A Visual Presentation of Rank-Ordered Sets

Year: 1998 Authors: Mara Alagić

Core claim

Rank-ordered sets with rank-preserving and rank-increasing functions form a cartesian closed categorical framework useful for interpreting programming entities.

Topics

category theory, rank-ordered sets, partial equivalence relations, cartesian closed category

Domains

category theory, order theory, domain theory, scientific visualization

Methods

categorical abstraction, inverse-limit construction, diagrammatic proof, definition-driven exposition

Media

commutative diagrams, set-theoretic notation, category diagrams

Paper text

The text below is the locally extracted OCR/Markdown version of the paper. Raw PDF files remain local and are not published here.

BRIDGES

Mathematical Connections

in Art, Music, and Science

A Visual Presentation of Rank-Ordered Sets

Mara Alagić Department of Mathematics and Statistics Wichita State University, Wichita 67226 - 0033 e-mail: mara@math.twsu.edu

Abstract

We are exploring two cartesian closed categories, rich with structure: rank-ordered sets, and its subcategory kinds. Categorical language is used for a conceptual clarification and imagination of these structures.

Motivation

The paper originated in an attempt to get the mathematical and conceptual understanding of some programming entities ([5], [1], [2]). The main contribution of this paper is a self-contained view to a specific cartesian closed category - rank ordered sets and rank-preserving functions. The fact is that this category can be also considered as a subcategory of the category of bounded ultrametrics with nonexpansive maps (see [5],[3]). That fact describes its geometric structure.

In the first section we present rank-ordered sets with rank-preserving functions starting from a category of sets, Set. The collection of rank-preserving functions is a rank-ordered set. The same is true for naturally defined product of two rank-ordered sets. Therefore we have in our hands a cartesian closed category of rank-ordered sets with rank-preserving functions.

In the second section two categories and of partial equivalence relations (per) defined on cpo’s (complete partial orders) are presented. Special collections of partial equivalence relations over certain domain are shown to be rank-ordered sets. Domain is introduced as a limit object of an inverse limit system in and each per in this collection is the limit of a sequence of pers of the approximating domains . Family of commutative diagrams proving this illustrates precisely corresponding relationships ([3]).

Imposing conditions that guarantee closure under limits and therefore induced cpo structure, and requiring that each per can be seen as a limit of a certain sequence of pers gives so-called acceptable collection of per’s over . The collection of rank-increasing functions has the same abstract structure as an acceptable collection of nice per’s ([4]).

The smallest subcategory of rank-ordered sets which contains a collection of acceptable pers and is closed under products and exponents is presented in Section 3. It is cartesian closed category, called category of kinds in order to suggest more complex structure than one of types.

238 Mara Alagic

1. Rank-ordered Sets and Rank-preserving Functions

In this section we have a close look at objects, rank-ordered sets and corresponding arrows, rank-preserving functions of a specific category.

We start with the category of sets and maps between sets, Set.

A set is rank-ordered if there is a collection of maps , , that assigns to each element from a sequence of elements (each from ) satisfying the following conditions:

(i) assigns to any the same element, denoted by . Thus, for all .

(ii) for all .

(iii) Let . If is a sequence from with the property then there is a unique such that for all , .

Let . If for some , then the rank of is defined by . If for all we say . If is finite and we say that approximates , denoted by .

In the next section we will see the construction of a specific rank-ordered set.

The arrows between rank-ordered sets are defined according to the following diagram (for all and all ):

A function is rank-preserving function if for all and all ,

The collection of rank-preserving functions from to , , is a rank-ordered set. and is defined by . Note that is a rank-ordered set.

Closure property of rank-preserving functions: If is rank-preserving function then

Thus, is rank-preserving iff it is rank-preserving in each of its arguments, separately.

> The category of rank-ordered sets with rank-preserving functions is a cartesian-closed category.

A Visual Presentation of Rank-Ordered Sets 239

1.1. Rank-increasing Functions.

The advantage of rank-increasing functions is that each endo-function has a unique fixed point. Also, the fixed point of a multivariable function that is rank-increasing in one argument retains its rank-related properties in other arguments.

A function is rank-increasing function if for all and all ,

The collection of rank-increasing functions from to is denoted by .

