S S symmetry
Article Centralising Monoids with Low-Arity Witnesses on a Four-Element Set
Mike Behrisch 1,*,† and Edith Vargas-García 2,†
1 Institute of Discrete Mathematics and Geometry, Technische Universität Wien, Wiedner Hauptstr. 8–10, 1040 Vienna, Austria 2 Departamento Académico de Matemáticas, ITAM, Río Hondo No. 1, Col. Tizapán San Angel, Del. Álvaro Obregón, Ciudad de México C.P. 01080, Mexico; [email protected] * Correspondence: [email protected] † These authors contributed equally to this work.
Abstract: As part of a project to identify all maximal centralising monoids on a four-element set, we determine all centralising monoids witnessed by unary or by idempotent binary operations on a four-element set. Moreover, we show that every centralising monoid on a set with at least four elements witnessed by the Mal’cev operation of a Boolean group operation is always a maximal centralising monoid, i.e., a co-atom below the full transformation monoid. On the other hand, we also prove that centralising monoids witnessed by certain types of permutations or retractive operations can never be maximal.
Keywords: centralising monoid; centraliser clone; commutation; endoprimal monoid; endomorphism monoid; maximal centralising monoid
MSC: 08A35; 08A02; (08A40, 08A60)
Citation: Behrisch, M.; Vargas-García, E. Centralising Monoids with Low-Arity Witnesses 1. Introduction on a Four-Element Set. Symmetry 2021, 13, 1471. https://doi.org/ There are various notions of symmetry in the sciences and in mathematics. Algebraic 10.3390/sym13081471 structures are usually considered symmetric if they have a lot of automorphisms or, more generally, endomorphisms. For a universal algebra (a structure without relations) the Academic Editor: Ivan Chajda automorphism group/endomorphism monoid consists of all those permutations/self- maps of the carrier set that commute with all fundamental operations of the algebra. Received: 23 July 2021 Studying commuting operations in more generality leads to the notion of the centraliser Accepted: 5 August 2021 clone of an algebra, or simply of a set of operations if the carrier set is understood from Published: 11 August 2021 the context. The unary part of the centraliser (clone) is then exactly the endomorphism monoid of the algebra, and the fundamental operations of the algebra are said to witness Publisher’s Note: MDPI stays neutral this monoid. with regard to jurisdictional claims in This article is concerned with algebras on a specific carrier set A that are ‘maximally published maps and institutional affil- symmetric’ in the sense that their endomorphism monoid is a co-atom in the lattice of all iations. possible endomorphism monoids of algebras on A. As the monoids in this lattice are the unary parts of (all) centraliser clones on A, they are called centralising monoids, and the co-atoms of this lattice are also referred to as maximal centralising monoids. The study of centralisers in algebra goes back to Cohn [1] (Chapter III.3), and notably Copyright: © 2021 by the authors. to Kuznecov [2] who first established logical methods for their investigation, exploiting Licensee MDPI, Basel, Switzerland. closure under operations whose graphs are primitive positively definable from given This article is an open access article operations. Kuznecov allegedly also discovered all 25 centraliser clones among Post’s distributed under the terms and lattice, a fact later re-proved by Hermann [3]. Danil’ˇcenko[4–7] continued the work of conditions of the Creative Commons Kuznecov by determining all 2986 centraliser clones on sets of three elements [8,9]. On Attribution (CC BY) license (https:// carrier sets of size four and beyond, currently no good overview of the lattice of centralisers creativecommons.org/licenses/by/ exists. Between 1974 and 1976 Harnau [10–15] worked on centralisers of unary operations, 4.0/).
Symmetry 2021, 13, 1471. https://doi.org/10.3390/sym13081471 https://www.mdpi.com/journal/symmetry Symmetry 2021, 13, 1471 2 of 40
which are dual to centralising monoids in terms of the Galois connection induced by commutation of finitary operations (see Section2). Centralising monoids of single unary operations, i.e., monounary algebras, were investigated in [16–18], showing, for example, which centralising monoids of this type are equal to the centralising monoid they describe as a witness [16] (Theorem 4.1, p. 8, Theorem 5.1, p. 10), and which of them have a unique unary operation as their witness [18] (Theorems 3.1 and 3.3, p. 4659 et seq.). Research on centralising monoids was further pushed forward in a series of papers [19–25] by Machida and Rosenberg, linking in particular maximal centralising monoids to the five types of functions appearing in Rosenberg’s Classification Theorem [26] for minimal clones. As a consequence, all 192 centralising monoids on three-element sets were determined in [21,23], and all 10 maximal centralising monoids among them were identified. This research is part of the project, begun in [27], to take the results of [23] to the next level, that is, to determine all maximal centralising monoids on four-element carrier sets. According to Proposition3 it is necessary for this to find all centralising monoids on {0, 1, 2, 3} induced by single functions of each of the five types of Rosenberg’s Theo- rem and to determine the ones among them which are maximal proper transformation monoids under set inclusion. As an initial step centralising groups of majority operations and semiprojections were studied in [28]. Extending the methods of [23], the case of ma- jority operations on {0, 1, 2, 3} was already completed in [27], and further investigated under a different aspect in [29]. In this article we tackle the cases of unary operations (permutations of prime order and non-identical retractive operations), binary idempotent operations and ternary minority operations arising as Mal’cev operations corresponding to Boolean groups. We determine all centralising monoids on 4 = {0, 1, 2, 3} witnessed by binary idem- potent operations, and we classify both the witnessing operations and the monoids up to conjugacy by inner automorphisms from Sym 4. We also exhibit the maximal monoids among them, in general, and again, up to conjugacy. With respect to unary operations, we establish that on every carrier set of size at least four, every single transposition or every product of disjoint transpositions without fixed points and every non-identical retraction witnesses a centralising monoid which is a co-atom in the lattice of those centralising monoids witnessed by sets of unary operations, that is, in the lattice of endomorphism monoids of unary algebras. However, we also show that, given at least four elements, no transposition and no permutation with a single fixed point, as well as no retraction fixing all but one element can ever witness a maximal centralising monoid. For {0, 1, 2, 3} we improve this by showing that no permutation at all will witness a maximal centralising monoid. Finally, based on results of Länger and Pöschel [30] about strongly constantively rigid relational systems, we give a new proof of the known fact [21] (Theorem 5.1) that the centralising monoid of a constant on an at least three-element set is indeed maximal. We use the same technique to prove that—with the exception of the two-element set—the centralising monoid of a Mal’cev operation of a Boolean group is always maximal.
