Lax orthogonal factorisations in monad-quantale-enriched categories
Maria Manuel Clementino, Ignacio Lopez Franco

TL;DR
This paper develops a framework for constructing weak factorisation systems in enriched categorical settings using lax orthogonal factorisation systems derived from presheaf monads on $( ext{T},V)$-categories, with applications to topological categories.
Contribution
It introduces a method to obtain lax orthogonal factorisation systems in $V$-enriched categories via presheaf monads, extending the theory to topological categories.
Findings
Presheaf monads are simple and induce lax orthogonal factorisation systems.
The approach applies to well-known topological categories over Set.
Presheaf submonads define additional LOFSs, enriching the categorical structure.
Abstract
We show that, for a quantale and a -monad laxly extended to -, the presheaf monad on the category of -categories is simple, giving rise to a lax orthogonal factorisation system (lofs) whose corresponding weak factorisation system has embeddings as left part. In addition, we present presheaf submonads and study the LOFSs they define. This provides a method of constructing weak factorisation systems on some well-known examples of topological categories over .
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsHomotopy and Cohomology in Algebraic Topology · Algebraic structures and combinatorial models
\lmcsheading
1–LABEL:LastPageJan. 25, 2017Sep 27, 2017
Lax orthogonal factorisations in
monad-quantale-enriched categories
Maria Manuel Clementino
CMUC, Department of Mathematics
University of Coimbra
3001-501 Coimbra
Portugal
and
Ignacio López-Franco
Department of Mathematics and Applications
CURE – Universidad de la República
Tacuarembó s/n
Maldonado
Uruguay
Dedicated to Jiří Adámek
Abstract.
We show that, for a quantale and a -monad laxly extended to -, the presheaf monad on the category of -categories is simple, giving rise to a lax orthogonal factorisation system (lofs) whose corresponding weak factorisation system has embeddings as left part. In addition, we present presheaf submonads and study the lofss they define. This provides a method of constructing weak factorisation systems on some well-known examples of topological categories over .
Key words and phrases:
Quantale, monad, enriched category, -category, presheaf monad, injective morphism
2010 Mathematics Subject Classification:
18A32, 18C15, 18C20, 06B35, 54B30
The authors acknowledge partial financial assistance by Centro de Matemática da Universidade de Coimbra – UID/MAT/00324/2013, funded by the Portuguese Government through FCT/MEC and co-funded by the European Regional Development Fund through the Partnership Agreement PT2020.
1. Introduction
In 1985 Cassidy-Hébert-Kelly [8] studied orthogonal factorisations systems induced by reflective subcategories, with particular emphasis in the case when the reflection is simple. Among the lax orthogonal factorisation systems (lofss), that generalise the orthogonal ones in 2-categories, those arising from simple monads – as defined by the authors of this paper in [14, 15] – have particular relevance. This paper intends to give a systematic way of producing simple monads in (some) topological categories over Set using the presheaf monads of -Cat studied in [23, 11]. Given a quantale and a well-behaved Set-monad , the category -Cat, of generalised -enriched categories and their functors, is topological and locally preordered (see [10, 16]). As crucial examples we mention the categories Ord of (pre)ordered sets and monotone maps, Top of topological spaces and continuous maps, Met of Lawvere generalised metric spaces and non-expansive maps [28], and App of Lowen approach spaces and non-expansive maps [30]. Equipping the quantale with a canonical -category structure, one gets naturally a Yoneda Lemma and a well-behaved presheaf monad that was shown to be lax idempotent in [23]. Here we show that it is simple, inducing a lax orthogonal factorisation system which underlies a weak factorisation system having embeddings as left part. (In order to avoid technicalities we restrict ourselves to separated, or skeletal, -categories, so that their hom-sets have an anti-symmetric order.) This encompasses the weak factorisation system in Ord studied by Adámek-Herrlich-Rosický-Tholen in [1].
These presheaf monads have interesting simple submonads, namely the ones which have as algebras the Lawvere complete -categories (see [12]), and that gives – as shown by Lawvere in [28] – Cauchy-complete spaces when one takes the identity monad and the complete half-real line. These also cover, following techniques developed in [11], the weak factorisation systems of studied in [5], having as left parts embeddings, dense embeddings, flat embeddings and completely flat embeddings.
