Non-unital polygraphs form a presheaf categories

Simon Henry
November 3, 2017
arXiv:1711.00744v1 [math.CT] 2 Nov 2017

Abstract
We prove, as claimed by A.Carboni and P.T.Johnstone, that the cat-
egory of non-unital polygraphs, i.e. polygraphs where the source and
target of each generator are not identity arrows, is a presheaf category.
More generally we develop a new criterion for proving that certain classes
of polygraphs are presheaf categories. This criterion also applies to the
larger class of polygraphs where only the source of each generator is not
an identity, and to the class of many-to-one polygraphs, producing a
new, more direct, proof that this is a presheaf category. The criterion it-
self seems to be extendable to more general type of operads over possibly
different combinatorics, but we leave this question for future work.
In an appendix we explain why this result is relevant if one wants to fix
the arguments of a famous paper of M.Kapranov and V.Voevodsky and
makes them into a proof of C.Simpsons semi-strictification conjecture.
We briefly present a program aiming at proving this conjecture.

Contents

1 Polygraphic preliminaries 2
1.1 -categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Polygraphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Classes of polygraphs . . . . . . . . . . . . . . . . . . . . . . . . . 7

2 Good class of polygraphs and pasting diagrams 8

2.1 Familially representable functors and presheaf categories . . . . . 10
2.2 Good class of polygraphs . . . . . . . . . . . . . . . . . . . . . . 14
2.3 Plexes and polyplexes in dimension n + 1 . . . . . . . . . . . . . 19
2.4 Automorphisms of polyplexes and the case of source-positive poly-
graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

A On C.Simpsons conjecture and the Kapranov-Voevodsky strat-

egy 35

Introduction
Some time ago I started analyzing in details the famous false proof by M.Kapranov
and V.Voevodsky in [12] that homotopy types can be represented by strict -
categories whose arrows are weakly invertible, hopping to prove the conjecture

1
by C.Simpson in [20] that this false proof can be made into a correct proof that
every homotopy types can be represented as an -groupoid whose associativity
and exchange rule holds strictly, but units and inverse are weak. This analysis
lead me to the conclusion that the main problem with [12] was in the choice
of the category of higher categorical diagrams they are using. Constructing a
category of diagrams having the correct properties for this proof to work, ap-
pears to be closely related to another problem: showing that certain classes of
polygraphs are presheaf categories. The present paper is devoted to this second
problem, but we have included in an appendix A a presentation of our ideas on
why we think the arguments of M.Kapranov and V.Voevodsky fails and how the
results of the present paper can rescue their arguments and maybe lead to proof
of C.Simpson conjecture. We also explain precisely while we have not succeeded
in proving this conjecture yet and what remains to be done. This appendix can
somehow be consider as an introduction to the present paper in the sense that
it explains its main motivations, but the paper can be read completely indepen-
dently of it and the results of the paper are interesting on their own, especially
for the theory of polygraphs. Moreover, around the end of the appendix some
understanding of the theory developed in the present paper will be needed to
follow the discussion.

The main result of this paper is that the category of polygraphs where the
source and targets of each generator is not an identity cell is a presheaf category.
This was claimed without a proof by P.T.Johnstone and A.Carboni in [6] after
they noted that the category of all polygraphs was not a presheaf category
contrary to what they claimed in their previous paper [5]. We prove more
generally that the category of polygraphs where the source (resp. the target)
of each generator is not an identity cell is a presheaf category, and we prove
that subclasses of polygraphs (in the sense of definition 1.3.1) of this class are
all presheaf categories. This in particular gives the first direct proof that the
category of many-to-one polygraphs is a presheaf category (which is known
because there is a rather indirect proof that it is equivalent to the category of
opetopic sets following from [9] and [8]). We postpone the introduction of the
main ideas involved in that proof to the beginning of section 2, after a short
introduction to strict -categories and polygraphs in section 1.

1 Polygraphic preliminaries
1.1 -categories
1.1.1. We will use R.Street one type definition of -category:

Definition : An -category is a set X together with unary operations sk and

tk for k > 0 and partially defined binary composition operations #n for n > 0
satisfying the following axioms:

1. x#n y is defined if and only if tn (x) = sn (y).

2. For every x X there exists an n such that sn (x) = tn (x) = x.

2
 3. For any x X, one has:

an x if n < m
an b m x =
bm x if n > m
where a and b are either s or t.
4. For any x X one has sn (x)#n x = x#n tn (x) = x.
5. For all x, y such that the composition below are defined,

sn (x#n y) = sn (x)
tn (x#n y) = tn (y)
And if k > n:

tk (x#n y) = tk (x)#n tk (y)

sk (x#n y) = sk (x)#n sk (y)
6. x#n (y#n z) = (x#n y)#n z when either side is defined.
7. If k < n

(x#n y)#k (z#n w) = (x#k z)#n (y#k w)

when the left hand side is defined.
A morphism of -categories is just a function commuting to all the structural
function (source, target and the partial composition).

A few remarks in order to clarify the intended meaning of the definition and to
relate it to the more intuitive globular definition:
In terms of the globular definition of -categories, the underlying set X
corresponds to the increasing union of the set of n-arrows for all n, where
n-arrows are seen as specific (n + 1)-arrows by identifying them with their
identities endomorphisms.
sn (x) and tn (x) denotes respectively the n-dimensional source and tar-
get of x, i.e. the first iterated source/target of x which is an n-arrow.
In particular the set of n-arrows is exactly the set of x X such that
sn (x) = x (or equivalently tn (x) = x) and if x is an n-arrow then its usual
source and target are the (n 1)-arrow sn1 (x) and tn1 (x).
The operation x#n y corresponds to the composition of two arrows along a
common boundary of dimension n, note that for -categories we use the
reverse or diagrammatic composition order: x#n y is defined when
the n-target of x is the n-source of y. The difference of conventions for
composition order between -categories and all other categories will not
be a problem as we will always consider strict -categories as just an
algebraic structure of interest, and not as actual categories (for example
equivalences of -categories will play no role in the present paper, only
isomorphisms).

3
 Note that this operation x#n y allows to compose arrow of different di-
mensions. In terms of the globular definition, if x is of dimension lower
than y then x#n y corresponds to the n-composition of y with the iter-
ated identity of x of same dimension as y. This features will become very
important when we discuss non-unital -categories in A.5.
We have not given rules for the value of sk (x#n y) and tk (x#n y) when
k < n but it follows from the axiom for sn (x#n y) that sk (x#n y) =
sk (x) = sk (y) when k < n and with the same formula for the target.

1.1.2. If X is an -category, elements of X will be called the arrows of X. An

n-arrow of X is an arrow x X such that sn (x) = x or equivalently tn (x) = x.
For any arrow x, the n-source and n-target sn (x) and tn (x) are n-arrows. Any
n-arrow can also be seen as an m-arrow for m > n and these corresponds to
identity arrows. The dimension of an arrow x is the smallest n such that x is an
n-arrow. In particular an arrow of dimension n is not the same as an n-arrow:
an arrow of dimension n is exactly a non-identity n-arrow.
Two n-arrows x and y are said to be parallel if sn1 x = sn1 y and tn1 x =
tn1 y. An (n + 1)-arrow z is said to be from x to y, written z : x y if sn z = x
and tn z = y, this implies that x and y are parallel.

1.2 Polygraphs
Polygraphs were first introduced under the name computads by R.Street in
[22] in the framework of 2-categories. The general n-categorical notion of com-
putad is hinted at, also by R.Street, in [23] and as far as I know first appears
explicitly in print in [19]. The name of polygraphs is due to A.Burroni, when
he re-introduced the notion independently in [4].

In this subsection we will just review the definition and basic property of poly-
graphs, without any proof.

The definition of polygraph is done by induction on the dimension, together

with the definition of the free -category X on a polygraph X.

1.2.1. Definition :
A 0-polygraph is just a set X. The free -category on a 0-polygraph X is
the discrete category with only 0-dimensional arrows given by the elements
of X. The elements of the set X are called the 0-cells of X.
An (n + 1)-polygraph X is the data of an n-polygraph Xn together with a
set of so-called (n + 1)-cells of X and two maps sn and tn which attach
to every (n + 1)-cell a pair of parallel n-arrows in (Xn ) . The free -
category generated by an (n + 1)-polygraph is the -category obtained by
freely adding an (n + 1)-arrow to (Xn ) for each (n + 1)-cell of X (with

4
 the source and target given by sn and tn ), i.e. it is defined by the following
universal property:
A morphism f from X to another -category Z is the data of a morphism
fn : (Xn ) Z together with the choice of an (n + 1)-arrow f (x) in Z
for all (n + 1)-cell x of X such that sn (f (x)) = fn (sn (x)) and tn (f (x)) =
fn (tn (x)).
A polygraph X is a tower of, for each n N, an n-polygraphs Xn where for
all n the underlying n-polygraph of Xn+1 is Xn . The free infinity category
X generated by a polygraph X is defined as the inductive limit of the
-categories Xn .

For example, a 1-polygraph is just an ordinary oriented graph, and the free
-category generated by it is the usual free category generated by a graph, i.e.
the category of paths in the graph. In a 2-polygraph one has additionally some
2-cells, each 2-cell being a 2-arrow between two (oriented) path in the graph
with the same sources and targets.
An n-polygraph will always be considered as an (n + 1)-polygraph with no
(n + 1)-cells, and more generally, as a polygraph with no k-cell for k > n. This
is compatible with the construction of the free -category.
The notion of morphisms of polygraphs is also defined by induction such that
the free -category construction become a functor: a morphism of 0-polygraphs
is just a function between sets (which can be seen as a functor between the
corresponding -categories). A morphism f between two (n + 1)-polygraphs
is the data of a morphism fn between their underlying n-polygraphs together
with a function f between their sets of (n + 1)-cells which are compatible to the
maps sn and tn : sn (f (x)) = fn (sn (x)) and tn (f (x)) = fn (tn (x)). Such a data
clearly induces a morphism between the corresponding free -categories.
This gives a category of n-polygraphs, and more generally a category of poly-
graphs. The free -category construction is a functor.
The category of all polygraphs is denoted by P, the (full) subcategory of n-
polygraphs by Pn .

1.2.2. If X is a polygraph, each n-cell x of X defines an n-arrow of X , also

denoted x. It is in fact an n-dimensional arrow of X as sn1 (x) and tn1 (x)
are elements of Xn1 and so cannot be equal to x.
These specific arrows of X will be called the generators of X . A generator of
X of dimension n is exactly the same as a n-cell of X. We want to avoid as
much as we can to use the word cell for anything related to X to avoid the
confusion between cells of X and the possible use of this word for the arrows of
an n-category.

In general, the set of all arrows of X admit a description as the set of expressions
formed using the composition operations ( #k ) and the cells of X, which are
well formed (syntactically) and well defined (i.e. such that the composition
involved are well defined) up to an equivalence relation generated by the axioms
of -categories. A precise formulation of this statement can be found in [18] or
in [14]. We will not really use this explicitly.

5
1.2.3. What will be more important for us are the following simpler properties,
which are all easy consequences of the above claim, but can all be proved more
directly:

The generators of X , i.e. the cells of X, are exactly the arrows that
cannot be written in a non-trivial way as a composite in X. In particular,
if an -category is free over a polygraph, then the polygraph is uniquely
determined up to unique isomorphism. Morphisms of polygraphs from X
to Y are the same as morphisms of -categories f : X Y which send
any generator of X to a generator of Y . In particular, any isomorphism
between X and Y comes from an isomorphism of polygraphs between
X and Y .
One can prove properties by induction on the arrows of a polygraph: If
some property P of arrows of X is such that P (a) holds for all the
generator of X and if for any x, y such that P (x) and P (y) hold and
x#k y is defined one also have P (x#k y) then P (x) holds for all arrow x of
X .
If a is an n-arrow of X and b is an n-cell of X, the number of times b
appears in any expression of a is a well defined number which does not
depend on the expression of a in terms of the generators. Moreover a is an
identity n-arrow (i.e. an (n1)-arrow) if and only if none of the generators
of dimension n appears in a. This can be deduced from the remark above
and the fact that none of the axioms of -category change that number of
appearances, or more directly using the linearization functor (sending
an -category to a chain complex it generates) which implement this
counting function (see subsection 4.3 of [16]). Composition is additive
with respect to this counting function, i.e. the number of times a appears
in x#k y is the sum of the number of time it appears in x and in y. It does
not seem possible to define such a number when the dimension of a and x
are different (although a counting function of this kind is defined in [14],
but it does not really corresponds to a number of appearances).

