Krn(G)K_{rn}(G) of an element of Kr(n+1)(G)K_{r(n+1)}(G), and then take its pnp^n-th root. ◻

Proposition 8. For every n≥0n\geq0, one has Pn(G^)=Kn(G)P_n(\widehat G)=K_n(G).

Proof. Every element of Qn(G)Q_n(G) has pnp^n-th power 11, so Pn(G^)≤Kn(G)P_n(\widehat G)\leq K_n(G). Conversely, let z∈Kn(G)z\in K_n(G). Apply (19) at level m=n+rnm=n+rn to obtain g∈Pn(G)g\in P_n(G) with

e=ηG(g)−1z∈Kn+rn(G)≤Krn(G).e=\eta_G(g)^{-1}z\in K_{n+rn}(G)\leq K_{rn}(G).

Lemma 7 writes e=hpne=h^{p^n}. The homomorphism ηG\eta_G sends Pn(G)P_n(G) into Pn(G^)P_n(\widehat G), and therefore

z=ηG(g)hpn∈Pn(G^).z=\eta_G(g)h^{p^n}\in P_n(\widehat G).

 ◻

Uniform completeness and the universal property

Lemma 9. For every group GG, the inverse limit (3) is complete and Hausdorff in its inverse-limit uniformity, and ηG(G)\eta_G(G) is dense in it. The pullback of that uniformity along ηG\eta_G is (2). Moreover,

G is complete⟺ηG is surjective,G is Hausdorff⟺ηG is injective,\begin{align*} G\text{ is complete}&\quad\Longleftrightarrow\quad\eta_G\text{ is surjective},\tag{27a}\\ G\text{ is Hausdorff}&\quad\Longleftrightarrow\quad\eta_G\text{ is injective},\tag{27b} \end{align*}

where the conditions on GG refer to its power-subgroup uniformity.

Proof. The product of the discrete groups Qn(G)Q_n(G) is complete and Hausdorff. The compatibility equations in (3) define closed subsets, so G^\widehat G is a closed, hence complete, subgroup. A finite set of coordinates is controlled by its largest index. Representing that coordinate by an element of GG proves density of ηG(G)\eta_G(G). Finally,

πn(ηG(a))=πn(ηG(b))⟺a−1b∈Pn(G),\pi_n(\eta_G(a))=\pi_n(\eta_G(b)) \quad\Longleftrightarrow\quad a^{-1}b\in P_n(G),

which proves the pullback assertion.

If GG is complete, its image under this uniformity-inducing map is complete, hence closed in the Hausdorff limit. Density then makes ηG\eta_G surjective. Conversely, a surjective uniformity-inducing map onto a complete space makes its domain complete. For (27b), the pullback uniformity separates points exactly when the map to the Hausdorff limit is injective. ◻

Proposition 10. Let G∈CpG\in\mathcal C_p, and let HH be any group complete and Hausdorff for its power-subgroup uniformity. Every homomorphism f:G→Hf:G\to H extends uniquely to a homomorphism f‾:G^→H\overline f:\widehat G\to H satisfying f‾ηG=f\overline f\eta_G=f.

Proof. A homomorphism f:G→Hf:G\to H induces quotient maps

fn:Qn(G)⟶Qn(H),gPn(G)⟼f(g)Pn(H),f_n:Q_n(G)\longrightarrow Q_n(H), \qquad gP_n(G)\longmapsto f(g)P_n(H),

which commute with the transitions. Coordinatewise application gives f^:G^→H^\widehat f:\widehat G\to\widehat H and f^ηG=ηHf\widehat f\eta_G=\eta_H f. Lemma 9 makes ηH\eta_H an isomorphism. Thus

f‾=ηH−1f^(30) \tag{30} \overline f=\eta_H^{-1}\widehat f

is an extension.

For uniqueness, let u,v:G^→Hu,v:\widehat G\to H satisfy uηG=vηGu\eta_G=v\eta_G. Fix z∈G^z\in\widehat G and n≥0n\geq0. Choose g∈Gg\in G representing πn(z)\pi_n(z). Proposition 8 gives

ηG(g)−1z∈Kn(G)=Pn(G^).\eta_G(g)^{-1}z\in K_n(G)=P_n(\widehat G).

Homomorphisms preserve generated power subgroups. Hence the images of u(z)u(z) and u(ηG(g))u(\eta_G(g)) in Qn(H)Q_n(H) agree, as do the corresponding images under vv. The assumed equality on ηG(G)\eta_G(G) gives equality of the nnth coordinates of ηH(u(z))\eta_H(u(z)) and ηH(v(z))\eta_H(v(z)). This holds for every nn, and ηH\eta_H is injective, so u(z)=v(z)u(z)=v(z). ◻

Proof of Theorem 1. Theorem 2, with q=pq=p, puts every nilpotent group in Cp\mathcal C_p. For fixed r,wr,w, identities (5) are preserved by quotient maps, products and subgroups. They therefore hold in every Qn(G)Q_n(G), in their product, and in the compatible-sequence subgroup G^\widehat G. Thus G^∈Cp\widehat G\in\mathcal C_p.

Proposition 8 identifies the two uniformity bases. Lemma 9 consequently makes G^\widehat G complete and Hausdorff also for its intrinsic uniformity, so it lies in Dp\mathcal D_p.

The coordinatewise maps f^\widehat f preserve identities and composition. They define the asserted functor, and f^ηG=ηHf\widehat f\eta_G=\eta_H f is naturality of its unit. Proposition 10 proves the bijections (7). These bijections are natural: postcomposition commutes with restriction, and precomposition is compatible with the displayed naturality square. Equivalently, uniqueness of the extensions identifies the corresponding composites. This proves the adjunction. ◻

References

Preprint · Lean (GitHub)

  1. T. H. Fay and G. L. Walls, Completions and categorical compactness for nilpotent groups, Quaestiones Math. 17 (1994), no. 4, 437–451, doi:10.1080/16073606.1994.9631776.
  1. D. Holgate, Completion and closure, Cah. Topol. Géom. Différ. Catég. 41 (2000), no. 2, 101–119, Numdam.
  1. E. I. Khukhro and V. D. Mazurov (eds.), The Kourovka Notebook: Unsolved problems in group theory, 21st ed., September 2026 update, Problem 10.52, arXiv:1401.0300v46.
  1. R. B. Warfield, Jr., Nilpotent Groups, Lecture Notes in Mathematics, vol. 513, Springer-Verlag, Berlin–Heidelberg, 1976, doi:10.1007/BFb0080152.