Category Free Category Theory and Its

arXiv:1602.01759v1 [[Link]] 4 Feb 2016

Philosophical Implications
Michael Heller∗
Copernicus Center for Interdisciplinary Studies
ul. Slawkowska 17, 31-016 Cracow, Poland
February 5, 2016

Abstract
There exists a dispute in philosophy, going back at least to Leibniz,
whether is it possible to view the world as a network of relations and
relations between relations with the role of objects, between which
these relations hold, entirely eliminated. Category theory seems to be
the correct mathematical theory for clarifying conceptual possibilities
in this respect. In this theory, objects acquire their identity either
by definition, when in defining category we postulate the existence of
objects, or formally by the existence of identity morphisms. We show
that it is perfectly possible to get rid of the identity of objects by
definition, but the formal identity of objects remains as an essential
element of the theory. This can be achieved by defining category ex-
clusively in terms of morphisms and identity morphisms (objectless,
or object free, category) and, analogously, by defining category the-
ory entirely in terms of functors and identity functors (categoryless,
or category free, category theory). With objects and categories elim-
inated, we focus on the “philosophy of arrows” and the roles various
identities play in it (identities as such, identities up to isomorphism,
identities up to natural isomorphism...). This perspective elucides a
contrast between “set ontology” and “categorical ontology”.
∗
Correspondence address: ul. Powstańców Warszawy 13/94, 33-110 Tarnów, Poland.
E-mail: mheller@[Link]

1
1 Introduction
There exists a tendency in philosophy, going back at least to Leibniz, to
view the universe as a network of relations and relations between relations
with the role of objects, between which these relations hold, minimalysed
or even reduced to null. If in Leibniz such a view should be possibly re-
stricted only to space and time1 , thinkers were not lacking who espoused this
standpoint in its full extent. It is interesting to notice that one of Leibniz’s
arguments against Newton’s idea of absolute space drew on his outspoken
Principle of the Identity of Indiscernibles which excludes the possibility for
two things being distinct that would not be different with respect to at least
one discernible property. Imagine now the world moved by some distance in
absolute space. Two such “configurations” (before and after the world has
been moved) would be indiscernible even for God. Therefore, the concept
of absolute space should be ruled out as contradicting the Principle of the
Identity of Indiscernibles. Essentially the same reasoning leads, according to
Leibniz, to rejection of absolute time. The concepts of identity and that of
relatedness of everything with everything are strictly interwoven with each
other. To see the interconnectedness of these concepts, let us introduce the
following terminology. There is a sense in which “we can talk about this
object, whatever properties it may have, as opposed to that object, whatever
properties this second object may have”[10, pp. 16-17]. In this sense, we
speak about the primitive thisness of this object. If this is not the case, i.e.
if the identity of the object depends on its properties related to other objects,
we can talk about the induced or contextual thisness of this object. If the
identity of every object is reduced only to the induced thisness, we can speak
about the complete relatedness of the system of objects. On the contrary, if
the identity of every object is reduced only to its primitive thisness, we could
speak about an “absolute” system in which relations between objects play
no role. Of course, “mixed situations” are also possible.
Category theory seems to be the correct mathematical theory for clarify-
ing conceptual possibilities in this domain (see, for instance, [7]). It seems
natural to identify objects of the preceding paragraph with objects of cat-
egory theory, and relations of the preceding paragraph with morphisms of
category theory, or objects with categories themselves and relations with
functors between categories (in this paper we do not consider n-categories).
1
It is debatable whether it can be extended to his monadology.

2
Is there a possibility to get rid of objects entirely and, in this way, to imple-
ment the idea of the complete relatedness of the system? To pose correctly
this question in the context of category theory, we must first look for the
identity of objects. In category theory objects acquire their identity on two
levels: first, by definition, when in defining category we postulate the ex-
istence of objects; second, formally by the existence of identity morphisms;
we can speak about the “identity by definition” and the “formal identity”,
respectively. In the following, we show that it is perfectly possible to get
rid of the identity of objects by definition, but the formal identity of objects
remains as an essential element of the theory.

