Structure

Finitely generated groups do not have ℵ1\aleph_1-filtered colimits

We show that the category Grpfg\Grp_\fg of finitely generated groups does not have ℵ1\aleph_1-filtered colimits.

Lemma 1.

Assume that an uncountable group HH exists with the following two properties:

  1. Every countable subset of HH is contained in a finitely generated subgroup of HH.
  2. HH is a subgroup of a product ∏n∈NHn\prod_{n \in \IN} H_n of finitely generated groups HnH_n.

Then Grpfg\Grp_\fg does not have ℵ1\aleph_1-filtered colimits.

Proof. Let II be the set of finitely generated subgroups of HH, partially ordered by inclusion, and let D:I→GrpfgD : I \to \Grp_\fg be the diagram which maps each U∈IU \in I to UU and each inclusion U⊆U′U \subseteq U' to the inclusion homomorphism U↪U′U \hookrightarrow U'. The poset II is ℵ1\aleph_1-filtered by (1): every countable subset of II has an upper bound in II, since the union of countably many finitely generated subgroups is countable.

Assume that DD has a colimit CC in Grpfg\Grp_\fg with colimit cocone λU:U→C\lambda_U : U \to C for U∈IU \in I. Since HH is the directed union of all U∈IU \in I, these homomorphisms glue to a homomorphism f:H→Cf : H \to C with f∣U=λUf|_U = \lambda_U. For each nn, the restrictions πn∣U:U→Hn\pi_n|_U : U \to H_n of the projection πn:H→Hn\pi_n : H \to H_n form a cocone of DD with tip HnH_n, so there is a homomorphism hn:C→Hnh_n : C \to H_n with hn∘λU=πn∣Uh_n \circ \lambda_U = \pi_n|_U for all U∈IU \in I, i.e. hn∘f=πnh_n \circ f = \pi_n. By (2), the πn\pi_n are jointly injective, so ff is injective. But HH is uncountable, whereas CC is countable as a finitely generated group. □\square

It remains to find such a group HH.

Idea

For (2), it is natural to look for HH of the form VN⋊GV^{\IN} \rtimes G, where VV is a countable vector space and G⊆Aut⁡(V)G \subseteq \Aut(V) is a finitely generated group acting on VNV^{\IN} coordinatewise. Such a group is uncountable and embeds into (V⋊G)N(V \rtimes G)^{\IN}. The real task is (1): given countably many w0,w1,…∈VNw_0,w_1,\dotsc \in V^{\IN}, we want to find them in a finitely generated subgroup. Our plan is to store all of them in a single element c∈VNc \in V^{\IN} and to compute them from cc using GG. In HH, the elements of GG act on VNV^{\IN} by conjugation, so everything we compute from cc in this way lies in the subgroup generated by GG and cc. This is the purpose of the semidirect product. Let us see which operations GG needs for this.

First, cc needs room for infinitely many vectors. So we let VV contain infinitely many independent copies V0,V1,V2,…V_0,V_1,V_2,\dotsc of one subspace, together with an element u∈Gu \in G mapping ViV_i to Vi+1V_{i+1}. We will put the data of wk,nw_{k,n} into the coordinate cnc_n. Since vectors in VV have finite support, cnc_n can only use finitely many copies, so we only put wk,nw_{k,n} for k≤nk \leq n into cnc_n.

Second, we need to read off a single copy. Moving ViV_i back to V0V_0 with u−iu^{-i}, it remains to discard everything outside of V0V_0, that is, to apply the projection PP onto V0V_0 along a complement WW. Since PP is not invertible, it cannot be an element of GG. But if s∈Gs \in G acts as −1-1 on V0V_0 and as 11 on WW, then s⋅v−v=−2P(v)s \cdot v - v = -2 P(v), which is P(v)P(v) when we work over F3\IF_3.

Third, wk,nw_{k,n} is an arbitrary vector of VV and does not fit into the single copy V0V_0. So we split it into two vectors of V0V_0 with an involution φ∈G\varphi \in G which interchanges V0V_0 and WW: both P(v)P(v) and P(φ⋅v)P(\varphi \cdot v) lie in V0V_0, and v=P(v)+φ⋅P(φ⋅v)v = P(v) + \varphi \cdot P(\varphi \cdot v). Thus, wk,nw_{k,n} takes up two copies, and we use V2k+1V_{2k+1} and V2k+2V_{2k+2}.

Finally, since cc only contains wk,nw_{k,n} for n≥kn \geq k, we only recover wkw_k up to finitely many coordinates. To correct these, we add an element b∈VNb \in V^{\IN} which, together with GG, generates all sequences with only finitely many non-zero coordinates. For this, it is convenient if a single basis vector of VV can be moved to all others by GG, which also makes V⋊GV \rtimes G finitely generated. This is the reason for a second shift r∈Gr \in G.