The examples of lofss and weak factorisation systems we consider are the result of a general construction distinct from, and in many ways orthogonal to, the more usual method of cofibrant generation. The construction of cofibrantly generated weak factorisation systems is usually known as Quillen’s small object argument [32]. A version for algebraic factorisation systems was introduced in [19]. The construction we employ, introduced in [14], gives rise to weak factorisations systems that need not be cofibrantly generated (see [29]).
This paper does not intend to be self-contained. In Section 2 and 3 we present the basic definitions and results on lax orthogonal factorisation systems and on -categories that are needed for this work. For a better understanding of these topics we refer to the papers mentioned there and to the monograph [24]. In Section 4 we study the presheaf monads on -categories and their simplicity. In Section 5 we explore the examples of lax orthogonal factorisation systems induced by these presheaf monads.
2. Lax orthogonal factorisation systems
Throughout we will be working in a category C enriched in posets, or -enriched category, so that each hom-set is equipped with an order structure that is preserved by composition: if , with , and , then and .
2.1. Weak factorisation systems
Given morphisms , we say that has the left lifting property with respect to , and that has the right lifting property with respect to , if every commutative square as shown has a (not necessarily unique) diagonal filler.
[TABLE]
A weak factorisation system (wfs) in a category is a pair of families of morphisms such that:
- •
consists of those morphisms with the left lifting property with respect to each morphism of .
- •
consists of those morphisms with the right lifting property with respect to each morphism of .
- •
Each morphism in the category factors through an element of followed by one of .
2.2. Algebraic weak factorisation systems
An Ord-functorial factorisation on an Ord-category C consists of a factorisation of the natural transformation with component at equal to , in the category of locally monotone functors . As in the case of functorial factorisations on ordinary categories, an Ord-functorial factorisation can be equivalently described as:
- •
A copointed endo-Ord-functor on with .
- •
A pointed endo-Ord-functor on with .
The three descriptions of an Ord-functorial factorisation are related by:
[TABLE]
An algebraic weak factorisation system, abbreviated awfs, on an Ord-category C consists of a pair , where is an Ord-comonad and is an Ord-monad on , such that and represent the same Ord-functorial factorisation on C (i.e., the equalities above hold), fulfilling a distributivity condition which we explain next.
Note that the components of and are as follows
[TABLE]
which together form a natural transformation , with as below.
[TABLE]
The distributivity axiom requires to be a mixed distributive law between the comonad and the monad , that reduces to the commutativity of the following diagrams.
[TABLE]
Algebraic weak factorisation systems were introduced by Grandis-Tholen in [20] under the name natural factorisation system; later, in [19], Garner added to this definition the distributivity conditions we described above.
Each awfs has an underlying wfs , with has an -coalgebra structure and has an -algebra structure. A coalgebra structure for , so that and , and an -algebra structure for , so that and , give a natural lifting for a commutative square :
[TABLE]
This lifting is unique – so that is an orthogonal factorisation system – if, and only if, and are idempotent. In fact idempotency of implies idempotency of and vice-versa, as shown in [4].
2.3. Lax orthogonal factorisation systems
Informally, a lax orthogonal factorisation system is an awfs whose liftings as in (b) have a universal property, as we explain next. First we recall that:
Definition** ([27]).**
An Ord-enriched monad is lax idempotent, or Kock-Zöberlein, if it satisfies any of the following equivalent conditions:
- (i)
; 2. (ii)
* (or, equivalently, );* 3. (iii)
* (or, equivalently, );* 4. (iv)
a morphism is an -algebra structure for if, and only if, with .
A lax idempotent -comonad is defined dually.
Lemma**.**
If is a lax idempotent monad on an Ord-category, the following conditions are equivalent, for an object of :
- (i)
* admits an -algebra structure;* 2. (ii)
* admits a unique -algebra structure;* 3. (iii)
* has a right inverse, i.e. admits an -algebra structure;* 4. (iv)
* is a retract of ;* 5. (v)
* is a retract of an -algebra.*
An awfs is a lax orthogonal factorisation system, abbreviated lofs, if and are lax idempotent. These factorisations were introduced by the authors in [14] and further studied in the Ord-enriched categories setting, as used here, in [15].
