A nonsplit extension 1→D+→E→F2×F2→11\to D^+\to E\to F_2\times F_2\to1 with the left-multiplication action

18.76

Problem

Let DD be a division ring and G≤D×G\le D^\times. Must every extension by D+D^+ split when the conjugation action is left multiplication in DD?

1⟶D+⟶E→πG⟶1,∃s:G→E: πs=1G  ?1\longrightarrow D^+\longrightarrow E\xrightarrow{\pi}G\longrightarrow1, \qquad\exists s:G\to E:\ \pi s=1_G\;?

The prescribed module structure is g⋅d=gdg\cdot d=gd; replacing it by the trivial action changes the problem.

The extension problem

Let DD be a division ring and G≤D×G\leq D^\times. Regard the additive group D+D^+ as a left GG-module by multiplication in DD. Kourovka Problem 18.76 concerns splitting of extensions with this action (1). The action is fixed throughout: replacing it by the trivial action changes the question.

Theorem 1. There exist a countable division ring DD of characteristic zero, an embedding F2×F2↪D×F_2\times F_2\hookrightarrow D^\times, and a countable torsion-free group EE fitting into a nonsplit exact sequence

1⟶D+→ιE→πF2×F2⟶1.(1) \tag{1} 1\longrightarrow D^+\xrightarrow{\iota}E\xrightarrow{\pi}F_2\times F_2\longrightarrow1.

Conjugation by any lift of g∈F2×F2g\in F_2\times F_2 acts on D+D^+ as u↦guu\mapsto gu.

We write P=F(a,b)P=F(a,b), Q=F(c,d)Q=F(c,d) and G=P×QG=P\times Q. Elements of the two factors commute in GG and therefore in every division ring containing its group algebra. Commutativity within a factor is neither assumed nor used.

A countable division ring containing the group algebra

Lemma 2. The group algebra Q[G]\mathbb Q[G] embeds in a countable division ring of characteristic zero.

Proof. Free groups are bi-orderable, and a lexicographic product of bi-orders orders GG. One construction for a free group uses the Magnus expansion in noncommuting variables: send a free generator xix_i to 1+Xi1+X_i and use

(1+Xi)−1=1−Xi+Xi2−Xi3+⋯ .(1+X_i)^{-1}=1-X_i+X_i^2-X_i^3+\cdots.

The expansion is injective. Ordering monomials first by degree and then lexicographically orders a nonidentity series by the sign of its first nonzero coefficient after the constant term. Multiplication and conjugation preserve that first term, so the induced order is invariant on both sides. This is the standard bi-order on a free group used in the ordered-series construction (2).

For a bi-ordered group, the Mal’cev–Neumann series ring Q((G))\mathbb Q((G)) consists of formal sums

∑g∈Grgg(rg∈Q)\sum_{g\in G}r_g g\qquad(r_g\in\mathbb Q)

with well-ordered support. Convolution is defined because only finitely many products from two well-ordered supports contribute to a given coefficient. This ring is a division ring (2; 3): after extracting the least nonzero term, a nonzero series has the form rg(1−u)r g(1-u) with supp⁡(u)>1\operatorname{supp}(u)>1, and its inverse is given by the summable geometric series ∑j≥0uj\sum_{j\geq0}u^j. Finite sums identify Q[G]\mathbb Q[G] with a subring.

Let D0=Q[G]D_0=\mathbb Q[G], and let Dr+1D_{r+1} be the subring generated by DrD_r and the inverses of all its nonzero elements inside Q((G))\mathbb Q((G)). Each DrD_r is countable. The union D=⋃r≥0DrD=\bigcup_{r\geq0}D_r is countable and closed under inversion of nonzero elements. It is the required division ring. ◻

Fix such an embedding and suppress it from the notation. In particular,

a−1≠0,b−1≠0,c−1≠0,d−1≠0(4) \tag{4} a-1\neq0,\quad b-1\neq0,\quad c-1\neq0,\quad d-1\neq0

in DD. Every element of Q[P]\mathbb Q[P] commutes with every element of Q[Q]\mathbb Q[Q]; inverses of nonzero elements retain this commutation property.

Derivations and the factor set

Lemma 3. There are maps α:P→Q[P]\alpha:P\to\mathbb Q[P] and β:Q→Q[Q]\beta:Q\to\mathbb Q[Q] satisfying

α(xx′)=α(x)+xα(x′),α(a)=0,α(b)=1,β(yy′)=β(y)+yβ(y′),β(c)=0,β(d)=1.\begin{align*} \alpha(xx')&=\alpha(x)+x\alpha(x'),& \alpha(a)&=0,&\alpha(b)&=1,\tag{5a}\\ \beta(yy')&=\beta(y)+y\beta(y'),& \beta(c)&=0,&\beta(d)&=1.\tag{5b} \end{align*}

Both maps vanish at the identity.

Proof. For a free group, prescribe values on the free generators and put α(x−1)=−x−1α(x)\alpha(x^{-1})=-x^{-1}\alpha(x) for their inverses. If x1⋯xrx_1\cdots x_r is a word, set

α(x1⋯xr)=∑j=1rx1⋯xj−1α(xj).\alpha(x_1\cdots x_r)=\sum_{j=1}^r x_1\cdots x_{j-1}\alpha(x_j).

Inserting or deleting adjacent inverse letters changes this sum by w(α(x)+xα(x−1))=0w\bigl(\alpha(x)+x\alpha(x^{-1})\bigr)=0. The value is therefore well-defined on the free group. Concatenating words proves (5a). Apply the same construction to QQ. ◻

For g=(x,y)g=(x,y) and h=(x′,y′)h=(x',y'), define

C(g,h)=α(x)yβ(y′)∈D.(7) \tag{7} C(g,h)=\alpha(x)y\beta(y')\in D.

