You are on page 1of 16

Imaging and imagining assemblages

E.V. Petracou and A. N. Yannacopoulos

Department of Geography, University of the Aegean,


and
Department of Statistics, Athens University of Economics and Business

Abstract
The notion of connectivity in its various forms is fundamental to the theory of assemblages.
Relations define connections, and different connectivities may lead to different structures with
varying properties as far as symbiosis or structural stability are concerned. We elaborate on an
imaging framework for the mapping of assemblages of wide and general applicability that at the
same time allows one to quantify and capture concepts such as connectivity, nonlinearity as well
as processes such as the dynamic changes of assemblages due to internal or external factors and
contingencies.

1 A brief introduction to assemblage theory


Assemblage theory is founded on the works of Deleuze and Guattari and has been further elaborated
by thinkers such as DeLanda, whose extensive work regenerated the interest in the field, Bonta and
Protevi, Phillips, Hillier, Robbins, Marks, Anderson and McFarlane. Assemblage theory is a concep-
tual toolbox which explores complex systems characterized by nonlinearity and far from equilibrium
operating trajectories. The most important feature of assemblages is the concept of connectivity of
heterogeneous elements, loosely connected subsystems or elements which operate together while keep-
ing their autonomy. The term element or subsystem is used in a very liberal sense, and the same
applies for connection which is used in a very loose sense, comprising any sense of relation between
elements, no matter how heteregeneous these may be. Furthermore, relations can appear on different
scales and the same element may participate in more than one such relations or connectivities, e.g. a
place can at the same time be a touristic place, a point of entry of refugees, a commercial center, a
cultural center and a national or a peripheral border, and an island. Relations are open and fluid, in
the process of becoming, and in continuous transformation.
Assemblage is a term employed to denote such a heterogeneous mixture of parts, and their multiple
connections at different scales, in a state of symbiosis and synthesis, while retaining their principal
identities, to comprise a from in constant transformation. Such forms are to be differentiated from
structures, as in philosophy elements participating in a structure lose their autonomy and function as
part of a whole while in an assemblage subsystems retain their autonomy and independent function-
ality. Assemblage theory is characterized by its high level of abstraction, elements can be practically
anything and the same applies for relations. Difference is considered as a creative force and the sig-
nificant point is not subjects, objects or concepts as such, but rather their connections, interactions
and flows (Bogue (2008) 84, 91). As Deleuze and Parnet mention “what matters is not the affiliations
but rather the alliances, co-functioning and symbiosis” Deleuze and Parnet (1987).
The transformation of form of an assemblage may come either from external or from internal
forces, and may possibly be subject to contingency. Deleuze and Guatari suggested the terms “double
articulation” and “ Body without Organs” in order to present a new model of the process of creation

1
AN
IT

TINA
N A
TI

Figure 1: Mobius band

of new strata, actualized systems with homogenized components operating at or near equilibrium
steady state (Protevi and Bonta (2004) p. 150). Such transformations are undertaken by the so called
abstract machine, an elusive concept under the action of which an assemblage evolves to new forms.
The abstract machine consists of the plane of content (which parts will be the basis of the formation
of a new entity) and the plane of expression (the parts that bring the new forms into existence).
Both planes cooperate in the transformation into a new form, i.e. in the process of actualization. As
DeLanda points out these planes are functioning simultaneously, in order to transform virtual forms
into actual forms DeLanda (2006).
Moreover, DeLanda argues that there is a distinction between properties of a part and its capacities.
The first refers to actualized features whereas the second to features that may be actualized as a result
of the action of the abstract machine, and consists of virtualities for the system. Based on that, along
with the crucial part that connectivity plays, and since connectivity is not related to distance at this
abstract level, (Marcus and Saka (2006) state that “assemblage is a topological concept that designates
the actualizations of the virtual”.
The distinction between actual and virtual plays a fundamental role in the theory of assemblages,
and emergence lies in between. As Hillier points out “the virtual is what conditions emergence and
the actual is what emerges” (Hillier and Abrahams (2013), p.40). The set of all virtual states, is what
Deleuze and Guattari refer to as the plane of immanence or the plane of consistency which is the virtual
realm of potentialities. The abstract machine acts on any one of these and will turn into assemblage
(i.e. become actualized) only the element of the set of virtual states on which it chooses to act upon.
The action of the abstract machine on the space of virtualities is what causes evolution, and multiple
actions (e.g. over time or over varying scales) produce curves of trajectories of the system, which
may be considered possible histories of the system, leading to the notion of steady states, singularities
or attractors, which may serve as skeletons of actual behaviour. When acting on certain states the
abstract machine will repeat itself or produce nearby states, thus creating persistent structures which
are likely to be observable. As assemblages may be endowed with parameters, for different choices
of the parameters the action of the abstract machine may vary, leading to different actualizations of
virtualities and diagrams essentially connecting the choice of parameters to the persistent outcome
(e.g. bifurcation diagrams) help understand the phenomenon and the mechanics of the transformation.
The outline of the traits of expression of an abstract machine is often called a virtual diagram.
An important dimension of any assemblage on any scale is territoralization which means the
continuous processes along which the assemblage is to be stabilized or established or destabilized and
perish (de-terorialization). One of the forces that destabilizes the identity of an assemblage is the lines
of flight. The lines of flight means change and transformation.

