On the spectra of rings of semialgebraic functions (Q1956304)

Throughout this review `maps' and `functions' are supposed to be continuous. Given a semialgebraic set \(M\subseteq\mathbb{R}^m\) (endowed with euclidean topology) the family of semialgebraic real valued functions defines an \(\mathbb{R}\)-algebra \(\mathcal{S}(M)\) containing the family of bounded functions \(\mathcal{S}^*(M)\) as a subalgebra. The authors introduce the notations \(\text{Spec}_S(M)=\text{Spec}(\mathcal{S}(M))\) and \(\beta_SM=\text{Spec}_{\text{max}}(\mathcal{S}(M))\) for the prime and maximal spectra of \(\mathcal{S}(M),\) and similarly \(\text{Spec}_S^*(M), \beta_S^*M\) in the case of \(\mathcal{S}^*(M).\) They write \(\text{Spec}_S^\diamond (M), \beta_S^\diamond M\) in statements that apply to \(\mathcal{S}(M)\) as well as \(\mathcal{S}^*(M).\) \quad Note that a basic open set for the Zariski topology of \(\text{Spec}_S(M)\) is of the form \(\text{Spec}_S(M)\setminus \mathcal{Z}(f),\) where for \(f\in \mathcal{S}(M)\) define \(\mathcal{Z}(f)=\{\mathfrak{p}\in \text{Spec}_S(M): f\in \mathfrak{p} \}.\) The article's goal is to prove the most important algebraic and topological properties of the Zariski and maximal spectra of the rings \(\mathcal{S}^\diamond(M)\) and the functorial properties of maps from \(\mathcal{S}^\diamond(M)\) to \(\mathcal{S}^\diamond(N)\) and from \(\text{Spec}_S^\diamond (M)\) to \(\text{Spec}_S^\diamond (N)\), naturally induced by semialgebraic maps \(N@>\varphi>> M\) (often the inclusion map \(N\hookrightarrow M\)). The authors say that their initial results could surely be obtained by Niels Schwartz's theory of real closed spaces [\textit{N. Schwartz}, Mem. Am. Math. Soc. 397, 122 p. (1989; Zbl 0697.14015)] but their approach, valid only for \(\mathbb{R}\) and not over arbitrary real closed fields, permits to use the theory of rings of continuous functions expounded in [\textit{L. Gillman} and \textit{M. Jerison}, Rings of continuous functions. The University Series in Higher Mathematics. Princeton-Toronto-London-New York: D. Van Nostrand Company, Inc. (1960; Zbl 0093.30001)] and admits a less involved presentation. Sections 2 and 3 recall and survey in a readable manner basic facts on semialgebraic sets and functions and about spectra of rings of such functions; but for six out of seven cited papers by Fernando and Gamboa, the reader is relegated to Fernando's web page. This hopefully justifies this long review of what might be called a research survey guide to papers by the authors. The main results are below cited as Theorem 4.8 and Theorem 5.1. In Section 2 properties of the closure \(\text{Cl}_{\mathbb{R}^n}(M)\) and the operators \(\rho_0,\rho_1\) defined by \(\rho_0(M)= \text{Cl}_{\mathbb{R}^n}(M)\setminus M\) and \(\rho_1=\rho_0\circ \rho_1\) are studied and a semialgebraic Tietze-Urysohn Lemma (see [\textit{J. Bochnak} et al., Real algebraic geometry. Transl. from the French. Rev. and updated ed. Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge. 36. Berlin: Springer. (1998; Zbl 0912.14023); \textit{H. Delfs} and \textit{M. Knebusch}, Pac. J. Math. 114, 47--71 (1984; Zbl 0548.14008)] ) is given: For example, if \(N\subset M \subset \mathbb{R}^n\) are semialgebraic, then the restriction \(\mathcal{S}(M) \ni f \mapsto f|_N \in \mathcal{S}(N)\) defines a homomorphism \(\phi\) which is surjective iff \(N\) is closed. Conditions for surjectivity in the case where \(\mathcal{S}^*(.)