Colloquium: Criticality and dynamical scaling in living systems

Miguel A. Muñoz
Instituto Carlos I de Fı́sica Teórica y Computacional and Departamento de Electromagnetismo y Fı́sica de la Materia,
Facultad de Ciencias Universidad de Granada 18071 Granada Spain

(Dated: May 16, 2018)

A celebrated and controversial hypothesis suggests that some biological systems –parts,
aspects, or groups of them– may extract important functional benefits from operating
arXiv:1712.04499v2 [[Link]-mech] 15 May 2018

at the edge of instability, halfway between order and disorder, i.e. in the vicinity of the
critical point of a phase transition. Criticality has been argued to provide biological sys-
tems with an optimal balance between robustness against perturbations and flexibility
to adapt to changing conditions, as well as to confer on them optimal computational
capabilities, huge dynamical repertoires, unparalleled sensitivity to stimuli, etc. Crit-
icality, with its concomitant scale invariance, can be conjectured to emerge in living
systems as the result of adaptive and evolutionary processes that, for reasons to be
fully elucidated, select for it as a template upon which further layers of complexity can
rest. This hypothesis is very suggestive as it proposes that criticality could constitute a
general and common organizing strategy in biology stemming from the physics of phase
transitions. However, despite its thrilling implications, this is still in its embryonic state
as a well-founded theory and, as such, it has elicited some healthy skepticism. From the
experimental side, the advent of high-throughput technologies has created new prospects
in the exploration of biological systems, and empirical evidence in favor of criticality has
proliferated, with examples ranging from endogenous brain activity and gene-expression
patterns, to flocks of birds and insect-colony foraging, to name but a few. Some pieces
of evidence are quite remarkable, while in some other cases empirical data are limited,
incomplete, or not fully convincing. More stringent experimental set-ups and theoretical
analyses are certainly needed to fully clarify the picture. In any case, the time seems
ripe for bridging the gap between this theoretical conjecture and its empirical validation.
Given the profound implications of shedding light on this issue, we believe that it is both
pertinent and timely to review the state of the art and to discuss future strategies and
perspectives.

CONTENTS B. Gene regulatory networks 17

1. Models of genetic regulatory networks 18
I. Introduction: Statistical physics of biological systems 2 2. Gene knock-out and damage spreading 18
3. Networks from DNA microarray data 19
II. Criticality and scale invariance 3
1. Scale-invariance and power laws 4 4. Zipf’s law in gene-expression data 19
2. Criticality in equilibrium systems and beyond 5 C. Collective behavior of cells 19
3. Non-equilibrium phase transitions: an example 5 1. Stem cell pluripotency 19
4. Self-organization to criticality 7 2. Morphogenesis I: Hydra regeneration 20
5. Classes of criticality 8 3. Morphogenesis II: Gap genes in Drosophila 20
6. Criticality on complex networks 8
D. Collective motion 20
7. Generic scale invariance 9
8. Statistical criticality 9 1. Flocks of birds 21
2. Insect swarms 21
III. Functional advantages of criticality 9 3. Mammal herds 21
A. Criticality in the auditory and other sensory systems 9 4. Social-insect foraging strategies 21
B. Exploiting criticality 10
1. Maximal sensitivity and dynamic range 10
V. Discussion 22
2. Large correlations 10
3. Statistical complexity and large repertoires 10
4. Computation exploiting criticality 11 Acknowledgments 23

IV. Alleged criticality and scaling in biological systems 11 Appendix A: Generic Scale invariance 23
A. Neural activity and brain networks 11
1. Spontaneous cortical activity 11 Appendix B: Probabilistic models and statistical
2. The edge of activity propagation: avalanches 12 criticality 24
3. The edge of neural synchronization 13
4. The edge of global stability 14
Appendix C: Adaptation and evolution towards criticality 24
5. The edge of percolation 14
6. The edge of a thermodynamic transition 14
7. Large-scale cortical dynamics 15 Appendix D: Other putatively critical living systems 25
8. Disruptions of criticality in pathological conditions 15
9. Mathematical models of neuro-criticality 15 References 25
 2

I. INTRODUCTION: STATISTICAL PHYSICS OF son et al., 1972; Bialek, 2012; Haken et al., 1985; Hop-
BIOLOGICAL SYSTEMS field, 1982; Kelso, 1984; Parisi, 1993; Schrödinger, 1967;
Sneppen, 2014).
One of the greatest challenges of Science is to shed light One of the most striking consequences of interac-
on the essence of the phenomenon that we call “life”, tions among elementary constituents of matter (atoms,
with all its astonishing diversity and complexity. Cells molecules, electrons...) is the emergence of diverse phases
–the basic building-blocks of life– are intricate dynam- whose behavior bears little resemblance with that of their
ical systems consisting of thousand types of interact- basic components or small groups of them (Anderson
ing molecules, being created, used and destroyed every et al., 1972; Chaikin and Lubensky, 2000; Stanley, 1987).
minute; multicellular organisms rely on the perfectly or- Systems consisting of very many (microscopic) compo-
chestrated motion of up to trillions of interacting cells, nents may exhibit rather diverse types of (macroscopic)
and communities assemble dozens of individuals, inter- collective behavior, i.e. phases, with different levels of
acting in countless ways, forming entangled ecosystems, internal order. Moreover, slight changes in external con-
and giving rise to a mind-blowing hierarchy of “complex- ditions (e.g. temperature, pression...) or in the strength
ity”. of interactions may induce dramatic structural rearrange-
The standard viewpoint in biology, stemming from ments, i.e. phase transitions.
the reductionist tradition, is that each molecular com- It is thus tempting to hypothesize that biological states
ponent (protein, nucleic acid, metabolite...) is specific might be manifestations of similar collective phases and
and requires individualized scrutiny. This one-at-the- that shifts between them could correspond to phase tran-
time approach has successfully identified and quantified sitions (Anderson et al., 1972; Hopfield, 1994). As a
most of the components and many of the basic inter- matter of fact, phase transitions are a common theme
actions of life as we know it, as stressed by the rapid in biology (Pollack and Chin, 2008; Solé, 2011), as illus-
advance of the “omics” sciences (genomics, proteomics, trated by the following non-exhaustive list of examples:
metabolomics...). Still, unfortunately, it offers no con- (i) synchronization phase transitions in collective biolog-
vincing explanation of how systemic properties emerge ical oscillators such as circadian clocks (Garcia-Ojalvo
(Sauer et al., 2007). Questions such as “how are those et al., 2004); (ii) percolation transitions of fibers in con-
myriads of elements and interactions coordinated to- nective tissues such as collagen (Alvarado et al., 2013;
gether in complex living creatures?” or “how does coher- Forgacs et al., 1991; Newman et al., 2004), (iii) melting
ent behavior emerge out of such a soup of highly heteroge- phase transition in DNA strands (Li and Retzloff, 2006;
neous components?” (Schrödinger, 1967) remain largely Poland and Scheraga, 1970); and (iv) transitions between
unanswered. different dynamical regimes (oscillations, bursting,...) in
A complementary strategy consists in looking at com- neuronal networks (Freeman, 2013; Freeman and Holmes,
plex biological problems from a global perspective, shift- 2005; Haken, 2013; Kelso, 1984; Rabinovich et al., 2006;
ing the focus from specific details of the molecular ma- Werner, 2007), etc.
chinery to integral aspects (Alon, 2006; Bialek, 2012; Life –guided by evolution– has found its way to ex-
Goldenfeld and Woese, 2011; Hartwell et al., 1999; ploit very diverse types of order: crystalline struc-
Kaneko, 2006; Sauer et al., 2007). System approaches tures (seashells, skeletons...), liquid states (blood, lymph,
to biology rely on the evidence that some of the most sap...), gels (vitreous humor, cell cytoplasm), etc. How-
fascinating phenomena of living systems –such as mem- ever, some aspects of biological systems –think e.g. of
ory and the ability to solve problems– are collective ones, neural networks or flocks of birds– exhibit intermediate
stemming from the interactions of many basic units, and levels of organization, half way between order and dis-
might not be reducible to the understanding of elemen- order, less regular than perfect crystals but more struc-
tary components on an individual basis (Bialek, 2018). tured than random gases. Remarkably, it has been con-
Theoreticians have long struggled to elucidate whether jectured that, under some circumstances, living systems
simple and general principles –such as those in physics– –i.e. parts, aspects, or groups of them– could draw im-
could be of any help in tackling biological complexity. portant functional advantages from operating right at the
More specifically, they have long been seduced by the borderline between ordered and disordered phases, i.e.
idea of adapting concepts and methods from statistical at the very edge of a (continuous) phase transition or
mechanics to shed light onto the large-scale organization critical point2 (Bak, 1996; Beggs, 2008; Chialvo, 2010;
of biological systems1 (Alon, 2006; Amit, 1992; Ander- Chialvo et al., 2008; Kauffman, 1993; Plenz, 2013; Plenz

1 The possibility that biological problems may stretch the frontiers 2 Phase transitions may occur in either a discontinuous/abrupt
of physics by uncovering phenomena and mechanisms unknown fashion (Binney et al., 1993) –with associate bistability of the
in purely physical systems is also inspiring (Frauenfelder, 2014; two different phases and an abrupt/discontinuous jump at the
Goldenfeld and Woese, 2011). transition point– or in continuous/progressive way with an asso-
 3

and Niebur, 2014). For instance, rather generically, liv- resolution (Sejnowski et al., 2014)– and, similarly, in
ing systems need to achieve a tradeoff between robustness genomics (Lesk, 2017) or in collective motion analyses
(resilience of the system state to external perturbations; (Cavagna et al., 2008). As a result, recent years have
which is a property of ordered phase), and flexibility (re- witnessed an upsurge of empirical works reporting on
sponsiveness to environmental stimuli, which is a feature putative scale-invariance and/or criticality in diverse bi-
of disordered phases). An optimal balance between these ological systems, supporting the above theoretical specu-
two conflicting tendencies can be accomplished by keep- lations. In some cases the evidence appears to be robust,
ing the system dynamical state at the borderline of an while in others it is marginal, incomplete, or, to say the
order-disorder phase transition, i.e. at criticality. Signa- least, doubtful. In any case, time seems to be ripe for
tures of criticality, such as the spontaneous emergence of bridging the gap between the theoretical hypothesis and
long-range spatio-temporal correlations and the exquisite its empirical validation.
sensitivity to stimuli are also susceptible to be exploited The purpose of the present Colloquium is to briefly re-
for functional purposes, e.g. to create coordinated global view the main ideas and motivation behind the criticality
behavior, as we shall discuss in what follows. The idea hypothesis as a possible guiding principle in the collec-
that –in some special circumstances– evolution might tive organization of living systems and to scrutinize and
have favored states close to the edge of a phase transi- discuss in a critical way the existing empirical evidence
tion is certainly tantalizing, as it suggests that operating and prospects. It also aims at providing the reader with
near criticality could be an overarching strategy in bi- a self-consistent view of what is criticality and what it
ological organization (Bak, 1996; Beggs, 2008; Chialvo, is not, as well as an overview of the literature on this
2010; Kauffman, 1993; Mora and Bialek, 2011; Plenz, active and fascinating research field with countless ram-
2013; Plenz and Niebur, 2014). ifications.
Critical points have long been appreciated to exhibit
Let us remark that there exist excellent articles re-
striking features. Still, given the need of careful fine
viewing some of these topics to different extents; the list
tuning for them to be observed, they were long treated
includes the very influential paper by Mora and Bialek
as rarities. The development of some of the most re-
(2011) which popularized the subject, and other focused
markable intelectual achievements of the second half of
on neural dynamics (Beggs, 2008; Chialvo, 2010; Chialvo
the 20th century, such as the scaling hypothesis and the
et al., 2008; Cocchi et al., 2017; Hesse and Gross, 2014;
renormalization group theory (Fisher, 1974; Wilson and
Massobrio et al., 2015; Plenz and Niebur, 2014; Shew and
Kogut, 1974), changed this view and led to an elegant
Plenz, 2013), gene networks (Roli et al., 2015), and col-
and precise theory of criticality, with unsuspected im-
lective motion (Vicsek and Zafeiris, 2012), respectively.
plications in many fields, from particle physics to poly-
The present paper aims at overviewing and complement-
mer science3 . A chief conclusion is that many features at
ing them, putting the emphasis on dynamical aspects,
critical points are quite robust and largely independent of
and discussing together empirical evidence and theoreti-
small-scale details, giving rise to universality in the large-
cal approaches.
scale behavior. This has very important consequences
for e.g. studies in biology, as criticality and its concomi-
tant scale-invariance can be understood through simple
stylized models –neglecting many irrelevant details of in-
II. CRITICALITY AND SCALE INVARIANCE
dividual components and putting the emphasis on how
they interact– paving the road to the understanding of
collective aspects of living systems in relatively simple Many discussions about “criticality” are semantic ones.
terms. Depending on authors and fields rather diverse contents
From the experimental side, the advent of high- are assigned to terms such as “critical”, “quasi-critical”,
throughput techniques and big-data analyses have cre- “dynamically critical”, “generically critical”, or “self-
ated new prospects in the exploration of biological sys- organized critical”. Given the broad audience this paper
tems. This is true, for example, in neuroscience –where is aimed at, we esteem that a section devoted to present
it is now possible to record activity from individual spik- a synthetic overview of basic concepts and to fix ideas
ing neurons to entire brains with previously-unthinkable and notation is necessary.4 Readers familiar with these
concepts can skip it.

ciated critical point. Our main focus here is on continuous ones,

but we will also encounter discontinuous transitions, which may
also play a relevant role in biology. 4 For a more exhaustive introduction to critical phenomena we
3 See, e.g. Binney et al. (1993); De Gennes (1979); Delamotte refer to the standard literature; e.g. Binney et al. (1993); Chris-
(2012); Henkel et al. (2008); Sethna (2006); Stanley (1987); tensen and Moloney (2005); Henkel et al. (2008); Marro and
Täuber (2017). Dickman (1999); Sethna (2006); and Stanley (1987).
 4

1. Scale-invariance and power laws

In a seminal paper entitled “Problems in Physics with

many scales of length” K. Wilson emphasized that “one
of the more conspicuous properties of nature is the great
diversity of size or length scales”, and cited oceans as
an example where phenomena at vastly disparate wave-
lengths coexist (Wilson, 1979). Different scales are usu-
ally decoupled and the “physics” at each one can be sep-
arately studied. However, there are situations –known as
scale-invariant or scale-free– where broadly diverse scales
make contributions of equal importance. A remarkable
FIG. 1 Random walks, such as the one illustrated in the left
instance of this –but just an example– are the critical panel, lack a characteristic scale. As a consequence, the distri-
points of continuous phase transitions where the micro- bution of return times to the origin, T , of the one-dimensional
scopic, mesoscopic and macroscopic scales are all alike. (unbiased) random walk obeys P (T ) ∼ T −α with α = 3/2 and
Power-law (or Pareto) distributions such as P (x) = the areas/sizes, S, covered by their excursions before return-
Ax−α , where α is a positive real number and A a normal- ing to the origin (i.e. “avalanches”) obey P (S) ∼ S −τ with
ization constant, are the statistical trademark of scale- τ = 4/3 (right panels) (Redner, 2001; di Santo et al., 2017b).
invariance or “scaling”5 . Actually, they are the only Some biological systems exhibit scaling as a consequence of
an underlying random-walk process; see e.g. Berg (1993) and
probability distribution functions for which a change Gerstein and Mandelbrot (1964).
of scale from x to Λx, for some constant Λ, leaves
the functional form of P (x) unaltered, i.e. P (Λx) =
A(Λx)α = AΛα xα = Λα P (x), in such a way that the 2011; Marsili and Zhang, 1998; Mora and Bialek, 2011;
ratio P (Λx)/P (x) = Λ−α does not depend on the vari- Sornette, 2006; Visser, 2013), Bendford’s law (Benford,
able x, i.e. it is scale invariant (Newman, 2005; Sornette, 1938; Pietronero et al., 2001), and Taylor’s law (Cohen
2006). As opposed to e.g. exponential distributions, et al., 2012; Giometto et al., 2015; Taylor, 1961)– stem
power-laws lack a relevant characteristic scale, besides from underlying power-law distributions.
natural cut-offs. Disputes on the validity and possible significance
Distributions with power-law tails appear in countless of power-laws have a long history in diverse research
scenarios, including the statistics of earthquakes, solar fields. For some authors they reveal fundamental mech-
flares, epidemic outbreaks, etc. (Mandelbrot, 1983; New- anisms, while some others perceive them as largely un-
man, 2005; Sornette, 2006; West, 2017). They are also informative (Kello et al., 2010; Stumpf and Porter, 2012)
a common theme in biology (Gisiger, 2001; Goldberger, or even “more-normal-than-normal” distributions (Will-
1992; Goldberger et al., 2002; Hu et al., 2012; West, inger et al., 2004). Still, in some cases, there is very ro-
2010). For example, physiological and clinical time-series bust evidence of scale invariance and it certainly provides
data have typically a spectrum that decays as a power valuable insight8 .
of the frequency (Mandelbrot, 2002) and mobility pat- The detection and statistical characterization of
terns often exhibit scale-free features (Barabasi, 2005; power-law distributions in real-world data is often hin-
Brockmann et al., 2006; Proekt et al., 2012). Moreover, dered by sampling problems since very rare but large
a number of commonly-observed statistical patterns of events control the statistics. Accordingly, the quality of
natural-world data –such as Zipf’s law6 ,7 (Baek et al., power-law fits to empirical data has been recently scru-
tinized, showing that many claims of scale-invariance ac-
tually lack statistical significance and, presently, more
stringent statistical tests have become a must (Clauset
5 A well-known example is the Guttenberg-Richter equation for the et al., 2009).
probability distribution of observing an earthquake of dissipated
energy E, P (E) ∝ E −α , (Corral, 2004).
6 This states that the frequency with which a given pattern is
observed declines as a negative power law of its rank, i.e. its
position in the list of possible patterns ordered from the most see also Tkačik et al. (2015) for a discussion of these issues and
frequent to the rarest one (Zipf, 1949). how can they influence the conclusions about statistical critical-
7 A very elegant and illuminating approach allowed Mora and ity of empirical data.
8 An important example are allometric scaling laws, which
Bialek to map the Zipf’s law to underlying statistical critical-
ity in a very precise way (Mora and Bialek, 2011). Within this are power-law relationships between different measures of
setting, it was observed, however, that the Zipf’s law (and its anatomy/physiology (Banavar et al., 2014, 2010b; Kleiber, 1932;
concomitant statistical criticality) may emerge rather generically West et al., 1997). These have been elegantly shown to stem
if there is a fluctuating unobserved (hidden) variable that affects from the constraint that living systems have an underlying opti-
the system, such as e.g a common input, even in systems not mal (e.g nutrient) transportation network (Banavar et al., 1999;
tuned to criticality (Aitchison et al., 2016; Schwab et al., 2014); Simini et al., 2010).
 5

From the mathematical side, very diverse explanatory (Henkel et al., 2008; Hinrichsen, 2000; Hohenberg and
mechanisms for the emergence of scaling in empirical Halperin, 1977; Kamenev, 2011; Marro and Dickman,
data have been put forward (Marković and Gros, 2014; 1999; Täuber, 2014, 2017). All along this paper, we adopt
Mitzenmacher, 2002; Newman, 2005; Simkin and Roy- a view of criticality and phase transitions focused mostly
chowdhury, 2011; Sornette, 2009). For example, ran- on dynamical and non-equilibrium aspects. This seems
dom walks give rise to power laws in the distribution to be the most natural choice to analyze living systems,
of return times and “avalanche” sizes as illustrated in which are dynamical entities kept away from thermal
Fig.1. Other examples are: (i) Underlying multiplica- equilibrium by permanently exchanging energy and mat-
tive processes (Reed and Hughes, 2002; Richmond and ter with their surroundings. It is important to underline
Solomon, 2001; Sornette, 1998; Sornette and Cont, 1997), that there exists an important alternative “statistical-
(ii) Preferential attachment processes (Barabási and Al- criticality” approach to the analysis of biological data. It
bert, 1999; Simon, 1955; Yule, 1925). (iii) Optimization focuses on the statistics of existing configurations (with-
and constrained optimization (Carlson and Doyle, 2000; out regard to the temporal order in which they appear,
Seoane and Solé, 2015). much as in equilibrium statistical mechanics) rather than
Even if –as the previous enumeration illustrates– em- on possible underlying dynamical processes, and it is only
pirical power-law distributions can in principle be as- briefly discussed here where, as said above, we choose to
cribed to a handful of possible different generative mech- focus on dynamical aspects.
anisms, in the forthcoming sections we discuss the most
prominent and general mechanism, able to account for
scale invariance both in space and time in a rather ro- 3. Non-equilibrium phase transitions: an example
bust, powerful, and universal way: criticality.
In order to turn the foregoing wordy explanations into
a more formal approach, we describe in detail –as a
2. Criticality in equilibrium systems and beyond guiding example– one of the simplest possible dynam-
ical models exhibiting a non-equilibrium phase transi-
The concept of criticality was born in the context tion. The contact process (CP) is a prototypical toy
of systems at thermodynamic equilibrium. A paradig- model to study the dynamics of propagation of some
matic example are ferromagnets. These exhibit a type of “activity” (as e.g. infections in epidemic spread-
continuous/second-order phase transition at a critical ing; see Fig.2) (Harris, 2002; Henkel et al., 2008; Hin-
temperature, Tc , below which the orientational sym- richsen, 2000; Marro and Dickman, 1999). At any given
metry of spins is spontaneously broken –i.e. a pre- time, each of the nodes i = 1, 2...N of a given network
ferred direction emerges– and, progressively, more or- (which in particular can be a lattice, a fully connected
dered/magnetized states emerge as the temperature is network, or one with a more complex architecture, de-
lowered. On the other hand, above Tc thermal fluctua- scribing the pattern of connections among units/nodes)
tions dominate and the system remains disordered. This is in a state si that can be either occupied/active (si = 1)
change in the collective state is usually encoded in an or empty/quiescent (si = 0). Occupied sites are emp-
order parameter (e.g. the overall magnetization) which tied at rate µ = 1 and new active nodes are created
measures the degree of order as the phase transition pro- at (empty) randomly-selected nearest neighbors of active
ceeds. ones at rate λ. Considering, for the sake of simplicity, a
The described symmetry-breaking is a collective phe- fully connected network with N nodes and performing a
nomenon that requires a system-wide coordination for large-N expansion of the corresponding Master equation
the global re-organization to emerge. This implies (Van Kampen, 1992), one readily obtains a “mean-field”
that the correlation length among individual components or deterministic equation:
needs to span the whole systems at criticality. Similarly,
when the system is becoming incipiently ordered, it is ρ̇(t) = λρ(t)(1 − ρ(t)) − ρ(t) = (λ − 1)ρ(t) − λρ2 (t) (1)
highly fluctuating in the orientation to be chosen. For
example, classical experiment with liquid-gas transitions where the dot stands
PN for time derivative of the activ-
(e.g. with CO2 ) shows that, right at criticality, light of ity density ρ = i=1 si /N . This simple one-variable
many different wavelengths scatters with internal struc- approximation already illustrates some of the essential
tures of the mixture (i.e. there are density fluctuations features of criticality. Eq.(1) reveals the presence of a
of all possible length scales), causing the normally trans- bifurcation at a value λc = 1, separating a subcritical
parent liquid to appear cloudy in a phenomenon called (also called “absorbing” or “quiescent”) phase (λ < 1)
critical opalescence (Binney et al., 1993; Stanley, 1987). in which transient activity decays to the only possible
Importantly, the concepts and methods developed steady-state, ρst = 0, from a supercritical (or “active”)
in the context of equilibrium systems were soon ex- one (λ > 1) with a sustained activity ρst = 1 − 1/λ (see
tended to time-dependent and non-equilibrium problems Fig.2). Thus, defining δ = |λ − 1| as the distance to
 6

criticality, ρst ∼ δ for small δ. In the quiescent (or ab-

sorbing) phase9 , an initial density decays exponentially,
ρ(t) = ρ(0) exp(−δt), implying that there is a character-
istic time scale proportional to δ −1 . Note that such time
diverges at criticality, i.e. it takes a huge time for the sys-
tem to “forget” its initial state, reflecting a generic fea-
ture of criticality: the so-called “critical slowing down”.
Indeed, right at the critical point, the activity decays
asymptotically as a power-law, ρ(t) ∼ t−1 .
Introducing an external field that creates activity at
empty sites at rate h, the overall response or “suscepti-
bility”, defined as Ξ = ∂ρ ∂h |h→0 , is Ξ ∝ δ
st −1
that, again,
diverges right at δ = 0, (i.e. λ = 1), illustrating the di-
verging response to infinitesimal perturbations, another
important generic feature of criticality.
A useful tool to analyze this type of transitions con-
sists in performing “spreading experiments” in which the
evolution of a single localized seed of activity in an other-
wise absorbing/quiescent state is monitored (see Fig.2C).
In this case, given the small number of active sites,
the dynamics is chiefly driven by fluctuations and can-
not be analyzed within the deterministic approximation
above. Stochastic cascades of spatio-temporal activity,
or “avalanches” of variable sizes and durations can be
generated from the initial seed before the system returns
to the quiescent state (extinction). In this framework the
critical point separates a regime of sure extinction (ab-
sorbing phase) from one of non-sure extinction (active
phase). Right at the critical point, the sizes and dura-
tions of avalanches are distributed as power-laws with
anomalously large (formally infinite) variance (Fig.2C)
10
. To understand this mathematically, one needs the
next-to-leading correction to Eq.(1) in the large-N ex-
pansion to include the effect of “demographic” fluctua-
√
tions. This leads to an additional term + ρη(t), where
η(t) is a Gaussian white noise of variance σ 2 = (λ+1)/N .
11