Corollary**.**
If is a lofs, then its underlying weak factorisation system consists of the class of the morphisms admitting a (unique) -coalgebra structure and the class consists of the morphisms admitting a (unique) -algebra structure.
As for orthogonal factorisation systems, lax idempotency of implies lax idempotency of and vice-versa. In fact:
Theorem** ([15]).**
- (1)
Given an awfs on an Ord-category C, the following conditions are equivalent:
- (i)
* is a lofs;* 2. (ii)
* is lax idempotent;* 3. (iii)
* is lax idempotent.* 2. (2)
Given a domain-preserving Ord-comonad and a codomain-preserving Ord-monad inducing the same Ord-functorial factorisation , the following conditions are equivalent:
- (i)
* is a lofs.* 2. (ii)
Both and are lax idempotent. 3. (iii)
One of and is lax idempotent and the distributive law axioms (a) hold.
2.4. Lifting operations
Diagram (b) shows that every functorial factorisation system induces a canonical lifting operation from the forgetful Ord-functor - to the forgetful Ord-functor -, meaning that every commutative diagram
[TABLE]
has a canonical diagonal filler so that , . Those fillers respect both composition and order in a natural way (see [15] for details).
A lifting operation from to is said to be kz if, for every commutative diagram (c) and every diagonal filler , one has .
Theorem** ([15]).**
For an awfs on an Ord-category C, the following conditions are equivalent:
- (i)
* is a lofs.* 2. (ii)
The lifting operation from -* to - is kz.*
2.5. Simple monads and their lofss
The notion of simple monad we present here, studied in [14, 15], is the Ord-enriched version of simple reflection of [8]. In an -enriched category with comma-objects, given an Ord-monad , we construct a monad on by considering the comma-object and defining as the second projection. Then is the unique morphism making the following diagram commute.
[TABLE]
The Ord-functorial factorisation defines a copointed endo--functor , with , and a pointed endo--functor , with . Moreover, underlies a monad on whose multiplication is defined by the unique morphism given by the universal property of the comma-object:
[TABLE]
(See [15] for details.)
Lemma**.**
Given a monad on , the following conditions are equivalent for a morphism in C:
- (i)
* has an -coalgebra structure.* 2. (ii)
* is a lari (=left adjoint right inverse111We use the terminology introduced by J.W. Gray in [21].), that is, it has a right adjoint such that .*
Proof.
(i)(ii): If is an -coalgebra structure for , then is a left inverse of :
[TABLE]
and, moreover, it is right adjoint to :
[TABLE]
(ii)(i): Let be a right adjoint left inverse of . By definition of comma-object, from there exists a unique such that and . To conclude that , compose with and :
[TABLE]
∎
A morphism in C such that is a lari is called an -embedding. Denote by - the category that has as objects pairs of morphisms of C such that with , and as morphisms morphisms in such that .
Definition**.**
The Ord-monad is said to be simple if the locally monotone forgetful functor - has a right adjoint and the induced comonad has underlying functor and counit .
As shown in [15]:
Proposition**.**
A lax idempotent monad on C is simple if, and only if, for every morphism , there is an adjunction
[TABLE]
Theorem**.**
If is a lax idempotent and simple monad, then is a lofs. Moreover, the left class of the weak factorisation system it induces is the class of -embeddings.
Proof (Sketch of the proof; for details see [15])..
Simplicity of gives the comonad structure for needed to define the awfs.
In order to show that is a lax idempotent monad, that is, , we denote by and note that, by definition of , and . Then
[TABLE]
by simplicity of . Now, by definition of comma-object and by the equalities and , it follows that as claimed.
The last assertion follows from the lemma above. ∎
2.6. Submonads of simple monads
Well-behaved submonads of simple monads are simple, as stated below.
Theorem** ([15]).**
Suppose that is a monad morphism between Ord-monads whose components are pullback-stable -embeddings, and that -embeddings are full. If is lax idempotent, then is simple whenever is so. Moreover, every -embedding is an -embedding.
(Here by full morphism in an Ord-category we mean a morphism such that, for every , implies .)
3. -categories
3.1. The setting
First we describe the setting where we will be working throughout the paper.