\) substitutes \(\mathcal{S}(.)\) are also given. An ideal \(\mathfrak{a}\) in \(\mathcal{S}(M)\) is a \(z\)-ideal if \(f\in \mathfrak{a}\) and \(Z_M(f)\subset Z_M(g)\) implies \(g\in \mathfrak{a}\). In earlier papers [\textit{J. F. Fernando} and \textit{J. M. Gamboa}, ``On Łojasiewicz's inequality and the Nullstellensatz for rings of semialgebraic functions'', \url{http://http://eprints.ucm.es/24697/}] and [``On the Krull dimension of rings of semialgebraic functions'', \url{arxiv:1306.4109}] the authors have shown a Nullstellenatz: If M is locally compact then a nontrivial ideal in \(\mathcal{S}(M)\) is \(z\)-ideal iff it is radical; concerning the semialgebraic depth \(d_M(\mathfrak{p})=\min\{\dim Z_M(f):f\in \mathfrak{p}\}\) they got for prime \(z\)-ideals \(\mathfrak{p}\supset \mathfrak{q},\) that \(d_M(\mathfrak{p})<d_M(\mathfrak{q})\). In Section 3 the usual relation between what in [Zbl 0912.14023] is called prime cone of an algebra and the orderings of associated quotient fields is recalled in the context of the algebra \(\mathcal{S}^\diamond(M).\) Next the following observations are made: a radical ideal \(\mathfrak{a}\) of the ring \(\mathcal{S}^\diamond(M)\) enjoys a convexity property: \(0\leq f \leq g \in \mathfrak{a}\) implies \(f\in \mathfrak{a};\) the set of prime ideals in \(\mathcal{S}^\diamond (M)\) containing a fixed prime form a chain with respect to inclusion; the quotient \(\mathcal{S}^\diamond(M)/\mathfrak{a},\) \(\mathfrak{a}\) radical is an \(f\)-ring; the usual spectral topology of the real spectrum of the algebra \(\mathcal{S}^\diamond(M)\) coincides with the Zariski topology. Furthermore the map \(M\ni p \mapsto \mathfrak{m}_p=\{f\in \mathcal{S}(M):f(p)=0 \}\in \text{Spec}_S(M)\) embeds \(M\) into \(\text{Spec}_S(M)\) as a subspace; in fact \(\beta_S^\diamond M\) is a Hausdorff compactification of \(M\). The main difference between \(\mathcal{S}(M)\) and \(\mathcal{S}^*(M)\) is that while a function \(f\in \mathcal{S}^*(M)\) with \(Z_M(f)=\emptyset\) is of course a unit in \(\mathcal{S}(M),\) it needs not be invertible in \(\mathcal{S}^*(M).\) Seeing why one finds with \(\mathcal{W}(M)=\{f\in \mathcal{S}^*(M): Z_M(f)=\emptyset\},\) that \(\mathcal{S}^*(M)\) is the localization of \(\mathcal{S}(M)\) at \(\mathcal{W}(M):\) \(\mathcal{S}(M)=\mathcal{S}^*(M)_{\mathcal{W}(M)}.\) This provides a correspondence between the primes (points) of the spectrum of \(\mathcal{S}(M)\) and those in the complement of \(\mathcal{W}(M)\) in \(\text{Spec}_S^*(M).\) The last part of Section 3 is dedicated to properties of \(\beta_S M.\) It is mentioned that the map \(\beta _S M \ni \mathfrak{m}\mapsto \mathfrak{m}^*:=\)unique maximal ideal containing \(\mathfrak{m}\cap \mathcal{S}^*(M)\) \quad provides a homeomorphism \(\beta_S M \cong \beta_S^* M. \) Maximal ideals defined by semialgebraic paths are also considered. Section 4 has title `Functoriality of \(\text{Spec}_S\)'. If \(N\subset \mathbb{R}^n\) and \(M\subset \mathbb{R}^m\) are semialgebraic and \(\varphi: N\rightarrow M\) a semialgebraic map, we have a homomorphism \(\phi:\mathcal{S}^\diamond(M)\ni f \mapsto f\circ \varphi \in \mathcal{S}^\diamond(N)\) which in turn defines the map \(\text{Spec}^\diamond(\varphi):\text{Spec}_S^\diamond(N)\ni \mathfrak{q} \mapsto \phi^{-1}(\mathfrak{q}) \in \text{Spec}_S^\diamond(M).