This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Algebraic structures on the Cantor set

Evgenii Reznichenko [email protected] Department of General Topology and Geometry, Mechanics and Mathematics Faculty, M. V. Lomonosov Moscow State University, Leninskie Gory 1, Moscow, 199991 Russia
Abstract

Below, by space we mean a separable metrizable zero-dimensional space. It is studied when the space can be embedded in a Cantor set while maintaining the algebraic structure. Main results of the work: every space is an open retract of a Boolean precompact group; every strongly homogeneous space is rectifiable. In this case, the space can be embedded in the Cantor set with the preservation of the algebraic structure. An example of a strongly homogeneous space is constructed which do not admit the structure of a right topological group.

keywords:
Baire space , strongly homogeneous space , rectifiable space , retract of group , separable metrizable zero-dimensional space , right topological group ,
MSC:
[2010] 54B10 , 54C30 , 54C05 , 54C20 , 22A05 , 54H11

1 Introduction

In this paper, we study what algebraic structures are possible on subspaces of the Cantor set 𝐂{\mathbf{C}}, that is, on separable metrizable zero-dimensional (SMZD) spaces. Any SMZD space is Mal’tsev, moreover, it is a retract of the topological group, that is, a retral space [1].

We are primarily interested in Mal’tsev, rectifiable, retral, homogeneous SMZD spaces, (right) topological groups. It is also studied when a SMZD space with algebraic structure can be embedded in SMZD a compact space with the same structure. In particular, when the SMZD space can be embedded in 𝐂{\mathbf{C}} with the algebraic structure preserved.

Main results of the work: each SMZD space is an open retract of a Boolean precompact SMZD group (Theorem 1); every SMZD strongly homogeneous space is rectifiable (Theorem 4).

Rectifiable spaces are also characterized as spaces on which there is a Mal’tsev operation with some additional identity (Φ3\Phi_{3}), the so-called homogeneous Mal’tsev operations (Definition 1). A strongly Mal’tsev operation is introduced and studied, which defines strongly rectifiable spaces. This class lies between rectifiable spaces and topological groups.

In the 6 section, an example of a strongly homogeneous SMZD space is constructed which do not admit the structure of a right topological group.

2 Definitions and notation

By a space we understand a regular topological space.

The weight and character of the space XX will be denoted by w(X)w(X) and χ(X)\chi(X), respectively.

Denote by Aut(X)\operatorname{Aut}(X) the set of all homeomorphisms of the space XX onto itself, Aut(X|M)={fAut(X):f(M)=M}\operatorname{Aut}(X|M)=\{f\in\operatorname{Aut}(X)\,:\,f(M)=M\} for MXM\subset X. Denote by idX\operatorname{id}_{X} the identity mapping of XX onto itself.

A space XX is called strongly homogeneous if every non-empty clopen subset of XX is homeomorphic to XX. Every strongly homogeneous SMZD space is homogeneous.

If f:XYf:X\to Y is a mapping and AXA\subset X, then f|A\left.f\right|_{A} denotes the restriction of ff to AA.

2.1 Non-Archimedean spaces

A subset MM of a linearly ordered set XX is called convex if abca\leq b\leq c holds for a,cMa,c\in M and bXb\in X, then bMb\in M.

A family of sets {\mathcal{B}} is called non-Archimedean if for U,VU,V\in{\mathcal{B}} the following holds: if UVU\cap V\neq\varnothing, then either UVU\subset V or VUV\subset U. Non-Archimedean base is the base {\mathcal{B}} of the space XX, which is a non-Archimedean family. Non-Archimedean bases called bases of rank 1 were introduced and studied in [2, 3].

Every non-Archimedean space has a base which is a tree by reverse inclusion (Theorem 2.9 [4]). Each a base which is a tree by reverse inclusion is a non-Archimedean base.

A linear order \leq on a space XX is said to be consistent with a non-Archimedean base {\mathcal{B}} if any element of UU\in{\mathcal{B}} is convex with respect to the order \leq.

2.2 Cantor set

Denote 𝐃={0,1}{\mathbf{D}}=\{0,1\}, 𝐃<ω=nω𝐃n{\mathbf{D}}^{<\omega}=\bigcup_{n\in\omega}{\mathbf{D}}^{n}. Here 𝐃0={}{\mathbf{D}}^{0}=\{\varnothing\}.

Denote by 𝐂{\mathbf{C}} the Cantor set, 𝐂=𝐃ω{\mathbf{C}}={\mathbf{D}}^{\omega}. Let x=(xn)nω𝐂x=(x_{n})_{n\in\omega}\in{\mathbf{C}}, kωk\in\omega and c=(cn)n<k𝐃k𝐃<ωc=(c_{n})_{n<k}\in{\mathbf{D}}^{k}\subset{\mathbf{D}}^{<\omega}. Denote

x|k\displaystyle x|k =(x0,,xk1),\displaystyle=(x_{0},...,x_{k-1}), x(k)\displaystyle x(k) =xk,\displaystyle=x_{k}, B(c)\displaystyle B(c) ={x𝐂:x|k=c}.\displaystyle=\{x\in{\mathbf{C}}\,:\,x|k=c\}.

The family

𝒞={B(c):c𝐃<ω}\mathcal{C}=\{B(c)\,:\,c\in{\mathbf{D}}^{<\omega}\}

is the standard non-Archimedean base 𝐂{\mathbf{C}}. The base 𝒞\mathcal{C} is a tree by reverse inclusion. On 𝐂{\mathbf{C}} we will consider the lexigraphic order: for different x,y𝐂x,y\in{\mathbf{C}}, x<yx<y if x|k=y|kx|k=y|k, x(k)=0x(k)=0 and y(k)=1y(k)=1 for some kωk\in\omega. The lexigraphic order is consistent with the non-Archimedean base 𝒞\mathcal{C}.