2. Preliminaries We start by introducing basic notation with respect to sets, functions and relations, followed by fundamental facts regarding their Galois theory based on preservation and commutation. We also give a brief overview of formal concept analysis in order to be able to switch between different Galois connections. In the second part of this section we present background theory on clones, in particular centraliser clones, and centralising monoids that we shall use extensively to derive our main results.
2.1. Notation and Basic Concepts We write N = {0, 1, 2 . . . } for the set of all natural numbers (finite ordinals) and N+ for the positive ones. It will be convenient for us to understand every n ∈ N as the set of its ordinal predecessors n = {0, . . . , n − 1} as in the von Neumann model of natural Symmetry 2021, 13, 1471 3 of 40
numbers. The cardinality (size) of a set A is denoted by |A|; if |A| < ℵ0, then we often pick a canonical representative for it, e.g., n = {0, . . . , n − 1} where n = |A|. Given sets A, B, C and functions f : A −→ B and g : B −→ C, we denote their composition by g ◦ f : A −→ C and mean the function mapping each element a ∈ A to (g ◦ f )(a) := g( f (a)) ∈ C. The set of all functions from A to B is symbolised as BA. If f ∈ BA and U ⊆ A, V ⊆ B, we write f [U] for the image { f (u) | u ∈ U} and f −1[V] for the preimage { x ∈ A | f (x) ∈ V}. The full image of f is also denoted as im( f ) := f [A]. The restriction of f to an arbitrary subset U ⊆ A is the function f |U : U −→ B mapping x ∈ U to f |U(x) := f (x). For n ∈ N we understand tuples x ∈ An as maps from n = {0, . . . , n − 1} to A that are simply written as x = (x0, ... , xn−1). Sometimes we allow ourselves to deviate from this standard notation if some other indexing like x = (a, b, c) or x = (x1, ... , xn) seems more n convenient. As tuples x ∈ A are maps, we have in particular im(x) = {x0,..., xn−1} and A I n we can compose with f ∈ B or α ∈ n , giving f ◦ x = ( f (x0), ... , f (xn−1)) ∈ B and I re-indexed tuples x ◦ α = xα(i) ∈ A . i∈I (n) An For a set A and n ∈ N, every f ∈ OA := A is said to be an n-ary operation on A. O = S O(n) We collect all finitary (non-nullary) operations on A in the set A n∈N+ A . For (n) (n) (n) (n) F ⊆ OA and n ∈ N we write F for F ∩ OA ; in particular OA = (OA) for n > 0. (n) (n) For i ∈ n ∈ N the n-ary projection onto the coordinate i is the operation ei ∈ OA given by (n) n (1) e (x) := xi for x ∈ A . We write shortly idA := e for the identity operation on A, and i n o 0 J = (n) ∈ ∈ ∈ we use A ei i n N for the set of all projections. For a A the n-ary constant (n) (n) (n) n with value a is the map ca ∈ O satisfying ca (x) = a for all x ∈ A . We collect all A n o (n) constant operations on A in the set CA = ca a ∈ A ∧ n ∈ N+ . We also need some more specific operations given by identities. An operation f ∈ OA is idempotent, if it satisfies (3) f (x, ... , x) = x for all x ∈ A. A ternary operation f ∈ OA is called a majority operation, if f (x, x, y) = f (x, y, x) = f (y, x, x) = x for all x, y ∈ A; it is called a minority operation if f (x, x, y) = f (x, y, x) = f (y, x, x) = y for all x, y ∈ A; finally, it is a Mal’cev operation if f (x, x, y) = f (y, x, x) = y for all x, y ∈ A. Among unary operations we need permutations (1) (1) f ∈ Sym A, i.e., bijective self-maps f ∈ OA , and retractive operations f ∈ OA satisfying f ◦ f = f (sometimes also called idempotent since they are idempotent elements in the D (1) E semigroup OA ; ◦ ). (m) m n For all m, n ∈ N and g1, ... , gn ∈ OA , we can form (g1, ... , gn) : A −→ A , called m n the tupling of g1, ... , gn, sending x ∈ A to (g1(x), ... , gn(x)) ∈ A . In this way, we can (m) (n) compose finitary operations g1, ... , gn ∈ OA with f ∈ OA as f ◦ (g1, ... , gn).A (concrete) clone of non-nullary operations on A is any set F ⊆ OA that is closed under this form of composition and satisfies JA ⊆ F. For m ∈ N an m-ary relation on a set A is any subset $ ⊆ Am of m-tuples. Any (m) m-ary operation f ∈ OA can be understood as an (m + 1)-ary relation via its graph • f := { (x1,..., xm, g(x1,..., xm)) | x1,..., xm ∈ A}. We extend this notation element- • • wise to sets of operations by putting F := { f | f ∈ F} for F ⊆ OA. A binary rela- tion $ ⊆ A2 is just a set of pairs; we define its inverse (sometimes also called dual) to be $−1 := { (y, x) | (x, y) ∈ $}. A binary $ ⊆ A2 is symmetric if $−1 ⊆ $, and reflexive if (x, x) ∈ $ for all x ∈ A. If $ is reflexive, symmetric