\) This map is the unique continuous map that extends \(\varphi;\) it maps prime \(z\)-ideals in \(\mathcal{S}(N)\) to \(z\)-ideals and, given \(\psi:M\rightarrow P\subset \mathbb{R}^p,\) there holds \(\text{Spec}_S^\diamond (\psi)\circ \text{Spec}_S^\diamond(\phi)= \text{Spec}_S^\diamond(\psi\circ \phi).\) Next natural questions concerning the closure operator in \(\text{Spec}_S^\diamond (M),\) are answered in corollaries 4.4, 4.5, 4.6. For example: given \(N\subset M,\) when is a primeideal in \(\text{Cl}_{\text{Spec}_S(M)}(N)\)?; what is \(\text{Cl}_{\text{Spec}_S(M))}(C_1\cap C_2) \)?; when do we have a homeomorphism \(\text{Cl}_{\text{Spec}_S^\diamond(M)}(N)\cong \text{Spec}_S^\diamond (N),\)?; etc. Then one of the main result of this paper is stated although its proof is deferred to still later when more lemmas are available - local compactness assumptions are made to ensure validity of the Lojasiewicz inequality. Theorem 4.8: Let \(N\subset M \subset \mathbb{R}^n\) be semialgebraic sets such that \(N\) is open in \(M\) and locally compact and let \(Y=M\setminus N.\) Let \(\mathcal{L}(Y)\) be the set of all prime ideals that contain some \(f\in \mathcal{S}(M)\) whose zero set in \(M\) is \(Y.\) Let \(j:N\hookrightarrow M\) be the inclusion map. Then \(\text{Spec}_S(j)\) yields a homeomorphism \(\text{Spec}_S(N) \cong \text{Spec}_S(M)\setminus \mathcal{L}(Y).\) Furthermore the preimage of a maximal ideal of \(\mathcal{S}(M)\) under this homeomorphism is a maximal ideal of \(\mathcal{S}(N)\) while for the image of maximal ideals under \(\text{Spec}_S(j)\) needs not be maximal. In the remainder of Section 4, lemmas 4.11, 4.13 and 4.15 characterize prime ideals in the spectral envelope \(\mathcal{L}(Y),\) explains the behaviour of the semialgebraic depth under extension and contraction of ideals, and shows that a \(\mathfrak{p}\in \mathcal{S}(M)\) so that every \(f\in \mathfrak{p}\) has a zero in the largest locally compact dense of \(M\) is a \(z\)-ideal. Sections 5 and 6 deal with the functoriality of \(\text{Spec}_S^*(.)\) and \(\beta_S^\diamond(.),\) respectively. Many results are formulated for what the authors call a `suitably arranged tuple' \((M,N,Y,j,i)\) determined by a pair \(N\subset M \subset R^n\) of semialgebraic sets where \(N\) is locally compact and dense in \(M,\) \(Y=M\setminus N,\) and \(j:N\hookrightarrow M\) and \(i:Y=M\setminus N\hookrightarrow M\) are inclusion maps. \quad It is shown in lemmas 5.5, 5.6, 5.7 e.g. that if \(\mathfrak{p}\in \text{Spec}_S^*(M)\) is so that \(\mathfrak{p}\mathcal{S}^*(N)=\mathcal{S}^*(N),\) then \(\mathfrak{p}\in \text{Cl}_{\text{Spec}_S^*(M)}(Y)\); that a radical ideal in \(\mathcal{S}^*(M)\) containing a prime is itself radical; that if \(N\subset M\) is dense in \(M\) and \(\mathfrak{q}\) minimal in \(\mathcal{S}^*(N),\) then \(\mathfrak{q}\cap \mathcal{S}^*(M)\) is minimal in \(\mathcal{S}^*(M).\) \quad In dramatic difference to the case of the \(\text{Spec}_S(j)\) map (see T4.1), given a semialgebraic map \(\varphi: N\rightarrow M,\) Lemma 5.9 says that the map \(\text{Spec}_S^*(\phi)\) maps (maximal) ideals in \(\beta_s^*N\) into \(\beta_s^*M.