2 Objectless Category Theory

In the present section, we briefly present, following [8, pp. 44-46] (see also [1,
pp. 41-43]), a formulation of category theory in terms of morphisms without
explicitly mentioning objects. As we shall see, this formulation is equivalent
to the standard one.

Definition 2.1 Let us axiomatically define a theory which we shall call an

objectless or object free category theory. In this theory, the only primitive
concepts (besides the usual logical concepts and the equality concept) are:
(I) α is a morphism,
(II) the composition αβ is defined and is equal to γ,
The following axioms are assumed:

1. Associativity of compositions: Let α, β, γ be morphisms. If the compo-

sitions βα and γβ exist, then

• the compositions γ(βα) and (γβ)α exist and are equal;

• if γ(βα) exists, then γβ exists, and if (γβ)α exists then βα exists.

2. Existence of identities: For every morphism α there exist morphisms ι

and ι′ , called identities, such that

• βι = β whenever βι is defined (and analogously for ι′ ),

• ιγ = γ whenever ιγ is defined (and analogously for ι′ ).
• αι and ι′ α are defined.

3
Lemma 2.1 Identities ι and ι′ of axiom (2) are uniquely determined by the
morphism α.

Proof. Let us prove the uniqueness for ι (for ι′ the proof goes analogously).
Let ι1 and ι2 be identities, and αι1 and αι2 exist. Then αι1 = α and (αι1 )ι2 =
αι2 . From axiom (1) it follows that ι1 ι2 is determined. But ι1 ι2 exists if an
only if ι1 = ι2 . Indeed, let us assume that ι1 ι2 exist then ι1 = ι1 ι2 = ι2 . And
vice versa, assume that ι1 = ι2 . Then from axiom (2) it follows that there
exists an identity ι such that ιι1 exists, and hence is equal to ι (because ι1 is
an identity). This, in turn, means that (ιι1 )ι2 exists, because (ιι1 )ι2 = ιι2 =
ιι1 = ι. Therefore, ι1 ι2 exists by Axiom 1. ✷
Let us denote by d(α) and c(α) identities that are uniquely determined
by a morphism α, i.e. such that the compositions αd(α) and c(α)α exist
(letters d and c come from “domain” and “codomain”, respectively).

Lemma 2.2 The composition βα exists if and only if c(α) = d(β), and
consequently,
d(βα) = d(α) and c(βα) = c(β).

Proof. Let c(α) = d(β) = ι, then βι and ια exist. From axiom (1) it
follows that there exists the composition (βι)α = βα. Let us now assume
that βα exists, and let us put ι = c(α). Then ια exists which implies that
βα = β(ια) = (βι)α. Since βι exists then d(β) = ι. ✷

Definition 2.2 If for any two identities ι1 and ι2 the class

hι1 , ι2 i = {α : d(α) = ι1 , c(α) = ι2 },

is a set then objectless category theory is called small.

Definition 2.3 Let us choose a class C of morphisms of the objectless cat-

egory theory (i.e. C is a model of the objectless category theory), and let
C 0 denote the class of all identities of C. If ι1 , ι2 , ι3 ∈ C 0 , we define the
composition
0
mCι1 ,ι2 ,ι3 : hι1 , ι2 i × hι2 , ι3 i → hι1 , ι3 i
0
by mC (α, β) = βα. Class C is called objectless category.

Proposition 1 The objectless category definition (2.3) is equivalent to the

standard definition of category.

4
Proof. To prove the theorem it is enough to reformulate the standard category
definition in the following way. A category C consists of
(I) a collection C 0 of objects,
(II) for each A, B ∈ C 0 ,
a collection hA, BiC 0 of morphisms from A to B,
(III) for each A, B, C ∈ C 0 , if α ∈ hA, BiC 0 and β ∈ hB, CiC 0 , the
composition
0
mC : hA, BiC 0 × hB, CiC 0 → hA, CiC 0
0
is defined by mC
A,B,C (α, β). The following axioms are assumed

1. Associativity: If α ∈ hA, BiC 0 , β ∈ hB, CiC 0 , γ ∈ hC, DiC 0 then

γ(βα) = (γβ)α.