Denote suppx={nω:x(n)=1}\operatorname{supp}x=\{{n\in\omega}\,:\,x(n)=1\} for x𝐂x\in{\mathbf{C}}, 𝐐={x𝐂:|suppx|<ω}{\mathbf{Q}}=\{x\in{\mathbf{C}}\,:\,|\operatorname{supp}x|<\omega\} and 𝐏=𝐂𝐐{\mathbf{P}}={\mathbf{C}}\setminus{\mathbf{Q}}. The space 𝐐{\mathbf{Q}} is homeomorphic to rational numbers, and the space 𝐏{\mathbf{P}} is homeomorphic to irrational numbers.

2.3 Rectifiable, Maltsev, retral spaces

Let XX be a space, eXe\in X, Ψ:X2X2\Psi:X^{2}\to X^{2}. A mapping is called rectification if it is a homeomorphism and there exists a mapping p:X2Xp:X^{2}\to X such that Ψ(x,y)=(x,p(x,y))\Psi(x,y)=(x,p(x,y)) and p(x,x)=ep(x,x)=e for x,yXx,y\in X. A space XX is called rectifiable if there exists a rectification on XX. We define the operation q:X2Xq:X^{2}\to X by the identity q(x,p(x,y))=yq(x,p(x,y))=y. The operation qq is continuous and the identities q(x,e)=xq(x,e)=x and p(x,q(x,y))=yp(x,q(x,y))=y hold for it. Also, Ψ1(x,y)=(x,q(x,y))\Psi^{-1}(x,y)=(x,q(x,y)). The operations pp and qq are called homogeneous algebra [5, 6].

A biternary algebra on a space XX is a pair of ternary continuous operations α,β:X3X\alpha,\beta:X^{3}\to X such that

α(y,y,x)=α(β(x,y,z),y,z)=β(α(x,y,z),y,z)=x,\alpha(y,y,x)=\alpha(\beta(x,y,z),y,z)=\beta(\alpha(x,y,z),y,z)=x,

for all x,y,zXx,y,z\in X (see [7]).

For a space XX the following conditions are equivalent: XX is a rectifiable space; XX is homeomorphic to a homogeneous algebra; there exists a structure of a biternary algebra on XX [5].

A mapping Φ:X3X\Phi:X^{3}\to X is called a Maltsev operation if Φ(x,y,y)=Φ(y,y,x)=x\Phi(x,y,y)=\Phi(y,y,x)=x for x,yXx,y\in X. A space XX is called Mal’tsev if there exists a continuous Mal’tsev operation on XX. Rectifiable spaces are Maltsev: Φ(x,y,z)=q(x,p(y,z))\Phi(x,y,z)=q(x,p(y,z)).

A space XX is called retral if there exists a topological group GG into which XX is embedded in such a way that XX is a retract of GG [8]. The retral space is Mal’tsev, the reverse is not true [1].

3 Non-Archimedean spaces

In this section, we prove a more detailed and specialized version of Theorem 7 (version 2) from [1] and apply it to 𝐂{\mathbf{C}}.

Let XX be a zero-dimensional space. Denote by B(X)B(X) the set of all finite subsets of XX. We consider B(X)B(X) as an adiative Boolean group with zero \varnothing and the group operation of addition is a symmetric difference. Denote by Γ(X)\Gamma(X) the family of all open partitions of XX. For γΓ(X)\gamma\in\Gamma(X) we set

H(γ)={gB(X):|gU| is even for Uγ}.H(\gamma)=\{g\in B(X)\,:\,|g\cap U|\text{ is even for }U\in\gamma\}.

The set H(γ)H(\gamma) is a subgroup of B(X)B(X). Let 𝒢Γ(X)\mathcal{G}\subset\Gamma(X). Denote by Bz(X,𝒢)B_{z}(X,\mathcal{G}) the group B(X)B(X) with the group topology whose prebase at unity is formed by sets of the form H(γ)H(\gamma) for γ𝒢\gamma\in\mathcal{G}. The set XX naturally embeds in B(X)B(X): i:XB(X),x{x}i:X\to B(X),\ x\mapsto\{x\}. This embedding is topological and i(X)i(X) is closed in B(X)B(X) if and only if 𝒢\bigcup\mathcal{G} is a subbase of XX. In what follows we will identify XX and i(X)i(X).

Denote by Bz(X)=Bz(X,Γ(X))B_{z}(X)=B_{z}(X,\Gamma(X)). The space XX is close embedded in Bz(X)B_{z}(X). Let YXY\subset X. Denote

Bz0(X)\displaystyle B_{z}^{0}(X) ={gBz(X):|g| is even},\displaystyle=\{g\in B_{z}(X)\,:\,|g|\text{ is even}\}, Bz1(X)\displaystyle B_{z}^{1}(X) ={gBz(X):|g| is odd},\displaystyle=\{g\in B_{z}(X)\,:\,|g|\text{ is odd}\},
Bz(X|Y)\displaystyle B_{z}(X|Y) ={gBz(X):gY},\displaystyle=\{g\in B_{z}(X)\,:\,g\subset Y\},
Bz0(X|Y)\displaystyle B_{z}^{0}(X|Y) =Bz(X|Y)Bz0(X),\displaystyle=B_{z}(X|Y)\cap B_{z}^{0}(X), Bz1(X|Y)\displaystyle B_{z}^{1}(X|Y) =Bz(X|Y)Bz1(X).\displaystyle=B_{z}(X|Y)\cap B_{z}^{1}(X).

The set Bz0(X)=H({X})B_{z}^{0}(X)=H(\{X\}) is a clopen subgroup of index two. The set Bz1(X)=Bz(X)Bz0(X)B_{z}^{1}(X)=B_{z}(X)\setminus B_{z}^{0}(X) is openly closed in Bz(X)B_{z}(X) and is a coset of the group Bz0(X)B_{z}^{0}(X) [1].

Let XX be a space with non-Archimedean base {\mathcal{B}} is a tree by reverse inclusion and some linear order << consistent with base {\mathcal{B}}.

For {\mathcal{B}}^{\prime}\subset{\mathcal{B}} denote by Λ()\Lambda_{*}({\mathcal{B}}^{\prime}) the set of maximal by inclusion elements {\mathcal{B}}^{\prime}. Then Λ()\Lambda_{*}({\mathcal{B}}^{\prime})\subset{\mathcal{B}}^{\prime}; the family {\mathcal{B}}^{\prime} is inscribed in Λ()\Lambda_{*}({\mathcal{B}}^{\prime}); Λ()\Lambda_{*}({\mathcal{B}}^{\prime}) is an open partition of \bigcup{\mathcal{B}}^{\prime}.

