Faithful inner-homogeneous R-representations via HNN extensions with cardinal bound max{ℵ0,∣R∣,∣G∣,∣V∣}
Kourovka17.101
Problem
Does every representation of a group over a field embed in a homogeneous representation: one in which every isomorphism between finitely generated subrepresentations extends to an automorphism?
Finite generation of U is over the group ring R[A]. The result below works over every nonzero commutative ring and makes the extending automorphism inner on both sorts.
Representations and the embedding theorem
Fix a commutative ring R with 1=0. Write X=(G,V,ρ) for an R-representation, where ρ:G→AutR(V), and abbreviate ρ(g)v to gv. An embedding (i,j):X↪Y=(H,W,σ) consists of
i:G↪H,j:V↪W,j(gv)=σ(i(g))j(v),(1)
with i a group homomorphism and j an R-linear map. All morphisms fix R. The action is faithful if kerρ=1.
A subrepresentation S=(A,U)≤X consists of A≤G and an R-submodule U≤V with AU⊆U. Write S≤fgX when
A=⟨a1,…,am⟩≤G,U=j=1∑nR[A]uj≤V(2)
for finite tuples (ai) and (uj). Finite generation is over R[A]: for a field R, the cyclic R[Z]-module R[z,z−1] has infinite R-dimension.
An isomorphism f:S∼T consists of
θ:A∼B,φ:U∼RU′,φ(au)=θ(a)φ(u).(3)
Put
D(X)={(S,T,f):S,T≤fgX,f:S∼T}.
For t∈G, the pair
(ct,Lt),ct(g)=tgt−1,Lt(v)=tv,(5)
is an automorphism of X, since Lt(gv)=ct(g)Lt(v). The representation is inner homogeneous if each member of D(X) is a restriction of some (ct,Lt).
Theorem 1. Every R-representation (G,V) embeds in a faithful inner homogeneous representation (Γ,W) with
∣Γ∣,∣W∣≤κ,κ=max{ℵ0,∣R∣,∣G∣,∣V∣}.(6)
Thus, for (S,T,(θ,φ))∈D(Γ,W), some t∈Γ satisfies
tat−1=θ(a)(a∈A),tu=φ(u)(u∈U).(7)
Plotkin’s Problem 17.101 asks for homogeneous extensions of group representations, with finite generation as in (2) (6). Aladova–Gvaramiya–Plotkin pose the fixed-ring version in (1, Definition 5.8 and Problem 5.11). Their discussion identifies the obstruction: extending G and inducing V need not extend a prescribed φ. Theorem 1 gives an inner extension on both sorts.
The group construction is due to Higman–Neumann–Neumann (4). Kegel’s permutation-action analogue already gives inner implementation and cardinal control (5, Theorem 1). For modules, one must also preserve the additive and scalar relations in V. Dicks develops the relevant Mayer–Vietoris constructions (2; 3); his induced grading applies to arbitrary modules for group-ring embeddings (3, Theorem 16 and p. 455). Here the relation module has the form of (3, §5, equation (16)), and the tree argument proves imδ∩(1⊗V)=0 for the prescribed linear relations.
The HNN group and its coset tree
Let A,B≤G and let θ:A→B be an isomorphism. Set
H=⟨G,t∣tat−1=θ(a)(a∈A)⟩,(8)
the quotient of G∗⟨t⟩ by the normal closure of these relations.
Lemma 2. The canonical homomorphism G→H is injective. The graph with vertex set H/G, edge set H/A, and endpoints
hA:hG⟶ht−1G(9)
is a tree.
Proof. For ε∈{+1,−1} put
C+=B,C−=A,ψ+=θ−1,ψ−=θ.
Choose sets Tε of representatives such that every g∈G has a unique expression g=rc, with r∈Tε and c∈Cε, and such that 1∈Tε represents Cε. Consider formal strings
r1tε1⋯rmtεmg,ri∈Tεi,g∈G,(11)
subject to
i>1,εi−1=−εi⟹ri=1.(12)
The case m=0 is the string g.
Write Pm=r1tε1⋯rmtεm, with P0=1. For g=rc, r∈Tε, c∈Cε, define
Both outputs satisfy (12). Since ψε(c)∈C−ε, reversing a noncancelling step deletes the added block and restores rψ−εψε(c)=rc. After a cancelling step, rmψε(c) is already in T−εC−ε form; reversal restores the deleted block. A second cancellation would violate (12). Hence the two stable-letter operations are inverse.
For a∈A, write the tail as rc, c∈B. Appending ta and θ(a)t uses the same representative r and the same case above; in either case the new tail is θ−1(c)a, preceded by rm in the cancelling case. Thus
(w⋅t)⋅a=(w⋅θ(a))⋅t.
Together with (w⋅b)⋅b′=w⋅(bb′), this defines a right H-action on the normal strings.
Every operation is valid in H, by
ctε=tεψε(c)(c∈Cε).