A simple analysis of the resulting stochastic equation12

shows that right at the critical point, the time required to
return to the quiescent state, i.e. the avalanche-durations
T are distributed as power laws: F (T ) ∼ T −α with
α = 2; similarly, avalanches sizes s obey P (S) ∼ S −τ ,
FIG. 2 Sketch of the main aspects of the contact process. with τ = 3/2 These mean-field exponents coincide with
(A) Dynamical rules. (B) Phase diagram, including a critical those of the (Galton-Watson) unbiased branching pro-
point. (C) Temporal raster plots of activity (avalanches) in cess (Harris, 2002; Liggett, 2004; Watson and Galton,
the different regimes, illustrating the complex patterns emerg- 1875), introduced to describe the statistics of family-
ing at criticality, which involve many different scales. (D) names, and often employed to illustrate the statistics of
Avalanche size distributions in the different phases (main)
and, right at the critical point for different system sizes (in-
set), illustrating finite-size scaling, i.e. the emergence at crit-
icality, of a straight line in a double-logarithmic plot, as cor- 9 A similar argument holds in the active phase.
10
responds to scale invariance (see also Fig.3). The large variability of possible patterns is a generic key feature
of criticality. In particular, in systems at equilibrium, the diver-
gence at criticality of the specific heat reflects the huge variability
of possible internal states (Binney et al., 1993).
11 The square-root noise stems from the central limit theorem
(Van Kampen, 1992).
12 See di Santo et al. (2017b) for a pedagogical derivation of this.
 7

critical avalanches. Away from criticality, as well as in Even if the actual dynamics might be much more com-
finite systems, cut-offs appear in the avalanche distribu- plicated, the resulting damage spreading process is sus-
tions (see Fig.2). In particular, as a reflection of the ceptible to be described in simple terms if local effective
underlying scale-invariance at criticality, the finite-size error “propagation” and error “healing” rates can be es-
cut-offs obey scaling laws such as timated.

P (S, N ) ∼ S −τ G(S/N ) (2)

4. Self-organization to criticality
where the power-law S −τ is cut-off by an unspecified
function, G, at an N -dependent scale (Binder, 1981; Bin- As we have seen criticality requires of parameter fine
ney et al., 1993; Stanley, 1987). This enforces that plot- tuning to a precise point to be observed. How it
ting P (S, N )S τ as a function of the rescaled variable is possible that natural systems (such as earthquakes,
S/N should give a unique curve into which all individual Barkhaussen noise, etc.) exhibit signatures of critical-
curves for different sizes N collapse. This finite-size scal- ity, but without any apparent need for parameter tun-
ing method constitutes an important tool for analyzing ing to settle them in at the edge of a phase transition?
critical phenomena (both in computer simulations and in To answer this question P. Bak and collaborators in-
experiments) as perfect power-laws/divergences can only troduced the important concept of “self-organized crit-
appear in the infinite-size limit, not reachable in biolog- icality” (SOC) through a series of archetypical models
ical problems. Indeed, while in finite systems true criti- (Bak, 1996; Bak et al., 1990; Bak and Tang, 1989; Cor-
cality does not exist, still, these may exhibit a progressive ral et al., 1995; Dhar, 1999; Drossel and Schwabl, 1992;
transition between order and disorder. This can be char- Frette et al., 1996; Olami et al., 1992), including its most
acterized by the existence of a peak in some quantity such famous representative: the sandpile model (Bak et al.,
as the susceptibility or the correlation length that usu- 1987).
ally diverge at (true) criticality; this is used as a proxy In the sandpile model a type of “stress” or “energy”
for “approximate” criticality in finite systems13 . (sandgrains) accumulates at a slow timescale at the sites
As a result of universality, all models exhibiting a phase of a (two-dimensional) lattice, and when the accumulated
transition to an absorbing/quiescent phase (without any stress overcomes a local instability threshold, it is in-
additional symmetry or conservation law) share the same stantaneously redistributed among nearest neighbor sites
set of critical exponents and scaling functions –i.e. the –and, possibly, released/dissipated at the system bound-
same type of scale-invariant organization– with the con- aries. This can create a cascade or “avalanche” of fur-
tact process (Henkel et al., 2008)14 ther instabilities. Remarkably, the durations and sizes
Even if the simple propagation model discussed above of such avalanches turn out to be distributed as power
is not intended as a faithful description of the actual dy- laws, i.e. the system becomes critical without any appar-
namics of any specific biological system, in some cases ent need for fine tuning16 (Bak, 1996; Bak et al., 1987;
–such as neural and gene regulatory networks– it can con- Christensen and Moloney, 2005; Dickman et al., 2000;
stitute an adequate effective representation of “damage Jensen, 1998; Pruessner, 2012; Turcotte, 1999; Watkins
spreading” experiments, in which two identical replicas et al., 2015). The mechanism for self-organization to crit-
of the same system are considered; a localized perturba- icality in sandpile models is described in Fig.3. It can
tion in the state of one unit/node is introduced in one of be seen that it is characterized by a dynamical feedback
the two, and the difference between both replicas is mon- that acts differentially depending on the actual system
itored as a function of time (Derrida and Pomeau, 1986). state. This is just an example of a broader class that
Depending on the the system dynamical state, such per- has been extensively analyzed in the context of control
turbations may grow (active phase), shrink (quiescent theory (Magnasco et al., 2009; Moreau and Sontag, 2003;
phase), or fluctuate marginally (critical point), provid- di Santo et al., 2016; Sornette, 1994), which is very likely
ing a practical tool to gauge the level of internal order15 . to emerge in biological systems, as we shall discuss. Two
important variants of this mechanism are as follows:
(a) Self-organized quasi-criticality is analogous to SOC
13 Similarly, systems in the presence of an external driving force but occurs when the dynamics is non-conservative and/or
are not truly critical; in these cases, the Widom line –signaling when the separation of timescales is not perfect (relevant
e.g. the position of maximal susceptibility or correlation– can be for biological problems). This self-organization mecha-
taken as a surrogate of criticality (Williams-Garcı́a et al., 2014).
14 To study spatial effects one needs to replace ρ(t) in Eq.(1) by a
nism drags the system back and forth around the critical
field ρ(x, t) and to introduce a diffusive coupling term (Henkel
et al., 2008; Hinrichsen, 2000; Ódor, 2008).
15 The precise relationship between the damage spreading threshold
and the system’s actual critical point is an important and subtle 16 Stochastic variants of the original (deterministic) sandpile model
issue (Coniglio et al., 1989; Grassberger, 1995; Hinrichsen and (Christensen et al., 1996; Manna, 1991) show much cleaner scal-
Domany, 1997). ing behavior than it (Bagnoli et al., 2003; Ktitarev et al., 2000).
 8

5. Classes of criticality

Not all dynamical phase transitions of relevance in bi-

ology occur between quiescent and active phases, nor can
be described by an associated activity-propagation pro-
cess, such as the contact process. Other important classes
of phase transitions to be found across this paper are:
(i) synchronization transitions, at which coherent behav-
ior of oscillators emerges, as described by the prototyp-
ical Kuramoto model (Acebrón et al., 2005; Kuramoto,
1975; Pikovsky et al., 2003). (ii) transitions to collec-
tive ordered motion, as represented for instance by the
Vicsek model (Vicsek et al., 1995; Vicsek and Zafeiris,
2012)) and its variants; and (iii) percolation transitions
FIG. 3 The self-organization-to-criticality (SOC) mechanism (Christensen and Moloney, 2005), and (iv) even (ther-
works by establishing a feedback loop between the dynamics
of the activity and that of the control-parameter (total ac-
modynamic) transitions such as that of the Ising model
cumulated energy/stress/sandgrains) at separated timescales. (Binney et al., 1993), to name but a few. Each of these
In particular, the control parameter itself becomes a dynami- classes has its own type of emerging ordering and its own
cal variable that operates in opposite ways depending on the scaling features. However, all of them share the basic fea-
system’s state: fast dissipation (negative force) dominates tures that constitute the fingerprints of criticality, such
while the control parameter lies within the active phase and as diverging correlations and response, large variability,
by slow driving dynamics (positive force) dominates in the scale invariance, etc.
absorbing/quiescent phase. This feedback self-organizes the
system to the critical point of its second-order phase tran-
sition if the separation between slow and fast timescales is
infinitely large and the dynamics is conservative (Bonachela 6. Criticality on complex networks
and Muñoz, 2009; Vespignani et al., 1998, 2000; Zapperi et al.,
1995; ?). Otherwise, the system is just self-organized to the Thus far we have discussed criticality in homogeneous
neighborhood of the critical point with excursions around it, systems. However, in many biological problems the sub-
i.e. “self-organized quasi-criticality” (Bonachela and Muñoz, strates on top of which dynamical processes run are
2009; Dickman et al., 2000). highly heterogeneous (Albert and Barabási, 2002; Cal-
darelli, 2007; Newman, 2003, 2010). In particular, com-
plex systems, including biological ones, can be described
point without sitting exactly at it, and is able to gen- as networks, where nodes represent units (neurons, genes,
erate effective scale-invariance across quite a few scales proteins, ...) and links stand for allowed pairwise inter-
(Bonachela and Muñoz, 2009; Kinouchi et al., 2018). actions among them. Such complex networks have been
(b) Adaptive criticality is a variant of SOC from a found to exhibit one or more of the following important
network perspective, in which connections among nodes architectural features: (i) large heterogeneity with a few
in a network are susceptible to be added, removed, or highly connected nodes and many loosely connected ones;
rewired depending on the system’s dynamical state, cre- actually the distribution of connection can be scale free
ating a feedback loop between network architecture and (Barabási and Albert, 1999), (ii) the small-world prop-
dynamics in a sort of co-adaptive process.17 This mecha- erty (Watts and Strogatz, 1998), (iii) modular organiza-
nism can drive the dynamics to criticality (Bianconi and tion18 , (iv) hierarchical organization, etc. (Corominas-
Marsili, 2004; Dorogovtsev et al., 2008; Liu and Bassler, Murtra et al., 2013). These structural features usually
2006) and, in parallel, the network architecture devel- entail profound implications on the dynamics of processes
ops a highly non-random structure, thus capturing the running on top of them (Barrat et al., 2008; Boccaletti
feedback between dynamics and architecture in actual et al., 2006; Dorogovtsev et al., 2008; Pastor-Satorras
biological networks. et al., 2015). For instance, synchronization transitions
proceed in a stepped way on modular networks (Arenas
et al., 2008), and broad critical-like phases can emerge
in hierarchical modular networks (as discussed e.g. in

17 Different variants of this idea have been proposed in the litera-

ture (Bornholdt and Rohlf, 2000; Dorogovtsev and Mendes, 2002;
Droste et al., 2013; Gros, 2008; Gross and Blasius, 2008; Kuehn, 18 Biology is “modular” in many aspects (Alon, 2006; Ravasz et al.,
2012; MacArthur et al., 2010; Meisel and Gross, 2009; Perotti 2002), meaning that some components in biological networks
et al., 2009; Rohlf, 2008; Rybarsch and Bornholdt, 2014; Saito (nodes) are connected among themselves more often or more
and Kikuchi, 2013; Solé et al., 2002b). strongly that they do with others (Alon, 2003).
 9

Moretti and Muñoz (2013) and Muñoz et al. (2010), Ap- III. FUNCTIONAL ADVANTAGES OF CRITICALITY
pendix A and in what follows).
Having discussed basic aspects of criticality and scale
invariance, we move on to ask: what are the poten-
tial virtues of them susceptible to be exploited by liv-
7. Generic scale invariance
ing systems to enhance their functionality? To shed
light onto this, we first describe a well-understood case
We have discussed the paradigm of a critical point –
in which both theoretical and empirical evidence match,
with its concomitant spatio-temporal scale-invariance–
and where the essential and beneficial role played by crit-
separating two alternative phases. However, in some
icality in a biological system is clear and illuminating.
systems with peculiar symmetries, conservation laws or
Later on we discuss a set of possible functional advan-
structural disorder, critical-like features may appear in
tages of criticality from a general perspective.
extended regions in the phase space and not just at a crit-
ical point. This is called generic scale invariance (Grin-
stein, 1991) and can account for empirically reported A. Criticality in the auditory and other sensory systems
scale-invariance in some biological problems without the
need to invoke precise tuning to criticality. Mechanisms The inner ear of vertebrates is able to detect acoustic
for the emergence of generic scale invariance are discussed stimuli with extraordinary sensitivity and exquisite fre-
in Appendix A. quency selectivity across many scales (Hudspeth, 2014).
At the basis of these exceptional features there are hair
cells, the ear’s sensory receptors, which oscillate spon-
8. Statistical criticality taneously even in the absence of stimuli, being able to
resonate with acoustic inputs (Choe et al., 1998; Gold,
To end this introductory section, we briefly discuss 1948; Martin et al., 2001). Intrinsic oscillations are either
an (already mentioned above) alternative perspective to damped or self-sustained depending on the concentration
criticality, particularly useful to analyze the wealth of of Calcium ions, with a Hopf bifurcation separating these
high-quality data now available for living systems (Mora two regimes. Empirical evidence reveals that the ion con-
and Bialek, 2011). It relies on the idea that some fun- centration is regulated in such a way that hair cells op-
damental questions in biology can be tackled within a erate in a regime very close to the Hopf bifurcation (Ca-
probabilistic setting (for instance, analyzing the statis- malet et al., 2000; Ospeck et al., 2001). This has been
tics of spiking patterns may help deciphering the way in argued to entail important consequences for signal pro-
which neurons encode information) (Rieke et al., 1995). cessing (Choe et al., 1998; Eguı́luz et al., 2000; Hudspeth
Bialek and coworkers developed a data-driven maximum et al., 2010; Martin et al., 2001), as we discuss now.
entropy (statistical physics) approach to biological prob- In the simplest possible setting, a hair cell can be ef-
lems, that consists in approximating the probability dis- fectively described as Hopf oscillator (Strogatz, 2014):
tribution of different patterns in a given dataset by a
φ̇(t) = (a + iω̃)φ(t) − |φ|2 φ(t) (3)
probabilistic model that consistently reproduces its main
statistical features (e.g. mean values and pairwise cor- where the φ is a complex number, ω̃ the resonance fre-
relations; see Appendix B). The resulting models are quency, and a is the control parameter (ion concentra-
akin to the Ising models.19 Rather remarkably, Bialek tion) setting the dynamical regime. Eq.(3)
and collaborators observed that the emerging probabilis- √ exhibits self-
sustained oscillations of the form φ(t) = aeiω̃t if a > 0,
tic models for a number of high-dimensional problems – while if a < 0 oscillations are damped.20 Introducing
including biological ones, from retinal neural populations stimuli of the characteristic frequency ω = ω̃ and small
(Schneidman et al., 2006; Tkačik et al., 2014, 2013, 2015) amplitude F (i.e. adding +F eiω̃t to Eq.(3)), and writing
to flocks of birds (Bialek et al., 2012) and the immune φ(t) = R(t)eiωt , one finds
system (Mora et al., 2010), for which excellent empiri-
cal data sets are available– have parameter values sitting Ṙ(t) = R(t)[a − R2 (t)] + F. (4)
close to the edge of a phase transition, i.e the emerging
probabilistic models seem to be critical in a very precise In the oscillatory regime, a > 0, the response R is pro-
sense (Mora and Bialek, 2011) (see Appendix B). portional to the input amplitude F . However, at the
bifurcation (or critical) point, a = 0, the response R is
strongly non-linear, as R = F 1/3 and, consequently, the

19 And since the inferred interactions among “spins” have both

signs, they are a sort of spin glasses (Tkačik et al., 2013; Tkacik
et al., 2009). 20 See Kern and Stoop (2003) from where this discussion is adapted.
 10

ratio response-to-signal R/F = F −2/3 diverges at F → 0,

leading to a huge response to tiny signals of the character-
istic frequency. On the other hand, if the input has some
other frequency ω 6= ω̃ the response is much smaller. This ρ Δ
entails an extremely efficient frequency-selection and am-
plification mechanism, vividly illustrating the advantage Δ
of working close to the instability point.
The described phenomenon involves a single hair-cell log(h) λ
with a specific intrinsic frequency and it is thus not a FIG. 4 Sketch of the behavior of the dynamic range, defined
collective critical phenomenon. However, the Cochlea as , near a critical point. (Left) Steady state density ρ as a
is arranged in such a way that it involves an (almost function of the driving force h (in log scale) for a given value
uni-dimensional) array of diverse and coupled hair cells. of the control parameter λ; the dynamic range, ∆, defined as
When coupling many different Hopf oscillators results in ∆ = 10 log[h(ρ = 0.9)/h(ρ = 0.1)], signals the interval where
the emergence of a true phase transition –i.e. a critical distinguishable responses (i.e. values of ρ) can be measured.
(Right) ∆ exhibits a pronounced peak at criticality.
point with scale-free avalanches– which entails sharpened
frequency response (Duke and Jülicher, 2003; Magnasco,
2003) and enhanced input sensitivity (Gomez et al., 2015;
Kern and Stoop, 2003; Stoop and Gomez, 2016). exploited by living systems in order to induce coordi-
Summing up, woking at criticality has been shown to nated behavior of individual units across space and time.
be essential to generate the extraordinary features of ver- This can be relevant for coordination purposes in e.g.
tebrate hearing, even the most intricate ones (Stoop and neural systems where coherent behavior across extended
Gomez, 2016). Similar virtues of criticality have been ex- areas is observed (Tagliazucchi et al., 2012), in flocks
plored in the olfactory system (Bushdid et al., 2014) and of birds (Cavagna et al., 2010) and in micro-organism
the visual cortex (Shew et al., 2015) (see also Chialvo colonies (De Palo et al., 2017). Similarly, the emergence
(2006) and Friston et al. (2012)). of very large correlation times and critical slowing down
may provide biological systems with a useful mechanism
for the generation of long-lasting and/or slow-decaying
B. Exploiting criticality memories at multiple timescales (see e.g. Deco and Jirsa
(2012)).
1. Maximal sensitivity and dynamic range

As discussed above, an important trademark of critical

points is the divergence of the response (or susceptibil-
ity) which is likely to be exploited in biological sensing 3. Statistical complexity and large repertoires
systems, needing to optimize their response to environ-
mental cues. To better quantify this, a related quantity, The variability of possible spatio-temporal patterns is
dubbed dynamic range, was introduced in Kinouchi and maximal at criticality (as illustrated in Fig.2); this may
Copelli (2006). Consider a model for activity propaga- allow biological systems to exhibit a very wide spectrum
tion (similar to the contact process) with a critical point of possible responses, sometimes called “dynamical reper-
(λc = 1) running on a random network, under the ac- toire” (Rämö et al., 2007, 2006; Yang et al., 2012). This
tion of an external stimulus, h, able to create activity at is consistent with the finding that e.g. models for brain
empty nodes. The dynamic range, ∆ (see Fig.4) gauges activity reach highest signal complexity, with a variety of
the range of diverse stimuli intensities where variations attractors and multistability when operating near criti-
in input h can be robustly coded by variations in the re- cality (Deco and Jirsa, 2012; Haimovici et al., 2013). Sim-
sponse, discarding stimuli that give almost indistinguish- ilarly, (i) the number of metastable states (Haldeman and
able outputs. ∆ turns out to exhibit a marked peak at Beggs, 2005), (ii) the variability of attractors to support
λc = 1, indicating that, at criticality, discriminative out- memories (de Arcangelis and Herrmann, 2010; Krawitz
puts can be associated to a very large variety of inputs, and Shmulevich, 2007), and (ii) the diversity in structure-
with obvious functional advantages for signal detection dynamics relationships (Nykter et al., 2008b) have been
and processing. predicted to be maximized at criticality. All this sug-
gests that in order to spontaneously generate complex
patterns –required e.g. to store highly diverse tokens of
2. Large correlations information– operating near criticality can be an excel-
lent solution for living systems. As a consequence of this,
The emergence of arbitrarily large correlation lengths the capacity to store and process information is optimal
at criticality is an important feature susceptible to be at criticality, as we discuss in more depth in what follows.
 11