and transitive, that is (x, y), (y, z) ∈ $ implies (x, z) ∈ $ for all x, y, z ∈ A, then $ belongs to Eq(A), the set of all equivalence m relations on A. By RA = { $ ⊆ A | m ∈ N+} we denote the set of all finitary relations on A. While functions can be composed, new relations can be constructed from given ones using logical expressions. A primitive positive formula ϕ (in prenex normal form) over a given relational signature consists of a prefix of finitely many existentially quantified variables (possibly none) followed by a finite non-empty conjunction of atomic predicates that correspond to the given signature (relations) and have been substituted by some Symmetry 2021, 13, 1471 4 of 40
tuple of variables. If the relation symbols are associated with concrete relations, the set of all satisfying value assignments to (a superset of) the free variables of ϕ determines a finitary relation on A, which is called primitive positively definable from the given relations. For a finite carrier set A, a set Q ⊆ RA is said to be a relational clone on A if it contains the diagonal ∆A := { (x, x) | x ∈ A} and is closed under all relations that are primitive positively definable from members of Q (for infinite carrier sets stronger closure properties are required). As this is an implicational definition, all relational clones on A form a closure system. The least relational clone on A containing a particular set R ⊆ RA will be denoted by [R] and is computed by adding to R all relations that are primitive positively RA definable from R ∪ {∆A}. Clones of operations and relational clones are connected in the following way: for m, n ∈ N and f : An −→ A, $ ⊆ Am we say that f preserves $ if and only if $ is a subuniverse m n of the algebra hA; f i . This means that for every (r0, ... , rn−1) ∈ $ the m-tuple obtained by the composition f ◦ (r0, ... , rn−1) of f with the tupling of r0, ... , rn−1 stays inside $, that is, for every (m × n)-matrix R ∈ Am×n the columns of which all belong to $, the m-tuple obtained by applying f row-wise to R remains in $. We then call $ an invariant of f or f a polymorphism of $. The preservation relation induces a Galois connection between OA and RA, giving rise to the derivation operators Q 7→ PolA Q and F 7→ InvA F, collecting all polymorphisms in OA of every relation $ ∈ Q ⊆ RA and all invariant relations $ ∈ RA for all given operations f ∈ F ⊆ OA, respectively. Moreover, we declare for n ∈ N and (n) (n) (n) (n) F ⊆ OA, Q ⊆ RA that PolA F := (PolA F) and InvA Q := (InvA Q) . Now for every Q ⊆ RA the set PolA Q forms a clone on A, and for every F ⊆ OA the invariants InvA F are a relational clone containing ∅ (since we omit nullary operations in our clones). Therefore, Inv Pol Q ⊇ [Q ∪ {∅}] and Pol Inv F contains the clone generated by F. On finite A A RA A A carrier sets A, these inclusions are equalities [31,32]; for infinite A local interpolation operators (and a strengthened definition of relational clone) need to be added to close the gap [33,34]. Besides the Galois connection induced by preservation, we shall encounter several other Galois connections that differ by the inducing relation or by restrictions of the domains they link. Switching between them and the associated closure systems can be expressed well in the framework of formal concept analysis [35,36], providing terminology, notation and theory for the manipulation of Galois connections on a general level. The basic object of formal concept analysis is that of a formal context K = (G, M, I) where I ⊆ G × M is any binary relation between sets G (commonly called objects) and M (usually called attributes). A context K induces two derivation operators of a Galois connection between G and M in the natural way, and specifies exactly between which sets this Galois connection is to be understood. The Galois closed sets on the side of objects are called extents and the ones on the side of attributes are referred to as intents. When we keep the relation I, but pass on to a subset H ⊆ G of the objects or N ⊆ M of the attributes (or both), we form 0 a subcontext K = (H, N, I ∩ (H × N)). Though technically incorrect, it is customary to omit the intersection ∩ (H × N) when specifying a subcontext, as the restriction becomes clear from stating the sets of objects and attributes. How the lattices of Galois closed sets 0 (extents and intents) of K and K are related is discussed in Chapter 3 of [35] (p. 97 et seqq.). We will mostly aim for context manipulations