So to some extend, polygraphs are just particular -category, which are free.
In fact it has been proved by F.Metayer in [16] that polygraphs are exactly the
cofibrant objects of the folk model structure on -categories.
If X and Y are polygraphs a morphism from X to Y will be said to be
polygraphic if it send generators to generators. As mentioned above, polygraphic
morphisms are exactly the same thing as morphisms of polygraphs, and are in
general more restrictive than general morphisms of -category between X and
Y .

1.2.4. The functor X X which send a polygraph to the free -category it

generates has a right adjoint G : Cat P. This right adjoint produces
rather complicated polygraph, indeed if X is an -category then G(X), together
with the co-unit of adjunction G(X) X is constructed inductively as follow:

The 0-cells of G(X) are exactly the 0-arrows of X.

6
 Once G(X)n and the map f : G(X)n X are constructed. An (n+1)-cell
of G(X) is given by a triple of: a pair of parallel n-arrows s,t in (G(x)n ) ,
and an (n+1)-arrow in X between f (s) and f (t). The source and target of
the triple (s, t, x) are given by s and t, and f is extended by f (s, t, x) = x.

In particular, there is a terminal object in the category of polygraphs denoted

P1, and given by G() where denotes the terminal -category (which has
only one arrow). And it is a rather rich object: it has one 0-cell, one 1-cell and
then one n-cell for each pair of parallel (n 1)-arrows in (P1) . For example
P1 has a 2-cell An,m from the n-times composite to the m-times composite of
its unique 1-cell for each pair of integers n, m > 0. This object will play an
extremely important role in the theory of good class of polygraphs developed in
section 2.
More details about this adjunction, for example its monadicity can be found in
[17].

1.2.5. Finally, a morphism of polygraphs is a monomorphism if and only if it is

injective on cells, an epimorphism if and only if it is surjective on cells, and the
category of polygraphs admit epi-mono factorization, i.e. every morphism of
polygraphs factors uniquely as an epimorphism followed by a monomorphism,
and this is achieve by the corresponding factorization on the sets of n-cells. This
is proved for example in [14], see 5.(9).

1.3 Classes of polygraphs

1.3.1. Definition : A class of polygraphs, is a full subcategory J of the category

of polygraphs which has the following properties:

If X J and there exists an arrow f : Y X then Y J.

If Y J and there exists an epimorphism Y X then X J.
`
If Xi J for all i I, iI Xi J

It seems that last two conditions can be replaced by saying that J is stable
under all colimits, but this would require a more detailed investigation of limits
and colimits in the category of polygraphs that we would prefer to avoid.

1.3.2. Some examples of this notions:

The category P of all polygraphs.

The category Pn of polygraphs of dimension at most n.
The category G of globular polygraphs, i.e. the polygraphs where both
the source and target of every cell is a cell.

7
 The category OP of opetopic polygraphs which are those polygraphs for
which the target of every cell is again a cell (and not an arbitrary arrow).
The category P+ of positive polygraphs, i.e. the polygraphs such that
the source and the target of each cell are non-identity arrows.
The categories Ps+ and Pt+ respectively of source-positive and target-
positive polygraphs which are the polygraphs such that the source (re-
spectively the target) of each cells is a non-identity arrow.

1.3.3. The intersection of two classes of polygraphs is again a class of poly-

graphs, for example P+ = Ps+ Pt+ . In particular if J is a class of polygraphs
one defines Jn as J Pn , i.e. the class of polygraphs in J of dimension at most
n.

1.3.4. Theorem : Every class of polygraphs J admit a terminal object denoted

J1. Moreover, J1 is a sub-polygraph of the terminal polygraph P1, and J is
exactly the class of all polygraph X whose unique map to P1 factors into J1.
This induces a correspondence between classes of polygraphs and sub-polygraphs
of P1.

Proof :
For any polygraph V J the image (in the sense of the epi-mono factorization)
of the unique map V P1 is an element of J. The set of all sub-polygraphs
of P1 which belongs to J is stable under small union (because J is stable under
small co-product and epimorphic image), hence it has a maximal element J1.
As J1 is an element of J, any object whose unique map to P1 factor into J1 is
in J, and conversely, for any object X J, its image in P1 is a subpolygraph
of P1 in J and hence is included in J1.
This makes J1 the terminal object of J, and if W P1 is any sub-polygraph
of P1 then the category of polygraphs over W is a class of polygraphs whose
terminal object is W , and this proves the correspondence. 

2 Good class of polygraphs and pasting diagrams

The overhaul goal of this section (and in fact of this paper) is to provide tools to
show that certain classes of polygraphs are presheaf categories and to use these
tools to prove that the categories of positive, source-positive or target-positive
polygraphs are all presheaf categories.
For example, the category of 1-polygraphs, i.e. of ordinary oriented graph is
clearly a presheaf category, and more generally, the category G of globular poly-
graphs is equivalent to the category of globular sets and is a presheaf category.
It has been known for a long time that the category of 2-polygraphs is also a
presheaf category, this is originally due to S.Schanuel (unpublished) and first

8
appears in [5]. At some point, some people were lead to believe that the category
of all polygraphs was itself a presheaf category (this was erroneously claimed
in [5] and generalized in [1]) but this was proved to be false in [15] and with
a more direct argument in [7]. There is also a rather large class of polygraphs
which is known to be a presheaf category: the class OP of opetopic polygraphs,
or many-to-one polygraphs is equivalent to the category of opetopic sets, and
hence is a presheaf category, but the proof of this is rather indirect: this fol-
lows from [9] which proves that opetopic sets are the same thing as something
called multitopic sets and [8] which proves that these multitopic sets are them-
selves the same thing as many-to-one polygraphs. One could also show that
Streets orientals ([23]) generates a class of polygraphs which is equivalent to
the category of semi-simplicial sets, and hence is also a presheaf category.
Also the work of M.Batanin in [2] gives a criterion to show that certain analogue
of the notion of polygraphs for more general globular operads are presheaf cate-
gories, it applies in particular to the category of polygraphs for Gray categories,
or the category of polygraphs for any cellular globular operads. This criterion
of M.Batanin is in fact very similar to the one we develop in the present pa-
per, but cannot be used on the new examples of positive polygraphs we study
here. The relation between the two criterion is discussed in 2.4.9. Of course,
M.Batanin has developed his criterion in a more general framework of globular
operads while we work specifically with the operads for strict -categories, but,
as we will explained at the end of this introduction, we are convinced that the
methods used here can be generalized to encompass M.Batanin work, and we
hope to come back to this in a future work.

Our general strategy is as follows: One first introduces in section 2.2 a notion
of good class of polygraphs (see definition 2.2.4), a good class is essentially a
class which is a presheaf category and which admits a good notion of pasting
diagrams such that any arrow in the free -category X generated by a given
polygraph X in that class can be represented uniquely by a polygraphic map
from one of these pasting diagrams to X

Following a terminology suggested to us by A.Burroni (that he used in [3]), we

will call a representable object of a good class a plex ( M.Makkai also used the
name Computope in [14]) and a pasting diagram will be called a polyplex.

Section 2.1 is an introduction to the ideas of A.Carboni and P.T.Johnstone from

[5] that we will use constantly in the rest of the section. They essentially show
how from a good notion of pasting diagram in dimension n one can prove that
one has a presheaf category in dimension n + 1. We illustrate this idea in 2.1.7
on a well known example: the fact that there is good notion of pasting diagram
for the free category on a graph and that this allows to see that 2-polygraphs
form a presheaf categories. The way P.T.Johnstone and A.Carboni arrived from
there, in [5], at a false claim that the category of all polygraphs was a presheaf
categories was by believing that such a notion of pasting diagrams should exists
in all dimensions, while it is in fact false in general, exactly because of the
Eckmann-Hilton argument, and is an extremely subtle issue in restricted cases.

The pivotal results of this paper are those of section 2.3 and more precisely
theorem 2.3.9 which show that if some class of polygraphs is good up to dimen-
sion n, then in dimension n + 1 it is still a presheaf category and one can still

9
make sense of a notion of polyplex for this class. Those (n + 1)-dimensional
polyplexes can still fails to represent uniquely the arrows of a free -category
but only because they might have automorphisms which preserves the arrow
they represents, and the main results (theorem 2.3.9) is that J will still be a
good class in dimension n + 1 if and only if there is no such automorphisms of
its (n + 1)-polyplexes.

Somehow a polyplex can be thought of as encoding a generic composition

operations that one can perform in an -category, its underlying polygraph
being its arity, and these automorphisms of a polyplex preserving the arrow
it represents that we mentioned can be thought of as a permutation of the
variables of this composition operation that preserves the results, i.e. a form of
commutativity of the operation. This is exactly what happens in the situation
of the Eckmann-Hilton argument where the composition of two endomorphisms
of an identity cell happen to be commutative and this prevents the class of all 2-
polygraphs to be such a good class of polygraphs, and the class of 3-polygraphs
to even be a presheaf category.

Finally, in section 2.4 one investigates in more details these automorphisms of

polyplexes and shows that they indeed do not exist if we restrict ourselves to
source-positive (or positive or target positive polygraphs) and this concludes
the proof of our claims that those form presheaf categories.

Finally, it is highly probable that under its present form our theorem 2.3.9
has no other applications than proving that the class of source/target-positive
polygraphs is a good class of polygraphs. But most of its proof does not really
use specific properties of the notion of -category and we strongly believe that
it should have some very important generalization to similar situation based,
for example, on different globular operads and even on different, non-globular,
combinatorics. We hope to soon be able to extend this result to any finitary
parametric right adjoint cartesian monad on a presheaf category over a direct
indexing category (possibly a directed category having some automorphisms),
using the extended notion of polygraphs proposed by M.Shulman1 .

2.1 Familially representable functors and presheaf cate-

gories
In this subsection we recall well known results about familially representable
functors, mostly coming from [5]2 , that will be the main tools toward proving
that certain class of polygraphs are presheaf categories. All the proofs of the
claims made below can be found in [5].

1 This idea of M.Shulman unfortunately only appears on the nLab, on the entry com-

putads, section on inverse diagrams at https://ncatlab.org/nlab/show/computad#on_

inverse_diagrams from revision 26.
2 note that this reference contains two important mistake that are explained and corrected

in [6].

10
2.1.1. Definition : A functor F : C Sets is said to be familially representable
if up to natural isomorphism it is of the form:
a
F (X) Hom(ci , X),
iI

for a set I and a family of objects ci C.

2.1.2. Proposition : The family (I, (ci )) defining a familially representable

functor F is uniquely determined (up to unique isomorphism) by the functor F .
Natural transformation between such functors correspond to morphisms in the
category of family F am(C op ), i.e. a morphism (I, (ci )) (J, (dj )) is a pair of:

a function f : I J,
an I-indexed collection of morphisms fi : df (i) ci .

Composition is simply given by compositions of functions between sets and com-

positions of the arrows in C.

2.1.3. Proposition : For a natural transformation : F G between famil-

ially representable functors F (I, ci ) and G (J, dj ) the following conditions
are equivalent:

The natural transformation is cartesian, i.e. for any morphism f : X

Y in C the naturality square of :

F (f )
F (X) F (Y )
X Y
G(f )
G(X) G(Y )

is a pullback square.

All the morphisms i : d(j) ci of the morphism of family corresponding

to are isomorphisms.

11
2.1.4. Proposition : If the category C has all finite co-limits, then the category
of familially representable functor on C is stable under finite limit in the category
of all functors from C to Sets. Moreover in terms of family, these co-limits can
be described as follow:

The terminal object is the family with only one object, which is initial in
C.
`
If (I, ci ) and (J, dj ) are family their products is given by (I J, ci dj ).
If one has two arrows (I, ci ) (K, ek ) and (J, dj ) (K, ek ) their fiber
product is given by:
!
a
I K J, ci dj
ek

where I K J denotes the ordinary fiber product of sets, i.e. the sets of
pair (i, j) I J which have the same images in K, and the letter k just
denote this common image in K.
If one has two arrows f, g : (I, ci ) (J, dj ) their equalizer is given by:


{i I|f (i) = g(i)}, coeq(fi , gi : df (i) ci

The reason familially representable functors are important in order to prove that
certain classes of polygraphs are presheaf categories is the following theorem
due to A.Carboni and P.Johnstone in [5] (theorem 4.1 and 4.3 of this reference)
which we only state in the restricted case that is of interest to us:

2.1.5. Theorem : Let E be a category with a terminal object. Let F : E Sets

be a functor. The comma category Sets/F of triples e E , X a set and
f : X F (e) where morphisms are the pair of morphisms e e and X X
which make the natural square commutes is a presheaf category if and only if E
is a presheaf category and F is familially representable.

2.1.6. One can be a little more precise about this theorem: Assume that E =
P rsh(C) is a presheaf category, and let:
a
F (X) = Hom(ai , X)
iI

be a familially representable functor from P rsh(C) to sets, represented by the

family of presheaves (ai )iI . Then it is not very hard to see that the comma
category Sets/F is equivalent to the category of presheaves over the small cat-
egory C + described as follow:

12
 C + contains C as a full subcategory and has in addition a set of new objects
(ai ) for i I.
The only morphisms in C + are the morphisms in C, the identities of the
ai and for each c C and i I one has that:

Hom(c, ai ) = ai (c)

composition with morphisms in C is given by the functoriality of the ai .

Moreover a presheaf P on C + corresponds to an object (e, X, f ) of Sets/F as

follows: the object e E = P rsh(C) is the restriction of P to C C + , the sets
X is:
a
X= P (ai )
iI

and an element (i, v P (ai )) is sent to F (e) by restricting it along the natural
map ai ai to get a morphism ai e, corresponding to an element of F (e).
Moreover we will only use this direction of the theorem of A.Carboni and
P.T.Johnstone. The converse direction (that if F is not familially representable
then Sets/F is not a presheaf category) is only important to us at a philosoph-
ical level to show that our approach is essentially optimal. It will not play any
concrete role in the present work.

2.1.7. As an example, we will use this to show that the category P2 of 2-

polygraphs is a presheaf category following the proof given in [5] (example 4.6).
We include this proof only because it is the simplest non-trivial case of the
general strategy we will use in this paper and we believe it is an illuminating
example.
A 2-polygraph is given by the data of a 1-polygraph, i.e. an ordinary graph,
together with a collection of 2-cells, each 2-cells being attached to a pair of
parallel 1-cells, hence the category of 2-polygraphs can be described as the
category Set/D where D is the functor from the category of graphs to the
category of sets, which maps a graph G to:

D(G) = {Pair of Parallel arrows in the free category G of path in G}

Now, the category P1 of 1-polygraphs (i.e. of graphs) is equivalent to a presheaf

category: the category with two objects P0 and P1 and with only two non-
identity arrows s, t : PO P1 . So in order to conclude we need to prove that
the functor D above is familially representable.
One starts with the simpler functor P which maps a graph G to the set of all
arrows of free category G of paths in G. It is familially representable, indeed
the set of path of length n is the same as Hom(Pn , G), with Pn being the graph:

Pn : a0 a1 an
hence:

13

a
P (G) = Hom(Pn , G)
n=0

Similarly, the functor D is familially represented by the following objects:

as1 ... asn1

An,m : a0 asn = atm

at1 ... atm1

each object An,m representing the functor sending a graph G to a pair of

parallel arrows (s, t) in G with s of length n and t of length m.
This concludes the proof that the category of 2-polygraphs is a presheaf category,
one moreover gets explicitly the category on which they are presheaves, its
objects are:

P0 the polygraph with only one 0-cell: .

P1 the polygraph with two 0-cell and one 1-cell between them:

for all integer n, m > 0, An,m the polygraph

as1 ... asn1

a0 t asn = atm

at1 ... atm1

The morphisms being the morphism of polygraphs between them, but more ex-
plicitly there is no non-identity morphisms between the An,m and the morphisms
from P0 and P1 to An,m are just the graph morphisms to An,m .

2.2 Good class of polygraphs

2.2.1. As pointed out by M.Makkai in [14] we do not just want a class of poly-
graphs J to be a presheaf category, we want it to be what he calls an effective
presheaf categories. An effective category is a category C endowed with a func-
tor F : C Set called a forgetful functor. It is common to ask additional

14
conditions on F , like being faithful or an isofibration, but those assumptions
plays no role here (although they will be satisfied on all the examples we will
encounter). Two effective categories are said to be equivalent if there is an
equivalence between them which commutes to the forgetful functor up to natu-
ral isomorphism.
The category of polygraphs is an effective category with the forgetful functor
mapping a polygraph X to the set of all its cells. Presheaf categories are effective
categories under the forget full functor:
a
F 7 F (c)
cC

Definition : One say that a class of polygraphs is an effective presheaf category

if it is equivalent as an effective category to a presheaf category.

More explicitly this means that there is a full subcategory I of the class J such
that the natural functor from J to P rsh(I) is an equivalence and such that the
functor from J to sets sending a polygraphs to the set of its cell is familially
representable by the family I:
a
{Cells of X} X(i)
iI

For example, P1 is obviously an effective presheaf category, and the discussion

in 2.1.7 shows that P2 is an effective presheaf category.

2.2.2. Definition : If a class of polygraphs J is an effective presheaf category,

a representable object is called a J-plex.

As mentioned above, this terminology is due to A.Burroni in [3]. They also have
been called Computopes by M.Makkai in [14] who defined them for a general
class of polygraphs.
The proposition below will show that to each cell of the terminal J-polygraph
J1 corresponds a J-plex which is unique up to unique isomorphism, and that
J-plexes are classified by the cells of J1.

2.2.3. Proposition : If a class of polygraphs J is an effective presheaf category,

with I the category of J-plexes, i.e. J P rsh(I) then:

1. There is a bijection between the J-plexes and the cells of J1.

2. For each cell x J1 the corresponding J-plex [x] I is characterized by
the following universal property:

Hom([x], X) = {a X|(a) = x where is the unique map : X J1 }

15
 3. If x is a cell of J1 of dimension n then the corresponding J-plex [x] has
only one cell of dimension n denoted x , no cells of dimension higher than
n, and a finite number of cells in total. Moreover [x] is generated by its
n-dimensional cell x in the sense that there is no strict sub-polygraphs
that contains it.
4. The category I of J-plexes is a directed category: every non identity mor-
phism goes from an object of dimension n to an object of dimension strictly
higher. Moreover all the slices of I are finite directed categories.
5. Any subclass of polygraphs of J is also an effective presheaf category.
Moreover if J J then the J -plexes are exactly the J-plexes that be-
longs to J .

One will often not distinguishes between a cell x of J1 and the plex [x] it
represents and denotes them both x. The cell x of [x] is called the universal
cell of x (or sometimes the top cell as it is the only cell of maximal dimension).
Proof :
1. One has that for each polygraph X, the set of cells of X is (isomorphic)
to:
a
Hom(i, X)
iI

In particular, taking X = J1, one has that for all i I Hom(i, J1) = {}
as J1 is terminal in J. And hence the set of cells of J1 can be written as:
a
{} = |I|
iI

2. For a general polygraph X in J, the unique map from X to J1 is at the

level of the sets of all cells:
a
Hom(i, X) |I|
iI

hence Hom(i, X) identifies with the set of cells of X which are sent to the
cells of J1 corresponding to i.
3. The identity map of a J-plex [x] corresponds (by its universal property)
to a cell x of [x] which is mapped to x in J1, hence is n-dimensional if
x has dimension n.
Let V be the smallest sub-polygraph of [x] which contains x , and let
i : V [x] be the inclusion map, because V contains x there is a map
p from [x] to V corresponding to this cell by the universal property of
[x], but the functoriality of the universal property gives immediately that
i j = Id[x] and i j maps x to x , so i j = IdV as V is the smallest
sub-polygraph containing x . This proves that [x] is generated by x in
the sense that every sub-polygraphs that contains x is equal to [x].
In particular [x] is finite, has no cells of dimension greater than n and x
is its only cell of dimension n.

16
 4. Morphisms from an object y I to a fixed object x I are in one to
one correspondence with the cells of x. So for all of them except one, y
have dimension strictly smaller than the dimension of x, and the last one
corresponds to the cell x , i.e. the identity of x. Moreover there is only a
finite number of them because x only have a finite number of cells, so the
slice category is finite.

5. If J J is a sub-class of polygraphs then we define I to be the full

subcategory of I of objects which correspond to cells x J 1 J1, those
are objects of J because their map to J1 factors into J 1. The category
I is downward closed in I (i.e. if f : y x with x I then y I )
hence presheaves over I can be identified with those presheaves F over I
such that x F (i) i I . But objects of J are exactly the objects of
J such that their map to J1 factor into J 1, hence it corresponds exactly
to presheaf over I satisfying the condition above.

2.2.4. As we mentioned in 2.1.5, a key property for proving that some class
of polygraphs is a presheaf category is that the Free -category functors
should be familially representable on this class. For this reason we will introduce
the following notion of good class of polygraphs, which will be our typical
induction hypothesis in a proof that a given class of polygraphs is a presheaf
category.

Definition : A class of polygraphs J is said to be a good class of polygraphs if:

J is an effective presheaf category.

The functor X 7 X from J to sets which maps a polygraph X to the set
of all arrows of the free -category X is familially representable.

2.2.5. Definition : If J is a good class of polygraphs, the objects that familially

represent X 7 X , i.e. the j such that:
a
X Hom(j, X)
Are called the J-polyplexes or J-pasting diagrams.

Here again, the terminology is due to A.Burroni in [3].

17
2.2.6. Proposition : If J is a good class of polygraphs then:

The J-polyplexes are parametrized by the arrow of (J1) .

The J-polyplex [v] corresponding to an arrow v (J1) is characterized
by the universal property:

Hom([v], X) = {x X |s.t. (x) = v}

where denotes the unique polygraphic map : X (J1) .

If J is any subclass of polygraphs of J, then J is a good class of polygraphs
and the J -polyplexes are exactly the J-polyplexes that belong to J .

Here again we will often not distinguish between a polyplex and the corre-
sponding arrow of (J1) , note that a polyplex v is always endowed with an
arrow v v which corresponds to the identity of v under the universal prop-
erty of v given in the proposition. Similarly to plexes, this arrows will be called
the universal arrow of v, or sometimes its top arrow. The universal property of
polyplexes can be rephrased as: if X is a J-polygraph and a X then there
is a unique polyplex v and map f : v X such that f v = a. Note that v
is part of the structure of being a polyplex in the sense that it is the choice
of this arrow that explain how the functor Hom(v, ) appears as a subfunctor
of ( ) . In fact, in 2.4.8 we will construct an example of two polyplexes in the
(good) class of positive 3-polygraphs which are isomorphic as polygraphs but
with different universal arrows, which makes them non isomorphic as polyplexes.
Finally, if a is a cell of J1, one can see it as an arrow of J1 and the polyplex
[a] is the same polygraph as the plex [a], with the universal cell/arrow a being
the same thing in both case.
Proof :
This is very similar to the proof of proposition 2.2.3:
By assumption, the arrows of (J1) are in bijection with the coproduct over the
J-polyplexes of the Hom(v, J1) = {}, hence this defines a bijection between
arrows of (J1) and the set of J-polyplexes. Moreover for any object X one has
: X J1 which corresponds to
a
Hom([v], X) (J1)
v(J1)

which immediately gives the universal property of [v] claimed in the proposition.
Finally, if J J is any subclass of polygraphs of J, then J is an effective
presheaf category by 2.2.3, and for any X J one has that:
a
X = Hom(v, X)
v a J-polyplex

but any J-polyplex admitting a morphism to X is in J so one has that:

a
X = Hom(v, X)
v a J -polyplex

18
where the J -polyplexes are defined as being the J-polyplexes that belongs to
J .