2
Figure 2: A coffee cup stripped down to its bare essentials is equivalent to a dougnut

2 Imaging assemblages
As the main tool for imaging an assemblage we may consider the concept of the simplicial complex.
The simplicial complex has been used since the time of Poincaré as a tool for visualizing complex
geometric objects, by stripping them down to their very bare connectivity properties, thus giving rise
to the field of Analyse Situs which later developed to the fields of topology and combinatorial topology,
two of the cornerstone fields of modern mathematics. In topology sets which can be transformed
into one another via continuous transformations (i.e. transformations which do not allow a cutting
operation) are considered as equivalent, thus leading to the standard joke that to a topologist a cup
and a doughnut are the same object (to which we would like to comment that this is indeed so; who
can enjoy a cup of coffee without a dougnut?). In Figures 1 and 2 these ideas are illustrated. This
level of abstraction may be too much for some purposes, leading to content free statements, on the
other hand it may be a precious looking glass for other purposes leading to very general and profound
statements concerning the crux of the matter regarding general patterns of spatial arrangement or
connectivity. As such an example one may refer to the famous Euler charcteristic according to which
for any spherical polyhedron (or simplicial complex) the number of vertices V , edges E, and faces
F satisfy V − E + F = 2, whereas any simplicial complex is characterized by χ = V − E + F ,
the so called Euler characteristic of the complex. In the 1970’s R. Atkin, a mathematical physist
turned social scientist and planner, infuenced by the Bourbaki attitude towards axiomatization and
abstraction and its trasfer to other sciences, introduced the concept of the simplex and the simplicial
complex to social systems and planning (see e.g Atkin (1974), creating thus a new school of thought
which influenced other scientists such as Casti (see e.g. Casti (1979)) or Johnson (see e.g. Johnson
(2013)) who extended these ideas to the study of complexity and dynamics in complex systems.
We briefly introduce the concept of the simplicial complex focusing on its connection with various
key ideas that emerge within assemblage theory. Consider a set S of n entities each denoted by ei ,
i = 1, · · · , n. We are deliberately vague as to what these entities may be, as to the level of abstraction
that we choose to work at, they could literaly be anything, from agents taking place in a bargaining,
abstract concepts participating in the definition of a more complex concept etc. A complex abstracts
the connectivity between the various entities of S (hereafter called vertices ei ) in the following way:
If vertex ei is not connected to any other vertex we denote that as the 0-simplex [e0 ], if vertex ei
is connected with vertex ej , then we denote that as the 1-simplex [ei , ej ], if the vertices ei , ej , ek
are connected we denote that as the 2-simplex [ei , ej , ek ] etc. Unlike a graph, which only allows
connectivities pairwise, i.e., connectivities of two objects only in a direct fashion, a simplex allows
direct connectivities of more than two objects, i.e. three objects in the case of a 2-simplex, four
objects in the case of a 3-simplex etc. This is convenient framework for considering possibly complex
interactions and connections in fields like semantics, social science, planning etc, and introduces a
trully nonlinear aspect as a higher dimensional simplex cannot be simply be perceived as the sum of
lower dimensional simplices. An nice example comes from Katz (Katz (1979)) who trying to define
what vanilla ice cream is all about, describes it as the 4-simplex [e0 , e1 , e2 , e3 , e4 ] where the vertices
ei correspond to the concepts of cold, sweet, vanilla aroma, softness and yellow respectively, and this

3
4-simplex cannot be considered as the sum of smaller simplices such as [e0 , e1 ]+[e2 , e3 ]+[e4 ] . Another
example comes from an interaction of 5 agents in a decision procedure or a discussion, an interaction
of all five, modelled by a 4-simplex is not equivalent as pairwise interactions, even if word of what has
been discussed in each group is revealed by common agents,

[e0 , e1 , e2 , e3 , e4 ] 6= [e0 , e1 ] + [e1 , e2 ] + [e2 , e3 ] + [e3 , e4 ].

A system may consist of a collection of simplices, each modelling subsystems which may or may not
be connected between themselves and such a collection of simplices C is called a simplicial complex.
We may therefore define an abstract simplicial complex as simply a collection of subsets pf S with the
property that if it contains a set X ⊂ S then it also contains all the subsets of X. A subset X ⊂ S
contained in this collection C is called a simplex (of dimension 1 less than its cardinality) and all its
elements are assumed as directly connected. An abstract simplicial complex will be an abstract model
for any system providing information on the important aspects of the system such as the connectivity
between the various subsystems. Since the elements of the set S can literally be anything and there
is no need to specify the nature of the relation which drives us into constructing the subsets of S
that we consider as simplices, and furthermore, since for the same choice of S certain subsets may be
constructed with a different relation leading to affinities than others, it is very tempting to consider
an abstract simplicial complex as a model, a depiction of an assemblage. The generality of the above
construction allows full freedom and a unified description of possibly very diverse objects, focusing
on the aspects that assebmlage theory consider as crucial, namely connectivity of possible diverse
subsystems, out of which different forms of organization may arise.
Does that help us in imaging a particular assemblage? Since topological properties (and connec-
tivity is one of them) are invariant under continuous mappings we may consider that we map each
element ei of S to a point (denoted the same for simplicity) of some Euclidean space, therefore map-
ping 0-simplices to points, 1-simplices to lines, 2-simplices to triangles (trully 2 dimensional objects),
3-simplices to tetrahedra (trully 3 dimensional objects) and so forth. Through this mapping procedure
a particular assemblage can thus first be abstracted to an abstract simplicial complex and then mapped
to a geometric simplicial complex, a collection of polyhedra viewed as a subset of a Euclidean space,
in which human perception is well versed and trained. The complexity and geometry of connections
in the assemblage are thus easy to be perceived via this map, which creates a visual image of the
assemblage. Of course the human eye is restricted to 3 dimensional space, so that simplices of dimen-
sion 3 and higher are not directly visualizable however higher dimensional objects may be visualized
either by projection techniques on 3d space or by appropriate intersection with lower dimensional
objects (techniques akin to tomography). This imaging procedure may be supported by invoking the
geometric realization theorem according to which any abstract simplicial complex of dimension k has
a geometric realization in R2k+1 , the Euclidean space of dimension 2k + 1.
Even if visualization becomes a bit difficult when dimensions higher than 3 are needed, the theory of
combinatorial topology and its various invariants may offer other alternatives in describing structure
of an assemblage viewed in terms of (or identified with) a simplicial complex and importantly its
connectivity properties when immediate perception fails us. To this task we can seek help from
homology theory, an algebraic theory which allows the classification of simplicial complexes. Consider
the simplex [e0 , e1 , e2 ], depicted as a triangle, which abstracts the direct interaction between 3 entities
e0 , e1 , e2 . As noted this is different than [e0 , e1 ]+[e1 , e2 ]+[e2 , e0 ] which abstract the pairwise interaction
of the entities (as a concrete example consider 3 agents discussing in the same pannel as opposed to
0 discussing with 1, 1 discussing with 2 and 2 discussing with 0, with 1 possibly communicating to 2
aspects of the discussion with 0, etc.).
In Figure 4 we provide the geometric representation of the simplex [e0 , e1 , e2 ] (left) and [e0 , e1 ] +
[e1 , e2 ] + [e2 , e0 ] (right), and note the visual and (as will be revealed soon) topological difference be-
tween these two: The left diagram does not exhibit a hole, in contrast to the right diagram which