where the closure system of intents (and thus the lattice structure) does not change, such as object clarification and object reduction. We shall explain more details at the appropriate place in the text and give pointers to the literature there. Centrally for this paper will be the Galois connection of commutation given by the (n) context Kc = (OA, OA, ⊥). In this respect we say that an n-ary operation f ∈ OA (m) T commutes with an m-ary operation g ∈ OA , if g( f (X)) = f g X holds for any matrix m×n X ∈ A , where f (X) := ( f (X(i, ·)))i∈m denotes the tuple obtained by applying f row- wise to the matrix X, and similarly for g and the transposed matrix XT. This commutation condition will be denoted by f ⊥ g; it is easy to see that it is a symmetric property, i.e., we have f ⊥ g if and only if g ⊥ f . Therefore, the two Galois derivatives induced by Kc Symmetry 2021, 13, 1471 5 of 40
∗ coincide: they map F ⊆ OA to its centraliser F = { g ∈ OA | ∀ f ∈ F : g ⊥ f }, and the associated closure operator sends F to F∗∗, the bicentraliser of F. A routine verification ∗ • • shows that g ∈ F ⇐⇒ g ∈ PolA F ⇐⇒ g ∈ InvA F holds for F ∪ {g} ⊆ OA, so ∗ • F = PolA F , i.e., commutation can be rephrased in terms of preservation of graphs of operations. Hence, up to isomorphism, Kc is the subcontext of finitary operations and • relations with preservation, where RA is restricted to [OA] . As mentioned in the introduction, when studying centralising monoids we only look at the unary parts F∗(1) of centraliser clones, that is to say, we are dealing with the (1) (1) (1) subcontext K := OA, OA , ⊥ ∩ OA × OA , briefly OA, OA , ⊥ , of Kc. The intents ∗(1) of K, i.e., Galois closed sets M = F for some F ⊆ OA, are exactly all centralising monoids on A. Moreover, every set F ⊆ OA describing M in this way, is a witness of M, and we can equivalently express that M is witnessed by F via saying that M = End hA; Fi.A (1) centralising monoid M ⊆ OA is maximal if it is a co-atom in the intent lattice of K, i.e., a proper centralising monoid which is maximal under set inclusion.
2.2. Fundamental Results on Clones, Centralisers and Centralising Monoids The following simple observation exhibits necessary conditions that can be used to describe functions in a particular centraliser clone (or its unary part). The utility of said conditions was already observed by Harnau, cf. Lemma 2.6 and Satz 2.15 of [10], in the (1) context of centralisers of single unary operations f ∈ OA .
Lemma 1. For every set F ⊆ O we have F∗ = Pol [F•] . In particular, we have A A RA
∗ F ⊆ PolA{ ker( f ) | f ∈ F}, ∗ F ⊆ PolA{ im( f ) | f ∈ F}, ∗ F ⊆ PolA{ fix( f ) | f ∈ F},
(n) where for n ∈ N and f ∈ OA n o 2n ker( f ) := (x1,..., xn, y1,..., yn) ∈ A f (x1,..., xn) = f (y1,..., yn) , im( f ) := f [An], fix( f ) := { x ∈ A | f (x,..., x) = x}.
Proof. We have F∗ = Pol F• ⊇ Pol [F•] ⊇ Pol Inv Pol F• = Pol F•, following A A RA A A A A directly from the definition of commutation. Clearly, fix( f ) is primitive positively definable from the graph of f , im( f ) = { y ∈ A | ∃x1,..., xn ∈ A : f (x1,..., xn) = y} is too, and ker( f ) = (x, y) ∈ A2n ∃z ∈ A : f (x) = z ∧ f (y) = z as well. Second, we recollect the following helpful characterisation of centralising monoids.
(1) Lemma 2 (cf. [21] (Lemma 2.2)). For a set M ⊆ OA the following facts are equivalent: ∗(1) (a) M is a centralising monoid, i.e., there is F ⊆ OA such that M = F . (b) M = M∗∗(1), i.e., M is witnessed by M∗.
∗(1) ∗∗(1) ∗(1) ∗ ∗(1) ∗(1) Proof. If M = F for some F ⊆ OA, then M = ((F ) ) = F = M by virtue of the Galois derivatives ∗ and ∗(1). Conversely, if M = M∗∗(1), then M is a centralising monoid witnessed by F = M∗.
Lemma2 shows that every centralising monoid M = F∗(1) has a largest possible witness, as generally M ⊆ F∗, so F ⊆ M∗. We now turn to maximal centralising monoids. (1) Being co-atoms in the lattice of intents of K = OA, OA , ⊥ , they correspond via that Galois connection of commutation to an atom F in the lattice of extents of K. Such an Symmetry 2021, 13, 1471 6 of 40
(1)∗ atom must be generated (as an extent) by a single non-trivial object f ∈ OA \OA , i.e., F = { f }∗(1)∗. For the co-atom M = F∗(1) this means M = F∗(1) = { f }∗(1)∗∗(1) = { f }∗(1) by virtue of the Galois connection given by K. So M is singly witnessed by a non-trivial operation f ; however, even more is known about these witnesses: they can always be chosen as a generator of a minimal clone of minimum arity, a so-called minimal function.