2.2.7. The functor X 7 X does not just take values in the category of sets,
but in the category of -category. In particular this functor is endowed with all
the structure of an -category, and this structure manifest itself as operations
on the J-polyplexes. At the level of indexing set of the family those operations
corresponds to the structure of -category on (J1) under the identification
of polyplexes with arrows of (J1) , but one also get morphisms between the
polyplexes:

The source and targets maps sk and tk : X X are represented by mor-

phisms of families, i.e. for each polyplex v one has two polyplexes sk (v) and
tk (v) and two morphisms from sk (v) and tk (v) to v. Such that if an arrow of
X is represented by a morphism v X then its k-source and k-targets are
represented by the maps sk (v) v X and tk (v) v X.

The composition operation #k can be seen as an operation X X X

X where the fiber product is over the two maps tk and sk from X to X .
Because of proposition 2.1.4, this fiber`product is a familially representable
functor, represented by the family of v c w where v, w and c are polyplexes
such that tk (v) = sk (w) = c and the coproduct is taken along the maps c =
sk (v) v and c = tk (w) w mentioned above. And ` hence the composition
operation produced is encoded by maps v#k w v d w where v#k w is the
polyplex corresponding to the composition v#k w as arrows of (J1) . These
operations satisfies some associativity, exchange and compatibility conditions
just translating the corresponding axioms on X which we will not list.

2.2.8. We will prove at some point latter `(corollary 2.3.11) that in a good
class of polygraphs, the maps v#k w v d w mentioned above are always
isomorphisms, i.e. that all the compositions operations:

#k : X X X X
are cartesian natural transformations in X. In the meantime we will say that a
class of polygraphs J satisfies `
condition (C) if the compositions are cartesian,
i.e. if all the maps v#k w v d w mentioned above are isomorphisms.

2.3 Plexes and polyplexes in dimension n + 1

2.3.1. In this subsection, we consider a class of polygraphs J such that Jn is a
good class of polygraphs, (satisfying condition (C) of 2.2.8) and we will show,
using the theorem of P.T.Johnstone and A.Carboni that we quoted in 2.1.5,
that Jn+1 is an effective presheaf category (proposition 2.3.3) and that one can
define a notion of Jn+1 -polyplexes (this will be done in 2.3.7).

19
It will be clear from definition 2.3.7 that the Jn+1 -polyplexes of dimension
smaller than n are just the Jn -polyplexes in the sense of definition 2.2.5, and at
the end of this subsection it will appears that when Jn+1 is indeed a good class
of polygraphs the Jn+1 -polyplexes of definition 2.3.7 are also the same as the
polyplexes in the sense of definition 2.2.5. In the meantime, as we do not know
yet whether or not Jn+1 is a good class of polygraphs, whenever we speak of
Jn+1 -polyplexes we are always referring to the definition of the present section
(2.3.7), and for Jn -pasting diagrams the two definitions are already known to
be equivalent.
A Jn+1 -polyplex will be a special type of Jn+1 -polygraph v endowed with an
arrow v v , one will always represent a Jn+1 -polyplex as a pair (v, v ) of a
polygraph and its so called top arrow or universal arrow, in particular:

Definition : An automorphism of a Jn+1 -polyplex (v, v ) is an automorphism

w of v such that w v = v . One generally denotes by Gv the group of auto-
morphisms of a polyplex (v, v ).

Note that if these were the polyplexes of a good class of polygraphs (in the
sense of definition 2.2.5) one will always have Gv = {1} because of the universal
property of polyplexes.
The main result of this section (theorem 2.3.9) will be that under the assump-
tions that Jn is a good class of polygraphs satisfying condition (C), one has for
any Jn+1 -polygraphs X that:
a
X = Hom(v, x)/Gv
v a Jn+1 -polyplex

For the notion of Jn+1 -polyplex that we will define. To put it another way:
for each cell f X there is a unique Jn+1 -polyplex ([f ], f ) and a unique
map t : [f ] X such that f = t f , but where the uniqueness is only up to
non-unique isomorphism of the polyplex ([f ], f ).
In particular Jn+1 will be a good class of polygraphs if and only if all the
automorphisms groups Gv are trivial. Moreover it will also be the case that
when this happens condition (C) automatically holds for Jn+1 which will prove
by induction that this condition holds for all good class of polygraphs (Corollary
2.3.11).
In the next subsection we will study in more details the properties of automor-
phisms of polyplexes and we will show in particular that they are indeed trivial
for the class of source-positive polygraphs.

2.3.2. For the rest of this section one will typically assume that J is a class of
polygraphs such that Jn is a good class satisfying condition (C) of 2.2.8.

2.3.3. We will first prove that Jn+1 is an effective presheaf category:

Proposition : Let J be a class of polygraphs such that Jn is a good class of

polygraphs, then Jn+1 is an effective presheaf category.

20
Moreover every Jn+1 -plex is either a Jn -plex or obtained from a pair of parallel
Jn -polyplexes (a, a ) and (b, b ) glued together along the polyplexes sn1 a and
tn1 a with just an additional (n + 1)-cell between a to b .

Proof :
Let T be the class of all (n + 1)-polygraphs whose underlying n-polygraph
belongs to Jn . Then Jn+1 T so that it is enough to prove that T is an effective
presheaf category because of 2.2.3. We will use the theorem of A.Carboni and
P.T.Johnstone that we quoted in 2.1.5, indeed T can be described as the category
Sets/F where F is the functor from Jn to sets which maps each polygraph X
to the set of pair of parallel n-arrows of X , this can be described as:

 sn (v) = v ; sn (w) = w
F (X) := (v, w) (X )2  sn1 (v) = sn1 (w)

 tn1 (v) = tn1 (w)
As all the sk and tk are natural transformations of the functor X 7 X , so this
functor F is familially representable as a finite limit of familially representable
functors (see 2.1.4). Using the description of finite limit of familially repre-
sentable functor given in 2.1.4, the family representing F is exactly given by the
set of pairs of Jn -polyplexes a, b such that sn1 a = sn1 b and tn1 a = tn1 b,
with as objects the gluing of a and b along sn1 a and tn1 a.
Hence ones immediately deduces that T is a presheaf category, we need to show
that it is an effective presheaf category. This follows from the discussion in 2.1.6:
If X is a T -polygraph represented as an object of Sets/F with Xn J`n and
S F (X) being the set of (n + 1)-cells of X the total set of cells is S |Xn |
(where |Xn | denotes the set of cells of Xn ). The category that we get from
2.1.6 has for objects the Jn -plexes and one additional object for each pair of
polyplexes that appears in the representation of F . Then |Xn | corresponds to
the coproduct over all the Jn -plexes of Hom(v, X) because Jn is an effective
presheaf category and S corresponds to the coproduct over the new objects w
of Hom(w, X) by 2.1.6.
Moreover the T -plexes are exactly the objects described in the proposition: the
Jn -plexes and one additional object for each pair (u, v) of parallel Jn -polyplexes
which represent exactly the (n + 1)-cell glued between cells of the type of u
and v hence is exactly as described in the proposition. The Jn+1 -plexes are the
T -plexes which belong to Jn+1 by proposition 2.2.6 and so are a sub-collection
of these.


2.3.4. We still assume that Jn is a good class of polygraphs satisfying condition

(C). The idea general of our definition of Jn+1 -polyplexes is quite simple: one
wants them to be defined inductively by the fact that each Jn+1 -plex is a Jn+1 -
polyplex and that if two Jn+1 -polyplexes are composable then their composite
(constructed as a gluing in the spirit of condition (C) of 2.2.8) should also be
a Jn+1 -polyplex. This will only be possible because the boundary of a Jn+1 -
polyplex will be made of two Jn -polyplexes, which have no automorphisms and
so when we compose two Jn+1 -polyplex together, there is no ambiguity on how

21
we glue their boundary together, without this assumption there would be several
ways to form a same composition x#n y by twisting by an automorphism the
identification tn (x) sn (y).
So essentially one wants to have one Jn+1 -polyplex for each arrow of (Jn+1 ) ,
they will be constructed as the image of a certain morphism of -categories
F : (Jn+1 ) D where D is an -category, in fact an (n + 1)-category, whose
(n + 1)-arrows are pre-polyplexes, i.e. polygraphs with a marked (n + 1)-
arrow which are composed by gluing, and F will just be defined as sending the
generators of (Jn+1 ) to the corresponding Jn+1 -plexes. So the main point is
to construct this -category D
For technical reason, we will need an -category DX depending functorially
on a polygraph X Jn+1 1 whose (n + 1)-arrows are pre-polyplexes with a
map to X, and which will comes with natural morphism X DX . The D
we mentioned above corresponds just to the case X = (Jn+1 ) , but this more
general DX will be convenient to show that any arrow of X can be represented
by a Jn+1 -polyplex.

2.3.5. Definition of DX :
The arrows of DX are equivalences classes of triples (v, , ) where:

v is a finite polygraph in Jn+1 ,

is an arrow of v of dimension at most n + 1,

is a polygraphic morphism : v X.

Two such triples (v, , ) and (v , , ) are equivalent if there exists an isomor-
phism : v v such that () = and = .

Sources and targets in DX are defined as follows: If (v, , ) is an arrow of DX

and i 6 n then the arrow si () is of dimension at most n, so it belongs to
the n-skeleton of X which is an object of Jn , hence there is a uniquely defined
Jn -polyplex [si ()] and a map s : [si ()] v corresponding to this arrow, one
keep calling it its universal arrow si ().
One defines: si (v, , ) = ([si ()], si (), s), and similarly, ti (v, , ) =
([ti ()], ti (), s).
For any arrow x one defines sn+1 (x) = tn+1 (x) = x (i.e. DX will be an (n + 1)-
category).

All compositions x#i y for i > n are trivial as DX is an (n + 1)-category. Com-

positions for i 6 n are defined as follow:
If ti (v, , ) = si (v , , ), it means that there is an isomorphism between the
Jn -polyplexes [ti ()] and [si ( )] which send the universal arrow to the universal
arrow, but such an isomorphism, when it exists, is in fact an equality as Jn is a
good class of polygraphs and those are Jn -polyplexes, hence one can define:

a
(v, , )#i (v , , ) := v v , #i , (, )
[ti ()]

22
The composition #i makes sense because  ` and

 are arrows of v and v

respectively hence they both belong to v [ti ()] v , and as ti () and si ()
both lives in the image of [ti ()] and are equal there, it makes sense to compose
them in the pushout. Similarly and coincide through the identification of
[ti ()] with [si ( )] so (, ) is a well defined application on the pushout.

Proposition : DX defined above is an (n + 1)-category. Its underlying n-

category is isomorphic to the underlying n-category of X .

Proof :
If we restrict ourselves to the arrows of DX which are of the form ([], , v :
[] X) for a Jn -polyplex one gets exactly the k-arrows of X for k 6 n,
with the correct sources and targets, and the compositions law is correct exactly
because we assumed that (C) satisfied condition (C) of 2.2.8. Moreover sk and
tk for k 6 n takes values in this class of arrows, so the underlying n-category
of DX is indeed an -category simply because it is isomorphic to the underlying
n-category of X .
Moreover, as all the other sk and tk (for k > n) are the identity on DX all the
globular relations between the sk and tk are automatically deduced from the
globular relations in X . At this point we only have to check that the compo-
sition operations on DX are associative, compatible to the units, and satisfies
the exchange law but this is completely trivial when we write the correspond-
ing co-limits expressing these compositions, and the relation giving the value of
sk (an b) follow also directly from the definitions.


2.3.6. One can construct two morphisms X DX and DX X as follow:

To any (v, , ) DX one can associate () X . It is immediate that this
is a morphism of -categories.
X DX is constructed as follow: on the underlying n-category it is the natural
isomorphism between their underlying n-categories mentioned in proposition
2.3.5. Because of the universal property of X we only need to say what are
the images of the (n + 1)-generators of X . For each (n + 1)-generator a of X
there is a Jn+1 -plex [a] of dimension n + 1 and a map i : [a] X. One defines
the image of a in DX as the triple ([a], [a] , i), which can easily be checked to
have the correct source and target.
Finally, it is immediate to check that the composite X DX X is the
identity of X , and that these two constructions are functorial in X. One hence
has in particular proved that:

Proposition : X is a retract of DX , functorially in X.

23
2.3.7. Definition : A Jn+1 -polyplex is a pair (v, ) with v Jn+1 and
v whose isomorphism class is in the image of the morphism (Jn+1 1) D
constructed in 2.3.6.

Note that as (Jn+1 1) is a retract if D, this already shows that isomorphisms

classes of Jn+1 -polyplexes are exactly indexed by arrows of (Jn+1 1) and form
a sub--category of D.

2.3.8. Proposition : The image of X in DX is exactly the set of triples

(v, , ) such that (v, ) is a Jn+1 -polyplex.

Proof :
First we observe that the set W of (v, , ) such that (v, ) is a polyplex is
indeed an -category. Indeed as the map from Jn+1 1 D is a monomorphism
(it has a retraction) its image is a sub--category, and hence its pre-image along
the morphism DX D is also an -category, and this is exactly the subset W .
This implies that X is included in W , indeed, all the arrows of dimension 6 n
and all the generators of dimension n + 1 are sent into W , so one has indeed
X W .
Finally we will prove by induction on the arrows of (Jn+1 1) that if x is an
arrow of (Jn+1 1) corresponding to a polyplex (v, ) and if is any morphism
from v to X then (v, , ) X DX .
If x is a generator of (Jn+1 1) , then the corresponding polyplex is given by the
corresponding plex ([x], x ), a morphism : [x] X is the data of a cell of
type x of X and (x, x , ) is the image in DX of the corresponding generator
of X .
Assume now that x = a#k b in (Jn+1 1) and a and b satisfies the induction
hypothesis. a and ` b corresponds to polyplexes ([a], a ) and ([b], b ) and x
corresponds to ([a] c [b], a #k b ) where c is the ` polyplex corresponding to
the k-source of a and the k-target of b. If : [a] c [b] X is any map then it
is exactly the data of two maps a : [a] X and b : [b] X which agrees on
c. Both ([a], a , a ) and ([b], b , b ) are arrows of DX which are in the image
of X by the induction hypothesis, and x is exactly their k-composite, so it also
belongs to the image of X . 

2.3.9. Theorem : Let J be a class of polygraphs such that Jn is a good class

of polygraphs satisfying condition (C) of 2.2.8. Then for X Jn+1 one has:
a
X = (Hom(v, X)/Gv )
v a Jn+1 -polyplex

If all the Gv are trivial then Jn+1 is a good class of polygraphs satisfying condi-
tion (C), with the two notions of Jn+1 -polyplexes (from 2.2.6 and 2.3.7) being
equivalent. Conversely, if Jn+1 is a good class of polygraphs then all the Gv are
trivial.

24
Proof :
We have seen in 2.3.8 that X can be described as the subset of DX of equiva-
lences classes of triples (v, a, ) where (v, a) is a polyplex. Two such triples are
equivalent exactly if the polyplex are the same and if and are conjugate un-
der the action of Gv , and this already proves that X has the given description.
If all the Gv are trivial then this proves that X 7 X is familially representable.
We have seen in 2.3.3 that Jn+1 is an effective presheaf category hence Jn+1 is
a good class of polygraphs. The explicit formula for composition in DX given
in 2.3.5 shows that composition of two polyplexes is given by a pushout, and
hence that Jn+1 satisfies condition (C). Conversely, if Jn+1 is a good class of
polygraphs, it means that:
a
X Hom(wi , X)
iI

By comparing this formula to the one we gave in the statement of the theorem
and looking at the value of this functor on the terminal Jn+1 -polygraph and on
the map to the terminal Jn+1 -polygraph one concludes that there is a bijection
between the wi and the Jn+1 -polyplexes as they were defined in this section
such that Hom(wi , X) Hom(vi , X)/Gvi . Lemma 2.3.10 below concludes the
proof.


2.3.10. Lemma : Let C be any category, v an object of C endowed with an

action of an ordinary group G. If the set valued functor:

X 7 Hom(v, X)/G
is representable then the action of G on v is the trivial action.

Proof :

Let w be the object such that one has an isomorphism X : Hom(v, X)/G
Hom(w, X) functorial in X. Let i : w v be the element v (1v ) and let
1
p : v w be any element of the G-class of w (1w ) Hom(v, w)/G.
By applying to 1v Hom(v, v)/G the naturality square:
p
Hom(v, v)/G Hom(v, w)/G
v w
p
Hom(w, v) Hom(w, w)
one gets : p i = 1w
By applying to 1w Hom(w, w) the naturality square:
i
Hom(w, w) Hom(w, v)
1
w v1
i
Hom(v, w)/G Hom(v, v)/G

25
One obtains that i p is equal to 1v in Hom(v, v)/G, i.e. it is in the image of G,
and hence it is invertible. This implies that i and p are inverse of each other.
Finally, by applying to 1v Hom(v, v)/G the following naturality square for
any g G:
g
Hom(v, v)/G Hom(v, v)/G
v v
g
Hom(w, v) Hom(w, v)

one gets that g i = i, and as i is an isomorphism this implies that g = 1 as an

automorphism of v, i.e. that the action of G on v is trivial.


2.3.11. Corollary : Condition (C) of 2.2.8 is satisfied by any good class of

polygraphs.

Proof :
Let J be a good class of polygraphs, we prove by induction on n that Jn satisfies
condition (C). Condition (C) is clearly satisfied for the class of all 1-polygraphs,
so J1 satisfies it. If Jn is a good class of polygraphs satisfying condition (C)
then one can apply theorem 2.3.9 to it. As Jn+1 is a good class of polygraphs
all the Gv are trivial and hence Jn+1 is a good class of polygraphs satisfying
condition (C). As condition (C) only involve finite polygraphs in J it is enough
to check it for Jn for all n, hence this concludes the proof. 

2.4 Automorphisms of polyplexes and the case of source-

positive polygraphs
In this section we will study in more details the automorphisms of a polyplex
in the sense of definition 2.3.1, i.e. automorphisms of the underlying polygraph
whose action on the free -category it generates preserves the universal cell.
We are still working under the same assumption as in the previous section:
J is a class of polygraphs such that Jn is a good class of polygraphs, and
we are interested in the Jn+1 -polyplexes as they were defined in 2.3.7, and
more precisely in their automorphisms. Our goal is to find a criterion under
which these automorphisms are all trivial. The main result being theorem 2.4.5,
showing that in the case where J is the class of source-positive polygraphs then
all these automorphisms are indeed trivial and hence the class of all source-
positive polygraphs is a good class of polygraphs.
All the polyplexes of dimension 6 n are in the good class Jn and hence have
no automorphism. Hence we are only interested in the polyplexes of dimension
n + 1.

26
2.4.1. Some conventions for the section: if X is an Jn+1 -polygraph and v X
is an arrow of dimension 6 n + 1 and a is some cell of X, one says that a is in
v or appears in v if it is in the image of the polyplex [v] X representing v.
If a and v have the same dimension this corresponds with the usual meaning of
a generator appearing in an arrow. This can also be checked to be equivalent
to M.Makkais notion of the content of an arrow introduced in [14] for general
polygraphs.

2.4.2. Lemma : Let (v, ) be a Jn+1 -polyplex of dimension n + 1 and e a cell

of v of dimension 6 n then one has either:

e appears in sn ().
e appears in tn (x) for x some (n + 1)-cell of v.

Moreover, if e is of dimension n exactly then the two possibilities above cannot

occur simultaneously and in the second case the (n + 1)-generator x such that e
appears in tn (x) is unique.

One also has a dual statement obtained by exchanging source and target every-
where whose proof is exactly the same.
Proof :
We will prove the lemma by induction on arrows of (Jn+1 1) (corresponding to
the polyplex v).
For a generator of (Jn+1 1) , the polyplex (v, ) is a Jn+1 -plex with its unique
(n + 1)-cell. As mentioned in proposition 2.3.3, a Jn+1 -plex of dimension n + 1
is always constructed by gluing two n-dimensional Jn -polyplexes v s and v t on
their common boundary and adding only one (n + 1)-cell whose source and
target are given by the universal arrows of (v s ) and (v t ) , hence any generator
e of dimension 6 n is either in v s , which corresponds to the first case of the
dichotomy in the lemma, or in v t which corresponds to the second case of
dichotomy. Moreover if e is of dimension n, then as v s and v t are glued together
along a polygraph of dimension < n, e can only belong to one of them and so
the two possibilities are indeed incompatible.
For a composite a#k b in (Jn+1 1) , it corresponds to a polyplex of the form:
a
(a b, a #k b )
c

where c is the polyplex corresponding to tk (a) and sk (b), in particular it is of

dimension 6 k.
We need to distinguish several cases, depending on the dimension of a and b
and the value of k:

If k = n and both a and b are of dimension n + 1, then the cell e belongs

to either a or b. If it belongs to a it is either in the source of a which is
also the source of a#k b or in the the target of one of the (n + 1)-cell of a,
which are also cells of a#k b. If e belongs to b then it is either in the source
of b, in which case it belongs to the target of a as well and one concludes

27
 by the previous case, or in the target of one of the (n + 1)-cell of b, which
are also cells of the composite.
If the cell e is of dimension n exactly, and if it belongs to both a and b it
means that it is both in the source of b and in the target of a, but being in
the source of b it cannot be in the target of any (n + 1)-generators of b so
it is either in the source of a or in the target of a unique (n + 1)-generator
in a and this proves the uniqueness part in this case. If it belongs to only
one of a and b then one gets the uniqueness directly from the inductive
application of the uniqueness result in a and b separately (it cannot appear
in the source of b in this case).
If both a and b are of dimension n + 1 but k < n: a cell e belongs to either
a or b, in both case it belongs to either the target of an (n + 1)-cell or the
source of a or b but in this case both the source of a and the source of b are
included in the source of a#k b so it concludes the proof immediately. If e is
of dimension n exactly then as a and b are glued together on something of
dimension < n, the generator e cannot be in both of them simultaneously
and so one gets the uniqueness result by simply applying the uniqueness
results inductively in a and b separately.
If k < n and either a or b is of dimension 6 n, for example assume b is of
dimension 6 n (the proof is exactly the same if a is of dimension 6 n). In
this situation sn (a#k b) = sn (a)#k b hence any cell e that belongs to b also
belongs to sn (a#k b) which proves the claim for such cells. For any cell e
in a, one can just apply the induction hypothesis inside a: either e is in the
source of a or in the target of some (n + 1)-cells of a, in both case it proves
the claim. If e is of dimension n exactly then e cannot be both in a and b
simultaneously as they are glued along a polygraphs of dimension k < n,
if e is in b then it is in the source and there is no (n + 1)-cell it could be in
because all the (n + 1)-cells are in a, and if it is in a then it is (uniquely)
either in the source of a or in the target of a unique (n + 1)-generator of
a and this concludes the proof of this case.
All the other cases are trivial: if k = n and a and b are not both of
dimension n + 1 then the composite is equal to a or b, if neither a nor b is
of dimension n + 1 then the composite is not of dimension n + 1 and this
also concludes the proof.


The next lemma is our key tool to show that the automorphisms of a given
polyplex are trivial. It shows that any automorphism have some fixed points,
and that being a fix point of an automorphisms tend to be contagious.

2.4.3. Lemma : Let (v, ) be a Jn+1 -polyplex of dimension n + 1 and w Gv

be an automorphism.
1. If an arrow t v of dimension 6 n is fixed by w, then all the cells that
appears in t are fixed.

28
 2. w fixes all the cells that appears in sn () or tn ().
3. If w fixes some (n + 1)-cell x then all the cells that appears in sn (x) and
tn (x) are fixed by w.
4. If w fixes an n-cell a that belongs to the source or the target of an (n + 1)-
cell x, then x is also fixed by w.

Proof :

1. let t v be an arrow fixed by w. As t is of dimension smaller than n

it belongs to n-th skeleton of v which is a Jn -polygraph. In particular
t is uniquely represented by a Jn -polyplex a : [t] v, in the sense that
a (t ) = t as w t = t one has that w a also represent t, so by uniqueness
of the representation w a = a and hence all cells that are in the image
of a are fixed by w.
2. w fixes so it also fixes sn () and tn () which are arrows of dimension
6 n, hence the point 1. applies.
3. If w fixes x then it also fixes sn (x) and tn (x) which are arrows of dimension
6 n, hence the point 1. applies.
4. If a appears in tn (x) then w(a) will appears in tn (w(x)). Hence if w(a) =
a it means that a appears in both tn (x) and tn (w(x)), but as a is of
dimension n the uniqueness part of lemma 2.4.2 shows that x = w(x). The
version with sn is proved in the exact same way using the dual version of
lemma 2.4.2.