4
P

Figure 3: A geometric realization of a simplicial complex

e2 e2

e0 e1 e0 e1

Figure 4: [e0 , e1 , e2 ] vs [e0 , e1 ] + [e1 , e2 ] + [e2 , e0 ]

has. This is a major difference between the two simplicial complexes, both are connected but their
connectivity is qualitatively different in that the first complex does not display a hole (corresponding
to direct connectivity of all 3 entities) whereas the second displays a void, an one dimensional hole
(corresponding to connectivity of all 3 entities never directly but rather always through an interme-
diary). This qualitative difference allows for a depiction of the two assemblages and a differentiation
between the two even if a geometric representation is not feasible on account of high dimensionality,
simply by counting the number of holes in each connectivity structure. Recall that earlier on we
identified the cofee cup with the doughnut as by a continuous transformation one could be brought
into the other. Why was that so? Simply because the number of one dimensional holes in both objects
were the same. This idea of counting holes as a means of imaging, mapping and visualizing simplicial
complexes (equiv. assemblages) can be made rigorous with the construction of the homology groups
for the system, an algebraic construction which allows one to distinguish closed boundaries into ones
that arise as boundaries of higher dimensional “compact” objects and closed boundaries which con-
tain a void. This construction, which can be turned into a fully algorithmic procedure results to the
definition of a family of numbers called the Betti numbers of dimension k, βk , which in non-technical
terms essentially count the number of k-dimensional holes in the assemblage (modelled as the simpli-
cial complex). The zeroth order Betti number β0 counts the number of connected components of the
complex and if β0 = 1 it means that all the entities in the complex/assemblage communicate albeit
indirectly through lower dimensional connections (faces of the simplices). Both complexes in Figure
4 have β0 = 0. The first order Betti number β1 counts the number of one dimensional holes of the
complex. The complex in Figure 4 on the left has Betti number β1 = 0 (there is no one dimensional
hole as it is a 2-simplex which geometrically corresponds to a filled triangle) whereas the complex in
Figure 4 on the right has Betti number β1 = 1 (there is an 1-dimensional hole as the 2-simplex has
dissapeared giving rise to three 1-simplices, geometrically depicted as lines, which do give a closed
boundary but do not fill the internal region, leaving an 1-dimensional hole). Since the simplices in
Figure 4 are at most 2 dimensional, we cannot discuss higher dimensional holes or void than dimension
1.
In Figure 5 we display [e0 , e1 , e2 , e3 ], solid tetrahedron (on the left) vs [e0 , e3 , e1 ] + [e1 , e2 , e3 ] +

5
e2 e2

e3 e3
e0 e1 e0 e1

Figure 5: [e0 , e1 , e2 , e3 ], solid tetrahedron (on the left) vs [e0 , e3 , e1 ]+[e1 , e2 , e3 ]+[e0 , e2 , e3 ]+[e0 , e1 , e2 ],
the shell or surface of the tetrahedron without the interior (on the right).

e2 e5

e0 e1 e3 e4

Figure 6: A simplicial complex consisting of

[e0 , e2 , e3 ] + [e0 , e1 , e2 ], the shell or surface of the tetrahedron without the interior (on the right).
The shell of the tetrahedron creates a single 2 dimensional void, in the place of the interior of the
tetrahedron that has been removed and that can be captured by the two dimensional Betti number β2 ,
which can be formally calculated to be for the particular simplicial complex [e0 , e3 , e1 ] + [e1 , e2 , e3 ] +
[e0 , e2 , e3 ] + [e0 , e1 , e2 ] equal to β2 = 1. The other Betti numbers for the ...
The Betti number can be calculated using the concept of the boundary operator which maps
a simplex to a properly defined (formal) algebraic sum of the simplices defining the faces of the
simplex. In the context in which a simplex is used here the boundary operator ∂n decomposes a higher
dimensional concept which can best be described by the direct interaction of n + 1 entitiesσn+1 :=
[e0 , e1 , · · · , en ] into the algebraic formal sum of its lower dimensional faces [e0 , · · · , ei−1 , ei+1 , · · · , en ],
i = 0, · · · , n, each one corresponding to concepts related with direct interactions of any of the n entities
remaining if entity ei is rejected from the original concept. For instance, if we try to understand the
dynamics of a particular agreement between n + 1 agents by applying the boundary operator ∂ on the
full agreement (grand coalition) we obtain all the possible agreements and negotiations between the
n − 1 agents remaining after the agent i has abandoned the grand coalition for any i = 0, · · · , n. One
the remaining formal sum of n − 1-simplices (possible direct agreements of n − 1 agents) we may once
more act the boundary operator to obtain all possible n − 2-faces of the n − 1-simpices which in this
example correspond to smaller agreements of n − 2 agents that may occur on the occasion of any of
the agents j leaving the remaining agreements etc.
As an example consider the simplicial complex of Figure 6. In this simplicial complex which can
be expressed as the collection of simplices

{[e0 , e1 , e2 ], [e2 , e3 ], [e3 , e4 ], [e4 , e5 ], [e3 , e5 ]}

which is further completed with all the lower dimensional faces of higher dimensional objects as