4. Computation exploiting criticality These ideas are corroborated by information-theoretic

analyses (Cover and Thomas, 1991), which have un-
It was long-ago conjectured that the extraordinary veiled that the overall transmission of information be-
“computational power” of living systems could be the tween units in a network –as measured by diverse indi-
result of collective behavior, emerging out of a large num- cators23 – is maximal if the underlying dynamical process
ber of simple components (Amari, 1972; Carpenter and is critical24 .
Grossberg, 2016; Grossberg, 1982; Hopfield, 1982). By Let us also mention that (i) state-of-the-art deep learn-
“computation” it is usually meant an algorithm or sys- ing machines (LeCun et al., 2015) may rely on some form
tem that –with the aim of performing some task– assigns of intrinsic scale invariance or even criticality (Lin and
outputs to inputs following some internal logic. Thus, Tegmark, 2017; Mehta and Schwab, 2014; Oprisa and
the computational power of a given device is quantified Toth, 2017a,b; Ringel and de Bem, 2018; Song et al.,
by estimating the amount and diversity of associations 2017), opening exciting research avenues to understand
of inputs to outputs that it can support. As first sug- how artificial-intelligence machines achieve their extraor-
gested in (Ashby, 1960; Turing, 1950) and much further dinary performance, and (ii) from the empirical side, re-
developed in the context of machine learning (Crutchfield cent work has revealed that a mechanism akin to reservoir
and Young, 1988; Langton, 1990; Li et al., 1990; Packard, computing enables neuronal networks of the cerebellum
1988) networked systems operating at criticality can have to perform highly complex tasks in an efficient way by
exceptionally high computational capabilities. In partic- operating at criticality (Rössert et al., 2015).
ular, Langton formulated the question: under what con-
ditions will physical systems support the basic operations
of information transmission, storage, and modification, IV. ALLEGED CRITICALITY AND SCALING IN
required to support computation?’ His answer was that BIOLOGICAL SYSTEMS
systems21 operating at the “edge of chaos” are especially
suitable to perform complex computations22 . The “edge Having discussed putative virtues of critical dynam-
of chaos” or critical point (as we rather call it here) is the ics, susceptible in principle to be exploited by biological
borderline between two distinct phases or regimes: the systems, we now start a trip through some of the most-
chaotic/disordered one in which perturbations and noise remarkable existing empirical evidence revealing signa-
propagate unboundedly (thereby corrupting information tures of criticality in such systems. We warn the reader
storage) and the frozen/ordered phase whereas changes that –even if the aim is to present a collection as exten-
are rapidly erased (hindering the capacity to react and sive and exhaustive as possible– the selection of topics as
transmit information). Therefore, the critical point con- well as the extent in which they are discussed might be
fers on computing devices composed of equivalent units biased by our own experience. Also, importantly, even
an optimal tradeoff between information storage and in- if some of the experiments and findings to be discussed
formation transmission, two of the key ingredients pro- are very appealing, evidence in many cases is not com-
posed by Turing as indispensable for universal computing plete and conclusions should be always taken with cau-
machines (Turing, 1950). tion. Indeed, for many of the forthcoming examples, we
In artificial intelligence, criticality is exploited in so- also discuss existing criticisms and potential technical or
called “reservoir computing” (Lukoševičius et al., 2012) interpretative problems.
that was developed independently in the fields of ma-
chine learning (“echo state networks” of Jaeger (2007))
and computational neuroscience (“liquid state machine” A. Neural activity and brain networks
in Maass et al. (2002)). These machines consist of a net-
1. Spontaneous cortical activity
work of nodes and links, “the reservoir”, where each node
represents an abstract “neuron” and links between them
The cerebral-cortex of mammalians is never silent, not
mimic the connectivity of actual biological circuits. A
even under resting conditions nor in the absence of stim-
series of seminal works showed that such machines can
uli; instead, it exhibits a state of ceaseless spontaneous
perform real-time computations –responding rapidly to
time varying input signals– in a coherent yet flexible way
if they operate near a critical point (Bertschinger and
Natschlager, 2004; Boedecker et al., 2012; Legenstein and 23 Such as the transfer entropy (Lizier et al., 2008b; Shriki and
Maass, 2007; Legenstein, 2005; Maass et al., 2002). Yellin, 2016; Solé and Miramontes, 1995), Fisher information
(Wang et al., 2011) and, more in general, statistical complex-
ity (as discussed above) (Krawitz and Shmulevich, 2007; Lizier
et al., 2008a; Rämö et al., 2007).
21 Cellular automata in this case (Wolfram, 2002). 24 See Beggs (2008); Li et al. (1990); Luque and Ferrera (2000);
22 This proposal triggered a heated debate; see, e.g. (Crutchfield, Prokopenko (2013); and Ribeiro et al. (2008); and Barnett et al.
2012; Crutchfield and Young, 1988; Melanie, 1993). (2013) and Toyoizumi and Abbott (2011) for a discrepant view.
 12

electro-chemical activity with very high variability and

sensitivity(Arieli et al., 1996; Fox and Raichle, 2007;

LFP
Raichle, 2011; Yuste et al., 2005). Understanding the
genesis and functionality of spontaneous cortical activ-
ity – which accounts for about 20% of the total oxygen t
75
consumption of a person at rest– is key to shedding light
onto how the cortex processes information and computes 60

unit label
(Arieli et al., 1996; Deco et al., 2011, 2013a; He, 2014). 45

Criticality might play a key role to generate such a vari- 30

able and sensitive activity as diverse empirical results 15

suggest. 0
2 3 4 5
t x 10 4
11
An adult human brain consists of almost 10 neurons Δt
and up to 1015 synaptic connections among them, form- 75

ing an amazingly complex network through which electric 60

unit label
signals propagate (Keenan et al., 2007). Neurons inte- 45
grate presynaptic excitatory and inhibitory inputs from 30
other neurons, and fire an action potential when a given 15
threshold is overcome, stimulating further activity. This 0
3.15 3.2 3.25
generates irregular cascades or outbursts of activity in- t x 10 4
terspersed by quiescent periods, as empirically observed
both in vitro (Eytan and Marom, 2006; Sanchez-Vives FIG. 5 Sketch illustrating how neuronal avalanches are mea-
and McCormick, 2000; Segev and Ben-Jacob, 2001; Segev sured from Local field potential (LFP). (Top) LFPs are mea-
sured at different locations; negative peaks of the time series
et al., 2001; Tabak and Latham, 2003) and in vivo (Meis-
correlate with large population spikes of the underlying neu-
ter et al., 1991; Steriade et al., 1993) (see Fig.5). Is this rons within each local region. (Middle) Raster plot illustrat-
activity related to inherent critical behavior? In what ing the times at which peaks of the LPF occurs for different
follows we discuss empirical pieces of evidence suggest- sites, revealing a high degree of temporal clustering. (Bottom)
ing diverse possible connections with different types of Enhancing the temporal resolution, it is possible to resolve
phase transitions. the spatio-temporal organization within apparently coherent
large-scale events; is occurs occur in the form of “neuronal
avalanches” (shaded columns) interspersed by periods of qui-
escence (white columns).

2. The edge of activity propagation: avalanches scaling25 ; see Beggs and Plenz (2003); Mazzoni et al.
(2007); and Petermann et al. (2009)). The observed
exponents coincide with those of the (mean-field) criti-
In a remarkable breakthrough, Beggs and Plenz (2003)
cal contact/branching processes as described above and,
succeeded at resolving the internal spatio temporal orga-
thus, seem to describe a marginal activity-propagation
nization of the above-mentioned outbursts of neuronal
process. Moreover, the mean temporal profile of neuronal
activity. They analyzed mature cultures as well as acute
avalanches of widely varying durations is quantitatively
slices of rat cortex, and recorded spontaneous local field
described by a single universal scaling function (Fried-
potentials (LFP) –which provide coarse-grained measure-
man et al., 2012; Sethna et al., 2001), and scaling rela-
ments of electrochemical activity– at different locations
tionships between the measured exponents are fulfilled
and times. Local events of activity are defined as (neg-
(Friedman et al., 2012). Similar avalanches have been
ative) peaks of the LFP signals, which are indicative of
observed in vitro (Mazzoni et al., 2007; Pasquale et al.,
local population spikes (Beggs and Plenz, 2003). As illus-
2008) and in vivo for different species (Gireesh and Plenz,
trated in Fig.5, events at different sites have a tendency
2008; Hahn et al., 2010; Petermann et al., 2009; Ribeiro
to cluster in time, producing network spikes of activity.
et al., 2010; Yu et al., 2011) and across resolution scales,
Each of these outbursts of activity when temporally re-
from single neuron spikes to rather coarse-grained mea-
solved, consists in a cascade of succesive local events,
surements.26 The fact that at quite different resolution
organized as neuronal avalanches interspersed by peri-
ods of quiescence (Beggs and Plenz, 2003, 2004). The
avalanche sizes (i.e. number of local events each one in-
25 Instead, if data are temporally reshuffled the distributions be-
cludes) and durations were found to be distributed as
come exponential ones, meaning that large coherent events dis-
power-laws with exponents τ ≈ 3/2 and α ≈ 2, respec- appear (Beggs and Plenz, 2003; Plenz and Thiagarajan, 2007).
tively, with cut-offs that increase with system size in a 26 This includes single unit recordings (Bellay et al., 2015), lo-
scale-invariant way (i.e. the distributions obey finite-size cal field potentials (LFP) (Beggs and Plenz, 2003; Petermann
 13

scales similar results are reported is, by itself, strongly critical dynamics (additionally driven my external forces)
supportive of the existence of underlying scale-invariant rather than by a critical one.
dynamical processes.27 (iv) Limited scales: In general, no more than two, at
All this evidence regarding neuronal avalanches seems most three, orders of magnitude in avalanche statistics
to make a strong case in favor of criticality. However, have been reported which is somehow unsatisfactory. Ob-
some caveats need to be made: taining much broader regimes of scale invariance is tech-
(i) Thresholding: A source of ambiguities in extracting nically challenging, but would make a stronger case for
(discrete) events from (continuous) time-series analyses actual scale-invariance (Yu et al., 2014).
comes from thresholding; i.e. signals at any given spatio- (v) Some authors support different interpretations of
temporal location need to overcome some threshold to be the observed power-laws, which are unrelated to critical-
declared an “event”of activity. Petermann et al. (2009) ity (Bédard et al., 2006; Destexhe, 2009; Touboul and
compared results for different thresholds in LFPs time se- Destexhe, 2010, 2017).
ries and found that exponent values remain unchanged, These series of observations, taken together, seem to
suggesting the existence of a truly scale-invariant orga- shed some doubts on evidence in favor of criticality re-
nization of events. However, a word of caution is still re- lying on avalanches. To further stregthen it, we now
quired as recent works have underlined the “perils” asso- discuss other complementary experimental signatures of
ciated with thresholding, which in some controlled cases criticality from different perspectives.
has been shown to generate spurious effects such as ef-
fective exponent values and correlations in the timings of
consecutive avalanches (Font-Clos et al., 2015; Janićević 3. The edge of neural synchronization
et al., 2016; Laurson et al., 2009). Further clarifying this
issue is key to make solid progress in the empirical anal- Much attention has been historically devoted to brain
ysis of avalanching systems. rhythms observed in EEG, MEG, and LFP measure-
(ii) Time binning: Avalanches can only be defined ments (Buzsaki, 2009). Such rhythms emerge owing to
by employing a criterion to establish when an avalanche the transient synchronization between different neural re-
starts and when it ends. This requires setting a discrete gions/circuits, and they play a key role in neural function
time binning to be applied to the data: an avalanche (Steriade et al., 1996). Clusters of neurons with coherent
starts when a time-bin with some activity within it fol- neural activity have a much stronger coordinated effect
lows a series of preceding consecutive quiescent ones, and on other neuronal assemblies than asynchronous neurons
ends when a new quiescent time-bin appears (Beggs and do (Brunel and Hakim, 2008; Kelso et al., 1986; Scholz
Plenz, 2003) (see Fig.5). This introduces some ambi- et al., 1987). Thus, phase synchrony is essential for large-
guity, and the measured avalanche exponents have been scale integration of information (Varela et al., 2001), and
shown to be sensitive to the choice of the time-bin. How- abnormalities in the level of synchronization –either by
ever, taking the time bin to coincide with the mean inter- excess or by defect– are a signature of pathologies such
event interval, the mean-field branching process expo- as epilepsy, Parkinson’s disease, schizophrenia, or autism
nents seem to be systematically recovered (Beggs and (Yang et al., 2012). Empirically, the measured level of
Plenz, 2003; Haimovici et al., 2013; Petermann et al., synchronization across (resting) brain regions and across
2009; Tagliazucchi et al., 2012). As above, further work time has been found to be highly variable and with strong
is needed to mathematically clarify this important issue. long-range correlations. Such spatio-temporal variability
(iii) Sub-sampling: A related problem is that of sub- can be interpreted as a template to codify vastly differ-
sampling as a result of observational and resolution lim- ent tasks and to allow for a large dynamical repertoire
itations. Owing to these factors the statistic is not com- (Arieli et al., 1996), and has been observed to diminish
plete, and this might affect the shape of the observed dis- when the subject is engaged in a specific task (Tinker
tributions. Priesemann et al. (2009, 2013, 2014) argued and Velazquez, 2014).
that –taking into consideration sub-sampling effects– em- The role that criticality might play in keeping inter-
pirical data are best characterized by a slightly sub- mediate and variable levels of synchrony –which could
for example be essential to achieve a good balance be-
tween integration and segregation (Tononi et al., 1994)–
has been empirically analyzed as we discuss now. Ana-
et al., 2009), electroencephalography (EEG) (Allegrini et al.,
2010; Freeman et al., 2003; Meisel et al., 2013), electrocorticog- lyzing spontaneous bursts of coordinated activity (as in
raphy (ECoG) (Solovey et al., 2012), magnetoencephalography Fig.5; (Segev and Ben-Jacob, 2001; Segev et al., 2001))
(MEG) (Novikov et al., 1997; Palva et al., 2013; Poil et al., 2012; the overall level of phase synchrony between different
Shriki et al., 2013), and functional magnetic resonance imaging
(fMRI) (Haimovici et al., 2013; Tagliazucchi et al., 2012).
electrodes has been recorded under different pharma-
27 Some studies suggest that even single neurons can be intrinsically cological conditions, ranging from excitation-dominated
critical to optimize their inherent excitability (Gal and Marom, to inhibition-dominated regimes. It was observed that
2013; Gollo et al., 2013). the is a critical point at which excitation and inhibition
 14

balance (Yang et al., 2012). At such a point –i.e. at colation transitions. These empirical data reveal that
“the edge of synchrony” (Brunel, 2000; Deco et al., 2014; the dynamics is close to the critical percolation density
Palmigiano et al., 2017)– the level of synchronization value, but with broad excursions to both, sub- and super-
variability is maximal and scale-free avalanches of activ- critical phases, suggesting that regulatory mechanisms
ity can be concomitantly observed (Gireesh and Plenz, keep the system hovering around a percolation transition
2008; Yang et al., 2012). Actually, a recent theoretical (much as suggested by the mechanism of “self-organized
work emphasizes that if the cortex operates at a crit- quasi-criticality” discussed in Sect.I). In other words, the
ical point, it should be a synchronization critical point, resting brain spends most of the time near the point of
where marginal synchronization and scale-free avalanches marginal percolation of activity, neither too inactive nor
emerge together (di Santo et al., 2017a). Last but not exceedingly active.
least, the amazingly detailed computational model built
within the large-scale collaborative Blue brain project
(Markram, 2006) also suggests that the cortical dynamics
operates at the edge of a synchronization phase transition
(Markram et al., 2015).
6. The edge of a thermodynamic transition

4. The edge of global stability The state of a neural network at a given small time
window can be represented by a binary vector encoding
High temporal-resolution electrocorticography data whether each individual neuron has spiked or remained
from human reveal time-varying levels of activity across silent within it (Tkačik et al., 2013). Questions of interest
different spatial locations (Magnasco et al., 2009; Solovey are, how often does a given simultaneous (i.e. within
et al., 2012). Representing the system’s state at a given a given time bin) spike pattern appear? What is the
time as a vector, its time evolution can be approximated simplest probabilistic model (in the sense of equilibrium
as a series of linear (matricial) transformations between statistical mechanics) able to reproduce such statistical
successive time-discrete vector states (Akaike, 1969). By patterns?
employing an eigenvector decomposition of each of such
Pioneering empirical studies obtained data from large-
matrices, it is possible to monitor the temporal dynam-
scale multielectrode array recordings to determine the
ics of the leading eigenvalues (Lyapunov exponents). In
statistics of patterns of neural activity in large popu-
awake individuals, the leading eigenvalue turns out to
lations of retinal (ganglion) cells of salamander (Marre
oscillate closely around the threshold of instability, indi-
et al., 2012). Employing such high-resolution data
cating that the dynamics is self-regulated at the edge of
and inferring from them maximum-entropy probabilistic
a phase transition between stable and unstable regimes.
(Ising-like) models (as briefly described in Appendix B),
Quite remarkably, in anesthetized subjects eigenvalues
Tkačik et al. (2013) observed that the associated spe-
become much more stabilized, suggesting that operating
cific heat diverges as a function of sample size as oc-
at the edge of stability is a property of functional brain
curs in thermodynamic critical points. Furthermore, in-
and that deviations from such point can be used as a
troducing an effective temperature they observed that
measure of loss of consciousness (Alonso et al., 2014).
empirical data are poised near the critical point of the
(temperature-dependent) generalized model suggesting
that the visual cortex might operate in a close-to-critical
5. The edge of percolation
regime (Mora and Bialek, 2011; Mora et al., 2015; Tkačik
et al., 2014, 2015).
Cortical dynamics can be viewed as a sort of perco-
lation phenomenon. Tagliazucchi et al. (2012) analyzed A possible interpretation of these results –backed by re-
functional magnetic resonance imaging (fMRI) time se- cent empirical evidence (Gautam et al., 2015; Shew et al.,
ries at different regions of (resting) humans. By thresh- 2015)– is that adaptation to sensory inputs has tuned the
olding them they obtained discrete spatio-temporal maps visual cortex to statistical criticality, thus optimizing its
of activity (much as in Fig.5). They found that –using performance. A competing view is that the observed sig-
the density of “active” sites at a given time as a con- natures of criticality could reflect an effective averaging
trol parameter, and the size of the largest connected over un-observed variables (such as common external in-
cluster at each time as a percolation order parameter– puts in the case of retinal populations), lacking thus any
there is a value of the control parameter nearby which relationship with possible functional advantages (Aitchi-
the dynamics spends most of the time and, remarkably, son et al., 2016; Nonnenmacher et al., 2017) (see Ap-
it corresponds to the value for which the total number pendix B). We refer to Tkačik et al. (2015) and Nonnen-
of different connected clusters as well as their size vari- macher et al. (2017) for a more thorough discussions on
ability are maximal, as happens at the threshold of per- these alternative viewpoints.
 15

7. Large-scale cortical dynamics 8. Disruptions of criticality in pathological conditions

Important pieces of evidence that scale invariance and

Large research initiatives have allowed for the measure-
criticality might be specific of awake and healthy brain
ment of network of physical (neuro-anatomical) connec-
activity emerge from experimental analyses of neural ac-
tions between different regions of the human brain, i.e.
tivity under pathology or modified physiological condi-
the “human connectome network ”28 .
tions. For example, signatures of criticality have been
On the other hand, functional magnetic resonance reported to fade away during epileptic seizures (Hobbs
imaging (fMRI) studies performed in the resting-state et al., 2010; Meisel et al., 2012) as well as during anoma-
–i.e., while the subject is awake not performing any spe- lously large periods of wakefulness (Meisel et al., 2013) or
cific task– reveal the emergence of spatio-temporal pat- while performing simple tasks (Hahn et al., 2017; Tomen
terns of strongly coherent fluctuations in the level of ac- et al., 2014). Also, long-range temporal correlations –
tivity at the large scale. This allows for the determination characteristic of the awake state (Expert et al., 2011; He,
of so-called “resting state networks”, encoding pairwise 2011)– break down during anesthesia (Bellay et al., 2015;
correlations between different brain regions, or in words, Ribeiro et al., 2010), unconsciousness (Tagliazucchi et al.,
brain regions that become active or inactive together, 2016) or under deep sleep (Tagliazucchi et al., 2013), sug-
and that are consistently found in healthy individuals29 . gesting that critical dynamics is specific to the state of
wakefulness. Interestingly, sleep has been interpreted as
Diverse studies of simple dynamical models on top of a mechanism to restore the overall dynamics to a critical
the empirically determined human connectome network state (Pearlmutter and Houghton, 2009).
it was found that spatio-temporal correlations similar to
By pharmacologically altering the ratio of excitation to
those of the empirically-measured in the resting state are
inhibition, i.e. breaking the balance condition that char-
reproduced only if the models operate close to critical-
acterizes functional neural networks (Barral and Reyes,
ity (Cabral et al., 2011; Fraiman et al., 2009; Haimovici
2016; Rosenbaum and Doiron, 2014; van Vreeswijk and
et al., 2013; Plenz, 2013), suggesting that resting-state
Sompolinsky, 1996)– induces a tendency to super-critical
spatio-temporal patterns of activity emerge from the in-
propagation of activity, including many large system-
terplay between critical dynamics and the large-scale un-
spanning avalanches, clearly disrupting scale-invariant
derlying architecture of the brain. Thus, resting state
behavior (Beggs and Plenz, 2003; Mazzoni et al., 2007).
networks reflect structured/critical fluctuations among a
Similarly, only during naturally balanced conditions the
set of possible attractors suggestive of a state of alertness
dynamic range (as defined above) is empirically observed
facilitating rapid task-dependent shifts (Deco and Jirsa,
to be maximal (Gautam et al., 2015; Shew et al., 2009).
2012; Deco et al., 2013b; Ghosh et al., 2008).
There is also experimental evidence supporting the
On the other hand, one could expect that scale- idea that developing cortical networks go through differ-
invariance emerges in broad regions of parameter space ent stages in the process of maturating: they shift from
and not just at critical points (see Appendix A), owing being supercritical, to subcritical, and then finally, con-
to the modular and highly heterogeneous architecture of verge towards criticality only when they become mature
structural brain networks. This has indeed been veri- (Stewart and Plenz, 2008; Tetzlaff et al., 2010).
fied to be true for models of neural activity propagation Taken together, these observations suggest that crit-
(Moretti and Muñoz, 2013) as well as for synchroniza- icality is the baseline state of mature, healthy, and
tion dynamics (Sadilek and Thurner, 2015; Shanahan, awake neural networks and that deviations from criti-
2010; Villegas et al., 2014), and implies that cortical dy- cality have profound functional consequences (Massobrio
namics might not be required to be exactly critical to et al., 2015).
reproduce empirical findings, but just to be located in a
broad region in parameter space exhibiting generic scale
invariance (e.g. in a Griffiths phase; see Appendix A). 9. Mathematical models of neuro-criticality

Since the idea that the computational power of the

brain could emerge out of collective properties of neu-
ronal assemblies (Hertz et al., 1991; Hopfield, 1982),
a large and disparate number of modeling approaches
28 The resulting human connectome turns out to be a network or- have been proposed to scrutinize neural dynamics (Amit,
ganized in hierarchical modular way (Betzel et al., 2013; Break- 1992; Amit and Brunel, 1997; Dayan and Abbott, 2006;
spear, 2017; Hagmann et al., 2008; Kaiser, 2011; Meunier et al.,
2010; Sporns, 2010; Sporns et al., 2004, 2005).
Izhikevich, 2004, 2007; Kandel et al., 2000; Mattia and
29 See the vast literature on this, e.g. Beckmann et al. (2005); Sanchez-Vives, 2012). These models uncovered a large
Biswal et al. (1995); Deco et al. (2011, 2013a); Diez et al. (2015); variety of phases and possible dynamical regimes such as
Greicius et al. (2003); and Raichle et al. (2001). up and down states (Hidalgo et al., 2012; Holcman and
 16