2.4.4. Proposition : If an automorphism w of a polyplex fixes all the (n + 1)-

cell then it is the identity.

In particular, Gv can be identified with a sub-group of the group of permutations

of the (n + 1)-cells.
Proof :
It follows immediately from lemma 2.4.2 together with points 2. and 3. of lemma
2.4.3. 

29
2.4.5. Theorem : The class of all source-positive polygraphs is a good class of
polygraphs.

Proof :
Let J be the class of all source-positive polygraphs. We proceed by induction
using theorem 2.3.9: J1 is a good class of polygraphs (it is the category of
graphs). We assume by induction that Jn is a good class of polygraphs, and
we need to show that for any Jn+1 -polyplex (v, ) of dimension n + 1, any
automorphism w Gv is the identity.

The general idea is to use lemma 2.4.3 to propagate the fact that w is the identity
from the source of v to all the (n + 1)-cells through the n-cells connecting them,
and then use lemma 2.4.4 to conclude that w is the identity.
In order to make this formal, one introduces the following notion: A chain of
cells in v is a finite sequence x1 , . . . , xk of (n + 1)-cells of v such that:

There is an n-cell which is both in the source of x1 and in sn ().

For each i > 0, there is an n-cell which is both in the source of xi+1 and
in the target of xi .

Lemma 2.4.3 shows that any such chain is fixed by w: indeed the n-cell in the
source of is fixed because of point 2. of 2.4.3 and this implies that x1 is fixed
because of point 4., and then inductively if xi is fixed then the n-cell which
belongs to both the target of xi and the source of xi+1 is fixed because of point
of 3. and this implies that xi+1 is fixed by point 4. of the lemma.

We will now show that any (n + 1)-cell of v appears in such a chain. More
precisely, we will prove by induction on arrows of (Jn+1 1) that in the corre-
sponding polyplex (v, ) any (n + 1)-cell of v appears in such a chain.
If (v, ) is a generator of (Jn+1 ) , i.e. v is a Jn+1 -plex, then the unique (n + 1)-
dimensional cell is the x1 of a chain exactly because one has assumed that its
source contains at least one n-cell.
If (v, ) corresponds to a composite a#k b in (Jn+1 1) , it is of the form:
a
(a b, a #k b )
c

Let x be an (n + 1)-cell in this polygraph v. Then x belongs to either a or b.

If x is in a it belongs to a chain in a which start on the source of a, and the
source of a is always included in the source of a#k b. If x is in b, there is a
chain (x1 , . . . , x) in b which start with an (n + 1)-cell x1 for which there is an
n-cell e both in its source and in the source of b. If k < n, then the source of
b is included in the source of a#k b and (x1 , . . . , x) is already the chain we are
looking for. If k = n then the source of b is exactly the target of a and by lemma
2.4.2 e is either in the source of a (in which case the proof is finished) or in the
target of some (n + 1)-cell y of a in which case there is a chain y1 , . . . , yi = y in
a and y1 , . . . , yn , x1 , . . . , x is a chain in a#k b which contains x.


30
2.4.6. Corollary : The class of positive polygraphs and the class of opetopic
polygraphs are both good classes of polygraphs. In particular they are presheaf
categories.

Proof :
They are both included in the class of target-positive polygraphs, which is a good
class of polygraphs exactly as the class of source-positive polygraphs. Hence by
proposition 2.2.6 they are also good classes of polygraphs. 

2.4.7. Some further questions that would be interesting:

Can we use the inductive description of plexes to show that the opetopic
plexes are the same as the opetopes ? This would give a direct proof that
opetopic polygraphs are the same as opetopic sets.
Can we give a more convenient description of the positive plexes or the
source-positive plexes ? We are after some generalization of the various
notion of n-pasting diagrams that have been devised by M.Johnson [10],
A.J.Power [19], R.Steiner [21] or R.Street [24].

2.4.8. We now give a promised examples of two positive 3-polyplexes with iso-
morphic underlying polygraphs (but different universal arrows). We also believe
that this example shows that the second question above is hard: the compo-
sition formed by these two polyplexes uses the exact same variable (meaning
they have the same underlying polygraphs) but produces different results which
hence only differ by the order in which certain cells are composed, but the key
point in all the theory of pasting diagrams mentioned is to restrict to situations
where there is a unique possible composition order or where the composition
order does not matter, and the following example seems to be far out of the
scope of such situations. This is not a clear counter example to the existence of
a nice description, but clearly an example we should meditate before attacking
this question. For example, there is still hope that things are still order inde-
pendent as long as the source and the target have been properly fixed as these
two examples also have different targets that even correspond to non-isomorphic
polyplexes.
We start with the following three 2-polyplexes:

D1 := A1,1 #0 P1 D2 := P1 #0 A1,1 A2,2

31
They all have for source and target the 1-polyplex , hence one can
consider the two 3-plexes U : D1 A2,2 and V : D2 A2,2 (any pair of
parallel 2-polyplexes defines a unique 3-plex between them). For example, U is
the 3-polygraph:

f
v

g

with one additional 2-arrow : f #0 v g#0 v and one additional (universal)

3-arrow U : #0 v .
We form the 3-polyplex U #0 A1,1 whose underlying polygraph is

f k
v

g h

with the same two additional cells and U as above. The 2-target of U #0 A1,1
is A2,2 #0 A1,1 , i.e:

f v k

g v h

and it is sent to U #0 A1,1 by sending each cell to the cell with the same name
(and both v and v are sent to v, in particular the map t2 (U #0 A1,1 ) U #0 A1,1
is not a monomorphism. Our example is obtained by post composing this 3-
polyplex with (a whiskering of) V , but there is two different way we can do
that:
We can either apply V on the sub-diagram corresponding to v, h, k and , which
means forming the composite:

(U #0 A1,1 )#2 [(f #0 V )#1 (A2,2 #0 h)]

or apply V to the subdiagram corresponding to v , h, k and , which means
forming the composite:

(U #0 A1,1 )#2 [(A2,2 #0 k)#1 (g#0 V )]

(in both these expressions, all the f, h, g and k should all be just P1 but
we thought that having the name of the precise arrow it corresponds in the
diagrams would help the reader to parse the notation). As v and v are actually
collapsed together in the underlying polygraph of U this makes the underlying
polygraphs of the polyplexes we obtain this way identical and equal to:

32
 f k
v

g h

with in addition:

: f #0 v g#0 v : v#0 k v#0 h

U : #0 v V : v#0
but they have different universal cells, which corresponds to the two different
composite given above, and have different 2-targets: the first one has for 2-
target ( #0 k)#1 (g#0 ) while the other has (f #0 )#1 ( #0 h) and these are
different, they even corresponds to different 2-polyplexes. The point of this
story being that and have disjoint boundary hence the vertical order does
not matter when we compose them, but once we replace them with and
we still have two way to vertically compose them but this time their boundaries
intersect and hence the vertical composition order matter.

2.4.9. We would like to finish this section with a rather technical observation for
the expert reader that play no concrete role in the paper: We want to show that
M.Batanin criterion in [2] for proving that the category of computads associated
to a globular operad is a presheaf category is insufficient for the main case of
interest to us, and try to explain the relation between our theorem and this
criterion. We refer the reader to [2] for the notion that we will mention below.
We start with a quick remark: in his paper M.Batanin quote a result of A.Carboni
and P.T.Johnstone from [5] saying that a finitary monad on the category of
sets is familially representable if and only if it corresponds to a strongly reg-
ular theory. This results is unfortunately the other false result of [5] (see [6]
for a counterexample): instead, strongly regular theories correspond to non-
symmetric operads while the condition of familial representability corresponds
to the weaker notion of -cofibrant symmetric operads. Due to the strong sim-
ilarities of M.Batanin results and the results of the present paper, we strongly
believe3 that all his results in [2] are corrects if we replace everywhere in his
paper strongly regular by this weaker condition of being a familially repre-
sentable monad or equivalently being a -cofibrant operads.

This being said, the class of positive polygraphs can be seen as the category
of polygraphs for the globular operad for non-unital -categories (see A.5 for
our notion of non-unital -categories) which is just the sub-monad of the free
-category monad D on the category of globular sets defined by

M (X) = {a D(X)|a is not an identity arrow in D(X).}

One easily check from an explicit description of D that M (X) defined this way
is indeed a sub-globular set of D(X) and that M is a sub-monad of D, with
the inclusion M D being cartesian. Hence M is a globular operad in the
3 We have unfortunately not being able to understand M.Batanin proofs well enough to be

sure of that, but he does not seem to be using strong regularity anywhere.

33
sense defined in [2], and the M -computads in his sense are exactly our positive
polygraphs. But the second slice P2 (M ) of M is not a -cofibrant operad.
Indeed a P2 (M )-algebra is the same as a non-unital 2-category (in the sense
of an M -algebra, or in the sense of A.5) with only one cell in dimension 0 and
dimension 1, but this is enough to form a Eckmann-Hilton collapse: if x and
y are two 2-arrows of such an -category and let e denotes the unique 1-arrow
of this category then, as e is a 1-arrow one has e#1 x = x#1 e = x and as there
is only one 1-arrow one has e#0 e = e, from there:

(e#0 x#0 e)#1 (e#0 y#0 e) = [(e#0 x)#0 e]#1 [e#0 (y#0 e)]
= [(e#0 x)#1 e]#0 [e#1 (y#0 e)]
= e#0 x#0 y#0 e
and starting from the other possible bracketing one gets:

(e#0 x#0 e)#1 (e#0 y#0 e) = [e#0 (x#0 e)]#1 [(e#0 y)#0 e]
= [e#1 (e#0 y)]#0 [(x#0 e)#1 e]
= e#0 y#0 x#0 e
hence one has a commutative operation on two variables:

e#0 y#0 x#0 e = e#0 x#0 y#0 e

which shows that the corresponding operad is not -cofibrant: it admits a
generic4 operation invariant under a non trivial permutation of its variables.
Informally, what happens here is that in M.Batanin criterion we look at how
the n-dimensional operations which are generic on variables of dimension n
but where all the lower dimensional variables are set to be equals behave un-
der permutation of the variables of dimension n, while in our framework we
are looking at the behavior under permutation of the variables of operations
which are globally generic i.e. where in some sense all the variables of all
dimension is used exactly once (this concept being made rigorous by the no-
tion of polyplexes). In both case the criterion is that in the absence of such
permutations one has a presheaf category. Moreover, we prove in our context
(lemma 2.4.4) that it is enough to show that such permutations acts trivially
on the n-dimensional arrows, which allows to recover M.Batanin criterion for
sub-operads of D. But on the other hand, our proof in sub-section 2.4 relies
heavily one carefully analyzing how such permutations of a polyplex acts on
(n 1)-variables (i.e. (n 1)-cells of the polyplexes), which would not be of
much help in the framework of M.Batanin criterion.

But obviously, while our criterion is more powerful in the sense that it is a nec-
essary and sufficient condition at least for detecting good classes of polygraphs,
M.Batanin criterion applies to the more general situations of any globular op-
erads. As mentioned earlier, we hope to extend our main theorem 2.3.9 to
any globular operads, or even to a more general notion of operads using other
4 By generic operations, we mean an operation where each variables that appears in it

appears exactly once. For an ordinary finitary algebraic theory, this only makes sense if the
theory corresponds to a symmetric operads (which is the case of such slice theory by results
of M.Batanin) and in this case the generic operations on n-variables are exactly the operation
in the set O(n) of the operads.

34
non-globular sort of combinatorics (typically to any parametric right adjoint
cartesian monad on a category of presheaf over a directed categories).

