Upload
uniud
View
1
Download
0
Embed Size (px)
Citation preview
Proceeding of the International Scientifical Conference.
Volume III.
189
UTOPIAS OF PERFECTION AND THEIR DYSTOPIAS
Livio C. Piccinini
Maria Antonietta Lepellere
Ting Fa Margherita Chang
University of Udine
Abstract. The search for optimality is a must in the study of biological evolution and many
models have been proposed in order to grasp its main operating gears. These models can also
give a deep comprehension of many social and economical phenomena and are studied in the
so called econophysics. Bak Sneppen is one of the most significant models because it balances
at best explication power and simplicity. Unlike cellular automata models, where phenomena
happen in a local framework without global supervision, Bak Sneppen models join locality
and globality, as many contemporary economical models require. The authors try to re-read
these models between utopic and dystopic frames, showing with simplified models some
behaviors that had not been considered in the previous literature. A search for uniformity
rather than optimality is also studied, since it seems to rule fashion evolution, as well as
political behavior of people (Band wagon).
Keywords: Optimality Utopias, Totality Utopias, Bak Sneppen Model, Modal Evolution,
Partitioned Frames and Evolution,
1. Introduction
The Bak-Sneppen model was originally introduced as a simple model of
evolution by Per Bak and Kim Sneppen [3] (compare also [2], [19]). Their
original model can be defined as follows. There is a circle of N nodes and they
are occupied by N different species, each of which has been assigned a random
fitness. The fitness values are independent and uniformly distributed on (0; 1).
At each discrete time step the system is updated by locating the lowest fitness
and replacing this fitness, and those of its two neighbors, by independent and
uniform (0; 1) random variables. Bak-Sneppen models can be defined on a wide
range of graphs using the same update rule as above. What the Bak Sneppen
model illustrates is that even random processes can result in self-organization to
a critical state (see [14] for a discussion). Threshold fitness rises, rapidly at first,
then exponentially slows until it reaches about 0.66, the critical state from which
level extinction sweep back and forth through the ecosystem. None of the
changes observed in the system are designed, however, to increase the critical
threshold or lead to extinction, but the dynamics of the model lead inevitably to
self organized criticality.
Brunk in [6] suggests that self-organized criticality is the sort of process that
should have great intuitive appeal to social scientists. Like earthquakes, the
precise prediction of riots; wars; and the collapse of economics, governments,
and societies is not possible in human systems in which catastrophic events can
be triggered by minuscule causes. A large bibliography of applications of self
organized criticality can be found in Turcotte [22] (see also [1] for an economic
Proceeding of the International Scientifical Conference.
Volume III.
190
application). The authors anyhow think that a sound basis for applying Bak-
Sneppen model of contact with neighbors is given by Duesenberry’s
demonstration effect. Its first presentation can be found in Duesenberry [9],
while many application in consumer’s economy and sociology can be found for
example in Cavalli [7].
Although the Bak-Sneppen model is extremely simple, it has not yet been
completely solved in spite of numerous analytical and numerical investigations.
Motivated by the difficulty of analyzing rigorously even the one-dimensional
version of the Bak-Sneppen model, J. Barbay and C. Kenyon [4] proposed a still
simpler model with discrete fitness values1. In their model each species has
fitness 0 or 1, and each new fitness is drawn from the Bernoulli distribution with
parameter p. Since there are typically several least fit species, the process then
repeatedly chooses a species at random for mutation among the least fit species.
They prove bounds on the average numbers of ones in the stationary distribution
and present experimental results. Parameter p can substitute up to a certain
level a plurality of values, but it cannot explain the staircase phenomenon found
by the authors in [18] section 5. Hence binary structure, though simple and
appealing is not sufficient for a thorough description of what may happen.