A. is a commutative and unital quantale, that is, a complete lattice equipped with a tensor product , with unit , and with right adjoint . We denote by the bicategory of -relations, having sets as objects, while morphisms are -relations, i.e. maps ; their composition is given by relational composition, that is, for and ,
[TABLE]
Every map is a -relation with if and elsewhere. This correspondence defines a bijective on objects and faithful pseudofunctor . is a locally ordered and locally complete bicategory, with if , for , , . It has an involution assigning to each the -relation defined by . For each both left and right compositions with preserve suprema, and therefore we have the following adjunctions
[TABLE]
so that, for every , , , ,
[TABLE]
B. is a non-trivial -monad that satisfies (BC); that is, preserves weak pullbacks and every naturality square of is a weak pullback. We point out that, in particular, the monad is taut in the sense of Manes [31] (see [13] for details).
C. is a -algebra structure on such that both and , , are -algebra homomorphisms, that is, the following diagrams
[TABLE]
are commutative, and, for all maps , and with for every , the following inequality holds
[TABLE]
for every . (For alternative descriptions of the latter condition see [22].)
D. Using we define, for each -relation , the -relation as the composite
[TABLE]
that is, for each , ,
[TABLE]
This defines a pseudofunctor that extends , so that is a natural transformation while is an op-lax natural transformation (see [22] for details).
3.2. -categories
Having fixed these data, a -category is a pair , where is a set and is a -relation such that
[TABLE]
Given -categories , , a -functor is a map such that
[TABLE]
We denote the category of -categories and -functors by . As defined in [16, Section 12], is (pre)order-enriched by:
[TABLE]
for . (This structure is in fact inherited from the order-enrichment of as explained in 3.5.) Identifying an element of with the -functor , , becomes (pre)ordered; is called separated, or skeletal, if, for , and implies . The category of separated -categories and -functors will be denoted by .
Examples**.**
Let be the identity monad and the identity map.
- •
When , -Cat is the category of (pre)ordered sets and monotone maps.
- •
Let be the complete half-real line ordered by the greater or equal relation, with and the truncated minus, so that , which is equal to if and [math] otherwise. As shown by Lawvere in [28], -Cat is the category of generalised metric spaces and non-expansive maps.
Let be the ultrafilter monad and be defined by .
- •
When – as shown by Barr in [3] – -Cat is the category of topological spaces and continuous maps.
- •
When – as shown in [10] – -Cat is the category of approach spaces and non-expansive maps [30].
3.3. The dual of a -category
When is the identity monad, is the category of -categories and -functors. In there is a natural notion of dual category, inducing a functor , with . To build a dual for a -category we first note that the -monad can be extended to , with , and make use of the following adjunction
[TABLE]
where, for a -category , a -algebra structure and a -homomorphism , and ; and, for a -category and a -functor , and . The functor lifts to a functor , with , and we define the dual of a -category as
[TABLE]
; that is, denoting its structure by ,
[TABLE]
for and (see [9]).
3.4. as a -category
As we have in both a -categorical structure and a -algebra structure , which is a -functor due to our assumptions, is a -category; this structure has a crucial role in our study, as we will see in the next section.
3.5. -bimodules
Given -categories and , a -bimodule (or simply a bimodule) is a -relation such that and , where the composition of two -relations and is given by the Kleisli convolution (see [25]), that is
[TABLE]
Under our assumptions bimodules compose, with the -categorical structures as identities for this composition. We denote by the category of -categories and -bimodules. is locally preordered by the preorder inherited from .
Every -functor induces a pair of bimodules and , defined by and ; that is, and , for , , and . The Kleisli convolution becomes simpler when composing with these bimodules: for any and , and . It is easy to check that and , that is, . The -functor is said to be fully faithful when , or, equivalently, , for every . The local (pre)order on corresponds to the local (pre)order on : for -functors ,
[TABLE]
4. The presheaf monad and its submonads
4.1. The Yoneda Lemma
The tensor product in defines a tensor product in , with , where , for , , . Its neutral element is . For each -category , the functor has a right adjoint .
Proposition** ([12]).**
For -categories and a -relation , the following conditions are equivalent:
- (i)
* is a bimodule;* 2. (ii)
* is a -functor.*
Since is a bimodule, this result tells us that is a -functor, and therefore, from the adjunction , induces the Yoneda -functor
[TABLE]
The following result provides a Yoneda Lemma for -categories.