For a family γ\gamma of open subsets XX, denote Λ(γ)=Λ()\Lambda(\gamma)=\Lambda_{*}({\mathcal{B}}^{\prime}), where ={U:UV for some Vγ}{\mathcal{B}}^{\prime}=\{U\in{\mathcal{B}}\,:\,U\subset V\text{ for some }V\in\gamma\}. Then Λ(γ)\Lambda(\gamma) is inscribed in γ\gamma; Λ(γ)\Lambda(\gamma) is an open partition of γ\bigcup\gamma; if VV\in{\mathcal{B}} and VWV\subset W for some WγW\in\gamma, then VUV\subset U for some UΛ(γ)U\in\Lambda(\gamma). If γ\gamma is an open cover of XX, then Λ(γ)Γ(X)\Lambda(\gamma)\in\Gamma(X) and Λ(γ)\Lambda(\gamma) are inscribed in γ\gamma. Therefore, Bz(X)=Bz(X,𝒫)B_{z}(X)=B_{z}(X,{\mathcal{P}}), where

𝒫={γΓ(X):γ}.{\mathcal{P}}=\{\gamma\in\Gamma(X)\,:\,\gamma\subset{\mathcal{B}}\}.

For an open WXW\subset X, denote Λ(W)=Λ({W})\Lambda(W)=\Lambda(\{W\}). Then Λ(W)\Lambda(W) is an open partition of WW; if VV\in{\mathcal{B}} and VWV\subset W, then VUV\subset U for some UΛ(W)U\in\Lambda(W).

Recall the construction of the retraction r:Bz1(X)Xr:B_{z}^{1}(X)\to X from [1]. For gBz1(X)g\in B_{z}^{1}(X) we set Υ(g)=Λ(γ)\Upsilon(g)=\Lambda_{*}(\gamma), where γ={U:gU and |gU| is even}\gamma=\{U\in{\mathcal{B}}\,:\,g\cap U\neq\varnothing\text{ and }|g\cap U|\text{ is even}\}. Let R(g)=gΥ(g)R(g)=g\setminus\bigcup\Upsilon(g) and r(g)=minR(g)r(g)=\min R(g). Theorem 7 (version 2) [1] proves that rr is a continuous retraction.

Lemma 1.

The mapping rr is open. If YXY\subset X and S=Bz1(X|Y)S=B_{z}^{1}(X|Y), then the mapping r|S:SY\left.r\right|_{S}:S\to Y is a continuous open retraction of SS to YY.

Proof.

Let gBz1(X)g\in B_{z}^{1}(X) and WBz(X)W\subset B_{z}(X) be a neighborhood of the point gg. Then g+H(γ)Wg+H(\gamma)\subset W for some γΓ(X)\gamma\in\Gamma(X). Let x=r(g)x=r(g). Let xVγx\in V\in\gamma. The family Υ(g)\Upsilon(g) is finite and xΥ(g)x\notin\bigcup\Upsilon(g). Take UU\in{\mathcal{B}} such that xUVΥ(g)x\in U\subset V\setminus\bigcup\Upsilon(g).

Let yUy\in U and h=g+{x,y}h=g+\{x,y\}. Then Υ(g)=Υ(h)\Upsilon(g)=\Upsilon(h), R(h)=R(g)+{x,y}R(h)=R(g)+\{x,y\} and r(h)=yr(h)=y. Let M={g+{x,y}:yU}M=\{g+\{x,y\}\,:\,y\in U\}. Then Mg+H(γ)M\subset g+H(\gamma) and r(M)=Ur(M)=U. Hence r(g)Ur(W)r(g)\in U\subset r(W).

Let gYg\in Y, Q=UYQ=U\cap Y and L={g+{x,y}:yQ}L=\{g+\{x,y\}\,:\,y\in Q\}. Then L=MSL=M\cap S and r(L)=Qr(L)=Q. Hence r(g)Qr(WS)r(g)\in Q\subset r(W\cap S). ∎

Lemma 1 and Theorem 7 (version 2) [1] imply the following assertion.

Theorem 1.

Let XX be a non-Archimedean space. Then there exists a continuous open retraction r:Bz1(X)Xr:B_{z}^{1}(X)\to X such that for any YXY\subset X and S=Bz1(X|Y)S=B_{z}^{1}(X|Y) the map r|S:SY\left.r\right|_{S}:S\to Y is a continuous open retraction.

If XX is a compact space, then Bz(X)B_{z}(X) is a precompact group. The set Γ(X)\Gamma(X) is countable. Hence Bz(𝐂)B_{z}({\mathbf{C}}) is a precompact SMZD group. Let’s apply the theorem 1 to 𝐂{\mathbf{C}}.

Theorem 2.

There exists a continuous open retraction r:Bz1(𝐂)𝐂r:B_{z}^{1}({\mathbf{C}})\to{\mathbf{C}} such that for any X𝐂X\subset{\mathbf{C}} and S=Bz1(𝐂|X)S=B_{z}^{1}({\mathbf{C}}|X) the map r|S:SX\left.r\right|_{S}:S\to X is a continuous open retraction.

The set Bz1(𝐂)B_{z}^{1}({\mathbf{C}}) is a clopen subset of the precompact SMZD group Bz(𝐂)B_{z}({\mathbf{C}}). Let yYy\in Y and

hy:Bz1(𝐂)Bz0(𝐂),gg+y.h_{y}:B_{z}^{1}({\mathbf{C}})\to B_{z}^{0}({\mathbf{C}}),\ g\mapsto g+y.

The mapping hyh_{y} is a homeomorphism and hy(Bz1(𝐂|X))=Bz0(𝐂|X)h_{y}(B_{z}^{1}({\mathbf{C}}|X))=B_{z}^{0}({\mathbf{C}}|X), i.e. XX is an open continuous retract of the precompact Boolean SMZD group Bz0(𝐂|X)B_{z}^{0}({\mathbf{C}}|X).