An important feature of discrete-value model is that it is possible to analyze the
mode, and that an ordered fitness function is no longer required. Discrete fitness
can be replaced by names of colors, a probability can be attached to each color
(maybe the same to all of them). As usual the modal color is the color that has
most items; we call antimodal the color that has at least one item, and whose
total number of items is the least. In case of ties one of the tied colors is chosen
at random. The process is as follows: at each discrete time step one of the
antimodal items and its two neighbors are replaced at random by a color chosen
according to the law of probability. We call it Modal Bak-Sneppen Process.
Some particular cases will be discussed in sections 2 and 3.
2. Globality and Uniformity Utopias
Utopias could work at their best when no competition arises. A global model of
perfection should foresee that all its members reach the top level. Similarly in a
modal process unanimity should be attained. At first it seems that utopia cannot
coexist with randomness. Strangely enough that is not true; simply, when
randomness is allowed, more time is required. A fundamental feature is
globality, so that the choice of the evolving node will be no longer sequential or
random, but will be connected to some global evaluation of all the nodes. This
happens for example in DNA analysis, where longest matching is chosen for
improvement.
1 Also this case is by no means trivial, as it was shown by Meester and Znamenski ([16]). The
reduction of number of species was recently considered by Schlemm in [20].
Proceeding of the International Scientifical Conference.
Volume III.
191
The simplest case is the random optimization: in a totally random process a
random node changes and assumes a random value, so that after a transitory
phase in which some memory of the original values still survives, it reaches a
purely random distribution of values, that coincides in probability with a casual
extraction with replacement. The process becomes much more interesting if we
add some rule for the choice of the evolution node. In ordered sets a good idea is
to choose the node (or one of the nodes) that attains the minimum value, and
replace its value by a random one. This process leads to optimality, since all
nodes that reach a value greater than the present minimum are left unchanged.
Those that already attain the maximum will undergo no change until all nodes
reach the maximum. At this utopian stage one of the optima, at random, will be
chosen for changing, and then it (and only it) will keep changing until it
resumes again the optimal value. It must be remarked that in order to reach
optimality it is not enough to state a fixed objective and to require that what lies
below changes. The final result would be only a distribution lying between the
objective and the potential maximum, hence the objective should be made
dynamical.
An interesting process can arise also without ordered structures of values. In
this case we deal with a finite (and small) set of values (colors). It is enough to
choose an antimodal node and replace it at random. We call this process simple
modal uniformization. The structure will converge anyhow to a mode involving
all nodes but one; in particular if at the initial time more than half of the nodes
have the same color this will also be the final modal color. For more detail the
authors have proved the following property:
“Suppose there are K colors, ordered in a sequence increasing according to their
cardinalities ci. Consider the inequalities
E1 c1+ c2 + c3 +...+ cK-1 cK
E2 c1+ c2 + c3 +...+ cK-2 cK-1
…
EK-3 c1+ c2 + c3 c4
EK-2 c1+ c2 c3
EK-1 c1 = c2
Suppose that En is the first inequality that does not hold. Then only the colors
labelled by K, K-1,…, K-n+1 may become final modal color. If all the
inequalities hold, then each color may become final”.
If the first inequality is false, color K is the only final color (as stated before)
since even if all non modal colors are changed into one and the same color (what
has a non zero probability) this color would still remain antimodal. Repeated use
of this argument achieves the proof. The computation of the probability of each
Proceeding of the International Scientifical Conference.
Volume III.
192
outcome uses Bellman’s technique of dynamic programming,2 but is somewhat
technical and cannot be reported here.
A somewhat opposite process arises when one of the modal nodes is chosen for
replacement; in this case the evolution tends to a division into groups that have
all the same cardinality (plus/minus 1). Both schemes arise in political and
social analysis: Band Wagon in one case, Biocultural Diversity in the other case.
A balanced division of items is also useful in the creation of optimal codes in
information theory, according to Shannon entropy principle3. This is a technical
utopia that scholars devoted to data mining must efficiently pursue, as Mc
Cormick recalls in [13] when he explains the reasons that made Google so
successful.
3. Antagonistic Forces
Random collateral actions disturb the utopian search of optimality or totality.
What is relevant is that they do not succeed in a complete destruction of the
optimizing process, but rather lead to a self organized system of critical states.