Tsodyks, 2006; Mattia and Sanchez-Vives, 2012; Mejias 2015; Bonachela et al., 2010). If the separation of
et al., 2010; Parga and Abbott, 2007), synchronous and timescales in these models is fixed to moderate (finite)
asynchronous phases (Abbott and van Vreeswijk, 1993; values, critical self-organization is not achieved; instead,
Brunel, 2000; Brunel and Hakim, 2008), as well as phase the system hovers around the critical point with excur-
transitions separating them. Our aim here is not to sions to both sides of it –as in the above-discussed self-
review them exhaustively but, rather, to discuss those organized quasi-criticality (Bonachela and Muñoz, 2009;
aimed at justifying the possible emergence of criticality Kinouchi et al., 2018)– or may become not critical at all
in actual neural networks. (Bonachela et al., 2010).
P. Bak and collaborators are to be acknowledged for To overcome these difficulties an influential model was
first proposing that concepts of self-organization to crit- proposed to explain self-organized criticality without as-
icality could play a role in neural dynamics30 . Herz and suming conservative dynamics nor an infinite separation
Hopfield (1995) realized that stylized (integrate-and-fire) of timescales (Millman et al., 2010). This model (con-
models of neuronal networks were mathematically equiv- sisting of a network of leaky integrate-and-fire neurons
alent to SOC archetypes. with synaptic plasticity) exhibits a discontinuous phase
Short-time synaptic depression (Markram and transition –rather than a continuous one with a criti-
Tsodyks, 1996; Sussillo et al., 2007; Tsodyks and cal point– between states of high and low activity, re-
Markram, 1997) was introduced in SOC-like neural- spectively. This is neurobiologically sound as similar
network models (in which some form of neural “stress” “up” and “down” states are empirically known to emerge
is accumulated and then released to connected units under deep sleep or anesthesia (Holcman and Tsodyks,
in a conserved way) as a mechanism to regulatory 2006; Steriade et al., 1993). Remarkably, the model was
mechanism able to auto-organize them to the edge of a also found to display scale-free avalanches all across its
phase transition (de Arcangelis et al., 2006; Gómez et al., active phase. This is puzzling from the viewpoint of
2008; Levina et al., 2007, 2009)31 ; synaptic resources models of activity propagation, which generate scale-free
become depleted owing to network activity and remain avalanches only at criticality.
so for a characteristic recovery period, while they slowly This apparent paradox has been recently solved:
recover to their baseline level. The alternation of these avalanches in the model of Millman et al. are not
activity-dependent mechanisms (i.e. slow charging the result of criticality; they appear owing to the ex-
and fast dissipation) generates a feedback loop that, istence of generic scale invariance, which is a conse-
allegedly, guides the networks to criticality, much as in quence of an underlying neutral dynamics (see Appendix
SOC (Fig.3). A). Importantly, such neutral avalanches are detected in
Alternative regulatory (homeostatic) mechanisms such computational models by employing information about
as spike-timing dependent plasticity (Effenberger et al., causal relationships on which neuron triggers the fir-
2015; Meisel and Gross, 2009; Rubinov et al., 2011; ing of which other (Martinello et al., 2017), and this
Shin and Kim, 2006), retro-synaptic signals (Hernandez- type of information is usually not accessible in experi-
Urbina and Herrmann, 2017), and Hebbian plasticity ments.32 Furthermore, if avalanches in the model of Mill-
(de Arcangelis and Herrmann, 2010; Uhlig et al., 2013), man et al. are measured as in experiments (employing a
have been proposed to explain self-organization to criti- time binning) they are not scale-free (Martinello et al.,
cality (Bienenstock and Lehmann, 1998). 2017). Thus, this model –as well as some other simi-
However, these SOC-like approaches might not be bio- lar ones (Stepp et al., 2015)– do not describe empirical
logically plausible, as they rely on conservative or almost- temporally-defined scale-free avalanches. More generally,
conservative dynamics (while neurons and synapses are these results reveal a gap in the literature between time-
leaky) and, even more importantly, they require of an binned defined avalanches (in experiments) and causally
unrealistically large (infinite) separation of timescales be- defined avalanches (in models).
tween dissipation and recovering to actually self-tune the All the above-discussed approaches have in common
dynamics to a critical state (de Andrade Costa et al., that they identify neural criticality with the edge of an
activity-propagation phase transition. Recently, some
other theoretical models have provided theoretical evi-
dence that neural dynamics should exhibit a synchroniza-
30 See e.g. Bak (1996); Bak and Chialvo (2001); Chialvo (2004); tion phase transition, at which neuronal avalanches and
Chialvo and Bak (1999); and Stassinopoulos and Bak (1995).
Also, early work by Haken, Kelso and coworkers brought about incipient oscillations coexist (Gireesh and Plenz, 2008;
the role that critical fluctuations and critical slowing-down might Poil et al., 2012; di Santo et al., 2017a; Yang et al.,
play in neural dynamics (Haken, 1977, 2013; Kelso et al., 1986; 2012). However, these models provide no explanation
Scholz et al., 1987).
31 This opened the door to studies of the interplay between criti-
cal dynamics, memory and learning (de Arcangelis, 2011, 2012;
de Arcangelis and Herrmann, 2010, 2012; de Arcangelis et al.,
2014). 32 See, however, Williams-Garcia et al. (2017).
 17

Egrla
–other than a possible fine tuning– of why the dynamics
should operate precisely at the edge synchronization. Zfp322a
Esrrb

MAPK signalling Sall4

Last but not least, the amazingly detailed computa- Oct4
FGF signalling

tional model built within the large-scale collaborative Rex1

Fgf4

Blue brain project (Markram, 2006) suggests that the Nanog Sox2

Jak/Stat3
cortical dynamics operates at the edge of a phase transi- LIF
C-Fos

tion between an asynchronous phase and a synchronous Average connectivity K IL-6

one with emerging oscillations (Markram et al., 2015). Ordered phase Disordered phase
The regulation of calcium dynamics has been cited as
a possible responsible mechanism for keeping the system Criticality
(K~2) i 1 i 2 i 3 OU T
close to such a critical state, operating at a point at which 0 0 0 0
0 0 1 1
a whole set of empirical results can be quantitatively ex- 0 1 0 0

plained by the model (Markram et al., 2015). 0 1 1 1

1 0 0 0
Finally, let us comment on two theoretical approaches 1 0 1 1
Boolean 1 1 0 1
–not relying on criticality– proposed to account for scale- approach 1 1 1 0
free neuronal avalanches. The first one is a mechanism
called “stochastic amplification of fluctuations” which is FIG. 6 The upper panels represent two gene regulatory net-
able to produce highly variable avalanches with an (ap- works: (Left) a large scale one (for E. Coli (Gama-Castro
proximate but not perfect/critical) balance between ex- et al., 2015)) and (Right) a small-scale one (mouse embry-
citatory and inhibitory couplings together with inherent onic stem-cell subnetwork (Parfitt and Shen, 2014)). In both
stochasticity (Benayoun et al., 2010; Murphy and Miller, cases, nodes stand for genes and links between them for regu-
2009). However, this mechanism is not able to reproduce latory interactions (see main text). The lower panel shows a
sketch of random Boolean networks as simple models of gene
the empirically observed exponent values (di Santo et al.,
regulation. For low (high) average connectivities they lie in
2018). the ordered (disordered) phase, with a critical point occurring
The second is a recent work, Touboul and Destexhe close to K = 2. The table illustrates a set of logical opera-
(2017), where it is proposed that scale-free avalanches tions (associating an output to a set of 3 different inputs) for
can naturally emerge in networks of neurons (described a given node in a Boolean setting.
e.g. as a balanced network with excitation and inhibition
(Brunel, 2000), or even as simple Poissonian point pro-
cellular states have been identified as attractors of the
cesses) operating in synchronous irregular regimes away
dynamics of gene regulatory networks, where the genes
from criticality. In our opinion, further work needs to be
are the network nodes and their mutual regulatory (ac-
done to understand how and under which circumstances
tivation/repression) interactions are represented as di-
this is true, and what are the corresponding values of
rected links between them. Cells can be thought as
the resulting avalanche exponents. Summing up, appeal-
“machines” executing complex gene-expression programs
ing empirical evidences as well as sound dynamical mod-
that involve the coordinated expression of thousands of
els supporting the idea of criticality in the brain exists;
genes33 (Alon, 2006; Buchanan, 2010; Crick, 1970; Ki-
however, in many cases empirical results are not fully
tano et al., 2001; Koonin, 2011; Koonin et al., 2006).
convincing and alternative theoretical interpretations are
Consequently, the study of information processing in
still under debate. Fully clarifying the nature of the over-
cells shifted progressively from single genes to increas-
all cortical dynamical state remains an open challenge.
ingly complex circuits/networks of genes and regulatory
interactions, shedding light on collective cellular states
(Garcia-Ojalvo, 2011; Hartwell et al., 1999; Shmulevich
B. Gene regulatory networks
and Dougherty, 2010). The development of powerful ex-
perimental high-throughput technologies in molecular bi-
Leaving aside neural networks, we move on to another
ology paved the way to the experimental investigation
type of biological information-processing networks that
of gene-expression patterns in large regulatory networks
also exhibit signatures of criticality: genetic networks.
(Filkov, 2005) and, in particular, provided empirical ev-
Living cells exhibit stable characteristic features which
idence that sequences of cell states (apoptosis, prolifer-
are robust even under highly variable conditions. In par-
allel, they also exhibit flexibility to adapt to environ-
mental changes. These two aspects are compatible ow-
33 Individual genes are the basic information units of the genetic
ing to the fact that a given set of genes (i.e. a “geno-
code and occupy a central role in biological inheritance and evo-
type”) can give rise to different cellular states (“phe- lution. Gene information is transcribed into RNA molecules and
notypes”), consisting of diverse gene-expression patterns from them translated into proteins (i.e. “expressed”) which are
in which some genes are differentially expressed or si- the final result of gene expression and the building blocks of
lenced. Since the pioneering work of Kauffman (1993), functionality (Crick, 1970).
 18

ation, differentiation, etc.) can be viewed as programs 1993), and that this might entail a large variety of essen-
encoded in the dynamical attractors of gene regulatory tial functional advantages (Aldana et al., 2007; Kauffman
networks (Albert and Othmer, 2003; Espinosa-Soto et al., et al., 2003; Krawitz and Shmulevich, 2007; Nykter et al.,
2004; Huang et al., 2005; Li et al., 2004). 2008b; Ribeiro et al., 2008; Torres-Sosa et al., 2012).
In the ordered regime, convergence in state space im-
plies that distinctions between different inputs are readily
1. Models of genetic regulatory networks erased, precluding reliable discrimination of them. In the
disordered phase, even small perturbations lead to a very
Many genes are empirically observed to exhibit bista- large divergence of trajectories in state space preclud-
bility, i.e. their gene-expression levels can be approxi- ing reliable action (Kauffman et al., 2003). Hence, crit-
mated as either “high” (on) or very “low” (off) depending icality might confer on such networks an optimal trade-
on conditions. Such binary states are believed to be the off between the robustness and accuracy that biological
building blocks of genetic logical circuits (Tyson et al., machinery demands and responsiveness to environmen-
2003). Thus, genetic regulatory networks have been tra- tal clues (Kauffman et al., 2003). At larger evolutionary
ditionally modeled as binary information-processing sys- scales, criticality might provide networks with an opti-
tems in which the expression level of each gene is repre- mal balance between robustness and evolvability under
sented by a Boolean (on/off) variable and their interac- changing conditions (Aldana et al., 2007; Gershenson,
tions are modeled as Boolean functions whose inputs are 2012; Kaneko, 2012; Torres-Sosa et al., 2012; Wagner,
the states of other genes (see Fig.6) (Kauffman, 1993; 2005).
Shmulevich and Dougherty, 2010).34 Boolean descrip- It remains to be clarified how could adaptive (Gell-
tions constitute the most basic and crudest approach to Mann, 1994; Gros, 2008) and/or evolutionary (Nowak,
gene regulatory networks; still, they are particularly ade- 2006) mechanisms, specific of living systems36 , lead bi-
quate to analyze large networks as they reduce the over- ological networks to operate close to criticality. Theo-
whelming complexity of the real problem to a logical one, retical approaches tackling this question are discussed in
and they have been shown to successfully explain e.g. cell Appendix C.
cycles (Aldana, 2003; Bornholdt, 2005, 2008; De Jong,
2002; Drossel, 2008; Gros, 2008; Kauffman, 1996, 1993;
Serra et al., 2007). 2. Gene knock-out and damage spreading
In the simplest setup, the network architecture is de-
scribed as a random directed network35 and regulatory DNA microarrays or “DNA chips” are devices allow-
interactions are described as random Boolean functions ing to measure the expression levels of large numbers of
(Albert, 2004; Alon, 2006; De Jong, 2002; Gros, 2008; genes within a cell as well as to quantify the differences
Kauffman, 1969, 1993) (see Fig.6). So defined ran- in expression levels between two cells (Brown and Bot-
dom Boolean networks (RBNs) can operate in different stein, 1999). Also, novel technologies made it possible to
regimes, depending on e.g. their averaged connectiv- perform gene knock-out experiments in which individual
ity. The ordered (low connectivity) is characterized by genes are silenced. Combining these two techniques it
a small set of stable attractors which are largely robust became possible to perform “damage spreading” exper-
to perturbations, while in the disordered phase (large iments (Derrida and Pomeau, 1986; Rohlf et al., 2007)
connectivity) perturbations rapidly propagate and pro- in which the difference in gene-expression levels between
liferate hindering the existence of truly stable states. perturbed and unperturbed cells in quantitatively moni-
Separating these two phases there is a critical point at tored. The statistics of the sizes of “avalanches” caused
which perturbations propagate marginally (Derrida and by single-gene knock-out experiments has been analyzed
Pomeau, 1986). More complex models, with e.g. stochas- using empirical gene-expression data from the yeast (Sac-
ticity and/or continuous levels of activity, exhibit also charomyces cerevisae) (Hughes et al., 2000), with the
such two alternative phases (Rohlf and Bornholdt, 2002). conclusion that the best correspondence between empir-
Kauffman conjectured that models operating at their ical results and (RBN) model predictions is obtained for
critical point might provide the best possible representa- the model operating close to its critical point (Rämö
tion of real gene regulatory networks (Kauffman, 1996, et al., 2006; Serra et al., 2007, 2004). However, as a
word of caution, let us stress that, given that expression
levels are noisy, it is necessary to introduce a threshold
34 expression level to declare when a gene is differentially
Alternatively, it is also standard to use continuous approaches,
based on reaction-kinetics differential equations (Furusawa and
Kaneko, 2012b; Kaneko and Ikegami, 1992). See De Jong (2002)
for a review.
35 More realistic network architectures including, for example, node 36 This is, beyond purely self-organization mechanisms, such as
heterogeneity and modularity have also been considered (Aldana, SOC, also exhibited by inanimate systems (Halley and Winkler,
2003; Poblanno-Balp and Gershenson, 2011). 2008).
 19

expressed in the two cells. A caveat is that it is not clear Furusawa and Kaneko (2012a) analyzed an abstract dy-
what the influence of thresholding is on the definition of namical (not Boolean) model describing a cellular net-
avalanches and on their associated (size and duration) work (the network formed by the set of molecules which
probabilities. Thus, even if results are promising, more interact with others to give products within the cell)
precise analyses are still much needed. with nutrient uptake, and showed that the Zipf’s law
is a universal feature of self-regulated cells optimizing
their growth rate in nutrient-rich environments. In this
3. Networks from DNA microarray data setting, cells seem to adapt to criticality to achieve the
maximal capacity to assimilate and use nutrients for re-
In a parallel endeavor, empirical analyses of hundreds cursive formation of other products (Erez et al., 2017;
of DNA microarray experiments allowed researchers to Furusawa and Kaneko, 2012a; Hanel et al., 2010; Kaneko,
infer the whole network of regulatory interactions among 2006; Stokić et al., 2008).
genes, i.e. who regulates whom in a given cell (Filkov,
2005). It has been consistently found that the in-degree
distribution of such regulatory networks is Poissonian, C. Collective behavior of cells
while the out-degree distribution is scale-free (see Al-
dana (2003), Drossel and Greil (2009), and refs. therein). We have discussed the possibility of criticality within
Performing damage-spreading computational analyses of the internal networks of individual cells. But, also ensem-
dynamical RBN models running on top of such networks bles of cells –both in social unicellular organisms and in
–with the empirically-determined architecture of diverse multicellular ones– may exhibit signatures of scale invari-
organisms such as S. cerevisiae and E. coli (Albert and ance and criticality in their collective behavior (Nadell
Othmer, 2003)– it was concluded that they all are indeed et al., 2013). For instance, in aggregates of the (uni-
very close to criticality, in the sense of marginal propaga- cellular) amoeba Dictyostelium discoideum (the “slime
tion of perturbations (Aldana et al., 2007; Balleza et al., mold”), local cell-cell coupling via secreted chemicals
2008; Chowdhury et al., 2010; Darabos et al., 2009). may be tuned to a critical value (De Palo et al., 2017),
Alternatively to inferring the architecture of the under- resulting in emergent long-range communication and en-
lying network of interactions –which is a difficult problem hanced sensitivity. In the following we discuss some other
(Filkov, 2005)– algorithmic information theory has also examples in multicellular organisms.
been employed to asses the dynamical state directly from
empirical measurements from DNA-microarray data in
a model-free way (Kauffman et al., 2003; Shmulevich
1. Stem cell pluripotency
et al., 2005).37 Analyses of empirical data (for, e.g. the
macrophage) following these protocols produced results
Large diversity in gene-expression levels has been ob-
compatible with marginality in the information flow, i.e.
served in multipotent stem-cell populations of multi-
with critical dynamics (Nykter et al., 2008a)38 .
cellular organisms (Goodell et al., 2015). Multipotent
(hematopoietic) stem cells can differentiate onto either
4. Zipf’s law in gene-expression data erythroid or myeloid blood cells –with rather different
functionalities– depending on the expression level of a
Inspection of gene expression databases of diverse gene called Sca1 (Ridden et al., 2015). The empirically
unicellular organisms (such as yeast) reveals that the measured distribution of expression levels of Sca1 within
(continuous-valued) abundances of expressed genes are a population of stem cells turns out to be very broad
distributed as a power-law with exponent close to −1, and with various local maxima. This has been modeled
obeying the Zipf’s law (Furusawa and Kaneko, 2003)39 . as a stochastic process, and it has been found that the
model can exhibit two different regimes: either a stable
low-Sca1 or a stable high-Sca1 regime. Separating these
37 For example, one such method relies on computing estimators of phases there is a line of discontinuous transitions (with
the Kolmogorov complexity (Ming and Vitányi, 2014) of sets of bistability) finishing at a critical point. Remarkably, the
gene-expression time series in diverse microarrays, and comput- best fit to gene-expression empirical data is obtained fix-
ing how the difference between the information content of any
ing model parameters close to criticality, where maximal
two system states (Benedetto et al., 2002) changes over time.
38 Similar analyses for Eukaryotic cells gave results compatible with variability of the two possible phenotypes is obtained.
the dynamics being either ordered or critical but not disordered Thus, it seems that by adjusting near to criticality, the
(Shmulevich et al., 2005). stem-cell population is prompt to react and produce ei-
39 Indeed, clonal populations of unicellular organisms such as
viruses or bacteria often exhibit phenotypic diversity, which con-
ther erythroid or myeloid cells in response to changing
stitutes a sort of “bet-hedging” strategy to cope with unpre- demands in an optimal way (Ridden et al., 2015). Similar
dictable environmental changes (Kussell and Leibler, 2005; Veen- ideas have been discussed in the more general context of
ing et al., 2008; Wolf et al., 2005). collective cell decision making (Garcia-Ojalvo and Arias,
 20

2012; Halley et al., 2009; Lopez-Garcia et al., 2010; Ya- gene dynamics operates at criticality. This suggests that
maguchi et al., 2017), as well as in cancer progression criticality helps defining patterns without a characteris-
(Tsuchiya et al., 2015, 2016). tic scale, as required for expanding/developing systems
(see Bose and Pal (2017) and Pal et al. (2014) for a ped-
agogical discussion of these ideas in the general context
2. Morphogenesis I: Hydra regeneration of cell differentiation).

Morphogenesis is the biological process at the basis

of the development of multicellular organisms. It is D. Collective motion
achieved by a precise control of cell growth, prolifera-
tion, and differentiation. As first suggested in the sem- Collective motion of large groups of individuals is
inal work of Turing (1952), morphogenesis involves the a phenomenon observed in a variety of social organ-
creation of self-organized patterns and shapes in the em- isms such as flocks of birds, fish schools, insect swarms,
bryo. A prototypical organism studied in this context is herds of mammals, human crowds (Berdahl et al., 2013;
the Hydra polyp, which has a remarkable regeneration Bonabeau et al., 1999; Couzin and Krause, 2003; Krause
power, as an entire new individual can be spontaneously and Ruxton, 2002; Sumpter, 2010) and also, at smaller
re-assembled even from dissociated cells from an adult scales, in bacterial colonies (Nadell et al., 2013; Peru-
individual (Bosch, 2007). Along such a regeneration pro- ani et al., 2012; Ramaswamy, 2010; Sokolov et al., 2007),
cess, first a cell bilayer is formed with a spherical (hollow) and groups of cells in general (Méhes and Vicsek, 2014).
shape. How does the spherical symmetry break down Flocking, schooling, swarming, milling, and herding con-
to form a well-defined foot-head axis in adults? Dur- stitute outstanding examples of collective phases where
ing this process, there is a gene called ks1 that becomes simple interactions between individuals give rise to fas-
progressively expressed and that can be transferred to cinating emergent behavior at larger scales, even in the
neighboring cells. It was empirically found that right at absence of central coordination. Flock of birds and fish
the time when the spherical symmetry is broken, the size schools behave as plastic entities able to exhibit coherent
distribution of ks1 -rich domains of cells across the sphere motion, including e.g. rapid escape manoeuvres when
becomes scale-free and that a spanning cluster emerges, attacked by predators, which confers obvious fitness ad-
much as in a percolation phase transition (Gamba et al., vantages to the group as a whole (Couzin, 2007, 2009).
2012; Soriano et al., 2006). Thus a critical percolation- Such collective phenomena have attracted the atten-
like state with collective fluctuations of gene-expression tion of statistical physicists who have tackled the problem
levels is exploited to break the symmetry, defining a head- employing: (i) individual-based models of self-propelled
tail axis (Soriano et al., 2006, 2009). particles such as the one in Vicsek et al. (1995) which
models collective motion by assuming that an individual
in a group essentially follows the trajectory of its neigh-
bors, with some deviations treated as noise,40 and (ii)
3. Morphogenesis II: Gap genes in Drosophila
continuum (hydrodynamic) theories, more amenable to
theoretical analysis (Toner and Tu, 1995; Toner et al.,
A set of so-called “gap” genes is responsible for the
2005). These approaches have in common the existence
emergence of spatial patterns of gene-expression, that
of phase transitions between phases of coherent and in-
are at the origin of the formation of different segments
coherent motion. For example, in the Vicsek model, a
along the head-tail axis in the development of the fruit-fly
phase transition from an ordered “flocking phase” to a
(Drosophyla) embryo. Empirical scrutiny of the expres-
disordered “swarming phase” occurs when the density of
sion levels of gap genes along the head-tail axis revealed
individuals goes below a given threshold or, for a fixed
a number of remarkable features that include: slow dy-
density, when the level of stochasticity is large. This is
namics, correlations of expression-level fluctuations over
consistent with experimental findings; e.g. Buhl et al.
large distances, non-Gaussianity in the distribution of
(2006) investigated the social behavior of locusts and re-
such fluctuations, etc. Krotov et al. (2014) proposed a
ported on a density-driven phase transition from disor-
simple dynamical model in which the process is controlled
dered movement of individuals to highly aligned collec-
by only two mutually repressing gap genes. Assuming
tive motion as density is increased (Dyson et al., 2015).
that a fixed point exists, and performing a linear stability
At a conceptual level, marginally coordinated (critical)
analysis to describe the fate of fluctuations, one readily
motion can be hypothesized to constitute an optimal
finds that there is an instability point as the interaction
strength between the two genes is varied. Krotov et al.
(2014) argued that if the dynamics of the coupled system
is tuned to operate at such an instability point, then it 40 See Chaté et al. (2008); Chaté et al. (2008); Ginelli (2016);
constitutes an excellent qualitative description of all the and Grégoire and Chaté (2004) for detailed statistical-mechanics
above-mentioned empirical findings, implying that the analyses of Vicsek models and variants of it.
 21

tradeoff to deal with conflicting imperatives such as e.g. Furthermore, performed