Theorem** ([12]).**
Let be a -category. For all and all ,
[TABLE]
where denotes the -categorical structure on . In particular, is fully faithful.
4.2. The presheaf monad
In order to work in an -enriched category, from now on we restrict ourselves to -. We remark that the results of the previous subsection remain valid when we replace -Cat by -. Denoting by , we point out that, via Theorem Proposition,
[TABLE]
Moreover, the Yoneda -functor turns out to assign to each , that is, to each -functor , the bimodule . Each -functor induces a -functor , assigning to the bimodule , that is . This defines an endofunctor on . From the adjunction , for every -functor one gets a right adjoint to , . In particular, has a right adjoint , which, together with and , defines a lax idempotent monad, the presheaf monad. Next we show that this monad is simple. In order to do that we use Proposition 2.5.
Theorem**.**
The presheaf monad on is simple.
Proof.
We need to show that, for any -functor , in the diagram below .
[TABLE]
First we recall that the comma object is given by , and , where is the structure on . On one hand, as we observed before, has as right adjoint the -functor . On the other hand, . Next we will show that , which concludes the proof. For each and ,
[TABLE]
while
[TABLE]
To show that , so that , note that
[TABLE]
∎
Proposition**.**
- (1)
A -functor is a -embedding if, and only if, it is fully faithful. 2. (2)
Fully faithful -functors are pullback stable.
Proof.
- (1)
If is a -functor, then has a right adjoint, . It remains to show that when is fully faithful; this means , and so, for any bimodule ,
[TABLE]
Conversely, if , then, for any ,
[TABLE]
that is, is fully faithful. 2. (2)
As in any topological category, (bijective, fully faithful -functors) is an orthogonal factorisation system in -Cat, and therefore fully faithful -functors are pullback-stable. ∎
4.3. Presheaf submonads
Let be a class of -bimodules satisfying the conditions:
- (S1)
is closed under composition. 2. (S2)
For every -functor , . 3. (S3)
For every -bimodule , provided that for every .
We call such a class saturated. There is a largest saturated class, of all -bimodules, and a smallest one, \{f^{*}\,|\,f\mbox{ is a (\mathbb{T},V)-functor}\}. In the last section we will present other saturated classes.
For each -category , we define
[TABLE]
equipped with the structure inherited from , and, to each -functor we assign
[TABLE]
Since for every , corestricts to ,
[TABLE]
Moreover, condition (S3) guarantees that (co)restricts to : by the Yoneda Lemma, for all , . So, is a submonad of .
Theorem**.**
If is a saturated class of bimodules, then the monad is lax idempotent and simple, and so it defines a lax orthogonal factorisation system.
Proof.
Since fully faithful -functors are pullback-stable and full, and the inclusion is clearly fully faithful, this result follows directly from Theorem 2.6. ∎
5. Examples: The induced lofss
5.1. General description
Now let us fix a saturated class of -bimodules as in 4.3. The presheaf submonad induces a lofs , and consequently a wfs where is the class of -embeddings.
Following [12], we say that a -functor is -dense if .
Lemma** ([12]).**
For a -functor , the following conditions are equivalent:
- (i)
* is -dense;* 2. (ii)
* is a left adjoint;* 3. (iii)
* is -dense.*
We note that has a right adjoint if and only if the right adjoint of can be (co)restricted to , which is the case precisely when .
Proposition**.**
For a -functor , the following conditions are equivalent:
- (i)
* belongs to ;* 2. (ii)
* is fully faithful and -dense.*
Proof.
(i)(ii): From Theorem Theorem we know that a -embedding is fully faithful, and, by definition, is a left adjoint. (ii)(i): If is -dense, then has a right adjoint , and so it remains to show that, when , : since for every , the proof follows the arguments used in Proposition Proposition(1). ∎
Corollary**.**
For every -category , is a -embedding.
The class is the class of morphisms with the right lifting property with respect to morphisms in ; that is, a morphism belongs to if, and only if, it is injective with respect to the class . Since is a lofs, these morphisms have the kz-lifting property with respect to morphisms in . Such morphisms encompass interesting properties, but are usually very difficult to identify.
5.2. Examples: the lari–opfibration lofs
When
[TABLE]
then , , is an isomorphism; that is, the monad is isomorphic to the identity monad. Therefore the corresponding lofs is the one studied in [15, Examples 4.7, 4.8], and the monad is the free (split) opfibration monad on -. Then is the class of laris and the class of split opfibrations.