Many different ways of collateral disturbance have been proposed; also cellular
automata or heat equation may enter in this scheme. Bak-Sneppen model was
proposed as a conceptual kernel of many description efforts; it supposes that the
two adjoining nodes (on the circle) are affected independently from their values,
and undergo a random change. Many other schemes have been proposed: the
graph setting can be changed, inasmuch one or more nodes may be involved, or
connected nodes may depend probabilistically according to an influence graph.
A limit, non local case, is given by a random choice of one or more nodes.
Incidentally we remark that this randomized case is much easier to be
analytically dealt with.
For any finite number of nodes and for a discrete number of fitness levels (or
colors, in the modal case) it is possible to describe explicitly the system using
Markov chains4. For each given set of probabilities the transition matrix can be
constructed by a computer program developed by the authors, but if the
dimension becomes too large too many states of the system must be looked for
and the search for average frequencies becomes very slow and numerically
unstable.
The examples are given for the simple case consisting of 5 nodes, a single
collateral disturbance and two possible fitness values, namely 0 and 1, each of
them with probability ½. A reduced state representation for the random neighbor
case is simply the number of 1’s present in the system. The transition matrix in
case of optimization is given in table 1.
2 We refer to [5]
3 A classical work is ([21]), but also [14] can be referred to.
4 Compare for example [23]
Proceeding of the International Scientifical Conference.
Volume III.
193
Table 1.
Transition matrix for random optimization with one collateral disturbance
0 1 2 3 4 5
0 1/4 1/16 0 0 0 0
1 2/4 5/16 1/8 0 0 0
2 1/4 7/16 3/8 3/16 0 0
3 0 3/16 3/8 7/16 1/4 1/4
4 0 0 1/8 5/16 2/4 2/4
5 0 0 0 1/16 1/4 1/4
A suitable rearrangement of the columns supplies the corresponding matrixes for
the uniformity processes. The average frequencies of the 6 states are
summarized in table 2, where F indicates the column where fitness is present,
BW and BD indicate the columns of modal processes, BW for the Band Wagon
and BD for the Biocultural Diversity. Column R represents the frequencies of
the states in a purely random process (binomial coefficients).
Table 2.
Averages frequencies of states of random optimization (one collateral disturbance)
State F BW BD R
0 0.002 0.0875 0.011 0.031
1 0.022 0.2125 0.137 0.156
2 0.113 0.2 0.352 0.313
3 0.323 0.2 0.352 0.313
4 0.385 0.2125 0.137 0.156
5 0.155 0.0875 0.011 0.031
Of course the last three columns are symmetrical, since they are modal, while
the first column is concentrated on the high value states since it corresponds to a
phenomenon oriented towards optimization. In the first column the average
frequency of 1-fitness can be computed, and is equal to 0.707 (to be compared
with the pure random frequency of 0.5). As for modal processes, Band Wagon
concentrates the frequencies upon the presence of a majority (either 1 or 4),
while on the opposite Biocultural Diversity concentrates in the middle of the
scale (a comparison may be made with column R)
In linear phenomena random disturbance is the most effective agent in
preventing optimization. Neighborhood contagion, as is well known in
epidemiology, slackens the diffusion rate, hence sharper results are to be
expected when Bak Sneppen processes are involved. The same case as before is
now considered, choosing as neighbor one node (say the left node). Even the
simplest representations of states must take into account the geometry of the
structure, but they can be invariant for rotation. Hence in the case of 5 nodes
and 2 fitness levels, the states with 2 and 3 nodes at level 1 must be split into
the case where these nodes are adjoining (states 2A and 3A) and the case where
they are split (states 2S and 3S). The transition matrix for the Markov chain is
given in table 3.
Proceeding of the International Scientifical Conference.
Volume III.
194
Table 3.
Transition matrix for unilateral Bak-Sneppen process with 5 nodes
0 1 2A 2S 3A 3S 4 5
0 2/8 1/16 0 0 0 0 0 0
1 4/8 5/16 1/12 2/12 0 0 0 0
2A 1/8 3/16 3/12 1/12 1/8 1/8 0 0
2S 1/8 4/16 1/12 4/12 0 1/8 0 0
3A 0 2/16 3/12 1/12 2/8 1/8 1/4 1/4
3S 0 1/16 2/12 3/12 1/8 3/8 0 0
4 0 0 2/12 1/12 3/8 2/8 2/4 2/4
5 0 0 0 0 1/8 0 1/4 1/4
The table of results (table 4) is built in the same way as the previous one, but in
order to simplify comparison the frequencies corresponding to the states 2A - 2S
and 3A – 3S have been summed together
Table 4.
Results of Bak Sneppen unilateral process and related problems. Case of 5 nodes.
State F BW BD R
0 0.001 0.094 0.011 0.031
1 0.014 0.219 0.136 0.156
2 = 2A+2S 0.085 0.187 0.353 0.313
3 = 3A+3S 0.308 0.187 0.353 0.313
4 0.414 0.219 0.136 0.156
5 0.177 0.094 0.011 0.031
The frequency of 1-fitness now has increased to 0.721. Also the majority effect
in Band Wagon modal optimization has been increased, while in the case of
Biocultural Diversity the neighborhood effect is less noticeable.
4. Partitions Defend Utopian Optimization
In this section the case of actual discrete two-sided Bak-Sneppen process will be
investigated. For a richer choice of cases a non prime number of nodes is
required; the choice is 6, so that it is easy to enumerate all the possible structures
(apart from rotations they are 14) and to calculate the corresponding transition
matrix. The generalization consists in changing the two connections of Bak-
Sneppen model in all possible (different) combinations, so that, in view of
symmetries, the number of different cases sums up to 6. In table 5 the two links
(according to positive rotation) are listed. Bak-Sneppen classic case lies in the
first row. In the third column, as a synthesis of the process, the average
percentage of 1’s is given.
Proceeding of the International Scientifical Conference.
Volume III.
195
Table 5.
Binary Bak-Sneppen on a frame of 6 nodes
1st
link 2nd
link Average
1 5 (=-1) 0.637389
1 2 0.620732
1 3 0.620732
1 4 0.607729
2 3 0.620732
2 4 0.75
random random 0.61229
In view of the increased collateral disturbance both the average of Bak-Sneppen
process (1st row) and of the random process (last row) are much lower than what
arose from tables 4 and 2, that amounted respectively to 0.721 and 0.707.
Usually the average of structured processes is higher than in random processes,
because long chains of maxima are structurally protected, but an exception is
shown for the case 1- 4 in which the two operating nodes are just opposite,
hence tend to break long chains of maximums wherever they can be located.
The table highlights an interesting phenomenon, that may arise only in the case
of a non prime number of nodes (hence the reason for choosing 6 nodes). The
average of the last case (2-4) is much greater than all the remaining averages. In
the transition matrix of the associated Markov chain there exists a persistent
group5 that is smaller than the whole set of 14 structures. Namely it contains
only the 4 structures 01 01 01, 01 01 11, 01 11 11, 11 11 11, while the remaining
10 form a transient group: this means that there exist sets of transformations that
allow to pass from any of them to any other, but there is the possibility of falling
outside into the persistent group without possibility of returning back. Nodes 2
and 4, together with the minimum conventionally placed at 0, form a subset that
has no interference with nodes 1-3-5, that on the contrary are activated when
minimum is attained in one of those nodes. Whenever an operation is performed,
one 3-element subset is left unchanged, while the other 3-element subset is
totally changed at random. The two subsets do not change between different
operations, what on the contrary happens in all the remaining cases. The process
cannot anyhow be divided into two independent subprocesses, since the
minimum must be looked for among both the subsets (for example a subset
containing 011 enters in this search, while a subset 111 enters in the search only
if also the other one is a 111 subset, in which case there are 6 minima of value
1). A further interaction is given by the number of minima; for example in the
case 000 011 the first subset has probability ¾ of being chosen, while the other
one only ¼.