2. Identities: For every B ∈ C 0 there exists a morphism ιB ∈ hB, BiC 0

such that
∀A∈C 0 ∀α∈hA,Bi 0 ιB α = α,
C
∀C∈C 0 ∀β∈hB,Ci 0 βιB = β.
C
To see the equivalence of the two definitions it is enough to suitably replace
in the above definition objects by their corresponding identities. ✷
This theorem creates three possibilities to look at the category theory:
(1) the standard way, in terms of objects and morphisms, (2) the objectless
way, in terms of morphisms only, (3) the hybrid way in which we take into
account the existence of objects but, if necessary or useful, we regard them as
identity morphisms. The hybrid way could be useful in some interpretative
issues.

3 Category Theory without Categories

It is almost trivial to see that the above move towards getting rid of objects
can be repeated on the level of categories themselves, i.e. towards formulating
category theory entirely in terms of functors. The idea would be the same
as above – to substitute identity functors for categories. What we need first,
is the objectless functor definition, but it is straightforward

5
Definition 3.1 Let C and D be objectless categories (definition (2.3)). The
objectless covariant (contravariant) functor Φ : C → D is an assignment of
morphisms of C to the morphisms of D in such a way that the compositions
and the identities are preserved, i.e if α, β are morphisms of C then
Φ(αβ) = Φ(α)Φ(β),
(Φ(αβ) = Φ(β)Φ(α)),
and if ι, ι′ ∈ C 0 then Φ(ι) and Φ(ι′ ) are identities associated with Φ(α).
Since functors can be composed and there exist identity functors, it is
straightforward to repeat the strategy of the preceding section and define
“categoryless, or category free, category theory”. The obviousness of the
definition allows us to do this in a sketchy way.
Definition 3.2 The only primitive concepts of the category theory without
categories are:
(I) Φ is a functor,
(II) the composition of functors ΦΨ is defined and is equal to Λ.
The axioms of the associativity of functor composition and of the existence
of functor identities (analogous to axioms (1) and (2) of definition (1.1)) are
assumed.
The results analogous to that of Lemma 2.1 and 2.2 are clearly true. Let
D(Φ) and C(Φ) be identity functors such that the compositions ΦD(Φ) and
C(Φ)Φ exist. Then the composition ΨΦ exists if and only if C(Φ) = D(Ψ).
The above results are almost trivial from the formal point of view, but
philosophically they are quite remarkable: the very concept of category is
not indispensable in developing the theory itself. This remains in agreement
with the practice of Eilenberg and Mac Lane who, in their seminal paper [4],
regarded categories as auxiliary constructs necessary only to ensure domains
and codomains for morphisms.

4 Contextual Thisness
What has been achieved by introducing objectless categories and category
theory without categories?2 Let us focus on objects (analogous things can
2
It is interesting to notice that William Lawvere in his seminal doctor thesis conse-
quently eliminated objects with the help of identity morphisms [6].

6
be said about categories). As we have seen, in the objectless category the-
ory, objects do not acquire their individuality by definition, but only formally
through the identity morphisms. Instead postulating the existence of objects,
one postulates the existence of morphisms which, when composed with other
morphisms, change nothing. One cannot speak here about the primitive
thisness in the sense described in Section 1, but only about the contextual
thisness, and the context is now given by those morphisms with which a given
identity morphism composes (to change nothing). I propose to call it com-
positional thisness. This is the closest (but still faraway) from the primitive
thisness (as it can be defined in set theory) that can be obtained in category
theory. Let us notice that the principle of indiscernibility of identities, in its
original form, does not hold for this kind of thisness – two identities could
be discernible through their compositions with different morphisms. We can,
therefore, speak of discernibility through compositions.
The elimination of objects is possible not only on the formal level, but
also on the level of logical language. Let us consider an example. For some
purposes it is important to define the concept of category in a first-order
language. The most obvious of such languages is the one in which there are
two sorts of variables: one ranging over objects and another over arrows.
However, the language becomes simpler and more elegant if we abolish this
dichotomy by adopting objectless category approach and replace objects by
their corresponding identity arrows. This requires some toil but is certainly
more in the spirit of category theory (see [5, 232-234]).