Corollary 1.

Any SMZD space is a continuous open retract of a precompact Boolean group.

4 Mal’tsev and rectifiable spaces

Let XX be a space with the operation Φ:X3X\Phi:X^{3}\to X. Consider the following conditions for the operation Φ\Phi:

  • (Φ1\Phi_{1})

    Φ(x,y,y)=Φ(y,y,x)=x\Phi(x,y,y)=\Phi(y,y,x)=x;

  • (Φ2\Phi_{2})

    Φ(x,y,Φ(y,z,u))=Φ(x,z,u)\Phi(x,y,\Phi(y,z,u))=\Phi(x,z,u);

  • (Φ3\Phi_{3})

    Φ(x,y,Φ(y,x,u))=u\Phi(x,y,\Phi(y,x,u))=u.

for x,y,z,uXx,y,z,u\in X. The Φ\Phi operation is a Mal’tsev operation if (Φ1\Phi_{1}) is satisfied.

Definition 1.

Let XX be a space and Φ:X3X\Phi:X^{3}\to X be a continuous map. We call a mapping Φ\Phi homogeneous Mal’tsev operation if Φ\Phi satisfies the conditions (Φ1\Phi_{1}) and (Φ3\Phi_{3}). We call a mapping Φ\Phi a strong Mal’tsev operation if the conditions (Φ1\Phi_{1}) and (Φ2\Phi_{2}) are satisfied for Φ\Phi. A space XX is said to be strongly rectifiable if there exists a continuous strong Mal’tsev operation on XX.

Proposition 1.

The space XX is rectifiable if and only if there exists a homogeneous Mal’tsev operation Φ\Phi on XX.

Proof.

Assume that XX is a rectifiable space. Let the operations pp and qq define the structure of a homogeneous algebra, Ψ\Psi is a rectification, Ψ(x,y)=(x,p(x,y))\Psi(x,y)=(x,p(x,y)). We put Φ(x,y,z)=q(x,p(y,z))\Phi(x,y,z)=q(x,p(y,z)). q(x,e)=xq(x,e)=x and p(x,x)=ep(x,x)=e imply Φ(x,y,y)=x\Phi(x,y,y)=x. q(x,p(x,y))=yq(x,p(x,y))=y implies Φ(y,y,x)=x\Phi(y,y,x)=x. Let’s check (Φ3\Phi_{3}). Since p(y,q(y,p(x,u)))=p(x,u)p(y,q(y,p(x,u)))=p(x,u) and q(x,p(x,u))=uq(x,p(x,u))=u, then

Φ(x,y,Φ(y,x,u))=q(x,p(y,q(y,p(x,u))))=u.\Phi(x,y,\Phi(y,x,u))=q(x,p(y,q(y,p(x,u))))=u.

Suppose that Φ\Phi satisfies the conditions (Φ1\Phi_{1}) and (Φ3\Phi_{3}). Take eXe\in X arbitrarily. Let p(x,y)=Φ(e,x,y)p(x,y)=\Phi(e,x,y) and q(x,y)=Φ(x,e,y)q(x,y)=\Phi(x,e,y). From (Φ1\Phi_{1}) follows p(x,x)=ep(x,x)=e and q(x,e)=xq(x,e)=x. From (Φ3\Phi_{3}) it follows

q(x,p(x,y))\displaystyle q(x,p(x,y)) =Φ(x,e,Φ(e,x,y))=y,\displaystyle=\Phi(x,e,\Phi(e,x,y))=y,
p(x,q(x,y))\displaystyle p(x,q(x,y)) =Φ(e,x,Φ(x,e,y))=y.\displaystyle=\Phi(e,x,\Phi(x,e,y))=y.

Proposition 2.

If XX is strongly rectifiable, then XX is rectifiable.

Proof.

The condition (Φ3\Phi_{3}) follows from (Φ1\Phi_{1}) and (Φ2\Phi_{2}), it suffices to substitute z=xz=x into (Φ3\Phi_{3}). It remains to apply the Proposition 1. ∎

Proposition 3.

Let YY be a Hausdorff space, XY=X¯X\subset Y=\overline{X}, mapping Φ^:Y3Y\widehat{\Phi}:Y^{3}\to Y is continuous, and Φ(X3)=X\Phi(X^{3})=X. Denote Φ=Φ^|X3\Phi=\left.\widehat{\Phi}\right|_{X^{3}}.

  1. 1.

    If Φ\Phi is a Mal’tsev operation, then Φ^\widehat{\Phi} is a Mal’tsev operation.

  2. 2.

    If Φ\Phi is a homogeneous Mal’tsev operation, then Φ^\widehat{\Phi} is a homogeneous Mal’tsev operation.

  3. 3.

    If Φ\Phi is a strong Mal’tsev operation, then Φ^\widehat{\Phi} is a strong Mal’tsev operation.

Proof.

If the identities (Φ1\Phi_{1}), (Φ2\Phi_{2}), and (Φ3\Phi_{3}) hold on a dense subspace, then they hold everywhere. ∎

Proposition 4.

Let XX be a space with continuous Mal’tsev operation Φ\Phi, Y1Y_{1} and Y2Y_{2} be compact Hausdorff extensions of XX, the operation Φ\Phi extends to Y1Y_{1} and Y2Y_{2} up to continuous operations Φ1\Phi_{1} and Φ2\Phi_{2}, respectively. Then Y1Y_{1} and Y2Y_{2} coincide, that is, there exists a homeomorphism f:Y1Y2f:Y_{1}\to Y_{2} for which f|X=idX\left.f\right|_{X}=\operatorname{id}_{X}.

Proof.