Proposition 3 (Theorem 3.2 in [21], Proposition 2.2 in [27]). For any maximal centralising (1) ∗ (1)∗ monoid M ⊆ OA on a finite set A there is a minimal function f ∈ M \OA (generating a minimal clone) such that M = { f }∗(1).
Minimal functions are separated by Rosenberg’s Classification Theorem [26] for min- imal clones into five distinct categories ((II) is a special case of (V), but is often listed on its own).
Theorem 4 (See [26]). On a finite set A every minimal function f is of one of the following types: (1) (I) f ∈ OA and f ∈ Sym A is a permutation of prime order or a retractive operation f ◦ f = f 6= idA; (2) (II) f ∈ OA is idempotent, i.e., f ◦ (idA, idA) = idA; (3) (III) f ∈ OA is a ternary (minority) Mal’cev operation arising as f (x, y, z) := x + y + z for x, y, z ∈ A from a Boolean group hA; +i; (3) (IV) f ∈ OA is a ternary majority operation; (n) (V) f ∈ OA is a proper semiprojection of arity n where 3 ≤ n ≤ |A|.
In this paper, we address centralising monoids witnessed by a minimal function of types (I)–(III) with a special focus on the set {0, 1, 2, 3}; those relating to type (IV) have already been considered in [27], the ones of type (V) are part of ongoing research. Before we can tend to this problem in more detail, we need further background information on clones, such as the centraliser of a constant operation.
Lemma 5 (Lemma 1.9 in [10], Lemma 5.2 in [21]). For any n ∈ N and a ∈ A we have n (n)o∗ ca = PolA{{a}} = { f ∈ OA | a ∈ fix( f )}.
(m) (n) Proof. Let m ∈ N and f ∈ OA . We have f ⊥ ca if and only if for all x11, ... , xmn ∈ A the condition
(n) (n) f (a,..., a) = f (ca (x11,..., x1n),..., ca (xm1,..., xmn)) (n) = ca ( f (x11,..., xm1),..., f (x1n,..., xmn)) = a
holds, that is, exactly if f (a,..., a) = a, i.e., precisely if a ∈ fix( f ), or f ∈ PolA{{a}}.
(1)∗ ∗ Corollary 6 (Lemma 5.3 in [21]). The centraliser CA = CA is the clone of all idempotent operations.
∗ (1)∗ T n (1)o T Proof. We have CA = a∈A ca = a∈A PolA{{a}} = PolA{{a} | a ∈ A} due to Lemma5.
The following characterisation is also very well known. Symmetry 2021, 13, 1471 7 of 40
Lemma 7. For a clone F on any set A the following statements are equivalent:
(a) F is a clone of idempotent operations, i.e., F ⊆ PolA{{a} | a ∈ A}. (b) For every f ∈ F and all a ∈ A we have f (a,..., a) = a. (1) (c) F = {idA}.
Proof. Condition (b) simply spells out the preservation of every singleton set {a} in (a). (1) From this it follows that every f ∈ F satisfies f (a) = a for every a ∈ A, i.e., f = idA. For F contains all projections, (b) implies (c). Now since F is a clone, with every f ∈ F (1) also f ◦ (idA, ... , idA) ∈ F . If (c) holds, then f ◦ (idA, ... , idA) ∈ {idA}, and this shows statement (b).
(1)∗ Lemma 8. For any set A we have (JA ∪ CA) ∩ CA = JA.
Proof. For |A| < 2 this is trivially true. For |A| ≥ 2, no constant map is idempotent, so, (1)∗ by Corollary6, (JA ∪ CA) ∩ CA can only consist of projections. The converse inclusion is trivial.
When proving that a certain centralising monoid is maximal, it will be necessary to ∗∗(1) (1) compute the centralising monoid M generated by some set M ⊆ OA and to show that it is the full transformation monoid on A. To do this we shall always demonstrate the seemingly stronger condition that the centraliser of M is trivial. The following lemma shows that this is in fact a necessary step to take whenever |A| ≥ 3.
Lemma 9. For any set A with |A| ≥ 3 and any subset M ⊆ OA the following statements are equivalent: ∗∗(1) (1) (a) M = OA . ∗∗ (b) M = OA. ∗ (c) M = JA.
Proof. The implications (c) =⇒ (b) =⇒ (a) are obvious as every function commutes with (1) ∗∗ ∗ (1)∗ projections. Now, if (a) holds, then OA ⊆ M , which is equivalent to F := M ⊆ OA . By Lemma1, this means the functions in F preserve the kernel of any unary operation, so they preserve every equivalence relation on A. Therefore, F ⊆ PolA Eq(A), and for |A| ≥ 3 the latter clone equals JA ∪ CA as is shown in Example 3.3 of [30] (p. 136). Now (1)∗ (1)∗ (1)∗ F ⊆ OA ⊆ CA implies that F ⊆ (JA ∪ CA) ∩ CA , so Lemma8 shows that (c) must be satisfied.
