1: \section{Introduction}
2:
3: Distributed synchronization phenomena are the subject of intense
4: research. In the brain, such phenomena are known to occur at different
5: scales, and are heavily studied at both the anatomical and
6: computational levels. In particular, synchronization has been
7: proposed as a general principle for temporal binding of multisensory
8: data~\cite{SinGra,Grossberg,Llinas,Mount,Tononi,Koch,NieEbi}, and as a
9: mechanism for perceptual grouping~\cite{YazGro}, neural
10: computation~\cite{Braitenberg,Brody,WangSlo5} and neural
11: communication~\cite{Kandel,IzhiDes,Schnitzler,Schoffelen}. Similar
12: mathematical models describe fish schooling or certain types of
13: phase-transition in physics~\cite{Strogatz}.
14:
15: In an ensemble of dynamical elements, {\it concurrent synchronization}
16: is defined as a regime where the whole system is divided into multiple
17: groups of fully synchronized elements\footnote{In the literature, this
18: phenomenon is often called \emph{poly-}, or \emph{cluster} or
19: \emph{partial} synchronization. However, the last term can also
20: designate a regime where the elements are not fully synchronized but
21: behave coherently~\cite{Strogatz}.}, but elements from different
22: groups are not necessarily synchronized~\cite{Belykh1,Zhang,PogSanNij}
23: and can be of entirely different dynamics~\cite{GolSte}. It can be
24: easily shown that such a regime corresponds to a flow-invariant linear
25: subspace of the global state space. Concurrent synchronization
26: phenomena are likely pervasive in the brain, where multiple
27: ``rhythms'' are known to coexist~\cite{Kandel,Schnitzler}, neurons can
28: exhibit many qualitatively different types of
29: oscillations~\cite{Kandel,Izhi}, and functional models often combine
30: multiple oscillatory dynamics.
31:
32: In this paper, we introduce a simple sufficient condition for a
33: general dynamical system to converge to a flow-invariant subspace. Our
34: analysis is built upon nonlinear contraction
35: theory~\cite{LohSlo,WangSlo}, and thus it inherits many of the
36: theory's features :
37:
38: \begin{itemize}
39: \item global exponential convergence and stability are guaranteed,
40: %, as
41: %opposed to~\cite{Zhang,GolSteTor} where only stability in the
42: %neighborhood of the invariant manifold is discussed,
43: \item convergence rates can be explicitly computed as
44: eigenvalues of well-defined symmetric matrices,
45: \item robustness to variations in dynamics can be easily quantified,
46: \item under simple conditions, convergence to a concurrently
47: synchronized state can be preserved through system combinations.
48: \end{itemize}
49:
50: %Besides being of interest in its own right, this result allows us
51: %to gain insight into synchronization issues and to generalize and
52: %strengthen existing results in the literature. In particular, the role
53: %of symmetries in the network is shown to play a central role.
54:
55: As we shall see, under simple conditions on the coupling strengths,
56: architectural symmetries~\cite{GolSteTor} and/or diffusion-like
57: couplings \emph{create globally stable concurrent synchronization
58: phenomena}. This is illustrated in figure 1, which is very loosely
59: inspired by oscillations in the thalamocortical
60: system~\cite{Llinas,SinGra,Koch,NieEbi,Schnitzler}. Qualitatively,
61: global stability of the concurrent synchronization is in the same
62: sense that an equilibrium point is globally stable -- any initial
63: conditions will lead back to it, in an exponential fashion. But of
64: course it can yield extremely complex, coordinated behaviors.
65:
66: \begin{figure}[ht]
67: \centering
68: \includegraphics[scale=0.6]{fig/thalamus.eps}
69: \caption[]{\small An example of concurrent synchronization.
70: Systems and connections of the same shape (and color) have
71: identical dynamics (except black arrows, which represent arbitrary
72: connections, and dashed arrows, which represent diffusive
73: connections). This paper shows that under simple conditions on the
74: coupling strengths, the group of (green) squares globally exponentially
75: synchronizes (thus providing synchronized input to the outer elements),
76: and so does the group of (yellow) diamonds, regardless of
77: the specific dynamics, connections, or inputs of the other systems.}
78: \label{fig:thalamus}
79: \end{figure}
80:
81: Section 2 recalls key concepts of nonlinear contraction theory and
82: derives a theoretical tool for studying global convergence to a
83: flow-invariant subspace. Section 3 presents the paper's main
84: mathematical results, relating stable concurrent synchronization to
85: coupling structures and flow-invariant subspaces created by symmetries
86: or diffusion-like couplings. Robustnesss of concurrent
87: synchronization to variations in individual dynamics is also
88: quantified, showing in particular how approximate symmetries lead to
89: quasi-synchronization. Section 4, motivated by evolution and
90: development, studies conditions under which concurrent synchronization
91: can be preserved through combinations of multiple concurrently
92: synchronized regimes.
93: %In a neural context, these conditions are
94: %facilitated by the high fan-out of typical neurons.
95: Finally, section 5 discusses potential applications of these results
96: to general questions in systems neuroscience and robotics.
97:
98: %that symmetry is not the sole systematic provider of flow-invariant
99: %subspaces $-$ in particular, symmetry can be broken by diffusive
100: %couplings (mediated e.g. by gap junctions in a neural context) while
101: %preserving concurrent synchronization.
102:
103: %%% Local Variables:
104: %%% mode: latex
105: %%% TeX-master: "neuro"
106: %%% End:
107: