Francesco Sorrentino Department of Mechanical Engineering



Similar documents
DATA ANALYSIS II. Matrix Algorithms

SALEM COMMUNITY COLLEGE Carneys Point, New Jersey COURSE SYLLABUS COVER SHEET. Action Taken (Please Check One) New Course Initiated

Rank one SVD: un algorithm pour la visualisation d une matrice non négative

NETZCOPE - a tool to analyze and display complex R&D collaboration networks

Linear Algebra Review. Vectors

Chapter 7. Lyapunov Exponents. 7.1 Maps

Part 2: Community Detection

x = + x 2 + x

Similarity and Diagonalization. Similar Matrices

Social Media Mining. Network Measures

Classification of Cartan matrices

USE OF EIGENVALUES AND EIGENVECTORS TO ANALYZE BIPARTIVITY OF NETWORK GRAPHS

Chapter 6. Orthogonality

Data Mining: Algorithms and Applications Matrix Math Review

MAT 242 Test 2 SOLUTIONS, FORM T

Nonlinear Iterative Partial Least Squares Method

Math 550 Notes. Chapter 7. Jesse Crawford. Department of Mathematics Tarleton State University. Fall 2010

Component Ordering in Independent Component Analysis Based on Data Power

Walk-Based Centrality and Communicability Measures for Network Analysis

is in plane V. However, it may be more convenient to introduce a plane coordinate system in V.

Similar matrices and Jordan form

Linear Algebra and TI 89

ON THE DEGREES OF FREEDOM OF SIGNALS ON GRAPHS. Mikhail Tsitsvero and Sergio Barbarossa

6. Cholesky factorization

Network (Tree) Topology Inference Based on Prüfer Sequence

Subspace Analysis and Optimization for AAM Based Face Alignment

Comparison of Non-linear Dimensionality Reduction Techniques for Classification with Gene Expression Microarray Data

Name: Section Registered In:

SECTIONS NOTES ON GRAPH THEORY NOTATION AND ITS USE IN THE STUDY OF SPARSE SYMMETRIC MATRICES

MATH 423 Linear Algebra II Lecture 38: Generalized eigenvectors. Jordan canonical form (continued).

University of Lille I PC first year list of exercises n 7. Review

Analysis of Internet Topologies: A Historical View

5.04 Principles of Inorganic Chemistry II

NEW VERSION OF DECISION SUPPORT SYSTEM FOR EVALUATING TAKEOVER BIDS IN PRIVATIZATION OF THE PUBLIC ENTERPRISES AND SERVICES

LINEAR ALGEBRA. September 23, 2010

USING SPECTRAL RADIUS RATIO FOR NODE DEGREE TO ANALYZE THE EVOLUTION OF SCALE- FREE NETWORKS AND SMALL-WORLD NETWORKS

State of Stress at Point

December 4, 2013 MATH 171 BASIC LINEAR ALGEBRA B. KITCHENS

Hello, my name is Olga Michasova and I present the work The generalized model of economic growth with human capital accumulation.

1 Sets and Set Notation.

by the matrix A results in a vector which is a reflection of the given

Analysis of Internet Topologies

System Identification for Acoustic Comms.:

A Simple Feature Extraction Technique of a Pattern By Hopfield Network

Review Jeopardy. Blue vs. Orange. Review Jeopardy

Lecture 5: Singular Value Decomposition SVD (1)

APPM4720/5720: Fast algorithms for big data. Gunnar Martinsson The University of Colorado at Boulder

Transportation Polytopes: a Twenty year Update

SHARP BOUNDS FOR THE SUM OF THE SQUARES OF THE DEGREES OF A GRAPH

Computation of crystal growth. using sharp interface methods

13 MATH FACTS a = The elements of a vector have a graphical interpretation, which is particularly easy to see in two or three dimensions.

Nimble Algorithms for Cloud Computing. Ravi Kannan, Santosh Vempala and David Woodruff

Applied Linear Algebra I Review page 1

How To Understand The Network Of A Network

Using the Theory of Reals in. Analyzing Continuous and Hybrid Systems

Introduction to Matrix Algebra

Segmentation & Clustering

Dynamical Systems Analysis II: Evaluating Stability, Eigenvalues

BOOLEAN CONSENSUS FOR SOCIETIES OF ROBOTS

A linear combination is a sum of scalars times quantities. Such expressions arise quite frequently and have the form

Orthogonal Diagonalization of Symmetric Matrices

Social Networks and Social Media

NOV /II. 1. Let f(z) = sin z, z C. Then f(z) : 3. Let the sequence {a n } be given. (A) is bounded in the complex plane

MATRIX ALGEBRA AND SYSTEMS OF EQUATIONS

How To Understand And Solve A Linear Programming Problem

Mining Social-Network Graphs

Direct Methods for Solving Linear Systems. Matrix Factorization

Mathematics Course 111: Algebra I Part IV: Vector Spaces

Lecture 1: Schur s Unitary Triangularization Theorem

Yousef Saad University of Minnesota Computer Science and Engineering. CRM Montreal - April 30, 2008

Notes on Symmetric Matrices

SYMMETRIC EIGENFACES MILI I. SHAH

Section Inner Products and Norms

Matrix Representations of Linear Transformations and Changes of Coordinates

[1] Diagonal factorization

Split Nonthreshold Laplacian Integral Graphs

Plate waves in phononic crystals slabs

Analysis of Algorithms, I

An Evaluation Model for Determining Insurance Policy Using AHP and Fuzzy Logic: Case Studies of Life and Annuity Insurances

Feature Point Selection using Structural Graph Matching for MLS based Image Registration

1 Introduction to Matrices

Understanding and Applying Kalman Filtering

3. Let A and B be two n n orthogonal matrices. Then prove that AB and BA are both orthogonal matrices. Prove a similar result for unitary matrices.