When one set reaches the configuration 111 (probability 1/8) it becomes stable,
in the sense that it can be changed only if the global minimum is 1, that is the
configuration is 111 111. In this case anyhow at least one of the two subsets will
5 We follow Hsu’s nomenclature [12], compare also Chang-Piccinini [8]
Proceeding of the International Scientifical Conference.
Volume III.
196
save the configuration 111, that can no longer be destroyed. The four structures
above in fact can be read, keeping the two subsets divided, as 000 111, 001
111, 011 111, 111 111. All computations become very simple since there is no
longer a Bak Sneppen structure. In particular the average delay time for reaching
the persistent group starting from any configuration of the transient group is 8,
and does not depend on the initial structure. The average time for reaching the
top configuration 111 111 starting from 000 000 (or any other transient) is 16,
while in the standard Bak Sneppen process is 23.07572. The increase in time is
due to the impossibility of protecting the structure 111 from decay.
5. Staircases and Sudden Balance Break
A richer example of Bak-Sneppen process will be shown. The simplest case that
shows the main features requires 3 subsets of 3 nodes each, and a ternary set of
values should be considered, say 0,1,2. This corresponds to 9 nodes and
displacements of 3-6. A simplified analysis is the following: label 0 is attached
to any subset in which the minimum is 0 (not regarding their number), label 1 to
any subset in which the minimum is 1 and label 2 to the set 222. Track of the
single subsets will not be kept, but they will be enumerated, getting the states
000, 100, 110, 111, 200, 210, 211, 220, 221, 222. Table 6 shows the transition
matrix.
Table 6.
Transitions in a 9 node tripartite Bak Sneppen set with 3 values.
A \DA 000 100 110 111 200 210 211 220 221 222
000 17/27
100 9/27 17/27
110 9/27 17/27 17/27
111 9/27 9/27
200 1/27 17/27
210 1/27 9/27 17/27 17/27
211 1/27 1/27 9/27 9/27
220 1/27 17/27 17/27 17/27
221 1/27 1/27 9/27 9/27 9/27
222 1/27 1/27 1/27
There are two transient groups (110, 111) and (210, 211) and one persistent
group (220, 221, 222). The final average is thus 1.667, since two subsets have
the form 222, and the third one is random on the three values 0, 1, 2. The
average times are 54 from 000, 100, 110,111 to the persistent group, 27 from the
groups 200, 210, 211 to the persistent group. The scheme of transitions is
represented in figure 1.
Proceeding of the International Scientifical Conference.
Volume III.
197
Fig. 1. Transition graph in a 9 node tripartite Bak Sneppen set with 3 values.
The transitions from 100 to 210 and from 110 to 211 show an interesting fact:
the subset that is changed into the optimal label 2 is not one labelled with 1, but
one that is labelled with 0, that is, its minimum must be lower than that of the
best subset. In Bak-Sneppen models, as sometimes also in real life, the principle
is “quieta non movere” (Let quiet things stay). For example in a situation
A=122, B=112, C=011, it is impossible that the top ones (A and B) reach the
state 222, while it is possible, even if unlikely, that this state is reached by C.
The idea is that movement requires dissatisfaction, while further movements are
caused by some form of nested neighborhood with unsatisfied people that can
in turn generate new dissatisfaction: We can remind the “happiness paradox” of
Easterlin ([10], [11]).
Examples more connected with economics and theory of decisions are given by
the evolution of road nets and urban systems. Here the influence of a single
intervention is somehow localized (hence “partitioned nets”), but usually the
decision is taken at a global level, choosing the worst section of a road system or
the most impoverished part of a city. An intervention on the set of roads perturbs
the neighbors in an unpredictable way, allowing a change in the overall structure
of the connexion system. The new road usually satisfies up-to-date standards
that are not satisfied by the unchanged sections. In cases of good planning also
the neighbor roads are improved, so that the overtaking phenomenon can arise.6
In urban systems an overtaking case is encountered when a large and central or
semicentral area is sold out at a very low price because of its decay, and is used
for a top level reconstruction. A classical case can be found in London’s East
End, especially around the Tower Bridge, but also Liverpool and Valencia
present interesting evolutions.