{[e0 , e1 , e2 ], [e2 , e3 ], [e3 , e4 ], [e4 , e5 ], [e3 , e5 ]; [e0 , e1 ], [e1 , e2 ], [e0 , e2 ], [e0 ], [e1 ], · · · , [e5 ]},

the auxiliary faces following after the semicolon. This simplicial complex is to be interpreted as
a direct interaction of entities e0 , e1 , e2 and then entities e2 , e3 , e3 , e4 , e4 , e5 and e3 , e5 interacting
pairwise, i.e. consists of one 2-simplex and four 1-simplices, and when augmented with the inclusion
of the auxiliary lower dimensional faces it consists of six 0-dimensional simplices, seven 1-dimensional

6
simplices and one 2-dimensional simplex. The the action of the operator ∂2 is only meaningful on
the 2-simplices of the complex and will yield the simplicial complex [e0 , e1 ] + [e1 , e2 ] + [e2 , e0 ] which
corresponds to all partial interactions by pairs obtained when the direct interaction [e0 , e1 , e2 ] fails
and which if visualised as a geometric object yields a closed curve (the blue curve on Figure 6). One
can observe that the simplicial complex of Figure 6 displays two such closed curves. The curves
[e0 , e1 ] + [e1 , e2 ] + [e0 , e2 ] (blue curve) and [e3 , e4 ] + [e4 , e5 ] + [e3 , e5 ] (red curve). The fact that the
curves are closed can be interpreted as that entities e0 , e1 , e2 will indirectly interact (even if their
direct interaction breaks down) through mutual pairwise interactions, and similarly entities e3 , e4 , e5
will indirectly interact through mutual pairwise interactions. However, even though are both curves
are closed there is a fundamental difference between then as the blue one derives as the remant of
a higher dimensional object (it is the boundary of the 2-simplex [e0 , e1 , e2 ]) whereas the red one
does not derive from such a higher dimensional object. The blue one therefore is an interaction
that bear some history, which may “flavour” it with special characteristics thus differentiating it
from the red one which does not carry any past history. This history may help explain patterns of
behaviour or even predict new ones. How would we see that the red and the blue curve are closed
without resorting to the geometric representation? We may act once more a boundary operator on
the simplices [e0 , e1 ] + [e1 , e2 ] + [e0 , e2 ] and [e3 , e4 ] + [e4 , e5 ] + [e3 , e5 ], now the boundary operator
∂1 which will provide information on what will remain if the present pairwise interactions between
these
 entities break down. Using  the conventions
 1 + 1 = 0 and ∂1[ei , ej ] = [ei ] + [ej ] we see that
∂1 [e0 , e1 ] + [e1 , e2 ] + [e0 , e2 ] = 0 and ∂1 [e3 , e4 ] + [e4 , e5 ] + [e3 , e5 ] = 0, which can be interpreted
as the closeness of the corresponding curve in the geometric representation of the complex. However,
note that in the initial simplicial complex of Figure 6 there exists a 2-complex [e0 , e1 , e2 ] such that
∂2 [e0 , e1 , e2 ] = [e0 , e1 ] + [e1 , e2 ] + [e0 , e2 ], whereas no such entity exists for [e3 , e4 ] + [e4 , e5 ] + [e3 , e5 ]; this
would have been [e3 , e4 , e5 ] but no such entity exists in our initial simplicial complex. To differentiate
between these two different types of closed 1-chains we need a construction called homology group
which factors out from the set of closed 1-chains, the contribution of these closed chains which arise
as boundaries of 2-simplices. This consists of closed 1-chains of the form c + b where b is any chain
arising as the boundary of a higher dimensional object and c is a closed chain which does not. We
try to express any closed chain in the simplicial complex in this form and we only keep track of the
c part, discarding the b part, and collect the c component into a set called the H1 homology group.
If that consists only of 0, then all closed 1-chains of the complex are boundaries of 2-simplices, if not
there exist other closed 1-chains except boundaries of higher dimensional simplices. The structure of
this group provides information on the number of distinct 1-dimensional holes in the structure. For
instance in the simplicial complex of Figure 6 there is only distince class of closed 1-chains that are
not boundaries and this is the red curve, and this is quantified by the Betti number β1 = 1. On the
other hand for the simplicial complex of Figure 7 there are two distinct classes of closed 1-chains that
are not boundaries and this is the red curve and the green curve, and this is quantified by the Betti
number β1 = 2.
Similar notions apply for higher dimensional objects if our simplicial complex carries such by
defining higher dimensional Betti numbers characterizing the number of higher dimensional voids.
Such voids, corresponds to gaps in the structure, and in the context where simplicial complexes are
used here provide information on the history or geneaology of the observed structures.
Given a set of entities V = {e0 , e1 , · · · , en }, we may now consider all possible interaction patterns
between these entities, by constructing the set of all possible simplicial complexes C that may arise
out of these vertices (entities). This is in general a big set consisting of complexes which correspond
to all possible interaction patterns. For example if we have the set of entities V = {e0 , e1 , e2 } then all

7
e2 e4
e3

e0 e1

Figure 7: A simplicial complex consisting β1 = 1.

e2 e4 e2 e4
e3 e3

e0 e1 e0 e1

Figure 8: β1 = 2 (left) β1 = 1 (right)

possible interactions (all possible simplicial complexes) that may be formed out of this are

c1 = {[e0 ], [e1 ], [e2 ]}, c2 = {[e0 ], [e1 , e2 ]}, c3 = {[e0 , e2 ], [e1 ]},
c4 = {[e0 , e1 ], e2 ]}, c5 = {[e0 , e1 , e2 ]}, c0 = ∅,