We identify XX with {(x,x)Y1×Y2:xX}\{(x,x)\in Y_{1}\times Y_{2}\,:\,x\in X\}. Let Y3Y_{3} be the closure of XX in Y1×Y2Y_{1}\times Y_{2}, πi\pi_{i} be the projection of Y1×Y2Y_{1}\times Y_{2} onto YiY_{i}, fi=πi|Y3f_{i}=\left.\pi_{i}\right|_{Y_{3}} for i=1.2i=1.2, Φ3=Φ1×Φ2|X3\Phi_{3}=\left.\Phi_{1}\times\Phi_{2}\right|_{X^{3}}. Then Φ3(X3)=X\Phi_{3}(X^{3})=X and, by the proposition 3, the operations Φi\Phi_{i} for i=1,2,3i=1,2,3 are a Mal’tsev operation and fif_{i} are morphisms with respect to them. Since the spaces YiY_{i} are compact, the mappings fif_{i} are factorial. Therefore, mappings fif_{i} are open (Mal’tsev’s Theorem 4.11 [9], [7]). An open continuous map that is a homeomorphism on a dense set is itself a homeomorphism. Hence the mappings fif_{i} are homeomorphisms. Therefore, the mapping f=f1f21:Y1Y2f=f_{1}\circ f_{2}^{-1}:Y_{1}\to Y_{2} is a homeomorphism. ∎

Remark 1.

Propositions 3, 4 and their proofs remain valid if we consider separately continuous Mal’tsev operations.

Definition 2.

Let XX be a space and Φ:X3X\Phi:X^{3}\to X be a continuous Mal’tsev operation. An operation Φ\Phi is called precompact if there exists a Hausdoff compact extension bXbX of the space XX onto which the mapping Φ\Phi extends to the mapping Φ^:bX3bX\widehat{\Phi}:bX^{3}\to bX.

A compact extension bXbX is uniquely defined up to isomorphism (Proposition 4), we denote it by X^\widehat{X} or (X,Φ)^\widehat{(X,\Phi)}.

  1. 1.

    A space XX is called precompact Mal’tsev if there exists a continuous precompact Mal’tsev operation on XX.

  2. 2.

    A space XX is called precompact rectifiable if there exists a continuous precompact homogeneous Mal’tsev operation on XX.

  3. 3.

    We call a space XX precompact strongly rectifiable if there exists a continuous precompact strong Mal’tsev operation on XX.

Proposition 5.

If XX is a topological group, then XX is strongly rectifiable.

If XX is a precompact topological group, then XX is precompact and strongly rectifiable.

Proof.

We put Φ(x,y,z)=xy1z\Phi(x,y,z)=xy^{-1}z. The operation Φ\Phi is the standard Mal’tsev operation on a group. Let’s check (Φ2\Phi_{2}).

Φ(x,y,Φ(y,z,u))=xy1(yz1u)=xz1u=Φ(x,z,u).\Phi(x,y,\Phi(y,z,u))=xy^{-1}(yz^{-1}u)=xz^{-1}u=\Phi(x,z,u).

If GG is precompact, then Φ\Phi extends to the completion of GG. ∎

Proposition 6.

Let XX be a Mal’tsev precompact space. Then w(X^)=w(X)=χ(X)w(\widehat{X})=w(X)=\chi(X).

Proof.

Let τ=χ(X)\tau=\chi(X). The compact Mal’tsev space X^\widehat{X} is a Dugundzhi compact (Theorem 1 [10]). The Dugunji compactum is a dyadic compactum. If a compact dyadic compact contains a dense subspace whose character does not exceed τ\tau, then its weight does not exceed τ\tau ([11]). ∎

Lemma 2.

Let Φ\Phi be a precompact homogeneous Mal’tsev operation on the space XX and eXe\in X. For any non-empty open UXU\subset X there exists a finite set MXM\subset X so that X=yMΦ(e,y,U)X=\bigcup_{y\in M}\Phi(e,y,U).

Proof.

Let WX^W\subset\widehat{X} be open and WX=UW\cap X=U. Take a non-empty open VV¯WV\subset\overline{V}\subset W. The family {Φ(e,y,V):yX^}\{\Phi(e,y,V)\,:\,y\in\widehat{X}\} forms an open cover of X^\widehat{X}. Then i=1nΦ(e,xi,V)\bigcup_{i=1}^{n}\Phi(e,x_{i},V) for some x1,x2,,xnX^x_{1},x_{2},...,x_{n}\in\widehat{X}. Then Φ(e,xi,V)Φ(e,yi,U)\Phi(e,x_{i},V)\subset\Phi(e,y_{i},U) for some yiXy_{i}\in X. Let M={y1,y2,,yn}M=\{y_{1},y_{2},...,y_{n}\}. ∎

5 Mal’tsev operations on the Cantor set 𝐂{\mathbf{C}}

Theorem 3.

Let XX be the SMZD space. Then XX is a pre-compact Mal’tsev space.

Moreover, there is a precompact Mal’tsev operation Φ\Phi on 𝐂{\mathbf{C}}, so that Φ(X3)=X\Phi(X^{3})=X for any X𝐂X\subset{\mathbf{C}}.

Proof.

Let r:Bz1(𝐂)𝐂r:B_{z}^{1}({\mathbf{C}})\to{\mathbf{C}} as in Theorem 2. Let Φ(x,y,z)=r({x,y,z})\Phi(x,y,z)=r(\{x,y,z\}). Then Φ(x,y,z){x,y,z}\Phi(x,y,z)\in\{x,y,z\} and hence Φ\Phi is a Mal’tsev operation. ∎

Theorem 4.

Let XX be a strongly homogeneous SMZD space. Then XX is a precompact strongly rectifiable space.

Moreover, there exists a precompact strongly Malcev operation Φ\Phi on XX, so that (X,Φ)^\widehat{(X,\Phi)} is homeomorphic to 𝐂{\mathbf{C}}.

Proof.

There is a compact zero-dimensional extension bXbX of the space XX homeomorphic to 𝐂{\mathbf{C}}, so that for any non-empty open-closed U,VbXU,V\subset bX, U,VYU,V\neq Y there exists fAut(Y|X)f\in\operatorname{Aut}(Y|X) so f(U)=Vf(U)=V, Lemma 2.2 [12]. We will assume that bX=𝐂bX={\mathbf{C}}.

Lemma 3.

For every non-empty open-closed U𝐂U\subset{\mathbf{C}} there exists a homeorphism fU:𝐂Uf_{U}:{\mathbf{C}}\to U, so that fU(X)=UXf_{U}(X)=U\cap X.