Reading a normal string from 1 returns that string. Equal elements of H therefore have equal normal strings, so G→H is injective. In particular, a word
g0tε1g1⋯tεmgm,m>0,(16)
with no segment tat−1, a∈A, or t−1bt, b∈B, lies outside G. Before a first cancellation, the coefficient carried past tεi lies in C−εi. For opposite successive signs, multiplication by this coefficient preserves membership in C−εi; cancellation would require one of the excluded segments. The normal string consequently retains all m stable letters.
They are distinct, since t∈/G. The graph is connected: its edges permit multiplication by a stable letter in either direction, while G fixes the base vertex, and G∪{t} generates H.
More explicitly, a step from wG can be written wG→wgtεG with g∈G. For ε=−1 the edge is wgA; for ε=+1 it is wgtA traversed backwards. Two consecutive steps containing tat−1, a∈A, traverse the same edge in opposite directions. The same holds for t−1bt, b∈B, using bt=tθ−1(b). A nonbacktracking path consequently gives a word of the form (16) with none of the excluded segments. A positive length closed path based at G would give such a word lying in G, contradicting the reduced-word assertion. By translating the base vertex, the same holds at every vertex. The graph is a tree. ◻
Modules on a forest
Let F be an oriented forest, with vertex set V, edge set E, and initial and terminal vertices o(e) and q(e) for an edge e. Assign an R-module Mv to each vertex and an R-module Ne to each edge, together with injective linear maps
αe:Ne⟶Mo(e),βe:Ne⟶Mq(e).(18)
Write ιv:Mv→⨁w∈VMw for the summand inclusion. The boundary map is
Consequently every canonical map Mv→coker∂ is injective.
Proof. For 0=x∈⨁eNe, put F=suppx. A component of the finite forest F containing an edge has two leaves, the endpoints of a longest simple path. At either leaf w, let e be its unique incident edge in F. Then
The last inequality uses xe=0 and the injectivity of the endpoint maps. Thus ∣supp(∂x)∣≥2, giving both ∂x=0 and ∂x∈/ιv(Mv) for every v. The induced map Mv→coker∂ is therefore injective. ◻
The argument requires neither finite generation of the modules nor finiteness of F. In particular, it does not choose complements to the endpoint images in (18).
Realizing a prescribed subrepresentation isomorphism
Theorem 4. Let (G,V) be an R-representation and let (θ,φ):(A,U)→(B,U′) be an isomorphism of subrepresentations. There are a representation (H,V,σ) with kerσ=1, an embedding (i,j):(G,V)→(H,V), and an element t∈H satisfying
ti(a)t−1=i(θ(a)),tj(u)=j(φ(u))(a∈A,u∈U).(22)
No finite-generation hypothesis is imposed, and
∣H∣,∣V∣≤max{ℵ0,∣R∣,∣G∣,∣V∣}.(23)
Proof. Take H from (8). By Lemma 2 we may identify G with its image in H. Form the induced modules
The formula is R-linear and satisfies δ(kh⊗u)=kδ(h⊗u), so it defines an R[H]-linear map.
For each x∈H/G choose a representative rx, with rG=1. The right R[G]-module decomposition R[H]=⨁xrxR[G] gives
M=x∈H/G⨁Vx,E=y∈H/A⨁Uy,(27)
where each Vx is a copy of V and each Uy is a copy of U. These are decompositions of R-modules; no basis of V or U is assumed.
Choose a representative sy for each y∈H/A. Write
sy=rxg0,syt−1=rzg1,g0,g1∈G,
where x=syG and z=syt−1G are the endpoints of y. The map from the summand Uy in (27) has exactly the two components
u⟼−g0u∈Vx,u⟼g1φ(u)∈Vz.(29)
Both endpoint maps are injective: the inclusions U,U′≤V are injective, φ is an isomorphism, and the actions of g0,g1 are invertible. By Lemma 2, the indexing graph in (27) is a tree. Lemma 3 therefore gives
kerδ=0,imδ∩(1⊗V)=0.(30)
Set
V0=M/imδ,j0(v)=[1⊗v].(31)
Equivalently, let D=⨁h∈HV with summand maps eh, and let C≤D be the R-submodule generated by
Then V0≅D/C. Indeed, quotienting by the first family gives M through eh(v)↦h⊗v; the inverse is the balanced map h⊗v↦[eh(v)]. Under this identification, the second family is the negative of δ(ht⊗u) and spans imδ. The submodule imδ is H-invariant, so V0 is an R[H]-module. Equation (30) makes j0 injective. Equivariance follows from
gj0(v)=[g⊗v]=[1⊗gv]=j0(gv).
For the stable letter,
0=[δ(t⊗u)]=[1⊗φ(u)]−[t⊗u]=j0(φ(u))−tj0(u).(34)
The conjugation equation comes from the presentation of H.
Put
V=V0⊕R[H],j(v)=(j0(v),0).(35)
For the diagonal action σ, the regular basis gives
with the additional intersection property (30). The target V0 need not itself be induced from G. For example, if G=A=1, U=V=R, and φ=id, then
M=R[t,t−1],V0=R[t,t−1]/(t−1)≅R.