5 Individuality of Categories
In mathematics we are usually interested in the individuality of entities “up
to isomorphism”. In this sense, mathematical entities change their individu-
ality depending on the context. For instance, in topology two homeomorphic
spaces are regarded as the same (homeomorphism is a topological isomor-
phism), but in differential geometry two diffeomorphic differential manifolds
are regarded as the same (diffeomorphism is an isomorphism of differential
manifolds). The concept of isomorphism has its categorical counterpart.
Let X and Y be objects in a category C. A morphism f : X → Y is said
to be isomorphism if there is a morphism g : Y → X such that g ◦ f = ιX
and f ◦ g = ιY . Since g is necessarily unique, we may write g = f −1 . As we
can see, this is fully compatible with the identification of objects in terms of

7
identity morphisms: the composition of the defining morphisms with their
respective inverses must give the identity arrow. We are entitled to say that
the context with respect to which the thisness of an object is determined is
now broadened to the extent the nature of mathematical reasoning demands:
thisness is not determined by the identity arrow of a given object alone, but
with the help of identities corresponding to suitable compositions (g ◦ f and
f ◦ g). I shall call it thisness up to isomorphism.
The isomorphism concept refers also to categories. A functor F : C → D
from a category C to a category D is said to be an isomorphism, if it has
the inverse G : D → C such that G ◦ F = idC and F ◦ G = idD (here we
assume that idC ∈ C 0 and idD ∈ D0 ). In such a case, we say that the
categories C and D are isomorphic, and we write C ∼ = D. It is not difficult to
see that what identity morphisms are for objects, the identity functors are for
categories. It turns out, however, that the concept of contextual thisness as
applied to categories (their isomorphism) is too rigid from the point of view
of category theory. Two categories can be equal (remaining in a bijective
correspondence), they can be unequal but isomorphic, or they can be not
even isomorphic but nevertheless equivalent from the categorical point of
view. To define the latter notion, we must first introduce the concept of
naturality.
Let us consider two functors F and G from a category C to a category D,
F : C → D and G : C → D. A natural transformation from the functor F to
the functor G is an assignment τ that associates every object X of C with the
arrow τX : F (X) → G(X) of D in such a way that for any arrow f : X → Y
of C one has τY ◦F (f ) = G(f )◦τX (to see this clearly draw the corresponding
commutative diagram). The arrows τX are said to be the components of τ .
We could imagine the functors F and G as giving two pictures of the category
C within the category D. If these two pictures are “faithful” to the original
and between themselves the transformation between the functors F and G
is natural. If every component τX of τ is an isomorphism in D, then τ is
said to be a natural isomorphism. If this is the case, two above mentioned
pictures of the category C within the category D are not only “faithful” but
also “exact”.
Let us now go back to the equivalence of categories from the categorical
point of view. A functor F : C → D is said to be an equivalence of categories,
if there is a functor G : D → C such that there exist natural isomorphisms
τ : idC ∼ = G ◦ F and σ : idD ∼ = F ◦ G; in such a case, we say that the
categories C and D are equivalent and write C ≃ D. In tis context, I shall