Properties of rank-increasing and rank-preserving functions:

  1. Every rank-increasing function is also rank-preserving function.
  2. The collection of rank-increasing functions from to is a rank-ordered set.
  3. Composition of a rank-preserving and a rank-increasing (or vice versa) functions is a rank-increasing function.
  4. If is rank-increasing function then
  1. Let and be rank-preserving functions. If one of them is rank-increasing then the composition is also rank-increasing.
  2. is rank-increasing (rank-preserving) iff it is rank-increasing (rank-preserving) in each of its arguments, separately.

Only rank-increasing subset of rank-preserving functions have unique fixed points. (The identity map from per’s to per’s is rank preserving but not rank-increasing function.) In addition, the fixed point of a multi-argument function that is rank-increasing in one argument retains its rank-related properties in other arguments.

2. Partial Equivalence Relations

The cartesian closed category of rank-ordered sets with rank-preserving functions provides an environment space for the construction of specific rank-ordered set - partial equivalence relations over a certain desired domain.

An complete partial order is a partial order (reflexive, antisymmetric, and transitive) with a least element , and such that for every countable directed collection .

There are two classes of morphisms that can be introduced between complete partial orders: continuous functions and embedding-projection pairs. Both of them give rise to a certain category.

240 Mara Alagić

A map is continuous iff for all directed where and is in . Note also that continuous maps on cpo’s are always monotonic.

An embedding-projection pair between cpo’s and consists of maps and such that and . Each determines the other, assuming both exist, so it suffices to name the embedding or the projection.

Let denotes the category of complete partial orders with continuous maps and let be the category of complete partial orders with projections as maps between them.

Example 1 Let denotes a set of counting numbers together with usual ordering and 1 as the initial object. Therefore, and (pointwise ordering) can be considered as ‘s. Given embedding , the corresponding projection is defined by , where now and .

Facts.

(i) The category CPO is a cartesian closed category: the singleton cpo is terminal object; is ordered coordinate-wise, and for each there exists a unique satisfying the adjointness condition.

(ii) Every has a fixed point.

(iii) Inverse limits exist in . For an inverse system , inverse limit, is the poset with coordinate-wise ordering, and with .

A new categories and may be defined in the following manner:

Let be a cpo. A relation is a partial equivalence relation, per, if it is symmetric and transitive relation on .

Example 2 On a set (or or ) there are binary relations, but only 14 pers:

A Visual Presentation of Rank-Ordered Sets 241

Objects of are pairs where is a cpo and is a partial equivalence relation, per, (symmetric and transitive) which is closed under sup’s of -chains.

Morphism in is given with continuous which respects the relation: .

is the subcategory of with the same objects and with arrows projection maps.

To construct a specific rank ordered set, we start with a continuous functor and construct a cpo as the inverse limit of the following -diagram.

Let (the one element cpo) be the terminal object of and the unique projection from to the terminal object.

A collection forms an inverse system:

Denote by

By this construction each is isomorphic to some subdomain , with projections . For let us denote by so that each is the limit of the -chain .

. So, ( continuous).

An example of such functor is where is some cpo of atomic values, is countably infinite cpo and is coalesced sum.

A continuous functor extends continuous functor if the following conditions are satisfied:

, ; ;

so that we may write shortly

The following shows that each per is the limit of a sequence of pers of the approximating domains . A family of commutative diagrams used to prove this illustrates precisely corresponding relationships among , and each per constructed over .

Let extends . Then the sequence in is a sequence of relations over in and, furthermore

242 Mara Alagić

is a relation over .

Consider the following family of commutative diagrams

“Fixed points” of each continuous F extending H are obtainable as limits of a diagram consisting of relations over ‘s. Note that if F extends H, the action of F on morphisms is completely determined by H. So, for fixed H, object maps on are maps from per’s to per’s.

Approximations ( is a rank ordered set.). A family of continuous mappings defined by for satisfies the following properties:

  1. Let . Then

if then .

  1. Let and where is the -th projection. Then

2.1. Acceptable collection of per’s over

(Construction of a rank-ordered set of per’s.)

We now consider partial equivalence relations over . Specifically, let us define an acceptable collection of per’s over imposing the following conditions.