In numerical simulations as soon as the dimension of the subsets is increased it
becomes more and more difficult to reach the persistent group; for example
6 Compare also the paper by two of the authors [17]
000 100 110 111
210 211200
220 221
222
Transient
group
Transient
group
Persistent
group
Proceeding of the International Scientifical Conference.
Volume III.
198
already subsets of 18 nodes (for a total of 36) very often require more than
100,000 iterations in order to reach the persistent group. The transition, when it
happens, is very similar to an “avalanche”, a typical structure of Bak Sneppen
processes and in general of punctuated equilibrium. In some cases the process is
not complete, hence it can be reverted, but finally it happens that the threshold is
reached. Increasing the number of values multiple thresholds arise; the lower
levels can be easily overcome, while the top levels can prove to be almost
unreachable. This is for example the case of ten values and forty nodes, in which
the top level has never been reached in ten simulations of 1,000,000 iterations.
Conclusions
While aiming at utopia one of the main indicators is the expected time to reach
it, should success be possible. Similarly the expected time of destruction by
dystopias is meaningful. Using the transition matrix one can compute the
average waiting time. In section 5 it has been shown that in partitioned frames a
final state can be reached which is not reversible, that is isolation pays, as all
authoritarian governments know, even if evolution can be hindered. On the
contrary when isolation is not achieved and random disturbances arise, all states
are again reachable, but the most frequent is an equilibrium distribution higher
than the average. On the contrary in modal uniformity search sudden changes of
mode can occur, as it happens with political majorities. Taking again into
account the 5-nodes model with one collateral disturbance it is possible to
compare the effects of protection against changes (table 7). The benchmark is
the purely random replacement of two nodes (R); the comparisons are with
antimodal choice plus random disturbance (RD) and antimodal choice plus next
neighbor disturbance (unilateral Bak-Sneppen, ND). The first columns list the
starting state, while table 7 shows the expected time to reach a new consistent
majority (namely 4 out of 5).
Table 7.
Expected time for the majority inversion in 5-nodes model
R RD ND
0 12.4 26.7 35.6
1 11.6 26.7 35.6
2A 10.1 22.7 28.3
2S 10.1 22.7 31.6
The greater protection given by Bak Sneppen model confirms that in linear
phenomena where no critical threshold arises a localized dissatisfaction is less
effective. As soon as the number of nodes increases the expected waiting times
grow faster and faster. For example for 13 nodes to pass from 3 to 10 in case RD
it amounts already to 944.0 (authors’ computation).
Proceeding of the International Scientifical Conference.
Volume III.
199
Summary
The paper considers some simple models of evolution tending to utopian development
towards perfection or uniformity, and shows how dystopic antagonistic forces prevent or
hinder the achievement of the goal. It is also shown that in linear phenomena, where no
threshold is available, locality of evolution hinders the destruction of achieved objectives, but
can also prevent further improvements (turris eburnea, ivory tower effect). The interesting
phenomenon of overtaking is considered in the last section, where a theoretical three level
model and real world examples are shown that satisfy the Bible sentence “the last will be the
first”. The examples show that this fact not necessarily happens, but show also that its
probability is not negligible. In figure 1, containing the theoretical scheme, the Bible is
satisfied along the diagonals, while vertical lines correspond to everyday life.
Bibliography
[1] Ausloos, Marcel; & Clippe, Paulette; & Pekalski, Andrzej. (2004). Evolution of economic
entities under heterogeneous political/environmental conditions within a Bak-Sneppen-
like dynamics Physica A 332, 394-402.
[2] Bak, Per. (1996). How Nature Works: The Science of Self-Organized Criticality. Springer.
[3] Bak, Per; & Sneppen, Kim. (1993). Punctuated equilibrium and criticality in a simple
model of evolution, Physical Review Letters. 71(24), 4083-4086.