Note that for |A| = 2, Lemma9 fails as, e.g., the clone L of all affine linear functions, which is the centraliser of the (ternary minority) Mal’cev function on the two-element set, ∗∗ ∗∗(1) (1) (1) contains all unary operations. So L = L ( OA, but L = L = OA . To establish for a subset M ⊆ OA condition (c) of Lemma9 or to prove at least ∗(1) M = {idA}, we shall exploit a theorem of Länger and Pöschel [30] on (strongly) constantively rigid systems of (binary) relations.
Proposition 10 (cf. [30] (Theorem 2.13, p. 136, Theorem 2.3, p. 133)). Let A be any (finite (2) or infinite) set with |A| ≥ 4, F ⊆ OA and let Q ⊆ InvA F satisfy the following two separation conditions (see Definition 2.7, p. 134 et seq. in [30]) (B) for all pairwise distinct x, y, z ∈ A there is $ ∈ Q such that (x, y) ∈ $ 63 (x, z); (D) for all pairwise distinct x, y, z, u ∈ A there is $ ∈ Q such that (x, y) ∈ $ 63 (z, u). (1) (1) Then F ⊆ {idA} ∪ CA ; moreover, if all relations in Q are reflexive, then F ⊆ JA ∪ CA. Symmetry 2021, 13, 1471 8 of 40
(1) (1) Proof. By our assumption we have F ⊆ PolA Q and, in particular, F ⊆ PolA Q. We (1) use Theorem 2.13 of [30] to infer that Q is constantively rigid, i.e., PolA Q ⊆ JA ∪ CA, (1) (1) wherefore F ⊆ {idA} ∪ CA . If, additionally, Q only contains reflexive relations, then Theorem 2.3 of [30] states that constantive rigidity of Q is equivalent to strong constantive rigidity, that is, F ⊆ PolA Q ⊆ JA ∪ CA.
Remark 11. There are two more conditions, (A) and (C), in [30], which can also be employed to en- sure (strong) constantive rigidity of systems Q of (binary reflexive) relations, see, e.g., Theorems 2.11 and 2.12 of [30]. In particular, via Theorem 2.12, (B) and (C) together may be used to obtain (strong) constantive rigidity also when |A| = 3, and (C) automatically follows from (B) if all the relations in Q are symmetric (see Lemma 2.8 in [30]), for example, equivalence relations. More concrete constantively rigid systems of relations are provided in Section 3 of [30], some of them also being strongly constantively rigid.
Corollary 12. Let A be any set with |A| ≥ 4 and M ⊆ OA be such that for each a ∈ A there is ( ) s ∈ M satisfying a ∈/ fix(s). Moreover assume that Q ⊆ [M•] 2 satisfies conditions (B) and RA ∗(1) ∗ (D), then M = {idA}; if all the relations in Q are reflexive, then M = JA; if, moreover, the relations used to satisfy (B) are reflexive and symmetric, then the result also follows for |A| = 3.
Proof. Let F := M∗ in Proposition 10 and note that [M•] ⊆ Inv Pol M• = Inv M∗. RA A A A ∗(1) (1) ∗ So Proposition 10 entails that M = F ⊆ JA ∪ CA and M = F ⊆ JA ∪ CA under the reflexive assumption. If the relations used in (B) are reflexive and symmetric, we combine the reasoning of Proposition 10 with Remark 11 to get F ⊆ JA ∪ CA also for |A| = 3. (n) ∗ Finally, F ∩ CA = ∅, for the supposition of ca ∈ F = M for some a ∈ A and n ∈ N n (n)o∗ would imply M ⊆ ca = { f ∈ OA | a ∈ fix( f )}, see Lemma5, which contradicts the assumption on M.
Corollary 13. Let A be any set with |A| ≥ 3 and M ⊆ OA be such that for each a ∈ A there is s ∈ M satisfying a ∈/ fix(s). Moreover assume that (B’) for all pairwise distinct x, y, z ∈ A there is s ∈ M(1) such that s(y) = s(x) 6= s(z), (D’) for all pairwise distinct x, y, z, u ∈ A there is s ∈ M(1) such that s(x) = s(y) and s(z) 6= s(u), ∗ then M = JA. n o ( ) ∈ (1) ⊆ (2) ∗ Proof. Use Corollary 12 on the equivalence relations ker s s M InvA M .
Corollary 14. Let A be any set with |A| ≥ 4 and M ⊆ OA be such that for each a ∈ A there is s ∈ M satisfying a ∈/ fix(s). Moreover assume that (B”) for all pairwise distinct x, y, z ∈ A there is s ∈ M(1) such that s(y) = s(x) 6= s(z) or s(x) = y or s(y) = x 6= s(z), (D”) for all pairwise distinct x, y, z, u ∈ A there is s ∈ M(1) satisfying one of the conditions s(x) = s(y) ∧ s(z) 6= s(u) or s(x) = y ∧ s(z) 6= u or s(y) = x ∧ s(u) 6= z, ∗(1) then M = {idA}. n o n o = ( ) ∈ (1) ∪ (1)• ∪ •−1 ∈ (1) ⊆ (2) ∗ Proof. Use the set Q : ker s s M M s s M InvA M in Corollary 12.