(i) the need to behave cohesively as a unique entity and
(ii) being highly responsive to information from transito-
rily well-informed individuals (Couzin et al., 2011, 2005; 2. Insect swarms
De Vincenzo et al., 2017; Vanni et al., 2011). Similar
dichotomies exist in the empirical examples we discuss Extensive field analyses of insect (midge) swarms –
now. which, unlike birds traveling in a flock, hover around
a spot on the ground– have also been performed (At-
tanasi et al., 2014b). By employing finite-size analyses
1. Flocks of birds of the data, Attanasi et al. showed that both the corre-
lation length and the susceptibility grow with the swarm
On the empirical side, pioneering work by Cavagna, size, while the spacing between midges decreases. More-
Giardina and collaborators (Ballerini et al., 2008; Cav- over, such changes with swarm size occur as in the Vicsek
agna et al., 2010) on starling flocks allowed to record model for finite-size systems sitting near the maximally
individual trajectories (with purposely devised tracking correlated point of their transition region at each finite
technology). By analyzing the fluctuations in individ- size42 . Thus, midges obey spatio-temporal scaling and,
ual velocity with respect to the average velocity of the to achieve it, they seem to regulate their average dis-
group, these studies provided remarkable evidence that tance or density (which acts as a control parameter) so
long-range scale-invariant correlations may be a general as to function close to criticality (Attanasi et al., 2014b;
feature in systems exhibiting collective motion. In par- Cavagna et al., 2017; Chaté and Muñoz, 2014). On the
ticular, experimentally measured correlations –both in contrary, laboratory experiments of small swarms do not
orientation and speed fluctuations– were found to grow indicate critical behavior, which may signal that it only
with flock size in large flocks, suggesting that a correla- arises in “natural conditions” or for larger sizes (Kelley
tion length much larger than the interaction range, could and Ouellette, 2013; Puckett and Ouellette, 2014).
be a common trait of self-organized groups needing to
achieve large-scale coordination (Cavagna et al., 2010).
Let us note that the scale-free correlations in the orien- 3. Mammal herds
tation might be attributed to the broken continuous (ro-
tational) symmetry, which as discussed in Appendix A Social herbivores (Merino sheep) have also been stud-
leads to generic scale-invariance. However, the presence ied in well-controlled environments, revealing the exis-
of scale-free correlations in the (scalar) speed fluctuations tence of two conflicting needs: (i) the protection from
cannot be explained in this way, suggesting that the flock predators offered by being part of large cohesive group
might be tuned to a critical point with maximal suscepti- and (ii) the exploration of foraging space by wandering
bility.41 Furthermore, experiments on starling flocks also individuals. Sheep resolve this conflict by alternating a
allow to measure how the information of the turning of slow foraging phase, during which the group spreads out,
one individual propagates across the flock, revealing that with fast packing events triggered by individual behav-
this occurs in a very fast and efficient way, which can be ioral shifts, leading to intermittent collective dynamics
taken as a direct evidence of the existence of scale-free with packing events of all accessible scales, i.e. a “near
correlations in flocks (Attanasi et al., 2014a). critical” state (Ginelli et al., 2015).
Bialek et al. (2012) applied a maximum entropy
method to construct a statistical model consistent with
4. Social-insect foraging strategies
the empirically measured correlations (see Appendix B).
They concluded that the interaction strength and the
Studies of ant foraging strategies have been recently
number of interacting neighbors do not change with flock
performed (Beekman et al., 2001; Bhattacharya and Vic-
size in the probabilistic model; and, more importantly,
sek, 2014; Gallotti and Chialvo, 2017; Li et al., 2014; Lo-
the model was able to reproduce scale-free correlations
engarov and Tereshko, 2008; Solé, 2011). For ant colonies
in velocity fluctuations. It was observed (i.e. inferred
to achieve an efficient foraging strategy, a tradeoff needs
from data) that this occurs as a result of the effective
to be reached between exploratory behavior of some indi-
model’s operating close to its critical point (Bialek et al.,
viduals and predominant compliance with the rules (Fein-
2014; Mora and Bialek, 2011; Mora et al., 2016).
erman and Korman, 2017). It has been found by using

41 Similar results have been obtained for aggregates of a social

amoeba (slime mold) (De Palo et al., 2017), as well as for colonies 42 The Vicsek model exhibits, at least for not-too-large sizes, a wide
of the bacteria Bacillus subtilis in the experimental setup of Chen regime where correlations peak at the transition and finite-size-
et al. (2012) (but not in that of Sokolov et al. (2007), which re- scaling holds (Baglietto et al., 2012; Chaté et al., 2008; Grégoire
veals only short-range correlations). and Chaté, 2004; Vicsek et al., 1995).
 22

a combination of experiments and theory that some ant can be very useful). Even if both approaches have deep
groups optimize their overall performance by sitting at interconnections, here we chose to focus mostly on the
the edge of a phase transition between random explo- dynamical one.
ration and gregarious strategies, thus resulting in effec- Synthesizing (maybe oversynthesizing), one could ar-
tive criticality. This entails efficient group-level process- gue that the ultimate reason why putative criticality
ing of information emerging out of an optimal amplifica- appears so often in the scrutiny of complex biologi-
tion of transient individual information (Gelblum et al., cal systems is that it constitutes the simplest phys-
2015). Similar ideas are being presently explored for the ical/dynamical mechanism generating complex spatio-
design of artificial systems, i.e. in swarm robotics (Beni, temporal patterns spanning through many different
2004; Erskine and Herrmann, 2014). scales, that are all correlated, implying system-wide co-
herence and large responses to perturbations. From this
To further enrich this bird’s-eye view over different as- perspective, critical-like behavior –and the nested hi-
pects of criticality in biological systems, a miscellaneous erarchy of spatio-temporal structures it spontaneously
collection of other examples from the literature is pre- generates– can be identified as a scaffold upon which
sented in Appendix C. (multiscale) biological systems may build up further com-
plexity.
Statistical physics teaches us that under some
V. DISCUSSION circumstances–including e.g. systems with some form of
heterogeneity (relevant for e.g. the study of brain net-
The hypothesis that living systems may operate in works), or in systems with continuous symmetries (rel-
the vicinity of critical points, with concomitant scale- evant in collective motion) the standard scenario of a
invariance, has long inspired scientists. From a theoreti- unique critical point separating diverse phases needs to
cal viewpoint this conjecture is certainly appealing, as it be replaced by that of extended critical-like regions (such
suggests an overarching mechanism exploited by biolog- as e.g. Griffiths phases discussed for the overall brain dy-
ical systems to derive important functional benefits es- namics) where some form of scale invariance emerges in
sential in their strive to survive and proliferate. The list a generic way. In such cases, it might suffice for biolog-
of possible critical features susceptible to be harnessed ical systems to operate in such phases without the need
by living systems include the unparalleled sensitivity to to invoke precise tuning to the edge of a phase transi-
stimuli, the existence of huge dynamical repertoires, max- tion to obtain functional benefits stemming from spatio-
imal transmission and storage of information, as well as temporal scale invariance.
optimal computational capabilities, among others. When From the experimental viewpoint, along the presen-
living systems are interpreted as information-processing tation we tried to summarize existing empirical pieces
devices –needing to operate robustly but, at the same of evidence for each of the discussed examples, stress-
time, having to cope with diverse environmental changes– ing possible drawbacks and interpretative problems, and
the virtues of critical behavior are undeniable. Criticality underlining criticisms raised in the literature. Readers
represents a simple strategy to achieve a balance between will extract their own conclusions on whether each of the
robustness (order) and flexibility (disorder) needed to de- examples is sufficiently convincing or not. Our general
rive functionality. Similar tradeoffs, as discussed along impression is that, in most of the cases, larger systems,
the paper (e.g. between stability and evolvability) un- more accurate measurements, and less ambiguous anal-
derline the potential of operating at the edge between yses would be needed to further confirm or disprove the
different types of order. existence of an underlying dynamical critical process. For
Throughout this essay we focused dynamical aspects most of the leading examples (i.e. neural systems, ge-
of criticality, meaning that in most of the discussed netic regulatory networks, and collective motion), our
examples it is assumed –either directly or indirectly– opinion is that, as of today, there is not a fully convinc-
that there is an underlying dynamical process at work, ing proof, where experimental evidence and mathemati-
and that such a process –susceptible to be mathemati- cal theory/modeling match perfectly; i.e. we still do not
cally modeled– operates in the vicinity of a continuous have a “smoking gun”. Still, the existing collection of
phase transition, at the borderline between two alterna- remarkable pieces of evidence is extremely appealing and
tive regimes. Such a dynamical perspective is essentially hard to neglect.
different from the purely statistical (or static) one, as de- Two important aspects should be considered in future
scribed e.g. in Mora and Bialek (2011). In this latter, the empirical analyses to make solid progress. One is that,
focus is on analyzing the statistics of existing patterns; it given that biological systems are finite, they cannot be
has the great advantage that it harnesses existing high- truly critical in the precise sense of statistical physics;
quality empirical datasets. On the other hand, it disre- thus it is important to perform, whenever possible, finite-
gards the possible dynamical generative mechanisms be- size analyses to prove the existence of scale-invariance
hind them and focused on an effective description (which within the experimentally accessible ranges. A second
 23

aspect is that the two alternative phases that the al- ACKNOWLEDGMENTS
leged criticality separates should be clearly identified in
each case. From this view, we find particularly appeal- I am deeply indebted to the collaborators with whom
ing pieces of evidence (e.g. in neuroscience) in which, I had the pleasure and privilege of studying some of the
by experimentally inducing alterations to standard con- topics discussed here; among them: G. Grinstein, P.L.
ditions, deviations from criticality are measured in oth- Garrido, J.A. Bonachela, A. Maritan, J. Marro, F. de
erwise critical-looking systems. los Santos, P.I. Hurtado, J.J. Torres, J. Cortés. S. John-
A general criticism can be raised to some of the anal- son, R. Burioni, S. Suweis, A. Gabrielli, C. Castellano,
yses discussed along this work, specifically, to those in S. Pigolotti, M. Cencini, A. Vespignani, R. Dickman, S.
which the evidence relies on the existence of a theoret- Zapperi, Y. Tu, D. Plenz, as well as to H. Chaté for his
ical model that provides, when tuned close to its criti- collaboration and encouragement in the early stages of
cal point, the best possible fit to empirical observations. this project. I am especially thankful to P. Moretti, J.
The criticism is that, if feature-rich empirical data with Hidalgo, P. Villegas, L. Seoane, J. Soriano, S. di Santo
structures spanning over broadly diverse scales are con- and V. Buendı́a, for extremely useful comments on early
sidered, then it seems almost a tautology to conclude versions of the manuscript. I acknowledge the Spanish-
that the best possible representation of them is obtained MINECO, grant FIS2017-84256-P (FEDER funds) for fi-
by fitting the proposed dynamical model to operate close nancial support.
its critical point, as this is typically the only region in
parameter space where complex (feature-rich) patterns,
with many scales, are generated. In contrast, from an
opposite perspective, if actual biological data are struc- APPENDIX A: GENERIC SCALE INVARIANCE
tured across many scales, it does not seem too far fetched
to assume –applying the Occam’s razor– that a general There are situations in which spatio-temporal scaling
common mechanism may underlie the emergence of such may emerge without the need of parameter fine tuning
a hierarchy of scales, and the main candidate mechanism (Grinstein, 1991). (i) A well-known example is the break-
for this consists in operating at the edge of a continuous ing of a continuous symmetry in low-dimensional sys-
phase transition, i.e. being close to criticality. Thus, we tems, as it happens e.g. in some models of magnetism
are confronted with a (epistemological) dichotomy: Is the in which each spin can point in any arbitrary direction
putative criticality of living systems just a reflection of in a plane (Binney et al., 1993). These systems, instead
the limitation of our models which can possibly resemble of the usual ordered phase at low temperature, exhibit
large levels of “complexity” only at criticality? or, on a broad “quasi-ordered” phase characterized by local or-
the contrary, is criticality actually a common organizing der together with generic power-law decaying correlations
principle at the roots of the generation of many levels of (Grinstein, 1991). This type of ordering is relevant for
organization required for complex biological behavior to bacterial-colony patterns (Ramos et al., 2008) as well as
emerge? Providing a satisfactory answer to these ques- in the analysis of collective motion (see Sect. IV-C).
tions is a problem of outmost importance to advance in (ii) Generic scale invariance can also emerge in the
the theoretical understanding and modeling of complex presence of structural disorder or heterogeneity. In sta-
living systems. tistical physics, one refers to “quenched disorder” as
Even if diverse biological systems were finally proved to the form of spatial-dependent heterogeneity which is
be genuinely critical, some researchers might still retain intrinsic to the microscopic components and remains
this conclusion as largely uninformative or even irrele- frozen in time, reflecting structural heterogeneities.
vant. It could be asked: “so what?”. What practical Quenched disorder can alter the nature of phase tran-
implications could be derived from such a knowledge? sitions (Villa Martı́n et al., 2014; Vojta, 2006) and can
In our opinion, the design of strategies to control neu- also induce novel phases absent in homogeneous systems.
ral/genetic networks –especially those aiming at resolv- For instance, in the contact process, quenched disor-
ing pathologies– based on notions of criticality, the con- der can be implemented by considering a lattice with
struction of algorithms of artificial intelligence exploiting some missing links, a more complex (disordered) net-
scale-invariance at different layers, or the application of work of connections, and/or a node-dependent propa-
ideas of collective motion/intelligence to the design of e.g. gation rate λ. In all these cases, a novel phase called
swarms of robots, could constitute important avenues to a Griffiths phase –characterized by critical-like features
provide constructive answers to the above question. appearing all across the phase and not just at a unique
Novel advances, both at the experimental and theo- point– emerges (Cafiero et al., 1998; Moreira and Dick-
retical sides, will help elucidating what is the actual role man, 1996; Muñoz et al., 2010; Vojta, 2006).
played by criticality and scale invariance in biological sys- (iii) Another mechanism that produces a type of
tems; meanwhile the mere possibility remains as inspiring generic scale-invariance relevant in biological systems
as ever and, definitely, worth pursuing. (see e.g. Sect. IV-A7) is “neutral dynamics”. Neutral
 24

theories play a key role in population genetics (Kimura, in the inferred model Eq.(5). Clearly, varying β a rela-
1984), population ecology (Azaele et al., 2016; Hubbell, tive change of the weights assigned to different configura-
2001), epidemics (Pinto and Munoz, 2011), etc. They tions is produced. In this way one generates a family of
have in common the fact that differences among possi- β-dependent probability distributions, P (s|βg), interpo-
ble coexisting “species” (let them be alleles of a gene, lating between the low and high temperature phases45 .
types of trees, bacterial strains,...) are neglected. In At some intermediate value, βc , there is a critical point
other words, all “species” are dynamically equivalent or (as identified e.g. by a peak in the susceptibility or the
neutral (Blythe and McKane, 2007; Dornic et al., 2001; specific heat). It has been found that diverse inference
Liggett, 2004). It has been recently shown that in a problems (from retinal populations (Schneidman et al.,
(“multispecies”) contact process that includes different 2006; Tkačik et al., 2014, 2013, 2015) to flocks of birds
types/species of activity, if a new species –neutral to the (Bialek et al., 2012), and the immune system (Mora et al.,
exiting ones– is introduced, it experiences a stochastic 2010)) produce models in which βc ≈ 1, –or converges to
process in which its total population does not have a net 1 as the system size is enlarged– i.e. that inferred models
tendency to either grow or shrink. This generates gener- appear to be close to the very critical point of the under-
ically scale-free avalanches of the focal species unfolding lying Ising-like problem (see Mora and Bialek (2011) for
in a sea of activity of the other species, without the need a clear and pedagogical discussion of these issues).
to invoque criticality (Martinello et al., 2017). Let us also mention that there is an ongoing debate
on the interpretation of these results. In particular, it
has been shown that signatures of criticality may emerge
APPENDIX B: PROBABILISTIC MODELS AND naturally in inferred models if there is a marginalization
STATISTICAL CRITICALITY
over non-observed variables, such as e.g. correlated ex-
ternal inputs, even without the need for direct interac-
Probabilistic models can be constructed such that they
tions among units Aitchison et al. (2016) and Schwab
match the statistics of observed empirical data (Rieke
et al. (2014). More in general, Marsili and collaborators
et al., 1995). Without loss of generality, an empirically
pointed out that the alleged criticality of such models
observed pattern at a given time window can be codified
can be a rather general consequence of the inference pro-
as a sequence of binary variables of length N : si = 0, 1
cedure, meaning that inferred models fitting real-world
for i = 1, 2, ...N . Denoting P (s) the (unknown) proba-
(“feature-rich” or “informative” ) data do, most likely,
bility of finding the system in the state s = (s1 , s2 , ...sN )
look critical when an effective probabilistic model is con-
it is possible to approximate it by a distribution function
structed (Haimovici and Marsili, 2015; Marsili et al.,
with the constraint that it reproduces the empirically-
2013; Mastromatteo and Marsili, 2011; Tyrcha et al.,
measured mean values hsi i for all i, as well as the covari-
2013). We shall not delve further into the controversy
ances hsi sj i for all i and j. Imposing a maximum entropy
about the meaning and significance of this type of purely
principle (i.e. selecting the model with the smallest num-
statistical approaches to criticality (see also the Discus-
ber of assumptions43 ) it is straightforward to derive the
sion section).
explicit form of the optimal model
 
1 X X
APPENDIX C: ADAPTATION AND EVOLUTION
P (s) = exp  Jij si sj + hi si  (5)
Z i<j i
TOWARDS CRITICALITY

where Z ensures normalization and which coincides with To shed light onto the general problem of how
the Boltzmann equilibrium distribution of the Ising-like information-processing (living) systems tune themselves
model, and where the free parameters hi and Jij need to operate near critical points, Goudarzi et al. (2012)
to be fitted, so that the imposed constraints are satisfied considered an ensemble of individuals or “agents”, each
(Ackley et al., 1985).44 represented as an internal RBN, including some input
Bialek and coworkers introduced an efective parameter nodes (able to read information from the environment)
β –much as an inverse temperature in equilibrium statis- and some readout nodes (providing outputs/responses).
tical mechanics– multiplying each estimated parameter Such agents evolve though a genetic algorithm (Goldberg
and Holland, 1988) that allocates larger “fitness” values
to agents that perform better a series of computational
43 In information theory, the entropy of a probability distribution tasks (each one consisting in assigning a given output to
quantifies the ignorance about the variable; thus, making no as-
sumptions about the distribution is equivalent to maximizing the
entropy (Banavar et al., 2010a; Cover and Thomas, 1991).
44 Obtaining the optimal parameter set –i.e. inferring effective in-
teractions from correlations– is a computationally costly task, 45 At β = 0 (infinite temperature) all configurations are equiprob-
usually referred as “inverse Ising problem” (Aurell and Ekeberg, able, while in the opposite limit all the weight concentrates on
2012; Cocco et al., 2009; Schneidman et al., 2006). the most likely (fully ordered) configuration.
 25

each specific input), which are alternated in time. The variability compatible with genotypic robustness46 .
conclusion is that agents converge to a state close to crit- Physiological rhythms. The presence of tempo-
icality; i.e. critical dynamics emerge as the optimal solu- ral scale-invariance in physiological rhythms of healthy
tion under the combined selective pressures of having to subjects, as well as its break-down in abnormal condi-
learn different tasks and being able to readily shift among tions, have been long explored (Bassingthwaighte et al.,
them, following changes in the tasks. Instead, in the pres- 1994; Goldberger et al., 2002; Losa, 1995). In particular,
ence of noise, optimal agents tend to be slightly subcrit- to mention one example, a specific connection between
ical, rather than critical, thus compensating for extrinsic the complex fluctuations of human heart-rate variabil-
sources of variability (Villegas et al., 2016). In a similar ity and criticality has been put forward (Ivanov, 2007;
approach, Hidalgo et al. (2014) showed that communities Ivanov et al., 1999; Kiyono et al., 2004, 2005). In the
of similar adaptive agents, whose task is to communicate related context of blood-pressure regulation, vaso-vagal
with each other (inferring their respective internal states) syncopes have been identified as large “avalanches” in a
in an efficient way, converge to quasi-critical states. This self-organized cardiovascular regulatory system poised at
result constitutes a possible parsimonious explanation for criticality (Fortrat and Gharib, 2016). In general, such a
the emergence of critical-like behavior in groups of in- regulation to scale-free behavior seems to impart health
dividuals coordinating themselves as a collective entity advantages, including system integrity and adaptability
(Hidalgo et al., 2016) (see also Iliopoulos et al. (2010)). (Goldberger et al., 2002).
Miscellanea. Criticality has also been claimed to
play a relevant role in the immune system (Burgos and
APPENDIX D: OTHER PUTATIVELY CRITICAL LIVING Moreno-Tovar, 1996; Mora et al., 2010), cancer and
SYSTEMS carcenogenesis (Davies et al., 2011; Rosenfeld, 2013; Solé
and Deisboeck, 2004; Solé, 2003), proteins (Phillips, 2009;
Here we briefly discuss, a sample of other biological Tang et al., 2017), mitochondria (Aon et al., 2004; Zam-
systems for which empirical evidences of criticality exist. poni et al., 2018), etc. Also, quantum criticality and its
Cell membranes. Cell membranes are not just rigid relevance for the origin of life at the microscopic scale
impenetrable walls separating the interior of cells from has been the subject of a recent proposal (Vattay et al.,
the outside environment; they regulate the kind, direc- 2015). Finally, let us mention that ecosystems as a whole
tion, and amount of substances that can pass across have been studied –from a macroevolutionary viewpoint–
them. Cell membranes are permeable only at some lo- as dynamical structures lying at the edge of instability
cations and, for this, their local composition needs to be (Adami, 1995; Bak and Sneppen, 1993; Biroli et al., 2017;
heterogeneous (Cicuta, 2013; Hyman and Simons, 2012; Sneppen et al., 1995; Solé et al., 2002a, 1999; Suweis
Lee et al., 2013). There is compelling empirical evidence et al., 2013), illustrating that the ideas discussed here
that the mixture of lipids that constitute the skeleton can be extended to larger scales in the hierarchy of bio-
of cell membranes operates very close to the (de-mixing) logical complexity.
phase transition at which their different components seg-
regate (Cicuta, 2013; Ehrig et al., 2011; Honerkamp-
Smith et al., 2008; Veatch et al., 2008, 2007). In this REFERENCES
way, composition fluctuations are extremely large, en-
Abbott, L., and C. van Vreeswijk (1993), Phys. Rev. E 48 (2),
abling very diverse structural domains to appear, thus 1483.
providing the membrane with a large spectrum of pos- Acebrón, J. A., L. L. Bonilla, C. J. Pérez-Vicente, F. Ritort,
sible local structures, at which different processes may and R. Spigler (2005), Rev. Mod. Phys. 77, 137.
occur, and entailing a rich repertoire of functionalities. Ackley, D., G. Hinton, and T. Sejnowski (1985), Cognitive
RNA viruses. RNA viruses are believed to repli- Science 9 (1), 147.
cate at the edge of an “error catastrophe”. If the error Adami, C. (1995), Phys. Lett. A 203, 29.
Aitchison, L., N. Corradi, and P. Latham (2016), PLoS
rates for copying the viral genome were very small RNA Comp. Biol. 12 (12), e1005110.
viruses would have little variability, hindering adaptation Akaike, H. (1969), Ann. Inst. of Stat. Math. 21 (1), 243.
and evolution. Instead, if they were too large then the Albert, R. (2004), in Complex networks (Springer) pp. 459–
fidelity of the replication machinery would be compro- 481.
mised and it would not be possible to maintain important Albert, R., and A.-L. Barabási (2002), Rev. Mod. Phys. 74,
genetic elements nor the identity of the (quasi)species it- 47.
self (Eigen et al., 1989; Eigen and Schuster, 1979). It was Albert, R., and H. Othmer (2003), [Link]. 223 (1), 1.
conjectured (Drake and Holland, 1999; Eigen, 2002; Solé
et al., 1999, 1996) and has been (partially) verified in
recent experiments (Crotty et al., 2001; Hart and Fergu- 46 Error catastrophe has been considered for treatment of viral in-
son, 2015) that RNA viruses might operate right at the fections employing drugs that push the error rate beyond this
edge of the catastrophe, providing them with maximal threshold; see Summers and Litwin (2006) for a critical review.
 26