Directed limits and cardinal bounds
Let I be a nonempty directed poset. Consider a system Xi=(Gi,Vi,ρi) with embeddings eij=(aij,bij) satisfying
eii=1Xi,ejkeij=eik(i≤j≤k).
Lemma 5. Every directed system of embeddings has a limit representation X=(G,V,ρ) and compatible embeddings ei:Xi→X such that every finite collection of group and module elements in X belongs to the image of one Xi. Moreover,
Passing to a common upper stage proves well-definedness and the representation identities. Injectivity follows from
[i,g]=[i,h]⟹aik(g)=aik(h)⟹g=h
for some k≥i, and the analogous calculation for bik. Finite sets have representatives in one stage. A compatible family (fi) factors uniquely by [i,x]↦fi(x) on each sort.
If a=b in Gi and kerρi=1, choose v∈Vi with av=bv. Then
[i,a][i,v]=[i,av]=[i,bv]=[i,b][i,v],
which proves kerρ=1.
Each quotient has cardinality at most that of the corresponding disjoint union. Under (42),
∣G∣≤i∈I∑∣Gi∣≤∣I∣κ≤κ,∣V∣≤i∈I∑∣Vi∣≤∣I∣κ≤κ.
These inequalities use only κ⋅κ=κ for infinite κ. ◻
Lemma 6. Suppose an isomorphism between subrepresentations of X is implemented in an extension Y by t∈GY. For every embedding (a,b):Y→Z, its transported isomorphism is implemented by a(t).
Proof. Apply a and b to the two implementation equations. The required identities follow from
a(tgt−1)=a(t)a(g)a(t)−1,b(tv)=a(t)b(v).
◻
Lemma 7. If κ is infinite and X=(G,V) satisfies ∣G∣,∣V∣≤κ, then
The codomain is (θ(A),φ(U)). There are at most κm+n≤κ choices for (52), hence ∣D(X)∣≤κ⋅κ=κ. ◻
Proposition 8. Suppose κ is infinite and
X=(G,V,ρ),∣R∣,∣G∣,∣V∣≤κ,kerρ=1.
There is an embedding X↪X+=(G+,V+,ρ+) with
∣G+∣,∣V+∣≤κ,kerρ+=1,
in which every member of D(X) is implemented by an element of G+.
Proof. By Lemma 7, write D(X)={dα:α<λ}, ∣λ∣≤κ. Construct compatible embeddings by transfinite recursion:
X0=X,Xα↪Xα+1,Xβ=α<βlimXα(β a nonzero limit).
At a successor, transport dα along X↪Xα and realize it by Theorem 4; at a limit, use Lemma 5. Induction gives
∣GXα∣,∣VXα∣≤κ,kerρXα=1(α≤λ),
since ∣β∣κ≤κ at every limit β≤λ. Lemma 6 preserves all previous realizations, so X+=Xλ suffices; for λ=0, take X+=X. ◻
Only κ⋅κ=κ is used; κ need not be regular.
The homogeneous representation
Proof of Theorem 1. Let κ be as in (6). The representation
X0=(G,V⊕R[G])
with diagonal action contains (G,V) by v↦(v,0) and satisfies kerρX0=1 and ∣GX0∣,∣VX0∣≤κ. Apply Proposition 8 successively to obtain
X0e0X1e1X2e2⋯,(59)
where D(Xn) is realized in Xn+1. Lemma 5 gives an extension (Γ,W,σ) with
kerσ=1,∣Γ∣,∣W∣≤κ.
Identify each stage with its image in this direct limit. These images are nested, so
Γ=n<ω⋃Gn,W=n<ω⋃Vn.(61)
Let (S,T,(θ,φ))∈D(Γ,W), with S=(A,U) and T=(B,U′). Choose finite group generating sets SA,SB and finite module generating sets TU,TU′ over R[A],R[B], respectively. By Lemma 5, some Xn contains all four finite sets. Closure of that stage under its operations gives
The stage actions are the restrictions of the limit action, so (S,T,(θ,φ))∈D(Xn). Choose the implementing t∈Gn+1. By Lemma 6, (7) holds in (Γ,W). The automorphism (ct,Lt), with inverse (ct−1,Lt−1), extends the given isomorphism. ◻
Corollary 9. If R is countable, every representation with countable group and module embeds in a countable inner homogeneous representation (Γ,W,σ) with kerσ=1.
Corollary 10. Let K be a field. Every representation of a group G on a K-vector space V embeds in an inner homogeneous representation (Γ,W,σ) satisfying
E. V. Aladova, A. A. Gvaramiya and B. I. Plotkin, Logic in representations of groups, Algebra and Logic 51 (2012), 1–27. doi:10.1007/s10469-012-9167-8. Russian original: Algebra i Logika 51 (2012), 3–40, MathNet:al520.
W. Dicks, Mayer–Vietoris presentations over colimits of rings, Proc. London Math. Soc. (3) 34 (1977), no. 3, 557–576. doi:10.1112/plms/s3-34.3.557.