Proof.

Let V1V_{1} and V2V_{2} be a partition of 𝐂{\mathbf{C}} into two non-empty open-closed subsets and U1U_{1} and U2U_{2} be a partition of UU into two non-empty open-closed subsets. For i=1,2i=1,2, let fi:ViUif_{i}:V_{i}\to U_{i} be a homeorphism such that fi(ViX)=UiXf_{i}(V_{i}\cap X)=U_{i}\cap X. Then fUf_{U} is the union of f1f_{1} and f2f_{2}. ∎

Denote 𝒞=𝒞{𝐂}\mathcal{C}^{*}=\mathcal{C}\setminus\{{\mathbf{C}}\}. For U,V𝒞U,V\in\mathcal{C}^{*} we set

Θ[U,V]=fUfV1:VU.\Theta\left[U,V\right]=f_{U}\circ f_{V}^{-1}:V\to U.

It follows from the construction

  • (Θ1\Theta_{1})

    Θ[U,V]:VU\Theta\left[U,V\right]:V\to U is a homeomorphism and Θ[U,V](VX)=UX\Theta\left[U,V\right](V\cap X)=U\cap X for U,V𝒞U,V\in\mathcal{C}^{*};

  • (Θ2\Theta_{2})

    Θ[U,U]=idU\Theta\left[U,U\right]=\operatorname{id}_{U} for U𝒞U\in\mathcal{C}^{*};

  • (Θ3\Theta_{3})

    Θ[U,V]Θ[V,W]=Θ[U,W]\Theta\left[U,V\right]\circ\Theta\left[V,W\right]=\Theta\left[U,W\right] for U,V,W𝒞U,V,W\in\mathcal{C}^{*}.

Let x=(xn)nω𝐂x=(x_{n})_{n\in\omega}\in{\mathbf{C}}, kωk\in\omega. Let’s put

Uk(x)\displaystyle U_{k}(x) ={y𝐂:x|k+1=y|k+1}=B(x|k+1),\displaystyle=\{y\in{\mathbf{C}}\,:\,x|k+1=y|k+1\}=B(x|k+1),
Vk(x)\displaystyle V_{k}(x) ={y𝐂:x|k=y|k and x(k)y(k)}=Uk1(x)Uk(x)=B(x~),\displaystyle=\{y\in{\mathbf{C}}\,:\,x|k=y|k\text{ and }x(k)\neq y(k)\}=U_{k-1}(x)\setminus U_{k}(x)=B(\tilde{x}),

where x~=(x0,,xk1,1xk)\tilde{x}=(x_{0},...,x_{k-1},1-x_{k}). Note that

  • (U1U_{1})

    Uk(x),Vk(x)𝒞U_{k}(x),V_{k}(x)\in\mathcal{C}^{*};

  • (U2U_{2})

    the family {Vn(x):nk}\{V_{n}(x)\,:\,n\leq k\} is an open partition of the set 𝐂Uk(x){\mathbf{C}}\setminus U_{k}(x);

  • (U3U_{3})

    the family {Vn(x):nω}\{V_{n}(x)\,:\,{n\in\omega}\} is an open partition of the set 𝐂{x}{\mathbf{C}}\setminus\{x\};

  • (U4U_{4})

    the family {Un(x):nω}\{U_{n}(x)\,:\,{n\in\omega}\} is the base at the point xx;

  • (U5U_{5})

    Uk(x)=Uk(y)U_{k}(x)=U_{k}(y), Vk(x)=Vk(y)V_{k}(x)=V_{k}(y) and xUk(y)x\in U_{k}(y) if yUk(x)y\in U_{k}(x);

  • (U6U_{6})

    Vk(x)=Uk(y)V_{k}(x)=U_{k}(y), Vk(y)=Uk(x)V_{k}(y)=U_{k}(x) and xVk(y)x\in V_{k}(y) if yVk(x)y\in V_{k}(x).

For x,y𝐂x,y\in{\mathbf{C}} we define the mapping hx,yAut(𝐂)h_{x,y}\in\operatorname{Aut}({\mathbf{C}}). Let z𝐂z\in{\mathbf{C}}. Let’s put

hx,y(z)={x,z=yΘ[Vk(x),Vk(y)](z),if zVk(x) for some kωh_{x,y}(z)=\begin{cases}x,&z=y\\ \Theta\left[V_{k}(x),V_{k}(y)\right](z),&\text{if $z\in V_{k}(x)$ for some $k\in\omega$}\end{cases}

It follows from the construction

  • (h1h_{1})

    hx,yAut(𝐂)h_{x,y}\in\operatorname{Aut}({\mathbf{C}}) for x,y𝐂x,y\in{\mathbf{C}} and hx,yAut(𝐂|X)h_{x,y}\in\operatorname{Aut}({\mathbf{C}}|X) for x,yXx,y\in X;

  • (h2h_{2})

    hx,x=id𝐂h_{x,x}=\operatorname{id}_{\mathbf{C}};

  • (h3h_{3})

    hx,yhy,z=hx,zh_{x,y}\circ h_{y,z}=h_{x,z};

  • (h4h_{4})

    hx,y(y)=xh_{x,y}(y)=x;

  • (h5h_{5})

    hx,y(Uk(y))=Uk(x)h_{x,y}(U_{k}(y))=U_{k}(x) and hx,y(Vk(y))=Vk(x)h_{x,y}(V_{k}(y))=V_{k}(x).

  • (h6h_{6})

    if xUk(x)x^{\prime}\in U_{k}(x) and yUk(y)y^{\prime}\in U_{k}(y) then hx,y|V=hx,y|V\left.h_{x,y}\right|_{V}=\left.h_{x^{\prime},y^{\prime}}\right|_{V}, where V=Vk(y)V=V_{k}(y).

Let’s put

Φ:𝐂3𝐂,(x,y,z)hx,y(z).\Phi:{\mathbf{C}}^{3}\to{\mathbf{C}},\quad(x,y,z)\mapsto h_{x,y}(z).

The condition (Φ1\Phi_{1}) follows from (h2h_{2}) and (h4h_{4}). The condition (Φ2\Phi_{2}) follows from (h3h_{3}). Hence the operation Φ\Phi is a strong Mal’tsev operation.