Lemma 4. The map CC is normalized and satisfies

C(g,h)+C(gh,k)=gC(h,k)+C(g,hk)(g,h,k∈G).(8) \tag{8} C(g,h)+C(gh,k)=gC(h,k)+C(g,hk) \qquad(g,h,k\in G).

Proof. Normalization follows from α(1)=β(1)=0\alpha(1)=\beta(1)=0. Write k=(x′′,y′′)k=(x'',y''). The left side of (8) expands as

α(x)yβ(y′)+α(x)yy′β(y′′)+xα(x′)yy′β(y′′).\alpha(x)y\beta(y')+\alpha(x)yy'\beta(y'') +x\alpha(x')yy'\beta(y'').

The right side has the same three terms, since yy commutes with α(x′)\alpha(x') and β(y′y′′)=β(y′)+y′β(y′′)\beta(y'y'')=\beta(y')+y'\beta(y''). ◻

Define a multiplication on E=D×GE=D\times G by

(u,g)(v,h)=(u+gv+C(g,h),gh).(10) \tag{10} (u,g)(v,h)=\bigl(u+gv+C(g,h),gh\bigr).

Equation (8) proves associativity, and (0,1)(0,1) is the identity. The inverse is

(u,g)−1=(−g−1(u+C(g,g−1)),g−1).(11) \tag{11} (u,g)^{-1}=\bigl(-g^{-1}(u+C(g,g^{-1})),g^{-1}\bigr).

It is a right inverse by direct substitution. The cocycle equation with (g,g−1,g)(g,g^{-1},g) gives C(g,g−1)=gC(g−1,g)C(g,g^{-1})=gC(g^{-1},g), which also verifies the left inverse.

Set ι(v)=(v,1)\iota(v)=(v,1) and π(u,g)=g\pi(u,g)=g. These are homomorphisms with ι\iota injective, π\pi surjective and ker⁡π=ι(D+)\ker\pi=\iota(D^+). Moreover, normalization gives

(u,g)(v,1)=(gv,1)(u,g),(u,g)(v,1)=(gv,1)(u,g),

which proves the prescribed conjugation action.

The obstruction to a section

Lemma 5. The projection π:E→G\pi:E\to G has no group-homomorphic section.

Proof. Suppose a section exists. Write its values on the four free generators as

(ua,a),(ub,b),(vc,c),(vd,d).(u_a,a),\quad(u_b,b),\quad(v_c,c),\quad(v_d,d).

Here a generator denotes its canonical image in P×QP\times Q. Each of a,ba,b commutes with each of c,dc,d. The corresponding lifts must therefore commute. From (7),

C(a,c)=C(a,d)=C(b,c)=0,C(b,d)=1,C(a,c)=C(a,d)=C(b,c)=0,\quad C(b,d)=1,

whereas C(c,a)=C(d,a)=C(c,b)=C(d,b)=0C(c,a)=C(d,a)=C(c,b)=C(d,b)=0. Substitution in (10) gives

(c−1)ua=(a−1)vc,(d−1)ua=(a−1)vd,(c−1)ub=(b−1)vc,(d−1)ub−(b−1)vd=1.\begin{align*} (c-1)u_a&=(a-1)v_c,\tag{15a}\\ (d-1)u_a&=(a-1)v_d,\tag{15b}\\ (c-1)u_b&=(b-1)v_c,\tag{15c}\\ (d-1)u_b-(b-1)v_d&=1.\tag{15d} \end{align*}

By (4), equations (15a)–(15c) imply

vc=(a−1)−1(c−1)ua,vd=(a−1)−1(d−1)ua,ub=(b−1)(a−1)−1ua.\begin{align*} v_c&=(a-1)^{-1}(c-1)u_a,\\ v_d&=(a-1)^{-1}(d-1)u_a,\\ u_b&=(b-1)(a-1)^{-1}u_a. \end{align*}

For the last equality, commute c−1c-1 across b−1b-1 and (a−1)−1(a-1)^{-1} and cancel it on the left. Similarly d−1d-1 commutes with both of these factors, so the left side of (15d) is

(d−1)(b−1)(a−1)−1ua−(b−1)(a−1)−1(d−1)ua=0.(d-1)(b-1)(a-1)^{-1}u_a -(b-1)(a-1)^{-1}(d-1)u_a=0.

This contradicts (15d). ◻

Proof of Theorem 1. The construction above gives the exact sequence and its action, and Lemma 5 proves that it does not split. Both DD and GG are countable, hence so is EE. Free groups are torsion-free: a nonempty cyclically reduced word has nonempty powers, and every nonidentity word is conjugate to one. Therefore P×QP\times Q is torsion-free. If an element of EE has finite order, its image in GG is trivial. It belongs to D+D^+, which is torsion-free since char⁡D=0\operatorname{char}D=0. Thus EE is torsion-free. ◻

References

Preprint · Lean (GitHub)

  1. E. I. Khukhro and V. D. Mazurov (eds.), Unsolved Problems in Group Theory: The Kourovka Notebook, Problem 18.76. arXiv:1401.0300.
  1. B. H. Neumann, On ordered division rings, Trans. Amer. Math. Soc. 66 (1949), 202–252. doi:10.1090/S0002-9947-1949-0032593-5.
  1. B. Poonen, Units in Hahn–Mal’cev–Neumann rings. https://math.mit.edu/~poonen/papers/malcev.pdf.