Aldana, M. (2003), Physica D 185 (1), 45. Bak, P., C. Tang, and K. Wiesenfeld (1987), Phys. Rev. Lett.
Aldana, M., E. Balleza, S. Kauffman, and O. Resendiz (2007), 59 (4), 381.
J. Theor. Biol. 245 (3), 433. Ballerini, M., N. Cabibbo, R. Candelier, A. Cavagna, E. Cis-
Allegrini, P., P. Paradisi, D. Menicucci, and A. Gemignani bani, I. Giardina, V. Lecomte, A. Orlandi, G. Parisi,
(2010), Front. Physiol. 1, 128. A. Procaccini, et al. (2008), Proc. Natl. Acad. Sci. USA.
Alon, U. (2003), Science 301 (5641), 1866. 105 (4), 1232.
Alon, U. (2006), An introduction to systems biology: design Balleza, E., E. R. Alvarez-Buylla, A. Chaos, S. Kauffman,
principles of biological circuits (CRC press, London). I. Shmulevich, and M. Aldana (2008), PLoS One 3 (6),
Alonso, L. M., A. Proekt, T. H. Schwartz, K. O. Pryor, G. A. e2456.
Cecchi, and M. O. Magnasco (2014), Front. Neural Circuits Banavar, J. R., T. J. Cooke, A. Rinaldo, and A. Maritan
8, 20. (2014), Proc. Natl. Acad. Sci. USA. 111 (9), 3332.
Alvarado, J., M. Sheinman, A. Sharma, F. C. MacKintosh, Banavar, J. R., A. Maritan, and A. Rinaldo (1999), Nature
and G. H. Koenderink (2013), Nat. Phys. 9 (9), 591. 399 (6732), 130.
Amari, S. (1972), IEEE Trans. Syst. Man. Cybern. 2, 643. Banavar, J. R., A. Maritan, and I. Volkov (2010a), J. of
Amit, D. (1992), Modeling brain function: The world of at- Phys.: Cond. Matt. 22 (6), 063101.
tractor neural networks (Cambridge University Press). Banavar, J. R., M. E. Moses, J. H. Brown, J. Damuth, A. Ri-
Amit, D., and N. Brunel (1997), Cerebral cortex 7 (3), 237. naldo, R. M. Sibly, and A. Maritan (2010b), Proc. Natl.
Anderson, P. W., et al. (1972), Science 177 (4047), 393. Acad. Sci. USA. 107 (36), 15816.
de Andrade Costa, A., M. Copelli, and O. Kinouchi (2015), Barabasi, A.-L. (2005), Nature 435, 207.
J. Stat. Mech. 2015 (6), P06004. Barabási, A.-L., and R. Albert (1999), Science 286 (5439),
Aon, M. A., S. Cortassa, and B. O’Rourke (2004), Proc. Nat. 509.
Acad. of Sci. USA 101 (13), 4447. Barnett, L., J. T. Lizier, M. Harré, A. K. Seth, and T. Bosso-
de Arcangelis, L. (2011), J. of Physics: Conf. Series 297 (1), maier (2013), Phys. Rev. Lett. 111 (17), 177203.
012001. Barral, J., and A. D. Reyes (2016), Nature Neurosci. 19 (12),
de Arcangelis, L. (2012), Eur. Phys. J. [Link]. 205 (1), 1690.
243. Barrat, A., M. Barthelemy, and A. Vespignani (2008), Dy-
de Arcangelis, L., and H. J. Herrmann (2010), Proc. Natl. namical processes on complex networks (Cambridge Univ.
Acad. Sci. USA. 107 (9), 3977. press).
de Arcangelis, L., and H. J. Herrmann (2012), Front. Physiol. Bassingthwaighte, J. B., L. S. Liebovitch, and B. J. West
3, 0062. (1994), in Fractal physiology (Springer) pp. 11–44.
de Arcangelis, L., F. Lombardi, and H. Herrmann (2014), J. Beckmann, C. F., M. DeLuca, J. T. Devlin, and S. M. Smith
Stat. Mech. 2014 (3), P03026. (2005), Philos. Trans. R. Soc. London, Ser. B 360 (1457),
de Arcangelis, L., C. Perrone-Capano, and H. J. Herrmann 1001.
(2006), Phys. Rev. Lett. 96, 028107. Bédard, C., H. Kröger, and A. Destexhe (2006), Phys. Rev.
Arenas, A., A. Dı́az-Guilera, J. Kurths, Y. Moreno, and Lett. 97 (11), 118102.
C. Zhou (2008), Phys. Rep. 469 (3), 93. Beekman, M., D. J. Sumpter, and F. L. Ratnieks (2001),
Arieli, A., A. Sterkin, A. Grinvald, and A. Aertsen (1996), Proc. Natl. Acad. Sci. USA. 98 (17), 9703.
Science 273 (5283), 1868. Beggs, J. M. (2008), Phil Trans R Soc A 366 (1864), 329.
Ashby, W. (1960), Design for a Brain. The origin of adaptive Beggs, J. M., and D. Plenz (2003), J. Neurosci. 23 (35),
behaviour (New York, Willey). 11167.
Attanasi, A., A. Cavagna, L. Del Castello, I. Giardina, T. S. Beggs, J. M., and D. Plenz (2004), J. Neurosci. 24 (22), 5216.
Grigera, A. Jelić, S. Melillo, L. Parisi, O. Pohl, E. Shen, Bellay, T., A. Klaus, S. Seshadri, and D. Plenz (2015), Elife
et al. (2014a), Nat. Phys. 10 (9), 691. 4, e07224.
Attanasi, A., A. Cavagna, L. Del Castello, I. Giardina, Benayoun, M., J. D. Cowan, W. Van Drongelen, and E. Wal-
S. Melillo, L. Parisi, O. Pohl, B. Rossaro, E. Shen, E. Sil- lace (2010), PLoS Comput. Biol. 6 (7), 13.
vestri, et al. (2014b), Phys. Rev. Lett. 113 (23), 238102. Benedetto, D., E. Caglioti, and V. Loreto (2002), Phys. Rev.
Aurell, E., and M. Ekeberg (2012), Phys. Rev. Lett. 108 (9), Lett. 88 (4), 048702.
090201. Benford, F. (1938), Proc. Am. Philos. Soc. , 551.
Azaele, S., S. Suweis, J. Grilli, I. Volkov, J. R. Banavar, and Beni, G. (2004), in International Workshop on Swarm
A. Maritan (2016), Rev. Mod. Phys. 88, 035003. Robotics (Springer) pp. 1–9.
Baek, S. K., S. Bernhardsson, and P. Minnhagen (2011), New Berdahl, A., C. J. Torney, C. C. Ioannou, J. J. Faria, and
J. Phys. 13 (4), 043004. I. D. Couzin (2013), Science 339 (6119), 574.
Baglietto, G., E. V. Albano, and J. Candia (2012), Interface Berg, H. C. (1993), Random walks in biology (Princeton Uni-
Focus 2 (6), 708. versity Press).
Bagnoli, F., F. Cecconi, A. Flammini, and A. Vespignani Bertschinger, N., and T. Natschlager (2004), Neural Comput.
(2003), EPL (Europhys. Lett.) 63 (4), 512. 16 (7), 1413.
Bak, P. (1996), How nature works: the science of self- Betzel, R. F., A. Griffa, A. Avena-Koenigsberger, J. Goñi, J.-
organized criticality (Copernicus, New York). P. Thiran, P. Hagmann, and O. Sporns (2013), Network
Bak, P., K. Chen, and C. Tang (1990), Phys. Lett. A 147 (5), Science 1, 353.
297. Bhattacharya, K., and T. Vicsek (2014), J. Royal Soc. Inter-
Bak, P., and D. R. Chialvo (2001), Phys. Rev. E 63, 031912. face 11 (100).
Bak, P., and K. Sneppen (1993), Phys. Rev. Lett. 71 (24), Bialek, W. (2012), Biophysics: searching for principles
4083. (Princeton University Press).
Bak, P., and C. Tang (1989), J. Geophys. Res 94 (15), 635. Bialek, W. (2018), Rep. Prog. Phys. 81 (1), 012601.
 27

Bialek, W., A. Cavagna, I. Giardina, T. Mora, O. Pohl, E. Sil- Carpenter, G. A., and S. Grossberg (2016), in Encyclopedia
vestri, M. Viale, and A. M. Walczak (2014), Proc. Natl. of Machine Learning and Data Mining (Springer) pp. 1–17.
Acad. Sci. USA. 111 (20), 7212. Cavagna, A., A. Cimarelli, I. Giardina, G. Parisi, R. Santa-
Bialek, W., A. Cavagna, I. Giardina, T. Mora, E. Silvestri, gati, F. Stefanini, and M. Viale (2010), Proc. Natl. Acad.
M. Viale, and A. M. Walczak (2012), Proc. Natl. Acad. Sci. USA. 107 (26), 11865.
Sci. USA. 109 (13), 4786. Cavagna, A., D. Conti, C. Creato, L. Del Castello, I. Giardina,
Bianconi, G., and M. Marsili (2004), Phys. Rev. E 70 (3), T. S. Grigera, S. Melillo, L. Parisi, and M. Viale (2017),
035105. Nat. Phys..
Bienenstock, E., and D. Lehmann (1998), Adv. Complex Sys- Cavagna, A., I. Giardina, A. Orlandi, G. Parisi, A. Procaccini,
tems 1 (04), 361. M. Viale, and V. Zdravkovic (2008), arXiv:0802.1668.
Binder, K. (1981), Z. Phys. B 43 (2), 119. Chaikin, P. M., and T. C. Lubensky (2000), Principles of
Binney, J., N. Dowrick, A. Fisher, and M. Newman (1993), condensed matter physics, Vol. 1 (Cambridge Univ Press).
The Theory of Critical Phenomena (Oxford University Chaté, H., F. Ginelli, G. Grégoire, F. Peruani, and F. Ray-
Press, Oxford). naud (2008), Europ. Phys. Jour. B 64 (3-4), 451.
Biroli, G., G. Bunin, and C. Cammarota (2017), Chaté, H., F. Ginelli, G. Grégoire, and F. Raynaud (2008),
arXiv:1710.03606. Phys. Rev. E 77, 046113.
Biswal, B., F. Zerrin Yetkin, V. M. Haughton, and J. S. Hyde Chaté, H., and M. Muñoz (2014), Physics 7, 120.
(1995), Magnetic resonance in medicine 34 (4), 537. Chen, X., X. Dong, A. Beer, H. Swinney, and H. Zhang
Blythe, R. A., and A. J. McKane (2007), J. Stat. Mech. (2012), Phys. Rev. Lett. 108 (14), 148101.
2007 (07), P07018. Chialvo, D. R. (2004), Physica A 340 (4), 756.
Boccaletti, S., V. Latora, Y. Moreno, M. Chavez, and D.-U. Chialvo, D. R. (2006), Nat. Phys. 2, 301.
Hwang (2006), Phys. Rep. 424 (4), 175. Chialvo, D. R. (2010), Nat. Phys. 6, 744.
Boedecker, J., O. Obst, J. T. Lizier, N. M. Mayer, and Chialvo, D. R., and P. Bak (1999), Neuroscience 90 (4), 1137.
M. Asada (2012), Theory in Biosci. 131 (3), 205. Chialvo, D. R., P. Balenzuela, and D. Fraiman (2008), AIP
Bonabeau, E., M. Dorigo, and G. Theraulaz (1999), Swarm Conference Proceedings 1028 (1), 28.
intelligence: from natural to artificial systems, 1 (Oxford Choe, Y., M. O. Magnasco, and A. Hudspeth (1998), Proc.
University Press). Natl. Acad. Sci. USA. 95 (26), 15321.
Bonachela, J., S. de Franciscis, J. Torres, and M. A. Muñoz Chowdhury, S., J. Lloyd-Price, O.-P. Smolander, W. C. Baici,
(2010), J. Stat. Mech. 2010 (02), P02015. T. R. Hughes, O. Yli-Harja, G. Chua, and A. S. Ribeiro
Bonachela, J. A., and M. A. Muñoz (2009), J. Stat. Mech. , (2010), BMC Systems Biology 4 (1), 1.
P09009. Christensen, K., Á. Corral, V. Frette, J. Feder, and
Bornholdt, S. (2005), Science 310 (5747), 449. T. Jøssang (1996), Phys. Rev. Lett. 77 (1), 107.
Bornholdt, S. (2008), J. R. Soc. Interface 5 (Suppl 1), S85. Christensen, K., and N. Moloney (2005), Complexity and
Bornholdt, S., and T. Rohlf (2000), Phys. Rev. Lett. 84 (26), criticality (Imperial College Press, London).
6114. Cicuta, P. (2013), in Encyclopedia of Biophysics (Springer)
Bosch, T. (2007), Devel. Biol. 303 (2), 421. pp. 387–390.
Bose, I., and M. Pal (2017), J. of Biosci. 42 (4), 683. Clauset, A., C. R. Shalizi, and M. E. J. Newman (2009),
Breakspear, M. (2017), Nature neuroscience 20 (3), 340. SIAM Rev. 51 (4), 661.
Brockmann, D., L. Hufnagel, and T. Geisel (2006), Nature Cocchi, L., L. L. Gollo, A. Zalesky, and M. Breakspear
439 (7075), 462. (2017), Progress in Neurobiology.
Brown, P. O., and D. Botstein (1999), Nature Genet. 21, 33. Cocco, S., S. Leibler, and R. Monasson (2009), Proc. Natl.
Brunel, N. (2000), J. Comput. Neurosci. 8, 183. Acad. Sci. USA. 106 (33), 14058.
Brunel, N., and V. Hakim (2008), Chaos 18 (1), 015113. Cohen, J. E., M. Xu, and W. S. Schuster (2012), Proc. Natl.
Buchanan, M. (2010), Networks in cell biology (Cambridge Acad. Sci. USA 109 (39), 15829.
University Press). Coniglio, A., L. de Arcangelis, H. J. Herrmann, and N. Jan
Buhl, J., D. J. Sumpter, I. D. Couzin, J. J. Hale, E. Despland, (1989), EPL (Europhys. Lett.) 8 (4), 315.
E. Miller, and S. J. Simpson (2006), Science 312 (5778), Corominas-Murtra, B., J. Goñi, R. V. Solé, and
1402. C. Rodrı́guez-Caso (2013), Proc. Natl. Acad. Sci. USA.
Burgos, J., and P. Moreno-Tovar (1996), Biosystems 39 (3), 110 (33), 13316.
227. Corral, A. (2004), Phys. Rev. Lett. 92 (10), 108501.
Bushdid, C., M. O. Magnasco, L. B. Vosshall, and A. Keller Corral, A., C. J. Pérez, A. Dı́az-Guilera, and A. Arenas
(2014), Science 343 (6177), 1370. (1995), Phys. Rev. Lett. 74 (1), 118.
Buzsaki, G. (2009), Rhythms of the Brain (Oxford University Couzin, I. (2007), Nature 445 (7129), 715.
Press, USA). Couzin, I. D. (2009), Trends in cognitive sciences 13 (1), 36.
Cabral, J., E. Hugues, O. Sporns, and G. Deco (2011), Neu- Couzin, I. D., C. C. Ioannou, G. Demirel, T. Gross, C. J.
roimage 57 (1), 130. Torney, A. Hartnett, L. Conradt, S. A. Levin, and N. E.
Cafiero, R., A. Gabrielli, and M. A. Muñoz (1998), Phys. Leonard (2011), Science 334 (6062), 1578.
Rev. E 57 (5), 5060. Couzin, I. D., and J. Krause (2003), Advances in the Study
Caldarelli, G. (2007), Scale-free networks: complex webs in of Behavior 32, 1.
nature and technology (Oxford University Press). Couzin, I. D., J. Krause, N. R. Franks, and S. A. Levin
Camalet, S., T. Duke, F. Jülicher, and J. Prost (2000), Proc. (2005), Nature 433 (7025), 513.
Natl. Acad. Sci. USA. 97 (7), 3183. Cover, T. M., and J. Thomas (1991), Elements of Information
Carlson, J. M., and J. Doyle (2000), Phys. Rev. Lett. 84 (11), Theory (Wiley).
2529.
 28

Crick, F. (1970), Nature 227 (5258), 561. Effenberger, F., J. Jost, and A. Levina (2015), PLoS Comput.
Crotty, S., C. E. Cameron, and R. Andino (2001), Proc. Natl. Biol. 11 (9), e1004420.
Acad. Sci. USA. 98 (12), 6895. Eguı́luz, V. M., M. Ospeck, Y. Choe, A. Hudspeth, and M. O.
Crutchfield, J. P. (2012), Nat. Phys. 8 (1), 17. Magnasco (2000), Phys. Rev. Lett. 84 (22), 5232.
Crutchfield, J. P., and K. Young (1988), in Entropy, Com- Ehrig, J., E. P. Petrov, and P. Schwille (2011), Biophys. Jour.
plexity, and the Physics of Information, edited by W. Zurek 100 (1), 80.
(Addison-Wesley, Reading). Eigen, M. (2002), Proc. Natl. Acad. Sci. USA. 99 (21), 13374.
Darabos, C., M. Giacobini, M. Tomassini, P. Provero, and Eigen, M., J. McCaskill, and P. Schuster (1989), Adv. Chem.
F. Di Cunto (2009), in European Conf. on Artificial Life Phys 75, 149.
(Springer) pp. 281–288. Eigen, M., and P. Schuster (1979), New York.
Davies, P. C., L. Demetrius, and J. A. Tuszynski (2011), Erez, A., T. A. Byrd, R. M. Vogel, G. Altan-Bonnet, and
Theoretical Biology and Medical Modelling 8 (1), 1. A. Mugler (2017), arXiv:1703.04194.
Dayan, P., and L. F. Abbott (2006), Theoretical Neu- Erskine, A., and J. M. Herrmann (2014), arXiv:1402.6888.
roscience: Computational and Mathematical Modeling of Espinosa-Soto, C., P. Padilla-Longoria, and E. R. Alvarez-
Neural Systems (Cambridge, MIT Press, USA). Buylla (2004), The Plant Cell 16 (11), 2923.
De Gennes, P.-G. (1979), Scaling concepts in polymer physics Expert, P., R. Lambiotte, D. R. Chialvo, K. Christensen, H. J.
(Cornell university press). Jensen, D. J. Sharp, and F. Turkheimer (2011), J. R. Soc.
De Jong, H. (2002), J. Comput. Biol. 9 (1), 67. Interface 8 (57), 472.
De Palo, G., D. Yi, and R. G. Endres (2017), PLoS biology Eytan, D., and S. Marom (2006), J. Neurosci. 26 (33), 8465.
15 (4), e1002602. Feinerman, O., and A. Korman (2017), J. [Link]. 220 (1),
De Vincenzo, I., I. Giannoccaro, G. Carbone, and P. Grigolini 73.
(2017), Phys. Rev. E 96, 022309. Filkov, V. (2005), Handbook of Computational Molecular Bi-
Deco, G., and V. K. Jirsa (2012), J. Neurosci. 32 (10), 3366. ology , 27.
Deco, G., V. K. Jirsa, and A. R. McIntosh (2011), Nature Fisher, M. E. (1974), Rev. Mod. Phys. 46 (4), 597.
Rev. Neurosci. 12 (1), 43. Font-Clos, F., G. Pruessner, N. R. Moloney, and A. Deluca
Deco, G., V. K. Jirsa, and A. R. McIntosh (2013a), Trends (2015), New J. Phys 17 (4), 043066.
in Neurosci. 36 (5), 268. Forgacs, G., S. A. Newman, S. P. Obukhov, and D. E. Birk
Deco, G., A. Ponce-Alvarez, P. Hagmann, G. L. Romani, (1991), Phys. Rev. Lett. 67, 2399.
D. Mantini, and M. Corbetta (2014), J. Neurosci. 34 (23), Fortrat, J.-O., and C. Gharib (2016), Front. Physiol. 7.
7886. Fox, M. D., and M. E. Raichle (2007), Nature Reviews Neu-
Deco, G., A. Ponce-Alvarez, D. Mantini, G. L. Romani, roscience 8 (9), 700.
P. Hagmann, and M. Corbetta (2013b), J. Neurosci. Fraiman, D., P. Balenzuela, J. Jennifer, and D. R. Chialvo
33 (27), 11239. (2009), Phys. Rev. E 79 (6), 061922.
Delamotte, B. (2012), in Renormalization Group and Effective Frauenfelder, H. (2014), Phys. Biol. 11 (5), 053004.
Field Theory Approaches to Many-Body Systems (Springer) Freeman, W. J. (2013), in Chaos, CNN, Memristors and Be-
pp. 49–132. yond: A Festschrift for [Link]., Vol. 1, edited by A. A.
Derrida, B., and Y. Pomeau (1986), EPL (Europhys. Lett.) et al. (World Scientific Publishing Co.) pp. 271–284.
1 (2), 45. Freeman, W. J., and M. D. Holmes (2005), Neural Networks
Destexhe, A. (2009), J. Comput. Neurosci. 27 (3), 493. 18 (5), 497.
Dhar, D. (1999), Physica A 263 (1), 4. Freeman, W. J., M. D. Holmes, B. C. Burke, and S. Vanhat-
Dickman, R., M. A. Muñoz, A. Vespignani, and S. Zapperi alo (2003), Clinical Neurophys. 114 (6), 1053.
(2000), Braz. J. Phys. 30 (1), 27. Frette, V., K. Christensen, A. Malthe-Sørenssen, J. Feder,
Diez, I., P. Bonifazi, I. Escudero, B. Mateos, M. A. Muñoz, T. Jøssang, and P. Meakin (1996), Nature 379 (6560), 49.
S. Stramaglia, and J. M. Cortes (2015), Sci. Rep. 5. Friedman, N., S. Ito, B. A. W. Brinkman, M. Shimono,
Dornic, I., H. Chaté, J. Chave, and H. Hinrichsen (2001), R. E. L. DeVille, K. A. Dahmen, J. M. Beggs, and T. C.
Phys. Rev. Lett. 87, 045701. Butler (2012), Phys. Rev. Lett. 108, 208102.
Dorogovtsev, S. N., A. V. Goltsev, and J. F. Mendes (2008), Friston, K., M. Breakspear, and G. Deco (2012), Front. Com-
Rev. Mod. Phys. 80 (4), 1275. put. Neurosci. 6, 44.
Dorogovtsev, S. N., and J. F. Mendes (2002), Adv. in Phys. Furusawa, C., and K. Kaneko (2003), Phys. Rev. Lett. 90,
51 (4), 1079. 088102.
Drake, J. W., and J. J. Holland (1999), Proc. Natl. Acad. Furusawa, C., and K. Kaneko (2012a), Phys. Rev. Lett.
Sci. USA. 96 (24), 13910. 108 (20), 208103.
Drossel, B. (2008), in Reviews of nonlinear dynamics and com- Furusawa, C., and K. Kaneko (2012b), Science 338 (6104),
plexity, Vol. 1, edited by H. G. Schuster, Chap. 3 (John 215.
Wiley & Sons) pp. 69–110. Gal, A., and S. Marom (2013), Phys. Rev. E 88 (6), 062717.
Drossel, B., and F. Greil (2009), Phys. Rev. E 80, 026102. Gallotti, R., and D. R. Chialvo (2017), arXiv:1707.07135.
Drossel, B., and F. Schwabl (1992), Phys. Rev. Lett. 69 (11), Gama-Castro, S., H. Salgado, A. Santos-Zavaleta,
1629. D. Ledezma-Tejeida, L. Muñiz-Rascado, J. S. Garcı́a-
Droste, F., A.-L. Do, and T. Gross (2013), J. R. Soc. Interface Sotelo, K. Alquicira-Hernández, I. Martı́nez-Flores,
, 20120558. L. Pannier, J. A. Castro-Mondragón, et al. (2015), Nucleic
Duke, T., and F. Jülicher (2003), Phys. Rev. Lett. 90 (15), Acids Res. 44 (D1), D133.
158101. Gamba, A., M. Nicodemi, J. Soriano, and A. Ott (2012),
Dyson, L., C. A. Yates, J. Buhl, and A. J. McKane (2015), Phys. Rev. Lett. 108 (15), 158103.
Phys. Rev. E 92 (5), 052708. Garcia-Ojalvo, J. (2011), Contemp. Phys. 52 (5), 439.
 29