8
speak about thisness up to equivalence.
The notion of the equivalence of categories is clearly more tolerant than
that of the isomorphism of categories, in the sense that isomorphic categories
are equivalent but not vice versa. The relationship between these two notions
can be made more transparent in the following way. A category C is said to be
skeletal if its isomorphic objects are identical, i.e. A ∼
= B iff A = B where A
and B are objects of C. In other words, for skeletal categories, “isomorphic”
means the same as “is equal”, that is to say primitive thisness means the
same as contextual thisness. A full subcategory3 C 0 of a category C is called
a skeleton of C if C 0 is skeletal and each object of C is isomorphic to the
exactly one object of C 0 . Every two skeletons of a category are isomorphic.
A skeleton C 0 of C is clearly equivalent to C, C 0 ≃ C, and if categories
have the same skeleton, they are equivalent4 .
Finally, let us quote a few examples. The category of all finite sets and
all functions between them has the subcategory of all finite ordinal numbers
as its skeleton. The category of all well ordered sets with suitable order pre-
serving morphisms has the subcategory of all ordinal numbers as its skeleton.
The category of all vector spaces over a fixed field K and K-linear transfor-
mations has the subcategory of all K n , for n any cardinal number, as its
skeleton.
It is interesting to notice that if the axiom of choice holds for a category
C then it has a skeleton. This statement can be proved in the following way.
The relation of isomorphism on the collection of objects of a category C is an
equivalence relation. Let us choose one object from each equivalence class.
The full subcategory of C formed from the chosen elements is a skeleton of
C.

6 The Arrow Philosophy

Categories are complex entities. They are composed of objects and arrows
subject to a few axiomatic exigencies which, although rather simple, lead to
a rich structural variety. Since, as we have seen, objects can be eliminated,
3
If C 0 is a subcategory of C then every object of C 0 is an object of C, and a subcategory
C 0 of a category C is a full subcategory of C, if C has no arrows other than the ones already
present in C 0 .
4
All mathematical facts quoted in this section are standard in category theory; see for
instance [5, pp. 200-202].

9
everything finally reduces to arrows. Often the “philosophy of arrows” is
contrasted with the “philosophy of elements” (each of them with its primitive
thisness) and it is said that they determine two different “ontologies” for
mathematics. The set ontology is of reductionist type: in it everything is
reduced to “being an element of”, typified by the functor “∈”. An element
either belongs to a set, or does not belong to it which is a symptom of the
two-valued classical logic. The categorical ontology is of referential type with
arrows, as fundamental entities, indicating an action or reference. Being an
element of an object is replaced by an arrow pointing to a given object. Such
an arrow could be understood as denoting an action that consists of picking
up something (an “element”) in this object. Moreover, this picking up can be
done with various degrees of “intensity” depending on the domain of a given
arrow: with maximal intensity (for sure) if the domain is the terminal object
of the category, and with lesser intensity (with lesser certainty) otherwise.
Something similar happens on the level of categories with functors playing
the role of arrows. The above “arrow strategy” works also at this level. As
we have seen, categories could be entirely eliminated with identity functors
effectively replacing them.
At the bottom of the “new perspective”, exploited in the present work,
lies the concept of individuality. In the “set theoretical ontology”, the indi-
viduality of a set element is given by its very “thisness”, independently of
its relationship to the environment. In the “ontology” determined by the
strategy of arrows, one can also speak on thisnesses of some structures, but
it is always contextual from the very beginning. As we have seen, it strongly
depends on its environment. What we have called compositional thisness
depends on arrows with which the corresponding arrow composes to change
nothing. Two antiparallel arrows (i.e. arrows pointing in two opposite di-
rections) can in various ways interact with each other to produce identities.
If they compose to produce just identity, they are said to define an isomor-
phism, and we have, in such circumstances, agreed to speak about thisness
up to isomorphism. If they compose to produce identity in a natural way,
we have agreed to speak about thisness up to equivalence.
To this list of producing identities “of various degrees” we can add some
more. Let us consider the pair of functors between categories C and D:
F : D → C and G : C → D, and two natural transformations