whiich are 6 in total including the empty set. The set

C = {c0 , c1 , c2 , c3 , c4 , c5 },

contains all possible scenaria regarding the connectivity and interaction between these 3 entities, and
contains 6 qualitatively different outcomes which for example in the case where we model a negoti-
ation process correspond to all possible cases from complete breakdown of the negotiation (c1 ) to a
complete unanimous agreement (c5 ). The set contains the virtual and a geometric realization and
depiction of the elements of this set may be considered as a depiction of a virtual diagram in the
terminology of DeLanda, in the sense that it allows the enumeration and visualization of all vir-
tual states of the system. Clearly the actual state of the system must be one of the elements of
C, however, at this stage we do not touch upon the problem of selection of an actual state from
the set of possible virtual states. An idea of the complexity of possible virtual patterns that may
arise is given by the rapidly growing number of elements of the set Cn as n grows. As an indi-
cation of these numbers we state that starting from n = 1 the number of elements of the set Cn
is 1, 2, 9, 114, 6894, 7785062, 2414627396434, 56130437209370320359966, providing a clear indication of
the huge number of potentialities and of various structures and forms that may arise.
In Figure 9 we display three possible elements of C8 , corresponding to 3 different possible virtual
states concerning the structures that may arise out of the interaction of 8 entities. With the dark colour
we denote 3-simplices (i.e. direct interaction of 4 entities), with the light color we denote 2-simplices
(i.e. direct interaction of 3 entities), with lines 1-simplices (i.e. direct interaction of 2 entities) and
with points 0-simplices (i.e. an entity that does not interact with any other entity). Going from the
top to the second figure we observe the breakup of a 3-simplex and the creation of another by the
joining up of adjacent 2-simplices (the faces) whereas going to the bottom we observe the generation of
a new 3-simplex by the joining up of previously unconnected 2 simplices. Such depictions allow us to
visualise the possibility of generation of higher order and structure out of local less structured entities
(as opposed to tree structure). A transformation of one element to another is what corresponds to a
line of flight, the creation of new structures and forms out of existing ones.

8
Figure 9: Three elements of C8 presenting different virtual states that may arise from local organization.

What is the process that may drive the system from the top configuration to the bottom con-
figuration or perhaps to any other of the huge number of possible configurations which are elements
of C8 ? It can be from deterministic dynamics to sheer luck or a mixture of both. Specifying this
process will alllow us to determine the configurations of the system which are more likely to appear,
or determine the singularities or the time that it will take before one form or structure is succeeded by
another. However, this requires a detailed understanding of the “mechanics” of the system, the forces
that bind the various entities together or push them apart, if any, possibly knowledge of the history or
geneaology of the system, which substructures evolved from other higher ordered structures (a piece
of information that can be provided by homology etc. We will provide certain proposals dealing with
this situation in the next section where a dynamical systems approach on the set Cn will be proposed.
Before turning to that however, we wish to address the problem of scale and how the imaging
procedure of the structures via simplicial complexes may allow the visualization of structures at
different scales. The concept of scale is important in complex systems theory and in the theory of
assemblages.
To introduce the concept of scale consider the following geometric realization of n vertices e0 , e1 · · · , , en .
In order to see whether the vertices will interact directly or not we propose the following construction
(known in algebraic topology as the Cech complex). With each of the vertices as centers, construct
balls of growing radii. Whenever the disks intersect then we assume a direct interaction between their
centers, and construct the corresponding simplex. In Figure 10 we present an example for 3 vertices
depicted on the plane. In panel (a) the radii of the balls are too small and there is no intersection
between them (the balls are too small for the centers to interact) and this gives rise to the simplicial
complex {[e0 ], [e1 ], [e2 ]} in panel (a-s). In panel (b) the radii of the balls are larger and the balls
centered and e0 and e2 intersect providing a direct interaction between the entities e0 , e1 , i.e. the 1-

9
e2 e2

e0 e1 e0 e1

(a) (a-s)

e2 e2

e0 e1 e0 e1

(b) (b-s)

e2 e2

e0 e1 e0 e1

(c) (c-s)

Figure 10: Visualizing at different scales

simplex [e0 , e1 ] and the balls centered and e1 and e2 intersect providing a direct interaction between
the entities e1 , e2 , i.e. the 1- simplex [e1 , e2 ]. Therefore at this level (scale) we obtain the simplicial
complex {[e0 , e1 ], [e1 , e2 ]; [e0 ], [e1 ], [e2 ]} (panel (b-s)). Increasing the radii of the balls even further
(panel (c)) all three balls have a non empty intersection so that the simplex [e0 , e1 , e2 ] is obtained
(panel (c-s)). The increase of the size of the balls corresponds to a change in scale going from finer
to coarser scale, and the simple example shown in Figure 10 shows that different patterns, different
forms, different topologies occur from the same system at different scales. This phenomenon resembles
our intuition from complex systems where looking at them on different scales produces different view
points, different localized structures etc some of which may be important and lead to new forms while
others may simply be short lived artifacts. To quote Hector Sussman, if you call a self adjoint operator
an elephant and a spectral expansion a trunk you can safely prove a theorem that any elephant has
a trunk, but (paraphrazing Sussman) this does not say much about large gray animals. To avoid
the risk of getting into such circular and void arguments we need, before processing any futher this
intriguing idea of scale, provide a meaningful interpretation for the concept of scale. For geographical
or planning applications where the concept of physical Euclidean space in meaningful (e.g. location of
facilities and their connection with other properties) the concept of scale can be connected with actual
distance. We provide an alternative example where physical space does not play a role. Consider
once more the example of a union such as for example the EU where each entity corresponds to a
member state that can be characterized by certain views on a manifold of issues. If there are m issues
of matter we can consider each entity (member state) as an m-tuple ei = (ei,1 , · · · , ei,n ). To simplify
the discussion we assume each issue ` to be described by a binary variable taking the value ei,` = 1 if
entity i is in accordance with this issue and ei,` = 0 if not. The distance between two entities (states) i
and j can be given by the Hamming distance (or any weighted version of it) . The Hamming distance
is a true metric in the space {0, 1}m . The larger the Hamming distance between two entities ei ans
ej the more separated in attitudes these two are. Performing the construction of the Cech complex
using the concept of the ball in the this metric space we may construct as above simplicial complexes
at different scales. When performing the construction for small radii we only join in the same simple
entities that are very close together in terms of this distance, i.e., entities that are in full agreement
in almost every issue of matter. Such simplices correspond to structures or direct interactions on a

10
very fine scale. As the radii of the balls increase we allow entities that are very far apart in terms of
this distance which quantifies affinities in beliefs or attitudes to join into the same simplex forming
a sinlge entity. This is an simplex consisting of rather greatly diversified units and as such it can be
considered as a structure observed at a rather large scale, where details in attitudes on certain issues
do not matter, pretty much like what happens if you look at a town from an airplane and individual
neighbourghoods and their characteristic differences are no longer visible. How does this help in the
imaging of the assembleges formed? Well, there is no way to plot the points in the space {0, 1}m
for m large and obtain a depiction of the structures and patterns formed, however, there is no real
need to perform plots similar to the left column of Figure 10 in order to check which vertices can be
joined to a single simplex. This step does not require any visual intuition at all and can be performed
algorithmically on a computer 1 to reduce the original collection of entities along with their fine char-
acteristics concerning beliefs and attitudes, to simplicial complexes similar to those shown onthe right
column in Figure 10. However, no matter how high m is, the simplicial complexes constructed with
this reduction at various scales can be imaged and contain important information on the connectivity
structures present at various scales.
Now comes the important notion of persistence. At different scales different connectivities which
are modelled as different simplicial complexes arise. As the scale changes different structures arise,
some are born and some dissappear. To quantify these metamorphoses, a concept proposed by Edels-
bruner and his co-workers in computational topology (see e.g. Edelsbrunner and Harer (2010)) called
persistent homology may be of great help. As already mentioned above, the construction of simplicial
homology and in particular the Betti numbers allows a quantification of structure in terms of the
number of holes or voids that may appear in various dimensions. In order to couple that with the
concept of scale we produce from the same assemblage simplicial complex imagesCi at different scales
i , i = 1, · · · , s with a procedure either related to the Cech complex construction or other related
procedures (e.g. the Rips-Vietoris complex or the Witness complex). Each Ci is to be considered as
a snapshot of the same assemblage at a different scale and to the level of abstraction that this con-
struction works the scale can be either spatial, temporal, social (e.g. the level of the family, village,
town ,factory etc). For each such complex Ci , i = 1, · · · , s one may perform a homology analysis
and record topological features (such as cycles, holes or voids) that arise at this scale thus providing
information for connectivity and cohesion at this scale. Performing the same analysis at a different
scale one may reveal the topological features at the new scale and by comparison figure out which
structures perished by the change of scale and which new structures appeared as a result of the new
scale at which the assembage is viewed. Topological features which are observed on more than one
scale are called persistent and are viewed as robust features of the system. In the context in which
these constructions are used here, persistent features may be considered as enduring structures or
formations in the social topology, structures that survive across scales, are stable and more likely to
be observed and which in certain applications they may be considered as universal connectivities valid
from the micro to the macro level. The bookeeping of structures that appear and dissapear can be
kept in what is called a persistence diagram or barcode in which at each dimension and for each void
the number of scales for which this particular void survived are recorded.
For example in Figure 10 the direct interaction of entities e0 , e1 , e2 only appears at the coarser
scale, as well as the interaction of e0 and e1 while at a finer scale they dissapear. On the other hand
the interactions of e0 , e2 and e1 , e2 persist both on the intermediate as well as on the coarser scale,
whereas they do not on the finer scale.
1
It may get a little tedius and ineffective for large values of m however, there are other alternatives and similar
constructions.

11
3 Imagining assemblages
We are now ready to formalize and quantify within this framework the concept of lines of flight which
was used by Deleuze and DeLanda for the phenomena of change of form or pattern or structure in an
asseblage. Within our construction, we have mapped an assemblage to a simplicial complex and given
the number of entities n which are the possible vertices of the assemblage, the set of all potentialities
of the system, is the set Cn which is the set of all possible simplicial complexes that can be constructed
out of these vertices. This is the set containing all virtual states of the system, each element of the set
is potentially a possible assemblage that can materialize (given the right conditions) out of the given
n entities. A line of flight then is nothing else but a transformation of the actual assemblage from
one element of the set Cn to another element. This transformation in the language of mathematics
is a dynamical system on the set of virtual states Cn , S : Cn → Cn , the action of which transforms
the current assemblage which is modelled by a simplicial complex c ∈ Cn to a new assemblage out
of the allowed virtual states c0 = S(c). The action of the transformation can be perceived as (the
realization) of a line of filght from one state to another. The fixed points of this map i.e. assemblages
c0 ∈ Cn such that c0 = S(c0 ) are what in assemblage theory are called singularities and are related
to the equilibria of the system. The exact form of the map S is important as it specifies different
possible future states the system may evolve to. In particular the determination of the singularities
of the system depends crucially on the form of the dynamical system T that models the evolution
of assemblages into different potentialities (or virtualities); different dynamics will provide different
singularities.
There is no need to assume determinism in the evolution of assemblages, this dynamical system S
may well be a trully random, solely subject to contingency process. In the same fashion that a coin out
of the two possible virtual states H and T upon tossing it will provided an actual state which is either
H or T with equal probability (here the dynamical system S is the action of tossing the coin in the air
and allowing it to land) and upon repeating will yield a new state among these two again with equal
probability, we may assume that luck chooses the next assemblage that is created out of the evolution
of the previous one. We may thus turn the set Cn into an event space (in complete analogy as we can
do with the coin and {H, T }) and assign a probability for a particular assemblage (simplicial complex)
c ∈ Cn to materialize out of the large number of virtual potential states. Unlike the case of the coin,
we do not expect here all virtual states to be equally probable, in terms of becoming materialized as
actualities. However, if the cardinality of the set Cn is m we may assign probabilities pi , i = 1, · · · , m to
each virtual assemblage ci ∈ Cn . Such probabilities may be subjective or objective and ideally should
be assigned after a careful qualitative study of the system. For example, the nature of the vertices
(entities) and the nature of what we perceive as the relation that allows us to establish a connection
or interaction between two or more vertices (i.e. what properties of the vertices establish whether a
simplex will be formed or not) is expected to play a very important role in assigning probabilities for
the occurence of certain simplicial complexes out of Cn . Returning back to our favourite example of
consensus in a “society” if this society is at the scale of a 4 member family then consensus is easier
so that the probability of occurence of a 3-simplex is high enough. If we consider the same example
at the scale of a country, viewing a referendum question as for example Brexit then the probability of
occurence of the simplex of higher dimensions (corresponding to the vast majority of voters agreeing
to one of the referendum questions) will be devastatingly low. With this probability we choose an
assemblage ci as the initial state of the system s0 . Once the initial actual assemblage has been chosen
out of Cn with this probability then the evolution process may begin. The evolution process may
be random, deterministic or a mixture of both. For instance, the next possible actual state s1 can
be chosen with the same probability distribution as cj with probability pj and continue like that.
Other possibilities are also available, which may take into account the actual qualitative nature of
the entities involved and the situation to be modeled. For instance, once a particular assemblage

12
st = ci has materialized out of the virtual set Cn the very fact that ci has materialized no longer
makes any assemblage cj likely to materialize with probability pj , but rather this probability changes
to pi,j 6= pj . It may be that some of the substructures already formed in ci are in conflict with some
of the substructures which dominate in cj therefore hindering and turning unlikely the formation of cj
once ci has been formed. Rather, the some of the structures already formed in ci could facilitate the
formation of some of the structures dominant in some other simplicial complex ck so that the system
is more likely to evolve to ck after c − i has materialized than to cj , so that pik > pij . For instance,
in Figure 9 we display an example of such a possible evolution paths on C9 , where we start with the
assemblage
c0 := {[e0 , e1 , e2 , e3 ], [e1 , e4 , e6 ], [e2 , e6 ], [e4 , e5 , e7 ], [e4 , e6 , e7 ], [e5 , e6 , e7 ], [e4 , e5 , e6 ]}
(where so as not to clutter notation too much we only note the primary simplices with the higher
dimension, omitting their faces which are of course tacitly assumed as included in the simplicial
complex), which evolves to
c1 = {[e0 , e1 , e2 ], [e0 , e1 , e3 ], [e0 , e2 , e3 ], [e1 , e2 , e3 ], [e2 , e6 ], [e1 , e4 , e6 ], [e4 , e5 , e6 , e7 ]}.
Why should the system choose this particular transition (line of flight) against any other of the
numerous possible transitions (as many as the cardinality of C8 )? If the image of the assemblage in c0
corresponds to some agreement, then the formation [e0 , e1 , e2 , e3 ] is likely to break up as interaction
may produce frictions between the members of an informal coalition, thus leading after the passage
of time (or at the next scale) to the formation [e0 , e1 , e2 ] + [e0 , e1 , e3 ] + [e0 , e2 , e3 ] + [e1 , e2 , e3 ] where
the entities which were originally joined in a solid 3-simplex now break up and retain relations in
which any 3 of them interact directly but there is no longer direct interaction of the 4 of them.
This line of flight produces a 2 dimensional void, which is quantified by an increase in the Betti
number 2 by 1. We can trace the ancestry of the closed surface [e0 , e1 , e2 ] + [e0 , e1 , e3 ] + [e0 , e2 , e3 ] +
[e1 , e2 , e3 ] as coming from the break up of a higher dimensional structure since ∂3 [e0 , e1 , e2 , e3 ] =
[e0 , e1 , e2 ] + [e0 , e1 , e3 ] + [e0 , e2 , e3 ] + [e1 , e2 , e3 ]. On the contrary the indirect interaction of e4 , e5 , e6 , e7
in [e4 , e5 , e7 ] + [e4 , e6 , e7 ] + [e5 , e6 , e7 ] + [e4 , e5 , e6 ] provided an incentive to these entities of forming the
direct interaction [e4 , e5 , e6 , e7 ] at the next time (or next scale) which leads to the disappearance of a
2-dimensional void from c0 to c1 , which is quantified by a decrease in the second Betti number by 1.
No other notable changes are made. At the next time instance (or the next scale) the frictions that
arose from the direct interaction of e4 , e5 , e6 , e7 led to the break up of the higher dimensional structure
[e4 , e5 , e6 , e7 ] while the members of the former union still enter in interactions by 3, and this changes
again the topological structure of the assemblage by the generation of one more 2 dimensional void,
increasing β2 by 1, and this 2-dimensional void can be traced back to the break up of the 3-simplex
[e4 , e5 , e6 , e7 ]. At the same time the entities e4 , e6 , e7 discover affinities with e1 (possibly through the
mediation of e4 ) and decide to form a direct interaction thus leading to the formation of the higher
dimensional structure [e1 , e4 , e6 , e7 ]. These changes lead to a line of flight that results in a change of
form of the assemblage and to the generation of a new assemblage
c2 = {[e0 , e1 , e2 ], [e0 , e1 , e3 ], [e0 , e2 , e3 ], [e1 , e2 , e3 ], [e2 , e6 ], [e1 , e4 , e6 , e7 ], [e4 , e5 , e7 ], [e5 , e6 , e7 ], [e4 , e5 , e6 ]}.
The topological changes are summarized in the changes of the Betti numbers, which from top to
bottom can be computed and summarized as
c0 β0 = 1 β1 = 1 β2 = 1
c1 β0 = 1 β1 = 1 β2 = 1
c2 β0 = 1 β1 = 1 β2 = 2
where as discussed above, eventhough from c0 to c1 the Betti number β2 is unchanged this is due to
the birth and death of two disctinct high dimensional structures, which were both short lived. On

13
e2 e6

e3 e7
e0 e1 e4 e5
e3 e1 e4 e7 e5

Figure 11: Three elements of C8 presenting different virtual states and possible lines of flight.