Garcia-Ojalvo, J., and A. M. Arias (2012), Current opinion 3312.

in genetics & development 22 (6), 619. Hahn, G., A. Ponce-Alvarez, C. Monier, G. Benvenuti, A. Ku-
Garcia-Ojalvo, J., M. B. Elowitz, and S. H. Strogatz (2004), mar, F. Chavane, G. Deco, and Y. Frégnac (2017), PLoS
Proc. Natl. Acad. Sci. USA. 101 (30), 10955. Comput. Biol. 13 (5), e1005543.
Gautam, S. H., T. T. Hoang, K. McClanahan, S. K. Grady, Haimovici, A., and M. Marsili (2015), J. Stat. Mech.
and W. L. Shew (2015), PLoS Comput. Biol. 11 (12), 2015 (10), P10013.
e1004576. Haimovici, A., E. Tagliazucchi, P. Balenzuela, and D. R.
Gelblum, A., I. Pinkoviezky, E. Fonio, A. Ghosh, N. Gov, Chialvo (2013), Phys. Rev. Lett 110, 178101.
and O. Feinerman (2015), Nat. Comm. 6. Haken, H. (1977), Physics Bulletin 28 (9), 412.
GellMann, M. (1994), Complexity: Metaphors, models and Haken, H. (2013), Principles of brain functioning: a syner-
reality , 17. getic approach to brain activity, behavior and cognition,
Gershenson, C. (2012), Theory in Biosci. 131 (3), 181. Vol. 67 (Springer Science & Business Media).
Gerstein, G. L., and B. Mandelbrot (1964), Biophys. Jour. Haken, H., J. S. Kelso, and H. Bunz (1985), Biological cy-
4 (1), 41. bernetics 51 (5), 347.
Ghosh, A., Y. Rho, A. R. McIntosh, R. Kötter, and V. K. Haldeman, C., and J. M. Beggs (2005), Phys. Rev. Lett.
Jirsa (2008), PLoS Comput. Biol. 4 (10), e1000196. 94 (5), 058101.
Ginelli, F. (2016), Eur. Phys. Jour. Special Topics 225 (11- Halley, J. D., F. R. Burden, and D. A. Winkler (2009), Stem
12), 2099. Cell Research 2 (3), 165.
Ginelli, F., F. Peruani, M.-H. Pillot, H. Chaté, G. Theraulaz, Halley, J. D., and D. A. Winkler (2008), Biosystems 92 (2),
and R. Bon (2015), Proc. Natl. Acad. Sci. USA. 112 (41), 148.
12729. Hanel, R., M. Pöchacker, and S. Thurner (2010), Philos.
Giometto, A., M. Formentin, A. Rinaldo, J. E. Cohen, and Trans. R. Soc. London, Ser. A 368 (1933), 5583.
A. Maritan (2015), Proc. Natl. Acad. Sci. USA 112 (25), Harris, T. E. (2002), The theory of branching processes
7755. (Courier Corporation).
Gireesh, E. D., and D. Plenz (2008), Proc. Natl. Acad. Sci. Hart, G. R., and A. L. Ferguson (2015), Phys. Rev. E 91 (3),
USA 105 (21), 7576. 032705.
Gisiger, T. (2001), Biol. Rev Cambridge Philosophical Soc. Hartwell, L. H., J. J. Hopfield, S. Leibler, and A. W. Murray
76 (02), 161. (1999), Nature 402, C47.
Gold, T. (1948), Proc. R. Soc. Ser. B 135 (881), 492. He, B. J. (2011), J. Neurosci. 31 (39), 13786.
Goldberg, D. E., and J. H. Holland (1988), Machine learning He, B. J. (2014), Trends in cognitive sciences 18 (9), 480.
3 (2), 95. Henkel, M., H. Hinrichsen, and S. Lübeck (2008), Non-
Goldberger, A. L. (1992), IEEE Engineering in Medicine and equilibrium Phase Transitions: Absorbing phase transi-
Biology Magazine 11 (2), 47. tions, Theor. and Math. Phys. (Springer London, Berlin).
Goldberger, A. L., L. A. Amaral, J. M. Hausdorff, P. C. Hernandez-Urbina, V., and J. M. Herrmann (2017), Front.
Ivanov, C.-K. Peng, and H. E. Stanley (2002), Proc. Natl. Phys. 4, 54.
Acad. Sci. USA. 99 (suppl 1), 2466. Hertz, J., A. Krogh, and R. G. Palmer (1991), Introduction
Goldenfeld, N., and C. Woese (2011), Annu. Rev. Condens. to the theory of neural computation, Vol. 1 (Basic Books).
Matter Phys. 2 (1), 375. Herz, A. V., and J. J. Hopfield (1995), Phys. Rev. Lett.
Gollo, L., O. Kinouchi, and M. Copelli (2013), [Link]. 3. 75 (6), 1222.
Gomez, F., T. Lorimer, and R. Stoop (2015), Hesse, J., and T. Gross (2014), Front. Comput. Neurosci. 8.
arXiv:1510.03241. Hidalgo, J., J. Grilli, S. Suweis, A. Maritan, and M. A. Muñoz
Gómez, V., A. Kaltenbrunner, V. López, and H. J. Kappen (2016), J. of Stat. Mech. 2016 (3), 033203.
(2008), in NIPS conference, pp. 513–520. Hidalgo, J., J. Grilli, S. Suweis, M. A. Muñoz, J. R. Ba navar,
Goodell, M. A., H. Nguyen, and N. Shroyer (2015), Nature and A. Maritan (2014), Proc. Natl. Acad. Sci. USA.
reviews. Molecular cell biology 16 (5), 299. 111 (28), 10095.
Goudarzi, A., C. Teuscher, N. Gulbahce, and T. Rohlf (2012), Hidalgo, J., L. Seoane, J. Cortés, and M. Muñoz (2012),
Phys. Rev. Lett. 108 (12), 128702. PLoS One 7(8), e40710.
Grassberger, P. (1995), J. Stat. Phys. 79 (1-2), 13. Hinrichsen, H. (2000), Adv. in Phys. 49, 815.
Grégoire, G., and H. Chaté (2004), Phys. Rev. Lett. 92 (2), Hinrichsen, H., and E. Domany (1997), Phys. Rev. E 56 (1),
025702. 94.
Greicius, M. D., B. Krasnow, A. L. Reiss, and V. Menon Hobbs, J. P., J. L. Smith, and J. M. Beggs (2010), J. of Clin.
(2003), Proc. Natl. Acad. Sci. USA. 100 (1), 253. Neurophys. 27 (6), 380.
Grinstein, G. (1991), J. Appl. Phys. 69 (8), 5441. Hohenberg, P. C., and B. I. Halperin (1977), Rev. Mod. Phys.
Gros, C. (2008), Complex and adaptive dynamical systems: A 49 (3), 435.
primer (Springer). Holcman, D., and M. Tsodyks (2006), PLoS Comput. Biol.
Gross, T., and B. Blasius (2008), J. R. Soc. Interface 5 (20), 2 (3), e23.
259. Honerkamp-Smith, A. R., P. Cicuta, M. D. Collins, S. L.
Grossberg, S. (1982), in A Portrait of Twenty-five Years Veatch, M. den Nijs, M. Schick, and S. L. Keller (2008),
(Springer) pp. 257–276. Biophysical journal 95 (1), 236.
Hagmann, P., L. Cammoun, X. Gigandet, R. Meuli, C. J. Hopfield, J. (1994), J. Theor. Biol. 171 (1), 53.
Honey, V. J. Wedeen, and O. Sporns (2008), PLoS Biol Hopfield, J. J. (1982), Proc. Natl. Acad. Sci. USA. 79 (8),
6 (7), e159. 2554.
Hahn, G., T. Petermann, M. N. Havenith, S. Yu, W. Singer, Hu, K., J. H. Meijer, S. A. Shea, H. Tjebbe vanderLeest,
D. Plenz, and D. Nikolić (2010), J. Neurophysiol. 104 (6), B. Pittman-Polletta, T. Houben, F. van Oosterhout, T. De-
 30

boer, and F. A. Scheer (2012), PLoS One 7 (11), e48927. Kinouchi, O., L. Brochini, A. A. Costa, J. G. F. Campos, and
Huang, S., G. Eichler, Y. Bar-Yam, and D. E. Ingber (2005), M. Copelli (2018), arXiv:1803.05537.
Phys. Rev. Lett. 94 (12), 128701. Kinouchi, O., and M. Copelli (2006), Nat. Phys. 2 (5), 348.
Hubbell, S. P. (2001), The unified neutral theory of biodiver- Kitano, H., et al. (2001), Foundations of systems biology (MIT
sity and biogeography (MPB-32), Vol. 32 (Princeton Uni- press Cambridge).
versity Press). Kiyono, K., Z. Struzik, N. Aoyagi, S. Sakata, J. Hayano, and
Hudspeth, A. (2014), Nature Rev. Neurosci. 15 (9), 600. Y. Yamamoto (2004), Phys. Rev. Lett. 93 (17), 178103.
Hudspeth, A., F. Jülicher, and P. Martin (2010), Journal of Kiyono, K., Z. R. Struzik, N. Aoyagi, F. Togo, and Y. Ya-
neurophysiology 104 (3), 1219. mamoto (2005), Phys. Rev. Lett. 95 (5), 058101.
Hughes, T. R., M. J. Marton, A. R. Jones, C. J. Roberts, Kleiber, M. (1932), Hilgardia 6, 315.
R. Stoughton, C. D. Armour, H. A. Bennett, E. Coffey, Koonin, E. V. (2011), The logic of chance: the nature and
H. Dai, Y. D. He, et al. (2000), Cell 102 (1), 109. origin of biological evolution (FT press).
Hyman, A. A., and K. Simons (2012), Science 337 (6098), Koonin, E. V., Y. I. Wolf, and G. P. Karev (2006), Power
1047. laws, scale-free networks and genome biology (Springer).
Iliopoulos, D., A. Hintze, and C. Adami (2010), PLoS Com- Krause, J., and G. D. Ruxton (2002), Living in groups (Ox-
put. Biol. 6 (10). ford University Press).
Ivanov, P. (2007), IEEE Engineering in Med. and Biol. Mag- Krawitz, P., and I. Shmulevich (2007), Phys. Rev. Lett.
azine 26 (6), 33. 98 (15), 158701.
Ivanov, P. C., L. A. N. Amaral, A. L. Goldberger, S. Havlin, Krotov, D., J. O. Dubuis, T. Gregor, and W. Bialek (2014),
M. G. Rosenblum, Z. R. Struzik, and H. E. Stanley (1999), Proc. Natl. Acad. Sci. USA. 111 (10), 3683.
Nature 399 (6735), 461. Ktitarev, D., S. Lübeck, P. Grassberger, and V. Priezzhev
Izhikevich, E. M. (2004), IEEE transactions on Neural Net- (2000), Phys. Rev. E 61 (1), 81.
works 15 (5), 1063. Kuehn, C. (2012), Phys. Rev. E 85 (2), 026103.
Izhikevich, E. M. (2007), Dynamical Systems in Neuroscience: Kuramoto, Y. (1975), Lecture Notes in Physics 39, 420.
The Geometry of Excitability and Bursting (MIT Press, Kussell, E., and S. Leibler (2005), Science 309 (5743), 2075.
Cambridge MA). Langton, C. (1990), Physica D 42 (1), 12.
Jaeger, H. (2007), Scholarpedia 2 (9), 2330. Laurson, L., X. Illa, and M. J. Alava (2009), J. Stat. Mech.
Janićević, S., L. Laurson, K. J. Måløy, S. Santucci, and M. J. 2009 (01), P01019.
Alava (2016), Phys. Rev. Lett. 117 (23), 230601. LeCun, Y., Y. Bengio, and G. Hinton (2015), Nature
Jensen, H. J. (1998), Self-organized criticality: emergent com- 521 (7553), 436.
plex behavior in physical and biological systems (Cambridge Lee, C. F., C. P. Brangwynne, J. Gharakhani, A. A. Hyman,
Univ. press). and F. Jülicher (2013), Phys. Rev. Lett. 111 (8), 088101.
Kaiser, M. (2011), NeuroImage 57 (3), 892. Legenstein, R., and W. Maass (2007), Neural Networks
Kamenev, A. (2011), Field theory of non-equilibrium systems 20 (3), 323.
(Cambridge University Press). Legenstein, T. (2005), in Advances in Neural Information
Kandel, E., J. Schwartz, and T. Jessel (2000), Principles of Processing Systems 17: Proceedings of the 2004 Confer-
Neural Science (McGraw-Hill, New York). ence, Vol. 17 (MIT Press) p. 145.
Kaneko, K. (2006), Life: an introduction to complex systems Lesk, A. (2017), Introduction to genomics (Oxford University
biology (Springer). Press).
Kaneko, K. (2012), J. Stat. Phys. 148 (4), 687. Levina, A., J. M. Herrmann, and T. Geisel (2007), Nat. Phys.
Kaneko, K., and T. Ikegami (1992), Physica D: Nonlinear 3 (12), 857.
Phenomena 56 (4), 406. Levina, A., J. M. Herrmann, and T. Geisel (2009), Phys.
Kauffman, S. (1969), J. Theor. Biol. 22 (3), 437. Rev. Lett. 102 (11), 118110.
Kauffman, S. (1996), At home in the universe: The search Li, F., T. Long, Y. Lu, Q. Ouyang, and C. Tang (2004), Proc.
for the laws of self-organization and complexity (Oxford Natl. Acad. Sci. USA. 101 (14), 4781.
university press). Li, L., H. Peng, J. Kurths, Y. Yang, and H. J. Schellnhuber
Kauffman, S., C. Peterson, B. Samuelsson, and C. Troein (2014), Proc. Natl. Acad. Sci. USA. 111 (23), 8392.
(2003), Proc. Natl. Acad. Sci. USA 100 (25), 14796. Li, W., N. H. Packard, and C. G. Langton (1990), Physica
Kauffman, S. A. (1993), The origins of order: Self organiza- D 45 (1), 77.
tion and selection in evolution (Oxford university press). Li, Y. C., and D. Retzloff (2006), Math. Biosci. 203 (1), 137.
Keenan, J. P., A. L. Krill, S. M. Platek, and T. K. Shackelford Liggett, T. (2004), Interacting Particle Systems, Classics in
(2007), Evolutionary Cognitive Neuroscience , 579. Mathematics (Springer).
Kelley, D. H., and N. T. Ouellette (2013), Sci. Rep. 3. Lin, H. W., and M. Tegmark (2017), Entropy 19 (7), 299.
Kello, C. T., G. D. Brown, R. Ferrer-i Cancho, J. G. Holden, Liu, M., and K. E. Bassler (2006), Phys. Rev. E 74 (4),
K. Linkenkaer-Hansen, T. Rhodes, and G. C. Van Orden 041910.
(2010), Trends in cognitive sciences 14 (5), 223. Lizier, J. T., M. Prokopenko, and A. Y. Zomaya (2008a), in
Kelso, J., J. Scholz, and G. Schöner (1986), Phys. Lett. A ALIFE, pp. 374–381.
118 (6), 279. Lizier, J. T., M. Prokopenko, and A. Y. Zomaya (2008b),
Kelso, J. A. (1984), Am J Physiol Regul Integr Comp Physiol Phys. Rev. E 77 (2), 026110.
246 (6), R1000. Loengarov, A., and V. Tereshko (2008), Artif Life 14 (1),
Kern, A., and R. Stoop (2003), Phys. Rev. Lett. 91 (12), 111.
128101. Lopez-Garcia, C., A. M. Klein, B. D. Simons, and D. J.
Kimura, M. (1984), The neutral theory of molecular evolution Winton (2010), Science 330 (6005), 822.
(Cambridge University Press). Losa, G. A. (1995), Pathologica 87, 310.
 31

Lukoševičius, M., H. Jaeger, and B. Schrauwen (2012), KI- Mora, T., and W. Bialek (2011), J. Stat. Phys. 144 (2), 268.
Künstliche Intelligenz 26 (4), 365. Mora, T., S. Deny, and O. Marre (2015), Phys. Rev. Lett.
Luque, B., and A. Ferrera (2000), Complex Systems 12, 241. 114 (7), 078105.
Maass, W., T. Natschlager, and H. Markram (2002), Neural Mora, T., A. Walczak, W. Bialek, and C. Callan Jr (2010),
Comput. 14, 2531. Proc. Natl. Acad. Sci. USA. 107 (12), 5405.
MacArthur, B. D., R. J. Sánchez-Garcı́a, and A. Maayan Mora, T., A. M. Walczak, L. Del Castello, F. Ginelli,
(2010), Phys. Rev. Lett. 104 (16), 168701. S. Melillo, L. Parisi, M. Viale, A. Cavagna, and I. Gia-
Magnasco, M. O. (2003), Phys. Rev. Lett. 90 (5), 058101. rdina (2016), Nat. Phys. 12 (12), 1153.
Magnasco, M. O., O. Piro, and G. A. Cecchi (2009), Phys. Moreau, L., and E. Sontag (2003), Phys. Rev. E 68 (2),
Rev. Lett. 102 (25), 258102. 020901.
Mandelbrot, B. (2002), Gaussian self-affinity and fractals: Moreira, A. G., and R. Dickman (1996), Phys. Rev. E 54 (4),
Globality, the earth, 1/f noise, and R/S, Vol. 8 (Springer R3090.
Science & Business Media). Moretti, P., and M. A. Muñoz (2013), Nature Comm. 4.
Mandelbrot, B. B. (1983), The fractal geometry of nature, Vol. Muñoz, M., R. Juhász, C. Castellano, and G. Ódor (2010),
173 (Macmillan). Phys. Rev. Lett. 105, 128701.
Manna, S. (1991), J. of Phys. A: Mathematical and General Murphy, B. K., and K. D. Miller (2009), Neuron 61 (4), 635.
24 (7), L363. Nadell, C. D., V. Bucci, K. Drescher, S. A. Levin, B. L.
Marković, D., and C. Gros (2014), Phys. Rep. 536 (2), 41. Bassler, and J. B. Xavier (2013), Proc. R. Soc. B
Markram, H. (2006), Nature Rev. Neurosci. 7 (2), 153. 280 (1755), 20122770.
Markram, H., and M. Tsodyks (1996), Nature 382, 807. Newman, M. E. J. (2003), SIAM Review 45 (2), 167.
Markram, H., et al. (2015), Cell 163 (2), 456. Newman, M. E. J. (2005), Contemp Phys 46 (5), 323, cond-
Marre, O., D. Amodei, N. Deshmukh, K. Sadeghi, F. Soo, mat/0412004.
T. E. Holy, and M. J. Berry (2012), J. Neurosci. 32 (43), Newman, M. E. J. (2010), Networks: An Introduction (Oxford
14859. University Press).
Marro, J., and R. Dickman (1999), Nonequilibrium Phase Newman, S. A., G. Forgacs, B. Hinner, C. W. Maier, and
Transition in Lattice Models (Cambridge University Press). E. Sackmann (2004), Phys. Biol. 1 (2), 100.
Marsili, M., I. Mastromatteo, and Y. Roudi (2013), J. Stat. Nonnenmacher, M., C. Behrens, P. Berens, M. Bethge, and
Mech. 2013 (09), P09003. J. H. Macke (2017), PLoS Comp. Biol. 13 (10), e1005718.
Marsili, M., and Y.-C. Zhang (1998), Phys. Rev. Lett. Novikov, E., A. Novikov, D. Shannahoff-Khalsa, B. Schwartz,
80 (12), 2741. and J. Wright (1997), Phys. Rev. E 56 (3), R2387.
Martin, P., A. Hudspeth, and F. Jülicher (2001), Proc. Natl. Nowak, M. A. (2006), Evolutionary dynamics (Harvard Uni-
Acad. Sci. USA. 98 (25), 14380. versity Press).
Martinello, M., J. Hidalgo, A. Maritan, S. di Santo, D. Plenz, Nykter, M., N. D. Price, M. Aldana, S. A. Ramsey, S. A.
and M. A. Muñoz (2017), Physical Review X 7 (4), 041071. Kauffman, L. E. Hood, O. Yli-Harja, and I. Shmulevich
Massobrio, P., L. de Arcangelis, V. Pasquale, H. J. Jensen, (2008a), Proc. Natl. Acad. Sci. USA. 105 (6), 1897.
and D. Plenz (2015), Criticality as a signature of healthy Nykter, M., N. D. Price, A. Larjo, T. Aho, S. A. Kauffman,
neural systems: multi-scale experimental and computa- O. Yli-Harja, and I. Shmulevich (2008b), Phys. Rev. Lett
tional studies , 4. 100 (5), 058702.
Mastromatteo, I., and M. Marsili (2011), J. Stat. Mech. Ódor, G. (2008), Universality in Nonequilibrium Lattice Sys-
2011 (10), P10012. tems: Theoretical Foundations (World Scientific, Singa-
Mattia, M., and M. V. Sanchez-Vives (2012), Cognitive neu- pore).
rodynamics 6 (3), 239. Olami, Z., H. J. S. Feder, and K. Christensen (1992), Phys.
Mazzoni, A., F. D. Broccard, E. Garcia-Perez, P. Bonifazi, Rev. Lett. 68 (8), 1244.
M. E. Ruaro, and V. Torre (2007), PloS One 2 (5), e439. Oprisa, D., and P. Toth (2017a), arXiv:1702.08039.
Méhes, E., and T. Vicsek (2014), Integrative Biol. 6 (9), 831. Oprisa, D., and P. Toth (2017b), arXiv:1705.11023.
Mehta, P., and D. J. Schwab (2014), arXiv:1410.3831. Ospeck, M., V. M. Eguı́luz, and M. O. Magnasco (2001),
Meisel, C., and T. Gross (2009), Phys. Rev. E 80 (6), 061917. Biophy. Jour. 80 (6), 2597.
Meisel, C., E. Olbrich, O. Shriki, and P. Achermann (2013), Packard, N. H. (1988), Adaptation toward the edge of chaos
J. Neurosci. 33 (44), 17363. (University of Illinois at Urbana-Champaign, Center for
Meisel, C., A. Storch, S. Hallmeyer-Elgner, E. Bullmore, and Complex Systems Research).
T. Gross (2012), PLoS Comput. Biol. 8 (1), e1002312. Pal, M., S. Ghosh, and I. Bose (2014), Phys. Biol. 12 (1),
Meister, M., R. Wong, D. A. Baylor, and C. J. Shatz (1991), 016001.
Science 252 (5008), 939. Palmigiano, A., T. Geisel, F. Wolf, and D. Battaglia (2017),
Mejias, J. F., H. J. Kappen, and J. J. Torres (2010), PLoS Nature Neurosci. 20 (7), 1014.
One 5 (11), e13651. Palva, J. M., A. Zhigalov, J. Hirvonen, O. Korhonen,
Melanie, M. (1993), Complexity: Metaphors, Models, and Re- K. Linkenkaer-Hansen, and S. Palva (2013), Proc. Nat.
ality. Acad. Sci. USA 110 (9), 3585.
Meunier, D., R. Lambiotte, and E. T. Bullmore (2010), Front. Parfitt, D.-E., and M. M. Shen (2014), Phil. Trans. R. Soc.
Neurosci. 4, 200. B 369 (1657), 20130542.
Millman, D., S. Mihalas, A. Kirkwood, and E. Niebur (2010), Parga, N., and L. F. Abbott (2007), Front. Neurosci. 1, 4.
Nat. Phys. 6 (10), 801. Parisi, G. (1993), Physics World 6 (9), 42.
Ming, L., and P. Vitányi (2014), Algorithms and Complexity Pasquale, V., P. Massobrio, L. Bologna, M. Chiappalone, and
, 187. S. Martinoia (2008), Neuroscience 153 (4), 1354.
Mitzenmacher, M. (2002), Internet Mathematics 1, 226.
 32