[4] Barbay, Jérémy; & Kenyon, Claire. (2001). On the discrete Bak-Sneppen model of self-
organized criticality. Proc. 12th Annual ACM-SIAM Symposium on discrete Algorithms,
Washington DC.
[4] Bellman, Richard E. (1957). Dynamic Programming. Princeton: Princeton University
Press.
[6] Brunk, Gregory G. (2000). Understanding Self-Organized Criticality as Statistical
Proces. John Wiley & Sons, Inc., 5 (3), 26-33.
[7] Cavalli, Alessandro. (1998). Sociologia Enciclopedia delle Scienze Sociali, VIII, Roma:
Enciclopedia Italiana, 213-239.
[8] Chang, Ting Fa Margherita; & Piccinini, Livio Clemente. (2000). A new Method for
Analyzing Urban-Rural Relations through Commutation Data. H. Hillebrand, R.
Goetgeluk, H. Hetsen (eds.) Plurality and Rurality (The role of the countryside in
urbanized Regions). Vol.1, LEI The Hague, 102-122.
[9] Duesenberry, James S. (1949). Income, Savings and Theory of Consumer Behaviour.
Cambridge Mass, 89-125.
[10] Easterlin, Richard A. (1974). Does economic Growth improve the human Lot? In P. A.
David and M. W. Reder eds. Nations and Households in Economic Growth. Acad. Press
N.Y.
[11] Easterlin, Richard A. (2001). Income and Happiness: Towards a Unified Theory. The
Econ. J. 465-484.
[12] Hsu, Chieh S. (1987). Cell to cell mapping. A method of global analysis for non linear
systems. Springer Vg, NY.
[13] MacCormick, John. (2012). Nine Algorithms that Changed the Future. Princeton:
Princeton University Press. 31-44.
[14] McEliece, Robert J. (1977). The theory of information and coding. Addison Wesley.
[15] Meester, Ronald; & C. Quant. (2005). Connections between `classical' and `self-
organized' criticality. Markov processes and related fields, 11 (2), 355-370.
[16] Meester, Ronald; & Znamenski, Dmitri. (2002). Non-triviality of a discrete Bak-
Sneppen evolution model. J. Stat. Phys. 109, 987-1004.
Proceeding of the International Scientifical Conference.
Volume III.
200
[17] Piccinini, Livio C.; & Chang, T. F. Margherita; & Taverna, Mario. (2012) Dalle reti di
infrastrutture alle reti virtuali Sebastiano Cacciaguerra (editor) Il L.I.M.E.S. Padano-
Veneto a Nord-Est , Roma: Gangemi Editore, 79-96.
[18] Piccinini, Livio C.; & Lepellere, M. Antonietta; & Chang, T. F. Margherita (2011).
Partitioned Frames in Bak Sneppen Models. mpra.ub.uni-muenchen.de/43852/
[19] Rotundo, Giulia; & Scozzari, Andrea. (2008). Coevolutive Models for Firms Dynamic.
Lecture notes in Economics and Mathematical systems, 613, 143-158.
[20] Schlemm, Eckhard. (2012). Asymptotic Fitness Distribution in the Bak-Sneppen Model
of Biological Evolution with four Species. J. Sta. Phys. 148, 191-203.
[21] Shannon, Claude E.; & Weaver, Warren. (1949). The Mathematical Theory of
Communication. University of Illinois Press.
[22] Turcotte, Donald L. (1999). Self-organized criticality. Rep. Prog. Phys. 62, 1377–
1429.
[23] Woess, Wolfgang. (2009). Denumerable Markov Chains. European Mathematical
Society, Freiburg.
Livio C.Piccinini Department of Civil
Engineering and Architecture,
University of Udine
E-mail: [email protected]
Maria Antonietta
Lepellere
Department of Civil
Engineering and Architecture,
University of Udine
E-mail: [email protected]
Ting Fa Margherita
Chang
Department of Civil
Engineering and Architecture,
University of Udine
E-mail: [email protected]