the other hand the one dimensional hole created by [e1 , e2 ] + [e2 , e6 ] + [e1 , e6 ] lasts on all three scales
and is a long lived, robust or persistent feature. This ancestry is captured in the persistence diagram
of Figure 12. For each dimension we follow the major structures in this dimension and record their
birth and death processes as time or scale evolves. This diagram is akin to a bifurcation diagram and
provides a depiction, a mapping of the topological changes of the assemblage as a parameter (time or
scale in this case evolves).
One possible fashion in which we can sample the space of all virtual states, is to randomly generate
patterns (assemblages) using various rules, that may change from application to application and may
reflect our understanding of the phenomenon we wish to model and/or understand. The random
generation of patterns allows for the simulation of the phenomenon, forcing the abstract machine to
randomly generate actual forms out of the space of virtual states following a consistent set of rules,
in an attempt to decipher the mechanics of the process of generation, and hoping to find persistent
features (topological structures) which are common in the majority of these experiments. For instance,
we may consider the question of whether it is typical for a contingently generated assemblage to present
connectivity properties of dimension 0 or not and if not what is the expected number of disconnected
substructures that may appear? The random generation of actual forms from virtual may have rules,
reflecting how the local structures may evolve or organize, and depending on the choice of rules (which
are often parametrized) different assemblages may form. Analysis or computer experiments may allow
the researcher to establish the persitent topological features of assemblages that will be more often
formed, by the action of this abstract machine. This process of simulation allows us to imagine
possible assemblages, thus expanding and transgressing the boundaries of reality and perception. At
the same time by comparing the outcomes of the actions of this imaginary abstract machine with the
actual assemblages observed in reality, we may tune the imaginary abstract machine so as to produce
outcomes as close to reality as possible and try to understand which are the rules of the abstract
machine that led to what is observed in practice.
The field of random simplicial complexes is relatively young, but it provides a promising tool
towards imagining assemblages and understanding the intricate dynamics of complex systems. Below

14
DIMENSION 0

0 1 2 SCALE

DIMENSION 1

0 1 2 SCALE

DIMENSION 2

0 1 2 SCALE

Figure 12: Persistence diagram for the evolution of the example in Figure 11.

we provide an example of an imaginary abstract machine (IAM) which may be parameterized. Start
from n vertices (entities). These vertices may form a set of virtual states Cn . In order to choose
which assemblage will be actualized from that, we first consider the question of connectivities by
two, and select a random graph among all possible elements of Cn which consist of graphs (simplicial
complexes of dimension at most one) as follows: Each edge (connection of two vertices) is included
in the graph with probability p1 , independently from any other edge. All graphs with n vertices and
n!
−M
M edges are equiprobable with probability pM 1 (1 − p1 )
2!(n−2)! . Once this graph has been chosen,
we look at the so called 3-cliques in the graph, i.e. simplices of the form [ei , ej ] + [ej , ek ] + [ek , ei ]
and for each 3-clique that we find with probability p2 we create the 2-simplex [ei , ej , ek ] or with
probability 1 − p2 we leave an one dimensional hole. At the next step we look for 2-faces of the form
[ei , ej , ek ] + [ei , ej , em ] + [ei , em , ek ] + [ej , ek , em ] and for each such structure we find with probabiity
p3 we create the 3-simplex [ei , ej , ek , em ] or with probability 1 − p3 we leave a 2 dimensional void.
We keep like that until we reach the end. This is an example of a parametric imaginary abstract
machine, depending on the parameters p1 , p2 , p3 , · · · , pn , the action of which on the set of virtual
states will actualize a contingent assemblage. Applying this IAM for a number of times we may
simulate the possible evolutions of the phenomenon and try to tune the parameters to such values
so as the persistent features of the produced assemblages resemble those observed in reality. Note
that this abstract machine highlights the importance of local structures on the generation of larger
structures as a k-simplex is only to be formed (with some probability) from a preexisting local k − 1-
simplex. If there is no set of parameters for which the outcomes of this IAM resembles the actual
then there is something in our assumptions of how the true abstract machine that has produced
the actual observed assemblage functions, most probably a different virtual diagram is to be used.
Topological properties such as connectivity and the Betti numbers can be computed for such models
as functions of the parameters, and analytic estimates can often be made. In Figure 13 we display
a random assemblage in R3 and its persistent topological properties at various scales. As mentioned
above setting the rules to the IAM requires some understanding of the system we wish to understand
and model.

Acknowledgements: All figures in this paper have been made using the Tikz Latex Package
and have been made by the authors with the exception of the Mobius strip and the coffee cup which
were based on Tikz code freely available on the net, while the calculations leading to Figure 13 we
performed using Javaplex and the authors wish to thank all those who contributed to these resources.

15
Figure 13: A random assemblage and its persistence diagrams

Many thanks also go to Professor S. Xanthopoulos algebraic topological brother in arms of ANY, for
enjoyable and enlightening discussions.
The authors wish to express their gratitude to Professor Helen Briasoulis for putting together
this exciting workshop who created a 2-simplex at a different scale between a social scientist and an
unsociable one or viewed otherwise a negative and a positive one, and for offering them the venue to
express these viewpoints.

References
Atkin, R. H. (1974). Mathematical structure in human affairs. Heinemann Educational London.

Bogue, R. (2008). Deleuze and Guattari. Routledge.

Casti, J. L. (1979). Connectivity, complexity and catastrophe in large-scale systems, Volume 7. John
Wiley & Sons.

DeLanda, M. (2006). A new philosophy of society: Assemblage theory and social complexity. A&C
Black.

Deleuze, G. and C. Parnet (1987). Dialogues, trans. Hugh Tomlinson and Barbara Habberjam (London:
Athlone, 1987).

Edelsbrunner, H. and J. Harer (2010). Computational topology: an introduction. American Mathe-


matical Soc.

Hillier, J. and G. Abrahams (2013). Deleuze and guattari: Jean hillier in conversation with gareth
abrahams.

Johnson, J. (2013). Hypernetworks in the science of complex systems, Volume 3. World Scientific.

Katz, D. (1979). Gestalt psychology, its nature and significance. Greenwood Pub Group.

Marcus, G. E. and E. Saka (2006). Assemblage. Theory, culture & society 23 (2-3), 101–106.

Protevi, J. and M. Bonta (2004). Deleuze and geophilosophy: A guide and glossary. Edinburgh:
Edinburgh University Press.

16

You might also like