A On C.Simpsons conjecture and the Kapranov-

Voevodsky strategy
In 1991, M.Kapranov and V.Voevodsky published a proof ([12]) of a form of
the homotopy hypothesis, claiming that the homotopy category of spaces is
equivalent to the homotopy category of a certain kind of -groupoids, which
are strict -categories where every arrow is weakly invertible in every degree (we
refer to [12] for the precise definition of weak invertibility). In 1998, C.Simpson
published a proof ([20]) that this cannot be true. He was not able to point out a
precise mistake in [12] but he conjectured that this has to do with how units are
handled and he formulated the conjecture now called the Simpsons conjecture or
Simpsons semi-strictification conjecture, that the homotopy category of spaces
is equivalent to the homotopy category of a notion of -groupoids where the
associativity and the exchange rule are strict, but units and inverses are weak.
He did not gave a precise definition of what this means, and he added that
he thinks that the argument of [12] [...] actually serves to prove the above
statement .
At the time the present paper is written it seems to be largely accepted that it is
C.Simpsons paper which is correct and M.Kapranov and V.Voevodskys paper
which is flawed, but as far as we know it is still unclear where is the mistake in
this paper. C.Simpsons conjecture is still open, and while it is still plausible
that the general strategy of [12] could gives a proof, it seems that it needs an in
depth reworking and some new ideas in order to achieve that. The results about
positive polygraphs that we proved in the present paper somehow steam from
our personal analysis of why the paper [12] fails and what should be done to
fix it in order to prove Simpsons conjecture.
The goal of this section is to sum up this analysis, both in order to motivate
the present paper and to explain our plan to attack this conjecture.

A.1. We start by explaining the general idea of [12], which we refer too as the
Kapranov-Voevodsky strategy. One start with a topological space X and we
want to define a fundamental -groupoid of X, whose objects are points of X, 1-
arrows are paths between points, 2-arrows are endpoints preserving homotopies,
and more generally n-morphisms are boundary preserving homotopies between
(n 1)-morphisms. In this groupoid compositions are given by composition
of homotopies, but we want to define it in such a way that associativity and
exchange rule holds strictly, and not up to higher homotopy as it would be the
case with a naive definition.

If we just look at 1-arrows, composition can be made strict by using Moore

paths i.e. by allowing the length of paths to vary: if one has two composable
1-arrows given by [0, n] X and [0, m] X and we define their composite as
[0, n + m] X then one gets a strictly associative composition. The starting

35
idea in [12] is to push this idea to all dimension: 2-arrows now need to be
homotopies between paths of possibly different length and if we want to compose
them in a strictly associative way (and with strict exchange rules) the only way
is to define the composition formally by just gluing together the spaces indexing
those homotopies. To generalize this in all dimension, we need to come up with
a notion of generalized Moore homotopies indexed by certain diagrams or
cell complexes.the introduction of [12] give a nice explanation of this idea.

M.Kapranov and V.Voevodsky propose to use M.Johnsons notion of pasting

diagram from [10] which they improve in a companion paper [11]. They define
a category of Johnsons diagrams a and geometric realization functor for such
diagrams. One can then attempt to make the above idea formal by saying that
we want the n-cell of the fundamental -groupoid of X to be pairs of a pasting
diagram K together with a continuous map |K| X where |K| denotes the
geometric realization of the diagram K, and composition should be defined by
pasting of diagrams. Unfortunately such a construction have no chance of being
either a left of right adjoint functor, and Proving that a functor induces an
equivalence between homotopy categories can be very hard is this functor is not
part of an adjunction.
To avoid this problem, as well as possibly some technical difficulties in making
the above definition formal, they move to a slightly different construction: if
D denotes their category of Johnson diagrams, then they have two functors
D cat and D Spaces respectively defined as the free -category
generated by a diagram and their geometric realization functor. Using usual Kan
extension techniques one gets two left adjoint functors P rsh(D) cat and
P rsh(D) Spaces. They then use the fact that P rsh(D) is somehow similar
to the category of simplicial sets or cubical sets to set up a homotopy theory
for this presheaf category, and they claim to prove that these two adjunctions
induce equivalences between the three homotopy categories of interest, which
concludes their proof.

A.2. If we forget the more indirect version of the construction they actually uses,
and come back to the initial idea of using some generalized Moore homotopies
indexed by some class of pasting diagrams, then one can see that in order to
define a well behaved fundamental -groupoid of X whose n-arrows are maps
|K| X for K a pasting diagram, it seem to us that one should at least expect
that the category of diagrams we are using satisfies the following properties:
(A) One should be able to compose the pasting diagrams that we use, and this
composition should be given by a gluing operation both on diagrams and
on geometric realization. We need this to defines the (partially defined)
composition operations on the set of maps |K| X.
(B) If K and K are two parallel n-dimensional pasting diagrams (their
boundaries are the same diagrams) then one should be able to construct a
new pasting diagram by gluing K and K together along their boundary
and adding one new n + 1-cell between them. Having this, allows us to
comfortably see inside this fundamental -groupoid of X that if two
parallel cells corresponding to maps |K| X and |K | X are homotopic
in X then this homotopy is detected inside the fundamental -groupoid.

36
It appears that Johnson diagrams fail to have either property (A) or (B).
Property B fail because of the following stupid example: If one consider
the diagram representing a path and the diagram representing the
constant path, then a 2-arrow between them would be diagram with a loop and
a contraction of that loop. But Johnsons diagrams are not allowed to contains
loop (they satisfies a certain condition called loop free which as the name
suggest in particular implies that the underlying 1-graph cannot have loop).
This first obstruction is clearly related to the presence of units and disappear
when we work in a non-unital framework as suggested by C.Simpson.
While this first observation is very encouraging for C.Simpson conjecture, there
is unfortunately, a second type of counterexample to property (B) that still
exists even when we restrict ourselves to non-identity arrow. Consider the fol-
lowing two Johnson 2-pasting diagrams:

They are both legitimate Johnson diagrams and they are parallel, but if we
glue them along their common boundaries, the two vertical arrows will again
form a loop, so this take us outside of the class of Johnson diagrams. More
problematically, for the exact same reason there can be no Johnson 3-diagrams
which has these two diagram as source and target. So if one form a fundamental
-groupoid of X whose arrows are only parametrized by Johnson diagram there
can never be a 3-arrow between two 2-arrows parametrized by the two pasting
diagram above.
Moreover, this also provides a counterexample for condition (A), one can con-
sider the following two 3-dimensional Johnsons diagram:

The diagram representing a single 3-arrow from the first of these two
diagrams to the diagram A2,2 with the same boundary by only one internal
2-cell from the source to the target.
The diagram representing a single 3-arrow from A2,2 to the second of these
two diagrams.

These two diagrams are indeed Johnson pasting diagrams, but we if we compose
them (along A2,2 ) one again gets a loop with the two vertical arrows, so the
composite is not a Johnson diagram ! Hence it is not even possible to define an
-category this way.
Although not explicitly claimed in their paper, M.Kapranov and V.Voevodsky
seem to use that Johnsons pasting diagrams can be composed, at least in the
proof of their lemma 3.4 of [12]. This might be one of the problems with their
proof.

37
A.3. We believe that at the end of the day, and even if that might not be the
exact technical reason which makes their proof incorrect, the reason why the
original form of the Kapranov-Voevodsky strategy cannot succeed is exactly
because of the failure of these two properties for Johnson diagrams, and that
fixing the proof in order to proves C.Simpsons conjecture requires to construct
a new class of diagrams which satisfies those two properties.

What the present paper achieve is exactly to construct such a category of dia-
grams in the non-unital case. We do not know how to do that in the unital
case, and believe it is impossible (and it is indeed provably impossible as soon
as we impose more precise conditions on how those diagrams should behave).
So as a first step one should replace the category of Johnson diagrams with the
category of positive polyplexes in the sense of the present paper, but we will
see below that there is something even more natural to do: to use P+ it self, i.e.
the category of presheaf on the category of positive Plexes and not presheaves
one the category of polyplexes.

A.4. One of the problem with C.Simpsons conjecture is that the notion of -
category with weak units, or of non-unital -categories does not have a unique
definition. For example, one could understand non-unital categories as being
a globular set with all the operations f #k g defined for f and g of dimension
n and k < n. Or one can also require to have compositions like f #k g defined
even when f and g have different dimensions, allowing to define whiskering as
a#0 f :

instead of using an identity arrow 1f to define it as a horizontal composite a#0 1f

of two 2-arrows. One can also use even different composition shapes, for example
one can consult [13] for a simplicially based definition of weakly unital -
categories, and all those notions are non-equivalent (in the 1-categorical sense).
So we need to choose such a notion.
Fortunately, the notion of positive polygraphs for which we prove the existence
of a well behaved notion of pasting diagrams (the positive polyplexes), are pre-
cisely a notion of polygraphs corresponding to a certain well defined non-unital
-category, which are globular sets where all the compositions f #k h are de-
fined even when f and g have different dimensions, and all associativities and
exchange rules holds. Almost surprisingly, when writing down the definition
of such non-unital -categories one sees that they are in fact ordinary -
categories satisfying two additional axioms:

A.5. Definition : A non-unital -category is an -category X in the sense

of definition 1.1.1 which satisfies the following two additional axioms:

If f is an arrow of dimension greater than n, then sn (f ) and tn (f ) are of

dimension n exactly.

38
 for any f, g, k such that f #k g is defined, its dimension is the maximum
of the dimension of f and the dimension of g.

A morphism of non-unital -categories is a dimension preserving morphism of

-categories. The category of non-unital -categories is denoted Catnu .

Of course the interpretation of the set X has changed: it is no longer the

increasing union of the set of n-arrows, but the disjoint union of the sets of
n-arrows. The fact that a non-unital -category in this sense can be identified
with a certain ordinary -category corresponds just to the fact that there is a
faithful unitarization functor from catnu to cat which just freely
add units.
We claim that those -categories are the algebras for a monad on globular sets
(in fact the monad M mentioned in 2.4.9), and that the polygraphs for this
monad (following [1]) are exactly the positive polygraphs. In particular, one
has an adjunction ( ) : P+ catnu : N , and ( ) is just the usual free
-category on a polygraph, which happens to takes values in catnu when
applied in the category of positive polygraphs.

Nonetheless, the fact that those non-unital categories are a (non-full) subcate-
gory of strict -category means that, if the form of C.Simpsons that we are
going to conjecture holds, this means that the main result of M.Kapranov and
V.Voevodsky was a lot closer to be true than what have thought: every homo-
topy type would be representable by a strict unital -category: the unitariza-
tion of its non-unital -groupoids, and this -category indeed compute the
correct homotopy group, we just need to use a definition of homotopy group
that do not use the canonical identity arrows of the category, but a weak
units instead.

A.6. Finally, following the footstep of Kapranov and Voevodsky, one defines
what we will call the Naive geometric realization functor: P+ Spaces which
send any polygraph P to the geometric realization of the category P lex+ /P of
cells of P . This is a left adjoint functor and one has a diagram of left adjoint
functor:

Spaces catnu

| | ( )
+
P

With P+ being a presheaf category as proved in the present paper.

This could be our new basis to make the Kapranov-Voevodsky strategy into
a proof of Simpsons conjecture, and it has a very nice new feature that is
not present in the original Kapranov-Voevodsky strategy: if one start with a
topological space, applies to it the right adjoint functor from Spaces to P+
and then the free -category functor, then we obtain exactly the -category
corresponding to the intuitive idea of generalized Moore spaces we started from:

39
Indeed if we denote by N the right adjoint to the geometric realization, a cell
of N (X) is a map from a polyplex p to N (X) which is exactly the same as a
continuous map from |p| to X. Hence cells of the corresponding -category are
exactly maps from the geometric realization of a pasting diagram (a positive
polyplex) to the space X as expected in the beginning.

A.7. But unfortunately there is a new problem that comes with the increased
complexity in the shapes of the diagrams that we use: Basically, the naive
geometric realization is too naive, and cannot be used in this pictures. To clarify
the following discussion we will admit the following:

One can construct a weak model structure on catnu where the fi-
brant objects are the non-unital -categories in which every arrow has
a weak identity endomorphism (which are defined as weakly idempotent
endomorphisms) and which satisfies all of Kapranov and Voevodsky di-
visibility condition of [12]. The weak equivalences between fibrant objects
being the map inducing a bijection on all the n .
One can construct a similar weak model structure on P+ where the
fibrant objects are the polygraphs P such that P is fibrant in the previous
sense and such that for each arrow f P there exists a cell f P parallel
to f and a cell a between f and f . Weak equivalences in P+ are the arrows
that induces equivalences in the sense of the above weak model structure
on catnu
The pair of adjoint functor between P+ and catnu defined above is
a Quillen equivalence (for a notion of Quillen equivalence adapted to
weak model structures).