A collection of per’s over is an acceptable collection subject to the following conditions:

(p1) For all .

(p2) If \{(d_i,d_i')\mid i > 0\} is an -chain in with limit then .

(p3) iff for all .

(a1) contains the per .

A Visual Presentation of Rank-Ordered Sets 243

(a2) is closed under function space, i.e. implies .

(a3) is closed under F-bounded quantification.

(a4) If \{R_i \mid i < \omega\} \subseteq \mathcal{R} satisfies the property that for all , , (where denotes the restriction of to the subdomain ) then there is a unique such that for all , .

Conditions (p1), (p2), and (p3) describe nice per. The first two conditions give closure under limits and the third is the requirement that each per can be seen as a certain limit of a sequence of pers.

The set is not only closed under function space but it also contains unique sups of increasing chains of per’s.

Conditions (p1), (p2), and (p3) allow the definition of a cpo structure on .

A notion of rank on a per from (even on a nice per) is given by the following definition.

For let . If for some , . If for all , . If is finite and for some , we say approximates , .

and it satisfies the properties (p1), (p2), and (p3) of an acceptable collection.

A collection of nice pers has a cpo structure.

It is enough to notice that the least upper bound of an -chain with is the relation . Therefore, each in is the limit of the corresponding ‘s.

An acceptable collection of per’s, , is a rank-ordered set because

(i)

(ii) By construction of as a for some functor that extends , we have for all .

(iii) .

3. CCC of Kinds

The kind structure generated from is the smallest subcategory of the category of rank-ordered sets which contains and is closed under products and exponents. A kind structure is a cartesian closed category.

The elements of each kind may be ordered in two ways:

For types we consider the subtype ordering,

For let iff .

For type functions we have the induced pointwise ordering:

Assuming and are defined for , we define

iff for all , where , and

Mara Alagic

iff and where , .

For types we may also consider the rank ordering (induced by approximations of rank-ordered sets) defined as follows

For iff rank(R)<\infty and .

The induced ordering for type functions is

iff for all where .

4. Summary

In this paper we have described a cartesian closed category of rank-ordered sets and its two subcategories of partial equivalence relations defined on a very specific inverse limit object as its domain. Although rank-preserving functions are convenient arrows for rank-ordered sets, the advantage of rank-increasing functions is the existence of a unique fixed point in this case.

Categories and of partial equivalence relations defined on cpo’s (complete partial orders) are an important step in this construction. Special collections of partial equivalence relations over certain domain are rank-ordered sets. Domain is a limit object of an inverse limit system in and each per in this collection is the limit of a sequence of pers of the approximating domains . These partial equivalence relations obtained by the inverse-limit construction over suitable domain structure are used as models interpreting types by many authors (M.Coppo, Amadio, Cardelli, Abadi & Plotkin).

Elements of the domain can be viewed according to the limit construction of and therefore some very natural properties follow. These are also properties of approximations on cpo’s introduced in [4, Cardelli]. Imposing conditions that guarantee closure under limits and therefore induced cpo structure, and requiring that each per can be seen as a limit of a certain sequence of pers gives an acceptable collection of nice per’s ([3]).

The smallest subcategory of rank-ordered sets which contains a collection of acceptable pers and is closed under products and exponents is category of kinds. It suggests more complex structure then one of types - both of them convenient for interpreting certain programing entities. The elements of each kind may be ordered.

References

  • [1] M.Alagić. Relations with I-structure in categories with pullbacks.

Publications de l’Institut Mathematique, Nouvelle serie, Belgrade 50(64), pp. 355–365. 1992.

  • [2] S. Alagić and M. Alagić. Order-sorted model theory for temporal executable specifications. Theoretical Computer Science 179, pp.273–299, 1997.
  • [3] K. Bruce and J. C. Mitchell. Per Model of Subtyping, Recursive Types and Higher–order Polymorphism. Proceedings of POPL-92, pp.316–327. 1992.
  • [4] L. Cardelli. A Semantics of Multiple Inheritance. Information and Computation 76, pp 138–164, 1988.
  • [5] M.Abadi and G.D.Plotkin. A per Model of Polymorphism and Recursive Types, Proc. IEEE Symp. on Logic in Computer Science, pp. 355–365, 1990.

0 items under this folder.