3. Monoids Witnessed by Unary Operations There are two types of unary minimal functions in Rosenberg’s Theorem4. The first are permutations of prime order, that is, their cycle structure consists only of fixed points and cycles of length p for some prime p. The second are idempotent or retractive Symmetry 2021, 13, 1471 9 of 40
(1) operations f ∈ OA , which satisfy f ◦ f = f 6= idA. These are exactly those non-identical unary operations which fix every point of their image.
3.1. Computational Results for {0, 1, 2, 3} We started by computing a commutation table of all 256 unary operations on the (1) (1) set A = {0, 1, 2, 3}, i.e., the formal context K1 = OA , OA , ⊥ in the language of [35,36]. Being a 256 × 256 Boolean matrix this table is already confusingly big and we there- fore do not present it here. For purposes of verification, in Listing1 we have instead added simple Python code that allows anybody to reproduce such a table if desired. Us- ing standard algorithms presented in Section 2.1 of [35] or Chapter 2 of [36], one may compute from K1 that there are exactly 1485 centralising monoids on {0, 1, 2, 3} that can be witnessed by sets of unary operations. However, we are only interested in the co- atoms of this large lattice. By virtue of the Galois connection represented by K1, every co-atom M is the Galois derivative N∗(1) of an atom N on the dual side of the Galois connection. Since in the lattice of closed sets of any closure operator the atoms must be ∗(1)∗(1) (1) singly generated, it follows that N = { f } for a single unary function f ∈ OA and so M = N∗(1) = { f }∗(1)∗(1)∗(1) = { f }∗(1). Hence, to obtain the co-atoms, we only need to (1) iterate over the 255 rows of K1 belonging to each f ∈ OA \ {idA} (or just the unary f from Rosenberg’s Theorem) and check, which { f }∗(1) are maximal under set inclusion. The result is that there are 49 co-atoms in the lattice of closed sets of K1, and sub- sequently we shall explain our computational findings in terms of the types of unary operations from Rosenberg’s Theorem, i.e., permutations of prime order and non-identical retractions. In the following two subsections we shall then give theoretical evidence why the 49 co-atoms arise on {0, 1, 2, 3}. It is worth noting that these 49 co-atoms in the lattice belonging to K1 are merely candidates for maximal centralising monoids, and we shall indeed demonstrate that some co-atoms of K1 fail to be maximal among all centralising (1) monoids, that is, are no co-atoms with respect to OA, OA , ⊥ .
Listing 1. Python code to print a commutation table (0-1-matrix) for all 256 unary operations f on {0, 1, 2, 3}, where rows and columns are enumerated in lectic order of the value tables f ◦ (0, 1, 2, 3) from (0, 0, 0, 0), (0, 0, 0, 1), (0, 0, 0, 2),..., (3, 3, 3, 3). #!/usr/bin/env python f = [0] * 4 g = [0] * 4 row = [0] * 256 separator = "␣" # could also be an empty string or ", " if desired f o r f[0] i n range (4): f o r f[1] i n range (4): f o r f[2] i n range (4): f o r f[3] i n range (4): j=0 f o r g[0] i n range (4): f o r g[1] i n range (4): f o r g[2] i n range (4): f o r g[3] i n range (4): fgcommute= True f o r x i n range (4): fgcommute= fgcommute and f[g[x]] ==g[f[x]] i f (fgcommute): row[j] = 1 # indicates commutation of f and g e l s e : row[j] = 0 # indicates non-commutation of f and g j=j+1 p r i n t (separator.join( map( s t r ,row))) # output table row Symmetry 2021, 13, 1471 10 of 40
On a four-element set there are three types of permutations of prime order: three- cycles, transpositions or products of two disjoint transpositions. Up to conjugacy these can be represented by the following three functions on {0, 1, 2, 3}: (0, 1, 2), (0, 1), (0, 1)(2, 3). 4 There are 2 · (1) = 8 three-cycles (4 choices of the unique fixed point and 2 for the image of the first cycle element), but it will turn out that centralising monoids witnessed by these operations are not maximal (Lemma 16), not even with respect to K1. However, the monoids described by the other prime permutations are maximal when considered among the monoids witnessed by only unary operations (i.e., the closure system of K1). Yet again, they will not be maximal in the lattice of all centralising monoids on {0, 1, 2, 3}, 4 see Corollary 22. We have (2) = 6 transpositions (choosing the two transposed elements) 1 4 and 2 · (2) = 3 products of disjoint transpositions since such a function can be obtained by choosing the first transposed pair or the second one. Altogether there are 9 prime permutations (of order 2) that describe maximal monoids of K1. The non-trivial retractive functions on a four-element set can be separated into the following types: being constant (one-element image, there are 4 such functions), having a 4 3 two-element image, or having a three-element image (there are (3) · (1) = 12 such functions, first choosing the image and then the value of the element outside the image within it). The retractions with a two-element image can moreover be split up into those mapping the two elements outside the image to the same value and those mapping them to both distinct 4 values of the image. In both subcases we have (2) · 2 = 12 functions as we need to choose the two-element image and then one of two ways how the elements outside the image can be mapped into it. Retractions of all four types witness centralising monoids that are maximal among those centralising monoids witnessed by unary functions, see Corollary 27, and there are 3 · 12 + 4 = 40 non-trivial retractions on {0, 1, 2, 3}. Moreover, it is known n (1)o∗(1) from Theorem 5.1 of [21] that ca with a ∈ A is a maximal centralising monoid in general (that is on any finite set A with |A| ≥ 3 in the lattice of all centralising monoids), (1) so the 4 centralising monoids of constants are co-atoms with respect to OA, OA , ⊥ , not just K1. In total, we have 40 + 9 = 49 centralising monoids on {0, 1, 2, 3} that are maximal in the 1485-element lattice of those monoids that can be witnessed by sets of unary operations. ∗( ) These can be grouped together into 6 conjugacy types that can be represented as {(0, 1)} 1 , ∗(1) ∗(1) n (1)o ∗(1) ∗(1) ∗(1) {(0, 1)(2, 3)} , c0 , { f24} , { f16} and { f17} where the retractions f24, f16 and f17 are given as f24 ◦ (0, 1, 2, 3) = (0, 1, 2, 0), f16 ◦ (0, 1, 2, 3) = (0, 1, 0, 0) and f17 ◦ (0, 1, 2, 3) = (0, 1, 0, 1). From the explicit calculations (and the characterisations in the following subsections) it follows that
∗(1) ∗(1) {(0, 1)} = 16 {(0, 1)(2, 3)} = 16 ∗(1) n (1)o ∗(1) c = 64 { f } = 36 0 24
∗(1) ∗(1) { f16} = 20 { f17} = 16.
