On the remainder of the semialgebraic Stone-Čech compactification of a semialgebraic set (Q2401905)

Throughout let \(M,N\) be semialgebraic sets and use the notations \(\mathcal{S}(M), \mathcal{S}^*(M), \mathcal{S}^\diamond (M)\), \( \mathcal{S}(M,N)\), \( \text{Spec}_s^\diamond (M), \) etc. for the algebras of continuous semialgebraic functions and maps, and associated spectra in the same sense as in other reviews of papers by the authors [Zbl 1291.14085, Zbl 1330.14095]. A semialgebraic compactification of \(M\) is a pair \((X,j)\) composed from a compact semialgebraic subset \(X\) in an Euclidean space and a semialgebraic embedding \(M\mathop{\to}\limits^{j} X\) such that the image of \(M\) under \(j\), i.e. \(j(M)\), is dense in \(X\). The paper continues the authors' article [Trans. Am. Math. Soc. 364, No. 7, 3479--3511 (2012; Zbl 1281.14046)] where it was proved that \(\beta^*_s M\), defined as the family of maximal ideals in the Zariski spectrum \(\text{Spec}_s^* (M)\) of prime ideals of \(\mathcal{S}(M)\) is a Hausdorff compactification of \(M\), called semialgebraic Stone-Čech compactification of \(M\). This name is chosen to distinguish it from its classical counterpart for it has some analog properties. In the article it was shown e.g. that \(\beta_s^* (M), \) is the inverse limit of all semialgebraic compactifications of \(M\) which, as in the classical situation, define a directed system with respect to domination. However, note that \(\beta_s^* (M)\) and the homeomorphic \(\beta_s (M)\) are usually \textit{not} semialgebraic sets. The paper examines in particular the remainder \(\partial M= \beta_s^* M\setminus M\). It strives to distinguish the points of \(\beta_s^* M\) from those of \(M\) -- this being a problem also in classical Stone-Čech compactifications. The paper characterizes the points that admit a metrizable neighbourhood in \(\beta_s^*M\) in terms of the largest locally compact dense subset \(M_{\text{lc}}\) in \(M\) as constructed by \textit{H. Delfs} and \textit{M. Knebusch} [Locally semialgebraic spaces. Berlin: Springer (1985; Zbl 0582.14006)]: this set of points in \(\beta_s^*M\) equals \(M_{\text{lc}} \cup (\text{Cl}_{ \beta_s^*M }(\overline{M}_{\leq 1})\setminus \overline{M}_{\leq 1})\). See section 2 below for notation. In sections 1 and 2 the authors make a valid effort to ease access to their highly specialized article which uses results from no less than six earlier articles by the authors (and cites eight) and five by another great name in this part of real algebraic geometry, Niels Schwartz. Section 1 presents Schwartz's argument establishing a bijection \(\mathcal{S}(M,N) \leftrightarrow \text{Hom}_{\mathbb{R}}(\mathcal{S}(N),\mathcal{S}(M))\) from which the fact that \(M\) and \(N\) are semialgebraically homoemorphic if and only if \(\mathcal{S}(M)\) and \(\mathcal{S}(N)\) are \(\mathbb{R}\)-algebra isomorphisms is deducible. One can also deduce that \(\mathcal{S}^*(M)\) and \(\mathcal{S}^*(N)\) are isomorphic if and only if \(M\setminus \eta(M)\) and \(N\setminus \eta(N)\) are semialgebraically homeomorphic. Here the set of endpoints of a topological space \(X\) is denoted \(\eta(X)\) and a point \(p\) is called an \textit{endpoint} of \(X\) if it has an open neighbourhood \(U\) which is semialgebraically homeomorph with \([0,1[\) with \(p\) corresponding to 0. A point of \(a\in M\) defines as in classical algebraic geometry the maximal ideal \(\mathfrak{m}_a^\diamond\), of all \(f\in \mathcal{S}^\diamond (M)\) vanishing on \(a\) and via \(M\ni a\mapsto \mathfrak{m}_a^\diamond \in \text{Spec}_s^\diamond(M)\) we have an embedding of \(M\) into \(\text{Spec}_s^\diamond(M)\) as a dense subpace. Since homeomorphisms between Zariski spectra induced by algebra homomorphisms are quite special, the authors mention some interesting facts for general homeomorphisms (see the cited reviews) and indicate instances where \(\beta_s^*M\) and \(\partial M\) have been useful for solving problems concerning \(M\) and the rings \(\mathcal{S}^\diamond (M)\). Section 2 defines the local dimension \(\dim(M_x)\) of \(M\) at a point \(x\in \text{Cl}(M)\) (the euclidean closure of \(M\)) in the expected way. Letting \(M_{\leq 1}:=\{x\in M: \dim(M_x)\leq 1\}\) and allowing other analogous notations, among a number of facts from their earlier papers or other articles the following ones are mentioned: \(\cdot\) \(M_{\text{lc}}=\text{Cl}(M)\setminus \text{Cl}(\text{Cl}(M)\setminus M);\) \(\cdot\) \(M_{\geq 2}\cap \overline{M}_{\leq 1}\) is a finite set; \(\cdot\) If \(M_{\geq 2}\) is compact, then \(M=M_{lc}\). Relations between maps from \(M\) to \(N\); from \(\text{Spec}_s^\diamond(N) \) to \( \text{Spec}_s^\diamond(M)\) and from \( \beta_s^*(N) \) to \( \beta_s^*(M) \) are also given. There exists a well-defined map \(\beta_sM \rightarrow \beta_s^*M\) via \( \mathfrak{m} \mapsto \mathfrak{m}^*:=\) the unique maximal ideal containing the prime ideal \(\mathfrak{m}\cap \mathcal{S}^*(M)\). See the cited earlier reviews. An important rôle will be later played by the maximal ideals of \(\mathcal{S}^\diamond (M)\) which are \textit{not} of the form \(\mathfrak{m}_a^\diamond;\) these are called \textit{free} maximal ideals. The presentation of essential new material begins in section 3: See the review [Zbl 1330.14095] for the relations between -- and definitions of -- \(M,N,M_F, N_F, i_{M,F}, f_F, \psi_p\), etc. The constructions there mentioned can be applied in particular to the real closed field \(F_1:=\mathbb{R}((t^*))\) of real Puiseux series and \(F_0= \mathbb{R}((t^*))_{\text{alg}})\) of real Puiseux series that are algebraic over \(\mathbb{R}(t)\) in place of \(F\). A \textit{formal path} is an \(m\)-tuple \(\alpha=(\alpha_1,\dots,\alpha_m)\in \mathbb{R}[[t]]^m\) and those formal paths that applied to a right neighbourhood of \(0\in [0,1]\) are semialgebraic maps are precisely those for which \(\alpha\in \mathbb{R}[[t]]_{\text{alg}}^m\). It is next shown that \(\psi_\alpha(\mathcal{S}^*(M))\subseteq \mathbb{R}[[t^*]]^m\). Hence the evaluation map \(\mathbb{R}[[t^*]]^m\ni \alpha\mathop{\mapsto}\limits^{ev_0} \alpha(0)\in\mathbb{R}\) can be applied, resulting in a map \(\varphi_\alpha=ev_0\circ \psi_\alpha |_{\mathcal{S}^*(M)}:\mathcal{S}^*(M)\rightarrow \mathbb{R}\) with kernel \(\mathfrak{m}_\alpha^*:=\text{ker}(\varphi_\alpha)\). In case \(\alpha(0)\not \in M\), this is a free maximal ideal associated with a formal path. One defines \(\hat{\partial} M:=\{\text{free maximal ideals associated with formal paths }\}\) and finds in this case with \(\mathfrak{m}_\alpha:= \text{ker} (\psi_\alpha)\) that \(\mathfrak{m}_\alpha\cap \mathcal{S}^*(M)\subset \mathfrak{m}_\alpha^*. \) Among other results in this section it is shown that \(\eta(\beta_s^*M)= \eta(M)\cup ( \text{Cl}_{\beta_s^*M}(\overline{M}_{\leq 1})\setminus \overline{M}_{\leq 1})\) is a finite set. For section 4 let now \(\tilde{\partial} M:=\{\)free maximal ideals associated with semialgebraic paths\(\}\). Then clearly \(\tilde{\partial} M \subset \hat{\partial} M \subset \partial M. \) It is shown in theorem 4.3 that \(\tilde{\partial} M\) is dense in \(\partial M\) and that under the restriction that \(M=M_{\geq 2}\) is not compact, then \(\partial M \setminus \hat{\partial} M\) is dense in \(\partial M\) and \( \hat{\partial} M\setminus \tilde{\partial} M\) is dense in \(\hat{\partial} M\). The proof of this makes necessary a number of lemmas. Among these one finds results for the operators \(\tilde \partial\) and \(\hat \partial\): for example \(\hat \partial Y=\hat \partial M \cap \partial Y\) whenever \(Y\) is a closed semialgebraic subset of \(M\); furthermore \(\hat \partial M=\hat\partial M_{\geq 2}\sqcup \hat\partial \overline{M}_{\leq 1}\) (presumably meaning the topological sum). The next subsection is dedicated to points with countable basis of neighbourhoods. Proposition 4.8 shows that if \(\{U_k\}_k\) is a countable basis of neighbourhoods of a point \(p\in M\), then \(\{\text{Cl}_{\beta_s^* M}(U_k)\}_k\) is a countable basis of neighbourhoods of \(p\) in \(\beta_s^* M\). This is followed by theorem 4.9 according to which each point of \(\hat \partial M\) has a countable basis of neighbourhoods in \(\beta_s^* M\). Corollary 4.11 gives a connection to algebra: the set of maximal ideals of \(\mathcal{S}(M)\) which are the Jacobson radical of a principal ideal of \(S^*(M)\) equals \(M\cup \eta(\beta_s^*M)\). In theorem 4.12 the fact announced in the beginning paragraphs is proved. In its proof enters that in their earlier paper [loc. cit.] the authors have proved that the semialgebraic sets whose maximal spectrum is a metrizable space are precisely those whose maximal spectrum is homeomorphic to a semialgebraic set. The article concludes with an appendix showing that the behaviour of a non-semialgebraic homeomorphism between semialgebraic sets can be unpredictable with respect to its extension to the semialgebraic Stone-Čech compactification. The book by Bochnak, Coste and Roy [\textit{J. Bochnak} et al., Real algebraic geometry. Transl. from the French. Rev. and updated ed. Berlin: Springer (1998; Zbl 0912.14023)] will be valuable for those who venture to detail or understand the proofs, but at the same time it will show how far -- perhaps overspecialized? -- this part or real algebraic geometry has come since then.