Construction

We now make this precise. Let Ω≔Z×Z\Omega \coloneqq \IZ \times \IZ, and let VV be the F3\IF_3-vector space of finitely supported functions Ω→F3\Omega \to \IF_3, with basis (ep)p∈Ω(e_p)_{p \in \Omega}. Every bijection σ:Ω→Ω\sigma : \Omega \to \Omega induces an automorphism of VV by σ⋅ep≔eσ(p)\sigma \cdot e_p \coloneqq e_{\sigma(p)}. For i∈Zi \in \IZ let Vi⊆VV_i \subseteq V be the subspace of functions supported on {i}×Z\{i\} \times \IZ, and let W⊆VW \subseteq V be the subspace of functions vanishing on {0}×Z\{0\} \times \IZ. Then V=V0⊕WV = V_0 \oplus W, and we let P:V→VP : V \to V be the projection onto V0V_0.

Let u,ru,r be the automorphisms of VV induced by the bijections u(i,j)≔(i+1,j)u(i,j) \coloneqq (i+1,j) and r(i,j)≔(i,j+1)r(i,j) \coloneqq (i,j+1) of Ω\Omega, and let φ\varphi be the automorphism induced by an involution of Ω\Omega which maps {0}×Z\{0\} \times \IZ onto its complement (both sets are countably infinite). Thus, we have u⋅Vi=Vi+1,φ⋅V0=W,φ⋅W=V0,φ2=idV.u \cdot V_i = V_{i+1}, \qquad \varphi \cdot V_0 = W, \qquad \varphi \cdot W = V_0, \qquad \varphi^2 = \id_V. Let ss be the automorphism of VV with s⋅x=−xs \cdot x = -x for x∈V0x \in V_0 and s⋅y=ys \cdot y = y for y∈Wy \in W. Let G⊆Aut⁡(V)G \subseteq \Aut(V) be the subgroup generated by u,r,φ,su,r,\varphi,s, and define the semidirect product H≔VN⋊G,H \coloneqq V^{\IN} \rtimes G, where GG acts on VNV^{\IN} coordinatewise.

Lemma 2.

The group HH is uncountable and satisfies (2) in Lemma 1.

Proof. Since VNV^{\IN} is uncountable, so is HH. The map (z,g)↦((zn,g))n∈N(z,g) \mapsto ((z_n,g))_{n \in \IN} embeds HH into (V⋊G)N(V \rtimes G)^{\IN}, and V⋊GV \rtimes G is generated by u,r,φ,s,e(0,0)u,r,\varphi,s,e_{(0,0)}, since uirj⋅e(0,0)=e(i,j)u^i r^j \cdot e_{(0,0)} = e_{(i,j)}. □\square

For the proof of (1), we regard VNV^{\IN} and GG as subgroups of HH via z↦(z,1)z \mapsto (z,1) and g↦(0,g)g \mapsto (0,g), and we keep writing the group VNV^{\IN} additively. We extend PP coordinatewise to VNV^{\IN}.

Lemma 3.

Let x∈V0x \in V_0 and d∈Zd \in \IZ. Then P(ud⋅x)=xP(u^d \cdot x) = x if d=0d = 0, and P(ud⋅x)=0P(u^d \cdot x) = 0 otherwise.

Proof. For d=0d = 0 this holds since PP is the identity on V0V_0. For d≠0d \neq 0, we have ud⋅x∈Vd⊆Wu^d \cdot x \in V_d \subseteq W, and PP vanishes on WW. □\square

Lemma 4.

For v∈Vv \in V we have v=P(v)+φ⋅P(φ⋅v).v = P(v) + \varphi \cdot P(\varphi \cdot v).

Proof. Write v=x+yv = x + y with x∈V0x \in V_0 and y∈Wy \in W, so that P(v)=xP(v) = x. Since φ⋅x∈W\varphi \cdot x \in W and φ⋅y∈V0\varphi \cdot y \in V_0, we have P(φ⋅v)=φ⋅yP(\varphi \cdot v) = \varphi \cdot y, and hence φ⋅P(φ⋅v)=φ2⋅y=y\varphi \cdot P(\varphi \cdot v) = \varphi^2 \cdot y = y. Therefore, P(v)+φ⋅P(φ⋅v)=x+y=vP(v) + \varphi \cdot P(\varphi \cdot v) = x + y = v. □\square

Lemma 5.

Let Γ⊆H\Gamma \subseteq H be a subgroup containing GG. Then Γ∩VN\Gamma \cap V^{\IN} is a subgroup of VNV^{\IN} which is closed under the action of GG and under PP.