Let us prove that Φ\Phi is continuous. Let x,y,z,u𝐂x,y,z,u\in{\mathbf{C}}, Φ(x,y,z)=u\Phi(x,y,z)=u, UU be a neighborhood of the point uu. We need to find a neighborhood W𝐂3W\subset{\mathbf{C}}^{3} of the point (z,y,z)(z,y,z), so that Φ(W)U\Phi(W)\subset U.

Consider the case y=zy=z. Then x=ux=u. Then S=Uk(x)US=U_{k}(x)\subset U for some kωk\in\omega. Let V=Uk(y)V=U_{k}(y) and W=S×V×VW=S\times V\times V. Let’s check Φ(W)U\Phi(W)\subset U. Let (x,y,z)W(x^{\prime},y^{\prime},z^{\prime})\in W. From (U5U_{5}) and (h5h_{5}) it follows that hx,y(V)=Sh_{x^{\prime},y^{\prime}}(V)=S. Hence Φ(x,y,z)U\Phi(x^{\prime},y^{\prime},z^{\prime})\in U.

Consider the case yzy\neq z. Then zVm(y)z\in V_{m}(y) for some mωm\in\omega. Let S=Um(x)S=U_{m}(x) and V=Um(y)V=U_{m}(y). Since the mapping hx,yh_{x,y} is continuous, then hx,y(Q)Uh_{x,y}(Q^{\prime})\subset U for some neighborhood QQ^{\prime} of the point zz. Let Q=QVm(y)Q=Q^{\prime}\cap V_{m}(y) and W=S×V×QW=S\times V\times Q. Let’s check Φ(W)U\Phi(W)\subset U. Let (x,y,z)W(x^{\prime},y^{\prime},z^{\prime})\in W. From (U6U_{6}) it follows that hx,y|Q=hx,y|Q\left.h_{x^{\prime},y^{\prime}}\right|_{Q}=\left.h_{x,y}\right|_{Q}. Hence Φ(x,y,z)U\Phi(x^{\prime},y^{\prime},z^{\prime})\in U.

The continuity of Φ\Phi is proved. From (h1h_{1}) it follows that Φ(X3)=X\Phi(X^{3})=X. ∎

Proposition 7.

Let XX be a zero-dimensional first countable space. Then Xω×𝐂X^{\omega}\times{\mathbf{C}} is strongly homogeneous.

Proof.

Let Y=X×𝐃Y=X\times{\mathbf{D}}. Then the spaces Xω×𝐂X^{\omega}\times{\mathbf{C}}, YωY^{\omega} and Yω×𝐃Y^{\omega}\times{\mathbf{D}} are homeomorphic. Therefore, YωY^{\omega} has a proper open-closed subspace homeomorphic to YωY^{\omega}. Proposition 24 (5) [13] implies that YωY^{\omega} is strongly homogeneous. ∎

Proposition 8.

Let XX be the SMZD space. There is a space SMZD space YY such that X×YX\times Y is strongly homogeneous and strongly rectifiable.

Proof.

Let Y=Xω×𝐂Y=X^{\omega}\times{\mathbf{C}}. Then, by the Proposition 7, YY is strongly homogeneous. The space X×YX\times Y is homeomorphic to YY. It remains to apply the Theorem 4. ∎

6 Homogeneous spaces that are not right topological groups

Van Mill in [14] constructed a homogeneous space AA, a subspace of the real numbers, which is not a topological group. In fact, the space AA is not a right topological group either.

We call a set AXA\subset X be a clumsy if h(A)Ah(A)\cap A is non empty for any homeomorphism h:XXh:X\to X.

Theorem 5.

Let XX be an space and AXA\subset X be a clumsy subset. If |A|<|X||A|<|X|, then XX is not a right topological group.

Proof.

From the contrary. Take gXA1Ag\in X\setminus A^{-1}A. Then AgA=Ag\cap A=\varnothing. Contradiction. ∎

Theorem 5 generalizes Lemma 3.4. from [14] and the proof of the theorem is much simpler.

There is a countable dense set EAE\subset A such that EE is clumsy in AA (Theorem 3.3 [14]). Therefore, by the Theorem 5, the space AA from [14] is not a right topological group.

In [15] Taras Banakh noticed that the zero-dimensional homogeneous first-countable van Douwen compactum [16] contains a countable clumsy set and, therefore, is not a right topological group.

Let 𝐓=𝐏×𝐏𝐐×𝐐{\mathbf{T}}={\mathbf{P}}\times{\mathbf{P}}\cup{\mathbf{Q}}\times{\mathbf{Q}} is van Douwen space and 𝐒=𝐂×𝐓{\mathbf{S}}={\mathbf{C}}\times{\mathbf{T}} is van Mill space. Spaces 𝐓{\mathbf{T}} and 𝐒{\mathbf{S}} are SMZD strongly homogeneous spaces [17, 18].

Statement 1.

𝐐×𝐐{\mathbf{Q}}\times{\mathbf{Q}} is clumsy in 𝐓{\mathbf{T}}.

Proof.

Assume the contrary, that is, h(𝐐×𝐐)𝐐×𝐐=h({\mathbf{Q}}\times{\mathbf{Q}})\cap{\mathbf{Q}}\times{\mathbf{Q}}=\varnothing for some homeomorphism h:𝐓𝐓h:{\mathbf{T}}\to{\mathbf{T}}. Let F=𝐐×{0}F={\mathbf{Q}}\times\{0\}. Then FF is closed in 𝐓{\mathbf{T}} and homeomorphic to 𝐐{\mathbf{Q}}. Hence h(F)h(F) is closed in 𝐏×𝐏{\mathbf{P}}\times{\mathbf{P}} and homeomorphic to 𝐐{\mathbf{Q}}. Contradiction. ∎

The Theorem 5 and Statement 1 imply

Proposition 9.

𝐓{\mathbf{T}} is not a right topological group.

It is not clear about the van Mill space 𝐒{\mathbf{S}} whether the structure a right topological group is allowed on it.