ǫ : F G → IdC ,

10
and
η : IdD → GF.
With the help of these data one defines the adjunction between categories C
and D; F is called the left adjoint to G, and G is called the right adjoint to F .5
ǫ and η are then called the counit and the unit of this adjunction, respectively.
Here we have also an interaction between natural transformations producing
suitable unit and counit. Adjunction can be regarded as a generalisation of
the equivalence of categories [2, p. 209]. This time unit and counit arise from
interaction between functors rather than between categories. Adjunction
turns out to be ubiquitous in mathematics; it unveils intimate relationships
between sometimes very faraway departments of mathematics. In this sense,
we could speak on non-local context dependence.
This “strategy of units” tells us something about the nature of the “field
of categories”; it is drastically unlike a collection of sets. For the sake of
concreteness let us specialise to toposes, and let us consider a pair of adjoint
functors f ∗ : D → C and f∗ : C → D such that f ∗ is left adjoint to f∗ , and f ∗ is
also left exact6 . Such functors define an admissible transformation between
two toposes, which in this context are called frameworks (the name being
clearly motivated by an analogy with reference frame used in physics)7 . The
shifting through an admissible transformation from one framework to another
framework can lead to a shift in understanding a mathematical concept. Here
is an example. Let us consider the concept of real-valued continuous function
on a topological space X. Such a function can be interpreted, in the topos S
of constant sets, as a continuously varying real number. Let us now shift, via
the admissible transformation, from S to the topos Shv(X) of sheaves over
X. The concept of real function in S shifts to the concept of real number
as it is interpreted in Shv(X). J. L. Bell [3] describes the situation in the
following way:

...shifting to Shv(X) from S essentially amounts to placing one-

self in a framework which is, so to speak, itself ‘moving along’
with the variation over X of the given variable real numbers.
This causes the variation of any variable real number not to be
5
For the definition see, for example, [2, chapter 9] or in a very accessible way [9, chapter
5].
6
A left exact functor is a functor that preserves finite limits.
7
More precisely, the name “framework” is used for any topos having the natural num-
bers object.

11
 ‘noticed’ in Shv(X); it is accordingly there regarded as being
constant real number.

This is only the illustration of the main Bell’s idea that “the absolute uni-
verse of sets [should] be relinquished in favour of a plurality of local math-
ematical frameworks”. The interpretation of any mathematical concept is
not fixed, but changes with the change of a local framework. It should be
emphasised that units and counits (defining adjointness of admissible trans-
formations) play here the key role. The concept of identity seems to lie at
the core of the categorical architecture.

References
[1] Adámek, J., Herrlich, H., Strecker, G., Abstract and Concrete Categories.
The Joy of Cats, [Link]/[Link] (originally pub-
lished by Wiley and Sons, New York, 1990).

[2] Awodey, S., Category Theory, second edition, Oxford University Press,
2011.

[3] Bell, I.L., “From Absolute to Local Mathematics”, Synthese 69, 1986,
409-426.

[4] Eilenberg, S., Mac Lane, S., A General Theory of Natural Equiva-
lences,Trans. Amer. Math. Soc. 58, 1945, 231-294.

[5] Goldblatt, R., Topoi. The Categorical Analysis of Logic, Revised Edition,
Dover, Mineola, 1984.

[6] Lawvere, F.W., Functorial Semantics of Algebraic Theries and Some Al-
gebraic Problems in the Context of Functorial Semantics of Algebraic
Theories, [Link]/tac/reprints/articles/5/[Link]

[7] Rodin, A., 2005, Identity and Categorification, Philosoophia Scientiae

11, 2007, 27-65; arXiv:math/0509596 [math CT].

[8] Semadeni, Z., Wiweger, A., 1978, Wstȩp do teorii kategorii i funktorw
(Introduction the the theory of categories and functors), second edition,
PWN, Warszawa, 1978 (in Polish).

12
[9] H. Simmons, An Introduction to Category Theory, Cambridge University
Press, 2011.

[10] Teller, P., An Interpretative Introduction to Quantum Field Theory,

Princeton University Press, Princeton, 1995.

13