Proof. By definition of the semidirect product, conjugation with g∈Gg \in G in HH acts on VNV^{\IN} by z↦g⋅zz \mapsto g \cdot z. Hence, Γ∩VN\Gamma \cap V^{\IN} is closed under the action of GG. For v=x+yv = x + y with x∈V0x \in V_0 and y∈Wy \in W, we have s⋅v−v=(−x+y)−(x+y)=−2x=x=P(v),s \cdot v - v = (-x + y) - (x + y) = -2x = x = P(v), since −2=1-2 = 1 in F3\IF_3. Hence, P(z)=s⋅z−zP(z) = s \cdot z - z for z∈VNz \in V^{\IN}, so Γ∩VN\Gamma \cap V^{\IN} is also closed under PP. □\square

Proposition 6.

The group HH satisfies (1) in Lemma 1.

Proof. Let X⊆HX \subseteq H be countable. Since every element of HH is a product of an element of VNV^{\IN} and an element of GG, there are w0,w1,…∈VNw_0,w_1,\dotsc \in V^{\IN} such that XX is contained in the subgroup generated by GG and w0,w1,…w_0,w_1,\dotsc. Write wk=(wk,n)n∈Nw_k = (w_{k,n})_{n \in \IN} and define b,c∈VNb,c \in V^{\IN} by bn≔un⋅e(0,0),cn≔∑k=0n(u2k+1⋅P(wk,n)+u2k+2⋅P(φ⋅wk,n)).\textstyle b_n \coloneqq u^n \cdot e_{(0,0)}, \qquad c_n \coloneqq \sum_{k=0}^{n} \bigl(u^{2k+1} \cdot P(w_{k,n}) + u^{2k+2} \cdot P(\varphi \cdot w_{k,n})\bigr). Let Γ⊆H\Gamma \subseteq H be the finitely generated subgroup generated by u,r,φ,s,b,cu,r,\varphi,s,b,c, i.e. by GG, bb and cc. By Lemma 5, Γ∩VN\Gamma \cap V^{\IN} is a subgroup of VNV^{\IN} containing bb and cc which is closed under the action of GG and under PP. We show that wk∈Γw_k \in \Gamma for all kk, which implies X⊆ΓX \subseteq \Gamma.

First, Γ\Gamma contains every z∈VNz \in V^{\IN} with zn=0z_n = 0 for almost all nn. Indeed, let n∈Nn \in \IN. The coordinate of P(u−n⋅b)∈ΓP(u^{-n} \cdot b) \in \Gamma at n′n' is P(un′−n⋅e(0,0))P(u^{n'-n} \cdot e_{(0,0)}), which is e(0,0)e_{(0,0)} for n′=nn' = n and 00 otherwise. Acting on this element with uirju^i r^j replaces e(0,0)e_{(0,0)} by e(i,j)e_{(i,j)}, and these elements generate all z∈VNz \in V^{\IN} with zn′=0z_{n'} = 0 for n′≠nn' \neq n.

Second, let k∈Nk \in \IN and n≥kn \geq k. All vectors P(wk′,n)P(w_{k',n}) and P(φ⋅wk′,n)P(\varphi \cdot w_{k',n}) lie in V0V_0. If we apply the linear map P∘u−(2k+1)P \circ u^{-(2k+1)} to cnc_n, Lemma 3 shows that only the summand with u2k+1u^{2k+1} survives, so that P(u−(2k+1)⋅cn)=P(wk,n)P(u^{-(2k+1)} \cdot c_n) = P(w_{k,n}). Similarly, P(u−(2k+2)⋅cn)=P(φ⋅wk,n)P(u^{-(2k+2)} \cdot c_n) = P(\varphi \cdot w_{k,n}). Therefore, by Lemma 4, the element yk≔P(u−(2k+1)⋅c)+φ⋅P(u−(2k+2)⋅c)∈Γy_k \coloneqq P(u^{-(2k+1)} \cdot c) + \varphi \cdot P(u^{-(2k+2)} \cdot c) \in \Gamma has coordinate P(wk,n)+φ⋅P(φ⋅wk,n)=wk,nP(w_{k,n}) + \varphi \cdot P(\varphi \cdot w_{k,n}) = w_{k,n} for all n≥kn \geq k. Hence, wk−ykw_k - y_k has only finitely many non-zero coordinates, so it lies in Γ\Gamma by the first step, and therefore wk∈Γw_k \in \Gamma. □\square

Corollary 7.

The category Grpfg\Grp_\fg of finitely generated groups does not have ℵ1\aleph_1-filtered colimits.

Proof. Apply Lemma 1 to the group HH, using Lemma 2 and Proposition 6. □\square

Context

This page is referenced by the following categories.