Pastor-Satorras, R., C. Castellano, P. Van Mieghem, and Redner, S. (2001), A guide to first-passage processes (Cam-
A. Vespignani (2015), Rev. Mod. Phys. 87 (3), 925. bridge University Press).
Pearlmutter, B. A., and C. J. Houghton (2009), Neural Com- Reed, W. J., and B. D. Hughes (2002), Phys. Rev. E 66 (6),
put. 21 (6), 1622. 067103.
Perotti, J. I., O. V. Billoni, F. A. Tamarit, D. R. Chialvo, and Ribeiro, A. S., S. A. Kauffman, J. Lloyd-Price, B. Samuelsson,
S. A. Cannas (2009), Phys. Rev. Lett. 103 (10), 108701. and J. E. Socolar (2008), Phys. Rev. E 77 (1), 011901.
Peruani, F., J. Starruß, V. Jakovljevic, L. Søgaard-Andersen, Ribeiro, T. L., M. Copelli, F. Caixeta, H. Belchior, D. R.
A. Deutsch, and M. Bär (2012), Phys. Rev. Lett. 108, Chialvo, M. A. L. Nicolelis, and S. Ribeiro (2010), PLoS
098102. One 5 (11), e14129.
Petermann, T., T. A. Thiagarajan, M. Lebedev, M. Nicolelis, Richmond, P., and S. Solomon (2001), Int. J. Mod. Phys. C
D. R. Chialvo, and D. Plenz (2009), Proc. Natl. Acad. Sci. 12 (03), 333.
USA 106 (37), 15921. Ridden, S. J., H. H. Chang, K. C. Zygalakis, and B. D.
Phillips, J. (2009), Proc. Natl. Acad. Sci. USA. 106 (9), 3107. MacArthur (2015), Phys. Rev. Lett. 115, 208103.
Pietronero, L., E. Tosatti, V. Tosatti, and A. Vespignani Rieke, F., D. Bodnar, and W. Bialek (1995), Proc. R. Soc.
(2001), Physica A 293 (1), 297. London, Ser. B 262 (1365), 259.
Pikovsky, A., M. Rosenblum, and J. Kurths (2003), Synchro- Ringel, Z., and R. de Bem (2018), arXiv:1802.02154.
nization: a universal concept in nonlinear sciences, Vol. 12 Rohlf, T. (2008), EPL (Europhys. Lett.) 84 (1), 10004.
(Cambridge university press). Rohlf, T., and S. Bornholdt (2002), Physica A 310 (1), 245.
Pinto, O. A., and M. A. Munoz (2011), PloS One 6 (7), Rohlf, T., N. Gulbahce, and C. Teuscher (2007), Phys. Rev.
e21946. Lett. 99 (24), 248701.
Plenz, D. (2013), Physics 6, 47. Roli, A., M. Villani, A. Filisetti, and R. Serra (2015), J.
Plenz, D., and E. Niebur (2014), Criticality in neural systems Systems Sciences and Complexity , 1.
(John Wiley & Sons). Rosenbaum, R., and B. Doiron (2014), Phys. Rev. X 4 (2),
Plenz, D., and T. C. Thiagarajan (2007), Trends Neurosci 021039.
30 (3), 101. Rosenfeld, S. (2013), Gene Regulation and Systems Biology
Poblanno-Balp, R., and C. Gershenson (2011), Artificial Life 7, 23.
17 (4), 331. Rössert, C., P. Dean, and J. Porrill (2015), PLoS Comput.
Poil, S.-S., R. Hardstone, H. D. Mansvelder, and Biol. 11 (10), e1004515.
K. Linkenkaer-Hansen (2012), J. Neurosci. 32 (29), 9817. Rubinov, M., O. Sporns, J.-P. Thivierge, and M. Breakspear
Poland, D., and H. A. Scheraga (1970), Theory of helix-coil (2011), PLoS Comput. Biol. 7 (6), e1002038.
transitions in biopolymers (Academic Press). Rybarsch, M., and S. Bornholdt (2014), “Self-organized crit-
Pollack, G. H., and W.-C. Chin (2008), Phase Transitions in icality in neural network models,” in Criticality in Neural
Cell Biology (Springer). Systems (Wiley-VCH Verlag) pp. 227–254.
Priesemann, V., M. H. Munk, and M. Wibral (2009), BMC Sadilek, M., and S. Thurner (2015), Sci. Rep. 5.
Neurosci. 10 (1), 40. Saito, N., and M. Kikuchi (2013), New J. Phys. 15 (5),
Priesemann, V., M. Valderrama, M. Wibral, and 053037.
M. Le Van Quyen (2013), PLoS Comput. Biol. 9 (3), Sanchez-Vives, M. V., and D. A. McCormick (2000), Nature
e1002985. Neurosci. 3 (10), 1027.
Priesemann, V., M. Wibral, M. Valderrama, R. Pröpper, di Santo, S., R. Burioni, A. Vezzani, and M. A. Muñoz (2016),
M. Le Van Quyen, T. Geisel, J. Triesch, D. Nikolić, and Phys. Rev. Lett. 116 (24), 240601.
M. H. Munk (2014), Front. Syst. Neurosci. 8. di Santo, S., P. Villegas, R. Burioni, and M. A. Muñoz
Proekt, A., J. Banavar, A. Maritan, and D. Pfaff (2012), (2017a), Submitted.
Proc. Natl. Acad. Sci. USA. 109 (26), 10564. di Santo, S., P. Villegas, R. Burioni, and M. A. Muñoz
Prokopenko, M. (2013), in ALIFE, pp. 140–144. (2017b), Phys. Rev. E 95 (3), 032115.
Pruessner, G. (2012), Self-organised criticality: theory, models di Santo, S., P. Villegas, R. Burioni, and M. A. Muñoz (2018),
and characterisation (Cambridge University Press). arXiv:1803.07858.
Puckett, J. G., and N. T. Ouellette (2014), J. R. Soc. Inter- Sauer, U., M. Heinemann, and N. Zamboni (2007), Science
face 11 (99), 20140710. 316 (5824), 550.
Rabinovich, M., P. Varona, A. Selverston, and H. Abarbanel Schneidman, E., M. J. Berry, R. Segev, and W. Bialek (2006),
(2006), Rev. Mod. Phys. 78, 1213. Nature 440 (7087), 1007.
Raichle, M. E. (2011), Brain connectivity 1 (1), 3. Scholz, J., J. Kelso, and G. Schöner (1987), Phys. Lett. A
Raichle, M. E., A. M. MacLeod, A. Z. Snyder, W. J. Powers, 123 (8), 390.
D. A. Gusnard, and G. L. Shulman (2001), Proc. Natl. Schrödinger, E. (1967), What is Life? (Cambridge University
Acad. Sci. USA. 98 (2), 676. Press).
Ramaswamy, S. (2010), Ann. Rev. Cond. Matt. Phys.. Schwab, D. J., I. Nemenman, and P. Mehta (2014), Phys.
Rämö, P., S. Kauffman, J. Kesseli, and O. Yli-Harja (2007), Rev. Lett. 113, 068102.
Physica D 227 (1), 100. Segev, R., and E. Ben-Jacob (2001), Physica A 302 (1), 64.
Rämö, P., J. Kesseli, and O. Yli-Harja (2006), J. Theor. Biol. Segev, R., Y. Shapira, M. Benveniste, and E. Ben-Jacob
242 (1), 164. (2001), Phys. Rev. E 64 (1), 011920.
Ramos, F., C. López, E. Hernández-Garcı́a, and M. A. Munoz Sejnowski, T. J., P. S. Churchland, and J. A. Movshon (2014),
(2008), Phys. Rev. E 77 (2), 021102. Nat. Neurosci. 17 (11), 1440.
Ravasz, E., A. L. Somera, D. A. Mongru, Z. N. Oltvai, and Seoane, L. F., and R. Solé (2015), arXiv:1510.08697.
A.-L. Barabási (2002), science 297 (5586), 1551. Serra, R., M. Villani, A. Graudenzi, and S. Kauffman (2007),
J. Theor. Biol. 246 (3), 449.
 33

Serra, R., M. Villani, and A. Semeria (2004), J. Theor. Biol. Sporns, O. (2010), Networks of the Brain (MIT Press, USA).
227 (1), 149. Sporns, O., D. R. Chialvo, M. Kaiser, and C. C. Hilgetag
Sethna, J. (2006), Statistical mechanics: entropy, order pa- (2004), Trends Cogn Sci 8 (9), 418.
rameters, and complexity, Vol. 14 (Oxford Univ. Press). Sporns, O., G. Tononi, and R. Kötter (2005), PLoS Comput.
Sethna, J. P., K. A. Dahmen, and C. R. Myers (2001), Nature Biol. 1 (4), e42.
410 (6825), 242. Stanley, H. E. (1987), Introduction to phase transitions and
Shanahan, M. (2010), Chaos 20 (1), 013108. critical phenomena (Oxford Univ. Press).
Shew, W. L., W. P. Clawson, J. Pobst, Y. Karimipanah, N. C. Stassinopoulos, D., and P. Bak (1995), Phys. Rev. E 51 (5),
Wright, and R. Wessel (2015), Nat. Phys. , 659. 5033.
Shew, W. L., and D. Plenz (2013), The Neuroscientist 19 (1), Stepp, N., D. Plenz, and N. Srinivasa (2015), PLoS Comput.
88. Biol. 11 (1).
Shew, W. L., H. Yang, T. Petermann, R. Roy, and D. Plenz Steriade, M., F. Amzica, and D. Contreras (1996), J. Neu-
(2009), J. Neurosci. 29 (49), 15595. rosci. 16 (1), 392.
Shin, C.-W., and S. Kim (2006), Phys. Rev. E 74 (4), 045101. Steriade, M., A. Nunez, and F. Amzica (1993), J. Neurosci.
Shmulevich, I., and E. R. Dougherty (2010), Probabilistic 13 (8), 3252.
Boolean networks: the modeling and control of gene regu- Stewart, C. V., and D. Plenz (2008), J. of Neurosci. Methods
latory networks (SIAM). 169 (2), 405.
Shmulevich, I., S. A. Kauffman, and M. Aldana (2005), Proc. Stokić, D., R. Hanel, and S. Thurner (2008), Phys. Rev. E
Natl. Acad. Sci. USA. 102 (38), 13439. 77 (6), 061917.
Shriki, O., J. Alstott, F. Carver, T. Holroyd, R. N. Hen- Stoop, R., and F. Gomez (2016), Phys. Rev. Lett. 117,
son, M. L. Smith, R. Coppola, E. Bullmore, and D. Plenz 038102.
(2013), J. Neurosci. 33 (16), 7079. Strogatz, S. H. (2014), Nonlinear dynamics and chaos: with
Shriki, O., and D. Yellin (2016), PLoS Comput. Biol. 12 (2), applications to physics, biology, chemistry, and engineering
e1004698. (Westview press).
Simini, F., T. Anfodillo, M. Carrer, J. R. Banavar, and Stumpf, M. P., and M. A. Porter (2012), Science 335 (6069),
A. Maritan (2010), Proc. Natl. Acad. Sci. USA. 107 (17), 665.
7658. Summers, J., and S. Litwin (2006), J. of Virology 80 (1), 20.
Simkin, M. V., and V. P. Roychowdhury (2011), Phys. Rep. Sumpter, D. J. (2010), Collective animal behavior (Princeton
502 (1), 1. University Press).
Simon, H. A. (1955), Biometrika 42 (3/4), 425. Sussillo, D., T. Toyoizumi, and W. Maass (2007), J. Neuro-
Sneppen, K. (2014), Models of Life (Cambridge Univ. Press). physiol. 97 (6), 4079.
Sneppen, K., P. Bak, H. Flyvbjerg, and M. H. Jensen (1995), Suweis, S., F. Simini, J. R. Banavar, and A. Maritan (2013),
Proc. Natl. Acad. Sci. USA 92 (11), 5209. Nature 8500, 449.
Sokolov, A., I. S. Aranson, J. O. Kessler, and R. E. Goldstein Tabak, J., and P. E. Latham (2003), Neuroreport 14 (11),
(2007), Phys. Rev. Lett. 98 (15), 158102. 1445.
Solé, R. (2011), Phase transitions (Princeton University Tagliazucchi, E., P. Balenzuela, D. Fraiman, and D. R.
Press). Chialvo (2012), Front. Physiol. 3, 15.
Solé, R., and T. Deisboeck (2004), J. Theor. Biol. 228 (1), Tagliazucchi, E., D. R. Chialvo, M. Siniatchkin, E. Amico, J.-
47. F. Brichant, V. Bonhomme, Q. Noirhomme, H. Laufs, and
Solé, R. V. (2003), The European Phys. Jour. B 35 (1), 117. S. Laureys (2016), J. R. Soc. Interface 13 (114), 20151027.
Solé, R. V., D. Alonso, and A. McKane (2002a), Philos. Tagliazucchi, E., F. von Wegner, A. Morzelewski, V. Brod-
Trans. R. Soc. London, Ser. B 357 (1421), 667. beck, K. Jahnke, and H. Laufs (2013), Proc. Natl. Acad.
Solé, R. V., R. Ferrer-Cancho, J. M. Montoya, and Sci. USA. 110 (38), 15419.
S. Valverde (2002b), Complexity 8 (1), 20. Tang, Q.-Y., Y.-Y. Zhang, J. Wang, W. Wang, and D. R.
Solé, R. V., S. C. Manrubia, M. Benton, S. Kauffman, and Chialvo (2017), Phys. Rev. Lett. 118, 088102.
P. Bak (1999), Trends in Ecology & Evolution 14 (4), 156. Täuber, U. C. (2014), Critical dynamics: a field theory ap-
Solé, R. V., S. C. Manrubia, B. Luque, J. Delgado, and proach to equilibrium and non-equilibrium scaling behavior
J. Bascompte (1996), Complexity 1 (4), 13. (Cambridge University Press).
Solé, R. V., and O. Miramontes (1995), Physica D 80 (1), Täuber, U. C. (2017), Ann. Rev. [Link]. Phys. 8 (1), 185.
171. Taylor, L. (1961), Nature , 732.
Solovey, G., K. J. Miller, J. G. Ojemann, M. O. Magnasco, Tetzlaff, C., S. Okujeni, U. Egert, F. Wörgötter, and M. Butz
and G. A. Cecchi (2012), Front. Integrative Neurosci. 6. (2010), PLoS Comput. Biol. 6 (12), e1001013.
Song, J., M. Marsili, and J. Jo (2017), arXiv:1710.11324. Tinker, J., and J. L. P. Velazquez (2014), Front. Syst. Neu-
Soriano, J., C. Colombo, and A. Ott (2006), Phys. Rev. Lett. rosci. 8, 73.
97 (25), 258102. Tkačik, G., O. Marre, D. Amodei, E. Schneidman, W. Bialek,
Soriano, J., S. Rüdiger, P. Pullarkat, and A. Ott (2009), and M. J. Berry II (2014), PLoS Comput. Biol. 10 (1),
Biophys. Jour. 96 (4), 1649. e1003408.
Sornette, D. (1994), J. de Physique I 4 (2), 209. Tkačik, G., O. Marre, T. Mora, D. Amodei, M. J. Berry II,
Sornette, D. (1998), Phys. Rev. E 57 (4), 4811. and W. Bialek (2013), J. Stat. Mech. 2013 (03), P03011.
Sornette, D. (2006), Critical Phenomena in Natural Sciences Tkačik, G., T. Mora, O. Marre, D. Amodei, S. E. Palmer,
(Springer). M. J. Berry, and W. Bialek (2015), Proc. Natl. Acad. Sci.
Sornette, D. (2009), in Encyclopedia of Complexity and Sys- USA. 112 (37), 11508.
tems Science (Springer) pp. 7009–7024. Tkacik, G., E. Schneidman, I. Berry, J. Michael, and
Sornette, D., and R. Cont (1997), J. de Physique I 7 (3), 431. W. Bialek (2009), arXiv preprint arXiv:0912.5409.
 34

Tomen, N., D. Rotermund, and U. Ernst (2014), Front. Syst. Villegas, P., P. Moretti, and M. A. Muñoz (2014), Sci. Rep.
Neurosci 8, 151. 4, 5990.
Toner, J., and Y. Tu (1995), Phys. Rev. Lett. 75 (23), 4326. Villegas, P., J. Ruiz-Franco, J. Hidalgo, and M. A. Muñoz
Toner, J., Y. Tu, and S. Ramaswamy (2005), Ann. Phys. (2016), Sci. Rep. 6, 34743.
318 (1), 170. Visser, M. (2013), New J. Phys. 15 (4), 043021.
Tononi, G., O. Sporns, and G. M. Edelman (1994), Proc. Vojta, T. (2006), J. Phys. A 39 (22), R143.
Natl. Acad. Sci. USA. 91 (11), 5033. van Vreeswijk, C., and H. Sompolinsky (1996), Science 274,
Torres-Sosa, C., S. Huang, and M. Aldana (2012), PLoS 1724.
Comput. Biol. 8 (9), e1002669. Wagner, A. (2005), Robustness and evolvability in living sys-
Touboul, J., and A. Destexhe (2010), PloS One 5 (2), e8982. tems (Princeton University Press Princeton).
Touboul, J., and A. Destexhe (2017), Phys. Rev. E 95 (1), Wang, X. R., J. T. Lizier, and M. Prokopenko (2011), Arti-
012413. ficial life 17 (4), 315.
Toyoizumi, T., and L. F. Abbott (2011), Phys. Rev. E 84, Watkins, N. W., G. Pruessner, S. C. Chapman, N. B. Crosby,
051908. and H. J. Jensen (2015), Space Sci. Rev. , 1.
Tsodyks, M., and H. Markram (1997), Proc. Natl. Acad. Sci. Watson, H. W., and F. Galton (1875), J. of the Anthropo-
USA 94, 719. logical Ins. of Great Britain and Ireland 4, 138.
Tsuchiya, M., A. Giuliani, M. Hashimoto, J. Erenpreisa, and Watts, D. J., and S. H. Strogatz (1998), Nature 393 (6684),
K. Yoshikawa (2015), PloS One 10 (6), e0128565. 440.
Tsuchiya, M., A. Giuliani, M. Hashimoto, J. Erenpreisa, and Werner, G. (2007), Biosystems 90 (2), 496.
K. Yoshikawa (2016), PloS One 11 (12), e0167912. West, B. J. (2010), Front. Physiol. 1, 12.
Turcotte, D. (1999), Rep. Prog. Phys. 62 (10), 1377. West, G. (2017), Scale (Orion Publishing Group, Limited).
Turing, A. M. (1950), Mind 59 (236), 433. West, G. B., J. H. Brown, and B. J. Enquist (1997), Science
Turing, A. M. (1952), Philos. Trans. R. Soc. London, Ser. B 276 (5309), 122.
237 (641), 37. Williams-Garcia, R., J. M. Beggs, and G. Ortiz (2017), EPL
Tyrcha, J., Y. Roudi, M. Marsili, and J. Hertz (2013), 119 (1), 18003.
Journal of Statistical Mechanics: Theory and Experiment Williams-Garcı́a, R. V., M. Moore, J. M. Beggs, and G. Ortiz
2013 (03), P03005. (2014), Phys. Rev. E 90 (6), 062714.
Tyson, J. J., K. C. Chen, and B. Novak (2003), Current Willinger, W., D. Alderson, J. C. Doyle, and L. Li (2004), in
opinion in cell biology 15 (2), 221. Proc. Winter Simulation Conf., Vol. 1 (IEEE).
Uhlig, M., A. Levina, T. Geisel, and M. Herrmann (2013), Wilson, K. (1979), Sci. Am. 241 (2), 158.
Front. Comput. Neurosci. 7 (87). Wilson, K., and J. Kogut (1974), Phys. Rep. 12 (2), 75.
Van Kampen, N. G. (1992), Stochastic processes in physics Wolf, D. M., V. V. Vazirani, and A. P. Arkin (2005), J. Theor.
and chemistry, Vol. 1 (Elsevier). Biol. 234 (2), 227.
Vanni, F., M. Luković, and P. Grigolini (2011), Phys. Rev. Wolfram, S. (2002), A new kind of science (Wolfram media
Lett. 107 (7), 078103. Champaign, IL).
Varela, F., J.-P. Lachaux, E. Rodriguez, and J. Martinerie Yamaguchi, H., K. Kawaguchi, and T. Sagawa (2017), Phys.
(2001), Nature Rev. Neurosci. 2 (4), 229. Rev. E 96 (1), 012401.
Vattay, G., D. Salahub, I. Csabai, A. Nassimi, and S. A. Yang, H., W. L. Shew, R. Roy, and D. Plenz (2012), J. Neu-
Kaufmann (2015), J. of Phys.: Conference Series 626 (1), rosci. 32 (3), 1061.
012023. Yu, S., A. Klaus, H. Yang, and D. Plenz (2014), PloS One
Veatch, S. L., P. Cicuta, P. Sengupta, A. Honerkamp-Smith, 9 (6), e99761.
D. Holowka, and B. Baird (2008), ACS Chem. Biol. 3 (5), Yu, S., H. Yang, H. Nakahara, G. S. Santos, D. Nikolić, and
287. D. Plenz (2011), J. Neurosci. 31 (48), 17514.
Veatch, S. L., O. Soubias, S. L. Keller, and K. Gawrisch Yule, G. U. (1925), Philos. Trans. R. Soc. London Ser.B 213,
(2007), Proc. Natl. Acad. Sci. USA. 104 (45), 17650. 21.
Veening, J.-W., W. K. Smits, and O. P. Kuipers (2008), Yuste, R., J. N. MacLean, J. Smith, and A. Lansner (2005),
Annu. Rev. Microbiol. 62, 193. Nat. Rev. Neurosci. 6, 477.
Vespignani, A., R. Dickman, M. A. Muñoz, and S. Zapperi Zamponi, N., E. Zamponi, S. A. Cannas, O. V. Billoni, P. R.
(1998), Phys. Rev. Lett. 81 (25), 5676. Helguera, and D. R. Chialvo (2018), Sci. Rep. 8 (1), 363.
Vespignani, A., R. Dickman, M. A. Muñoz, and S. Zapperi Zapperi, S., K. B. Lauritsen, and H. E. Stanley (1995), Phys.
(2000), Phys. Rev. E 62 (4), 4564. Rev. Lett. 75 (22), 4071.
Vicsek, T., A. Czirók, E. Ben-Jacob, I. Cohen, and Zipf, G. K. (1949), Human behavior and the principle of least
O. Shochet (1995), Phys. Rev. Lett. 75 (6), 1226. effort (Addison-Wesley, Cambridge).
Vicsek, T., and A. Zafeiris (2012), Phys. Rep. 517 (3), 71.
Villa Martı́n, P., J. A. Bonachela, and M. A. Muñoz (2014),
Phys. Rev. E 89 (1), 012145.