\) Given a suitably arranged tuple \((M,N,Y,j,i)\) and a chain of prime ideals \(\mathfrak{p}_0 \subset \cdots \subset \mathfrak{p}_r\) in \(\mathcal{S}^*(M),\) by Corollary 5.11 there exists a chain \(\mathfrak{q}_0 \subset \cdots \subset \mathfrak{q}_r\) in \(\mathcal{S}^*(N),\) so that \(\mathfrak{q}_k\cap \mathcal{S}^*(M)=\mathfrak{p}_k,\) for \(k=1,\dots,r;\) furthermore the map \(\text{Spec}_S^*(j)\) is surjective. Suppose now we have a semialgebraic situation \(\mathbb{R}^n \supset N @>\varphi>> M\subset \mathbb{R}^m\) and there is a \(Y\subset M\) so that \(M_1=M\setminus Y\) is locally compact and dense in \(M\) and the map \(\psi=\varphi_{|N_1}:N_1=N\setminus \varphi^{-1}(Y) \rightarrow M_1\) is a semialgebraic homeomorphism. Let \(Z=\text{Cl}_{\text{Spec}_S^*(M)}(Y).\) Then the second main result of the paper, Theorem 5.1, says that the map \(\text{Spec}_S^*(\varphi):\text{Spec}_S^*(N)\rightarrow \text{Spec}_S^*(M)\) is surjective and its restriction to \(\text{Spec}_S^*(N)\setminus \text{Spec}_S^*(\varphi)^{-1}(Z) \) is a homeomorphism onto \(\text{Spec}_S^*(M)\setminus Z.\) As Corollary 5.13 to lemmas leading up to the just cited main result one has e.g. a going up lemma for prime ideals for suitably arranged tuples \((M,N,Y,i,j)\): assume \(\mathfrak{p}_0 \subset \mathfrak{p}_1\) are prime ideals of \(\mathcal{S}^*(M)\) and \(\mathfrak{q}_0\) is a prime ideal of \(\mathcal{S}^*(N)\) so that \(\mathfrak{q}_0 \cap \mathcal{S}^*(M)=\mathfrak{p}_0.\) Then there exists a prime ideal \(\mathfrak{q}_1 \supset \mathfrak{q}_0 \) of \(\mathcal{S}^*(N)\) so that \(\mathfrak{q}_1 \cap \mathcal{S}^*(M)=\mathfrak{p}_1.\) \quad In the last part of Section 5, an analysis of the spectrum of \(\mathcal{S}^*(M)\) in case that \(M\) is not necessarily locally compact is given via stratifications of \(M\) into locally compact pieces constructed from the \(\rho_1\) operator defined above. It is also shown in Proposition 5.19 that given any semialgebraic set there exist semialgebraic sets \(A_1,\dots,A_r\) whose union is \(M\) and which are Nash diffeomorphic respectively to euclidean spaces \(\mathbb{R}^{d_i}\) with \(0\leq d_i \leq \dim M,\) \(i=1,\dots,r\) and so that \(\text{Spec}_S^*(M)\) is the disjoint union of open subsets that are homeomorphic to certain open subsets of \(\text{Spec}_S^*(\mathbb{R}^{d_i})\) constructed via the mentioned diffeomorphisms. In Section 6 results of sections 4 and 5 are transferred to results for \(\beta_S^*M.\) This is done via the known fact that the map \(s_M:\text{Spec}_S(M) \rightarrow \beta_S(M)\) which maps each prime ideal to the unique maximal ideal of \(\mathcal{S}(M)\) containing it is a continuous map. The map \(\beta_S \varphi:= s_M\circ \text{Spec}_S(\varphi)|_{\beta_SN}: \beta_SN \rightarrow \beta_S M\) has properties that can be obtained as translations of properties of the map \(\beta_S^*\varphi=\text{Spec}_S^*(\varphi)|_{\beta_S^*N}. \) As an example, from connectedness results in Section 4, in Corollary 6.6 it is shown that if \(M_1,\dots,M_k\) are the connected components of semialgebraic \(M\subset \mathbb{R}^n,\) then their closures \(\text{Cl}_{\beta_S^*M}(M_i)\cong \beta_S^*M_i\) are connected components of \(\beta_S^*M.\) Finally, an easy corollary to the main result of this section follows after obvious notational changes (e.g. change \(\text{Spec}_{S^*}(M)\) to \(\beta_S^* M\) etc.) from Theorem 5.1.