5.3. Examples: the presheaf lofs
Let us now take the largest saturated class
[TABLE]
From Theorem 2.5 we know that the presheaf monad defines a lofs in -, and, consequently, a wfs , where is the class of fully faithful -functors. It is easy to check that they coincide with extremal monomorphisms in -, that is, topological embeddings. Therefore, from Theorem 2.5 we conclude that, for every quantale and monad in the conditions of 3.1, - has a wfs where is the class of embeddings.
When is the identity monad and , that is, in the category of (anti-symmetric) ordered sets and monotone maps, the morphisms in were characterised by Adámek (as mentioned in [35]), as those monotone maps which are fibre-complete, fibrations and co-fibrations (among some other characterisations; see also [1]).
When is the identity monad and , that is, in the category of separated generalised metric spaces and non-expansive maps, a characterisation of the morphisms in is not known. It follows from [2] that a (non-expansive) map belongs to if, and only if, is an hyperconvex metric space (see also [26]).
When and , that is, in the category of T0-spaces and continuous maps, morphisms in were studied in a series of papers by Cagliari and Mantovani (see [7] and references there), and characterised in [5, 6]; in [5] they are identified via a way-below relation while [6] gives characterisations that extend those of mentioned above.
5.4. Examples: the Lawvere lofs
The choice of
[TABLE]
has particular relevance.
When and , the injective objects in with respect to -embeddings are the Cauchy-complete metric spaces, that is, a non-expansive map belongs to if and only if is Cauchy-complete (see [28, 12, 11]). Therefore, the morphisms in are good candidates for a fibrewise notion of Cauchy-completeness. This lofs was studied in [15]. The morphisms in are the embeddings (=isometries) such that, for every , for some Cauchy sequence in . We point out that the non-expansive maps in do not coincide with Sozubek’s L-complete maps [34]. Indeed, Sozubek defines them via an injective property, but his left part – the so called -equivalences – is a proper subclass of .
When and , this choice of gives also an interesting lofs. As shown in [12], the -algebras for this monad in are the sober spaces. Since sober spaces are also the algebras for the lax idempotent and simple monad of completely prime filters on , the wfs induced by coincides with the wfs studied in [5]; that is, is the class of completely flat embeddings and is the class of fibrewise sober continuous maps (see [17, 18, 33]).
5.5. Further examples
Using the techniques of [17, 18] and [11, 3.7], one can define saturated classes of -bimodules and so that the left parts of the corresponding wfs are dense embeddings and flat embeddings. The simple presheaf submonads they define induce lofs whose underlying wfs were studied in [5], where and give the fibrewise notions of Scott domains and stably compact spaces.
The reference list from the paper itself. Each links out to its DOI / PubMed record.
- 1[1] J. Adámek, H. Herrlich, J. Rosický, W. Tholen, Weak factorization systems and topological functors. Appl. Categ. Structures 10 (2002) 237–249.
- 2[2] N. Aronszajn, P. Panitchpakdi, Extension of uniformly continuous transformations and hyperconvex metric spaces. Pacific J. Math. 6 (1956) 405–439.
- 3[3] M. Barr, Relational algebras. In: Lecture Notes in Math. 137, Springer, Berlin 1970, pp. 39–55.
- 4[4] J. Bourke, R. Garner, Algebraic weak factorisation systems: I. J. Pure Appl. Algebra 220 (2016) 108–147.
- 5[5] F. Cagliari, M.M. Clementino, S. Mantovani, Fibrewise injectivity and Kock-Zöberlein monads. J. Pure Appl. Algebra 216 (2012) 2411–2424.
- 6[6] F. Cagliari, M.M. Clementino, S. Mantovani, Fibrewise injectivity in order and topology. Topology Appl. 200 (2016) 61–78.
- 7[7] F. Cagliari, S. Mantovani, Injectivity and sections. J. Pure Appl. Algebra 204 (2006) 79–89.
- 8[8] C. Cassidy, M. Hébert, G.M. Kelly, Reflective subcategories, localizations and factorization systems. J. Aust. Math. Soc. Ser. A 38 (1985) 287–329.
