Filters and semigroup properties (Q914852)

Let (S,\(\phi\)) be the Stone-Čech compactification of a discrete semigroup S; we endow \(\beta\) S with the (unique) multiplication such that \(\phi: S\to \beta S\) becomes a homomorphism and such that the translations \(\beta\) \(S\to \beta S\), \(x\mapsto xy\), and \(\beta\) \(S\to \beta S\), \(x\mapsto \phi (s)x\), are continuous for all \(y\in \beta S\) and \(s\in S\). In the present paper the author investigates closed subsemigroups and closed ideals of \(\beta\) S which are constructed as intersections D(\({\mathcal A})=^{def}\cap_{A\in {\mathcal A}}\overline{\phi (A)}\), where \({\mathcal A}\) is a filter of subsets of S. Her main result asserts that \(S^*=\beta S\setminus \phi (S)\), the set of free ultrafilters on S, contains a closed subsemigroup of \(\beta\) S (hence an idempotent, by the fundamental Lemma of Ellis) if S satisfies the following weak cancellation property: (A) For every finite subset F of S there is an element \(x\in S\setminus F\) such that the set \(x^{- 1}F=^{def}\{y\in S|\) xy\(\in F\}\) is finite. Reviewer's remark: (i) Condition (A) is equivalent to: \((A')\) For every finite subset F of S the set C(F)\(=^{def}\{x\in S|\) \(x^{-1}F\) is finite\(\}\) is infinite. Using this formulation, the above result is seen much more directly and with elementary arguments: We first note that \((A')\) implies that the sets C(F), F a finite subset of S, form a filter \({\mathcal C}\) in S; by construction the intersection D(\({\mathcal C})\) is closed, non-empty, and does not meet \(\phi\) (S). Also, if \(x\in C(F)\) and \(y\in C(x^{-1}F)\) then xy\(\in C(F)\). Thus, since left translations by elements of \(\phi\) (S) are continuous, we have \(\phi\) (x)\(\overline{\phi (C(x^{-1}F))}\subseteq {\bar \phi}(C(F))\) and therefore \(\phi\) (x)D(\({\mathcal C})\subseteq \overline{\phi (C(x^{-1}F))}\subseteq D({\mathcal C})\), for all \(x\in C(F)\), F finite. By the continuity of the right translations in \(\beta\) S this implies \(\overline{\phi (C(F))}D({\mathcal C})\subseteq \overline{\phi (C(F))}\) and therefore D(\({\mathcal C})D({\mathcal C})\subseteq D({\mathcal C})\), that is, D(\({\mathcal C})\) is a closed subsemigroup of \(S^*.\) (ii) It seems that the arguments of (i) carry over to the more general case of the Ellis compactification of a locally compact semitopological semigroup; we only have to substitute ``compact subset'' for ``finite subset'' in the formulation and proof of the result. (iii) Condition (A) is by no means necessary for the existence of idempotents in \(S^*\) (take, e.g., \(S=G\times N\), where G is an infinite group and N is infinite with constant multiplication \(xy=n_ 0\) for some fixed \(n_ 0\in N\) and x,y\(\in N\); then \(C(F)=\emptyset\) for \(F=\{(1,n_ 0)\})\). In fact, if \(S_ 1\) is an infinite subsemigroup of S then the injection \(S_ 1\to S\) induces an injection \(S^*_ 1\to S^*\), so \(S^*\) contains an idempotent if \(S^*_ 1\) contains one. This raises the question whether any discrete semigroup S with \(E(S^*)\neq \emptyset\) must contain an infinite subsemigroup \(S_ 1\) which satisfies (A) (or some still more restrictive condition).