Proposition 10 (Corollary 5.4 [19]).

If a topological group GG contains a dense Čech complete subspace, then GG is Čech complete.

Recall that Polish spaces are exactly Čech complete separable metrizable spaces.

The space 𝐒{\mathbf{S}} contains a dense subspace homeomorphic to 𝐏{\mathbf{P}} and is not Polish. Then 𝐒ω{\mathbf{S}}^{\omega} also contains a dense subspace homeomorphic to 𝐏{\mathbf{P}} and is not Polish. The following sentence follows from the Proposition 10.

Proposition 11.

𝐒ω{\mathbf{S}}^{\omega} is not a topological group.

7 Examples and questions

Question 1.

(A.V. Arhangelskii) Let GG be a right topological group.

Is it true that if GG contains a dense Čech complete subspace, then GG is Čech complete?

Is it true that if GG is a separable metrizable and contains a dense Polish subspace, then GG is Polish?

Example 1.

The van Mill space 𝐒{\mathbf{S}} and the van Dawn space 𝐓{\mathbf{T}} have the following properties:

  • 1.

    are strongly homogeneous [18];

  • 2.

    absolutely Borel [18, 17];

  • 3.

    coset sets of some absolutely Borel groups [12];

  • 4.

    have a dense Polish subspace and are not Polish [18];

  • 5.

    precompact strongly rectifiable (Theorem 4);

  • 6.

    𝐓{\mathbf{T}} is not a right topological group (Proposition 9);

  • 7.

    𝐒{\mathbf{S}} is not a topological group (Proposition 10);

Question 2.

Is the van Mill space 𝐒{\mathbf{S}} a right topological group?

Question 3.

Let X{𝐒,𝐓}X\in\{{\mathbf{S}},{\mathbf{T}}\}.

  1. 1.

    Is the space X×𝐐X\times{\mathbf{Q}} homeomorphic to a (boolean) (precompact) (right) topological group?

  2. 2.

    Is the space X×𝐏X\times{\mathbf{P}} homeomorphic to a (boolean) (precompact) (right) topological group?

  3. 3.

    Is there a SMZD space YY such that X×YX\times Y is homeomorphic to a (boolean) (precompact) (right) topological group?

A topological group GG is called locally precompact if there exists a neighborhood UU of the neutral element ee of GG such that UU can be covered by finitely many left and right translates of each neighborhood of ee in GG. A group is locally precompact iff its completion is locally compact.

Proposition 12 (Theorem 8 [20]).

Let GG be a SMZD topological group which is not locally precompact. Then the space GG is strongly homogeneous.

There exist a dense subgroup GMG_{M} of the real line [21, Theorem 5.1] and dense subgroup GDG_{D} of the circle [16], so G{GM,GD}G\in\{G_{M},G_{D}\} satisfies the condition:

  • if A,BG^A,B\subset\widehat{G} are open sets and AGA\cap G is homeomorphic to BGB\cap G, then μ(A)=μ(B)\mu(A)=\mu(B),

where G^\widehat{G} is the completion of GG (for GMG_{M} the completion is the real line, and for GDG_{D} the completion is the circle) and μ\mu is the standard invariant Lebesgue measure on G^\widehat{G}. The group GDG_{D} is precompact, the group GMG_{M} is locally precompact but not precompact.

Example 2.

SMZD the groups GDG_{D} and GMG_{M} are not strongly homogeneous, it suffices to take two segments A,BA,B of different lengths, the ends of which do not lie in the group [16]. The group GMG_{M} is not homeomorphic to a precompact rectifiable space, a finite number of shifts of the set (1,1)GM(-1,1)\cap G_{M} cannot cover GMG_{M} (see Lemma 2).

Proposition 13 (Corollary 5 [20]).

If XX is a homogeneous SMZD space, then X×𝐐X\times{\mathbf{Q}} is a strongly homogeneous space.

Question 4.

Let G{GM,GD}G\in\{G_{M},G_{D}\}. Which of the spaces listed below are strongly homogeneous: G×𝐏G\times{\mathbf{P}}, G×𝐂G\times{\mathbf{C}}, GωG^{\omega}?

Question 5.

Which of the following spaces are homeomorphic to (1) precompact group; (2) precompact (strongly) rectifiable space: GM×𝐐G_{M}\times{\mathbf{Q}}, GM×𝐏G_{M}\times{\mathbf{P}}, GM×𝐂G_{M}\times{\mathbf{C}}, GMωG_{M}^{\omega}?

Question 6.

Let G{GM,GD}G\in\{G_{M},G_{D}\}. Which of the spaces listed below are homeomorphic to a Boolean group: GG, G×𝐐G\times{\mathbf{Q}}, G×PG\times P, G×𝐂G\times{\mathbf{C}}, GωG^{\omega}?

Proposition 14 ([22, 23]).

Let XX be a SMZD space. Then XωX^{\omega} is homogeneous.

Question 7 (Terada, [24]).

Let XX be a SMZD space. Is it true that XωX^{\omega} is a strongly homogeneous space?

Question 8.

Let XX be the SMZD space.

  1. 1.

    Which of the following spaces are (1) (strongly) (precompactly) rectifiable spaces; (2) is homeomorphic to a right topological group: XωX^{\omega}, Xω×𝐐X^{\omega}\times{\mathbf{Q}}, Xω×𝐏X^{\omega}\times{\mathbf{P}}, Xω×𝐂X^{\omega}\times{\mathbf{C}}?

  2. 2.

    Is the space Xω×𝐐X^{\omega}\times{\mathbf{Q}} homeomorphic to a topological (boolean, precompact) group?

  3. 3.

    Is there a SMZD space YY such that X×YX\times Y is homeomorphic to a (right) topological group?

Question 9.

Let XX be a homogeneous SMZD space.

  1. 1.

    Will the space XX be a (strongly) (precompact) rectifiable space?

  2. 2.

    Is the space X×𝐐X\times{\mathbf{Q}} homeomorphic to a (precompact) (right) topological group?

Question 10.

In the class SMZD spaces, distinguish the following classes: rectifiable; strongly straightened; pre-compact straightening; strongly precompactly rectifiable.

References