We do have proof of these claims, but they are long and largely out of the scope
of the present paper, so we leave for a future work both the precise definition of
weak model structure (which is roughly a weaker notion than Quillen model
structures which includes both left and right semi-model structures) as well as
the proof of the claims above.

A.8. We can now explain the problem that remains to be solved in order to
prove Simpsons conjecture: the naive geometric realization is not even a Quillen
functors. The reason for this is that the naive geometric realization send any
plex to a contractible topological spaces. But quite surprisingly, and contrary
to what happen with Johnson diagrams in [12], general plexes are not at all nice
contractible balls ! The following is an example of a non-contractible 3-plex :
Consider first the following 2-polyplex:

x y z

One can form its unique 3-plex endomorphism, which has the following un-
derlying polygraph:

40
 It has three 0-cells x, y, z.
It has four 1-cells: f, g : x y and h, k : y z.
It has four 2-cells: , : f g, , : h k.
It has one 3-cells: : #0 #0

we claim that this plex is not homotopy equivalent to the point in the model
structure we mentioned above.
The reason for that is that removing a given n-cell h and an (n 1)-cell a
appearing exactly once in the source or the target of h (and not in the other)
does not change the homotopy type of a polygraph in this weak model structure
(the reader can note that those corresponds exactly to how the generating
trivial cofibration of [12] are defined in the framework of Johnsons diagram).
Admitting that, one can gradually remove the following pairs of cells to the plex
above without changing its homotopy type: (, ) , (, k) (h, z), after that it
only remains the following cells: two 0-cells: x, y; two 1-cells f, g : x y and
two 2-cells , : h k, i.e. it is exactly the globular polygraph corresponding
to a free pair of parallel 2-cells, whose geometric realization is the 2-sphere.
Obviously this makes no real sense as we have not defined the model structure
involved, but we can also see that this polygraph should have the homotopy
type of a 2-sphere in a less formal but more intuitive way: if we think of it
in topological terms, its boundary corresponds to two 2-sphere glued together
on a point, or equivalently a single bigger 2-sphere whose equator has been
contracted to a single point. Adding the unique 3-dimension cell should corre-
sponds to gluing a 3-ball on some 2-sphere, i.e. to fill the interior of that single
sphere. One hence obtains a 3-ball whose equator is contracted to a single point.
But only the boundary of the equatorial disk is contracted, not the whole 2-
dimensional disk. The resulting space can be deformed into its equatorial disk,
which is a two dimensional disk whose boundary has been contracted to point,
i.e. a 2-sphere as claimed above !
The fact that we are able to understand this homotopy type both topologically
and in terms of our weak model structure suggest that this problem is not an
obstruction to Simpsons conjecture, but only a sign that we need a more subtle
geometric realization functor, we hence propose the following conjectures, which
basically form a more precise version of C.Simpson conjecture:

A.9. Conjecture : There exists a geometric realization functor | | from P+ to

Spaces (topological spaces, or simplicial sets) such that:

The geometric realization of the polygraph P0 (with just a single cell) is a

point.
| | is a left adjoint functor.
| | send monomorphism in P+ to cofibration of spaces.
If P is an n-plex for n > 0 and A is the sub-polygraph of P obtained
by removing the unique n-dimensional cell of P and a single (n 1)-
dimensional cell of P , then the cofibration |A| |P | is a weak equivalence.

41
Such a functor would give a left Quillen functor from P+ to Spaces sending the
point to the point. Hence if our version of the Simpson conjecture holds, and
P+ is indeed Quillen equivalence to spaces, then the usual universal property
of the model category of spaces should implies that:

Conjecture : Any functor satisfying the condition of the first conjecture is a

left Quillen equivalence.

And these two conjectures, together with the unproved claim that we made
earlier on the existence of weak model structures, implies C.Simpsons conjec-
ture. We believe the hard part is the first conjecture, i.e. constructing a good
geometric realization functor

A.10. Another possible approach to completely circumvent this difficulty that

we are also investigating is to restrict the class of polygraphs we are using to
avoid these non-contractible plexes. This would corresponds to restricting the
type of composition allowed in our non-unital categories to diagrams which are
topologically balls, i.e. things like whiskerings and horizontal compositions:

x y z x y z

would not be defined, but composition of more complex shapes, possibly con-
taining them, but which are topologically balls with nice boundary inclusion
like:

x y z or x y z

would be defined, and associative in the sense that any two way of composing
such diagrams would be equals. One could recover weak form of whiskering and
horizontal composition by inserting weak units in such diagrams. This form
of the conjecture would be in some sense weaker and is probably easier. At
the present time, the main reason why we have not been able to prove this
form of the conjecture is that we are not yet able to extend our arguments
about existence of weak model structure on P+ and catnu to this weaker
framework. The point is that we use the Gray tensor product to construct the
weak model structure and it seems surprisingly difficult to show that this class
of polygraphs is indeed stable under the Gray tensor product.

A.11. Finally, we clarify the main three differences between our approach and
the original approach of [12]:
One has changed the category of diagrams.
We use the category of presheaf on P lex+ (i.e. the composition diagrams
with just a single (n + 1)-cell between two n-pasting diagrams) instead of
the category of all pasting diagrams (polyplexes).

42
 We are not introducing degeneracies in our category of diagrams.

We already discussed in length why the change in the category of diagram was
necessary, and we explained that there seem to be an incorrect assumption used
in lemma 3.4 of [12] regarding the fact that Johnsons diagrams can be composed.
Moreover as the goal is remove units, not introducing degeneracies seem very
natural as those essentially corresponds to unit. One should also mention that
the absence of degeneracies is the reason why we need to move to weak model
structures instead of Quillen model structures: for example it is well known
that there is no model structure on semi-simplicial sets (simplicial sets without
degeneracies) where the (trivial) fibrations are the usual Kan (trivial) fibrations
and the weak equivalences are the homotopy equivalence, but we leave as a very
nice exercise to an interested reader to show that there exists a right semi-model
structure on the category of semi-simplicial sets with these properties.
But one might wonder what is the meaning of replacing presheaves on the
category of all pasting diagrams to presheaves on this smaller category. We
believe that, if correctly taken into account this is a completely unessential
change. But it seems that it has not been correctly taken into account in [12]
and that using the category of all pasting diagrams is actually responsible for
at least one direct mistake in their paper: it seems that this makes the lemma
3.4 already mentioned above trivially false. This lemma claims in particular
that if F is a presheaf on their category of Johnson diagram and X denotes
the -category generated by X (the image of X by the left Kan extension of
the natural free category functor from Johnson diagrams to -cat) then any
arrow of X can be represented by a cell of X, i.e. an element of X(a) for a a
pasting diagram.
We already mentioned that the the proof of lemma 3.4 seem to use that Johnson
diagram can be composed, which is not the case, but another problem is that,
even in situation where all the Johnson pasting diagrams appearing can be
composed there is still no way to compose cells of a presheaf on the category of
Johnson diagrams, even in the 1-dimensional situation:
Take the presheaf C obtained as a gluing of two copies of the representable object
glued along the representable object . The free -category obtained
is just the 1-category with two arrow . But colimits in presheaf
category are computed objectwise, so if A is any Johnson pasting diagram any
map from A to C has to factor in one of the two map , so no such map can
ever represent the composite of the two arrow in the free -category generated
by C !
As far as we know there are at least two solutions to this problem: The first is the
one we proposed above to restrict to presheaves on the category of pasting dia-
gram with only one top dimensional cell, and having the other pasting diagrams
represented by some gluing of those representable pasting diagrams. In this
presheaf category the object C is the same as the pasting diagram
and hence the problem disappear. Another alternative would be to restrict to
fibrant objects (Kan complexes in the terminology of [12]) in lemma 3.4 (which
would be sufficient for the rest of the argument) and to add to in the definition
of fibrant object a condition forcing cells to be weakly composable. Not that
this change in the definition of fibrant objects is probably necessary for the re-
sults of their section 2 to be true without restricting the category of diagram,
as we do not see how they obtains a group structure on the n without any

43
assumption of this kind.
While this problem seems easily fixable, we believe that fixing it would proba-
bly only makes the real problem of this lemma 3.4 appears: as we mentioned
above, one cannot compose Johnsons diagrams in general, and that would pro-
vide other kind of counterexamples to this lemma, and secondly, degeneracies
seems to allows to construct a presheaf such that an Eckmann-Hilton collapse
happen in the free -category generated by this presheaf, in this case the iden-
tity cell going between some u#0 v to a v#0 u which are equal because of an
Eckmann-Hilton collapse would not be representable by a single Johnson di-
agram with correct boundary. We initially wanted to give explicit examples
of this two phenomenons, but unfortunately the first type of counterexamples
related to the complete absence of compositions of cells whether the diagram
compose or not appears so often that it seems nearly impossible to actually con-
struct any interesting other kind of examples without first choosing a solution
to this problem.

References
[1] Michael A. Batanin. Computads for finitary monads on globular sets. Con-
temporary Mathematics, 230:3758, 1998.
[2] Michael A. Batanin. Computads and slices of operads. arXiv preprint
math/0209035, 2002.
[3] A Burroni. Automates et grammaires polygraphiques. Diagrammes,
67(68):932, 2012.
[4] Albert Burroni. Higher-dimensional word problems with applications to
equational logic. Theoretical computer science, 115(1):4362, 1993.
[5] Aurelio Carboni and Peter Johnstone. Connected limits, familial repre-
sentability and artin glueing. Mathematical Structures in Computer Sci-
ence, 5(4):441459, 1995.
[6] Aurelio Carboni and Peter Johnstone. Corrigenda for connected limits,
familial representability and artin glueing. Mathematical Structures in
Computer Science, 14(1):185187, 2004.
[7] Eugenia Cheng. A direct proof that the category of 3-computads is not
cartesian closed. arXiv preprint arXiv:1209.0414, 2012.
[8] Victor Harnik, Michael Makkai, and Marek Zawadowski. Computads and
multitopic sets. arXiv preprint arXiv:0811.3215, 2008.

[9] Claudio Hermida, Michael Makkai, and John Power. On weak higher
dimensional categories i: Part 1. Journal of pure and applied algebra,
154(1):221246, 2000.
[10] Michael Johnson. The combinatorics of n-categorical pasting. Journal of
Pure and Applied Algebra, 62(3):211225, 1989.

44
[11] Mikhail M Kapranov and Vladimir A. Voevodsky. Combinatorial-geometric
aspects of polycategory theory: pasting schemes and higher bruhat or-
ders (list of results). Cahiers de Topologie et Geometrie differentielle
categoriques, 32(1):1127, 1991.
[12] Mikhail M Kapranov and Vladimir A. Voevodsky. -groupoids and homo-
topy types. Cahiers de Topologie et Geometrie Differentielle Categoriques,
32(1):2946, 1991.
[13] Joachim Kock. Weak identity arrows in higher categories. International
Mathematics Research Papers, 2006, 2006.
[14] Michael Makkai. The word problem for computads. Available on the au-
thors web page http://www. math. mcgill. ca/makkai, 2005.
[15] Mihaly Makkai and Marek Zawadowski. The category of 3-computads is not
cartesian closed. Journal of Pure and Applied Algebra, 212(11):25432546,
2008.
[16] Francois Metayer. Cofibrant objects among higher-dimensional categories.
Homology, Homotopy & Applications, 10(1), 2008.
[17] Francois Metayer. Strict -categories are monadic over polygraphs. Theory
and Applications of Categories, 31(27):799806, 2016.

[18] Jacques Penon. Approche polygraphique des -categories non strictes.

Cahiers de Topologie et Geometrie Differentielle, 40(1):3180, 1999.
[19] A John Power. An n-categorical pasting theorem. In Category theory, pages
326358. Springer, 1991.
[20] Carlos Simpson. Homotopy types of strict 3-groupoids. arXiv preprint
math/9810059, 1998.
[21] Richard Steiner. Omega-categories and chain complexes. Homology, Ho-
motopy and Applications, 6(1):175200, 2004.
[22] Ross Street. Limits indexed by category-valued 2-functors. Journal of Pure
and Applied Algebra, 8(2):149181, 1976.
[23] Ross Street. The algebra of oriented simplexes. Journal of Pure and Applied
Algebra, 49(3):283335, 1987.
[24] Ross Street. Parity complexes. Cahiers de topologie et geometrie
differentielle categoriques, 32(4):315343, 1991.

45