∗(1) As (0, 1)(2, 3) has no fixed points but (0, 1) and f17 have, the monoid {(0, 1)(2, 3)} ∗(1) ∗(1) does not contain constant operations (cf. Lemma5) but {(0, 1)} and { f17} do. There- D ∗(1) E ∼ D ∗(1) E ∼ D ∗(1) E fore, {(0, 1)} ; ◦ 6= {(0, 1)(2, 3)} ; ◦ 6= { f17} ; ◦ . Furthermore, computa- D ∗(1) E D ∗(1) E tions show that {(0, 1)} ; ◦ has 96 binary term functions, but { f17} ; ◦ has 262 of D ∗(1) E ∼ D ∗(1) E them, so {(0, 1)} ; ◦ 6= { f17} ; ◦ . Therefore, the co-atoms of the lattice of intents of K1 fall into exactly 6 distinct isomorphism classes of monoids. Moreover, we computed that all 1485 monoids in the lattice of intents of K1 can be separated into 106 classes up to element-wise conjugacy (cf. Section4 for more explanation). Symmetry 2021, 13, 1471 11 of 40
3.2. Monoids Witnessed by Permutations In Lemma 4.11 of [27] (see also Lemma 4.10 of [37]) we provided a characterisation of when a finitary operation f ∈ OA on a finite set A commutes with a permutation s ∈ Sym A, based on examining orbits of the permutation group generated by s. Under the assumption that f is a function the given condition was sufficient to ensure that f ⊥ s, but blindly fulfilling the orbit condition could also lead to some value assignments for f that would contradict the assumption of f being a function. Thus, some care had to be taken when working with Lemma 4.11 from [27], but this was never much of an issue when studying majority operations f on small sets as in [27]. Here we improve the mentioned characterisation by placing an additional necessary condition on the choice of the function values such that no contradictions can occur.
Lemma 15. For a finite set A, n ∈ N+ and s ∈ Sym A let S := h{s}iSym A be the permutation group generated by s and T ⊆ An be a transversal of the orbits of the action of S on An via (n) (s˜, x) 7→ s˜ ◦ x. For f ∈ OA we have f ⊥ s if and only if for every x ∈ T the length of the orbit of f (x) under S divides the length ` of the orbit of x under S and f (sj ◦ x) = sj( f (x)) for all exponents j satisfying 1 ≤ j < `. In particular if for all x ∈ T a value f (x) is chosen such that the size of the orbit of f (x) is a divisor of the size ` of the orbit of x, then there is a unique extension of this partial definition to an n-ary function f ∈ {s}∗ by defining f (sj ◦ x) := sj( f (x)) for all 1 ≤ j < ` and x ∈ T.
Proof. Let m := |S| be the order of s. Lemma 4.11 of [27] shows that f ⊥ s if and only if for all x ∈ T the condition f (sj ◦ x) = sj( f (x)) holds for all 0 ≤ j < m. The length ` of the orbit of x is the least common multiple of the lengths of the orbits of the entries of x and each of these lengths divides m, so ` divides m. So f (sj ◦ x) = sj( f (x)) for all 1 ≤ j < ` follows from f ⊥ s. If t is the size of the orbit of f (x), then this orbit is sj( f (x)) 0 ≤ j < t and st( f (x)) = f (x). Writing ` = q · t + r with 0 ≤ r < t, it follows from the definition of ` and the commutation condition that
f (x) = f (s` ◦ x) = f (s ◦ s`−1 ◦ x) = s( f (s`−1 ◦ x)) = s(s`−1( f (x))) = s`( f (x)) = sr( f (x)),
= j( ( )) ⊥ so r 0 as the elements s f x 0≤j