A Tutorial on Spectral Clustering

Group Theory. 1 Cartan Subalgebra and the Roots. November 23, Cartan Subalgebra. 1.2 Root system

4: EIGENVALUES, EIGENVECTORS, DIAGONALIZATION

Distributed network topology reconstruction in presence of anonymous nodes

Lecture 15 An Arithmetic Circuit Lowerbound and Flows in Graphs

Convolution. 1D Formula: 2D Formula: Example on the web:

SCAN: A Structural Clustering Algorithm for Networks

Parallel Data Selection Based on Neurodynamic Optimization in the Era of Big Data

UNCOUPLING THE PERRON EIGENVECTOR PROBLEM

8 Square matrices continued: Determinants

Sketch As a Tool for Numerical Linear Algebra

Transcription:

Master stability function approaches to analyze stability of the synchronous evolution for hypernetworks and of synchronized clusters for networks with symmetries Francesco Sorrentino Department of Mechanical Engineering 203 IEEE International Workshop on Complex Systems and Networks, Vancouver, December the 2 th 203

Synchronization of Networks x ( t) F( x ( t)) x ( t) F( x2( 2 t )) x i ( t) F( x ( t)) i x N ( t) F( x ( t)) N The systems are characterized by the same What is the effect of the network structure on the dynamics (possibly chaotic), when uncoupled. dynamics? How do faults affect the network dynamics?

Synchronization of Complex Networks The dynamics of each node in the network can be described by the following equation:

Synchronization of Complex Networks The dynamics of each node in the network can be described by the following equation: x i ( t) F( x ( t)) i Individual dynamics (chaotic)

Synchronization of Complex Networks The dynamics of each node in the network can be described by the following equation: x i ( t) F( x ( t)) i N j A ij t Hx t H x j i Coupling term The matrix A{ A ij } is the network (weighted) adjacency matrix. The parameter measures the strengths of the network connections. The function H is an output function at each node. In principle H can be any function.

Synchronization of Complex Networks The dynamics of each node in the network can be described by the following equation: is the Laplacian matrix N j j ij i i t x H L t x F t x )) ( ( ) ( { ij } L L N N N N N d A A A d A A A d L 2 2 2 2 2 N j d i A ij

Synchronization of Complex Networks The dynamics of each node in the network can be described by the following equation: 2 3 4 is the Laplacian matrix N j j ij i i t x H L t x F t x )) ( ( ) ( { ij } L L 2 0 2 0 3 0 0 L L0 N N 2 0 The network equations admit a synchronous solution: t x t x t x t x s N 2. t F x t x s s where

Master Stability Function Reduced Linearized equation: [L. M. Pecora, T. L. Carroll, 98] s i s The Master Stability Function (MSF) approach studies transversal stability of the synchronous trajectory The sync solution is linearly stable iff the MSF is negative. Depending on the individual nodes dynamics, we can have three different kinds of MSF. Maximum Transverse Lyapunov exponent Λ(ν) DF( x ( t)) DH( x ( t)) x, x σλ 2,...,σλ N i

Parallel and transversal perturbations By construction there is always one eigenvalue 0 This eigenvalue is associated with the eigenvector [,,,], that is parallel to the synchronization manifold In order to evaluate stability, we are only concerned with transversal perturbations

Other Master Stability Functions Similar approaches have been used to analyze other types of synchronization. EXAMPLES )Pinning control of networks. F. Sorrentino, M. di Bernardo, F. Garofalo, and G. Chen, Phys. Rev.E, 75, 4, 04603 (2007).

Other Master Stability Functions Similar approaches have been used to analyze other types of synchronization. EXAMPLES 2) Network Synchronization of Groups F. Sorrentino and Edward Ott, Network Synchronization of Groups, Phys. Rev. E, 76, 0564 (2007).

Other Master Stability Functions Similar approaches have been used to analyze other types of synchronization. EXAMPLES 3) Adaptive synchronization of a time-evolving dynamical networks/ sensor network application F. Sorrentino, G. Barlev, A. B. Cohen, and E. Ott, Chaos, 20, 0303 (200) A. Cohen, B. Ravoori, F. Sorrentino, T. Murphy, E. Ott, and R. Roy, Chaos, 20, 04342 (200).

Other Master Stability Functions Similar approaches have been used to analyze other types of synchronization. EXAMPLES 4) Networks with couplings of different types F. Sorrentino, Synchronization of hypernetworks of coupled dynamical systems, New Journal of Physics, 4, 033035 (202) D. Irving and F. Sorrentino, Phys. Rev. E, Phys. Rev. E, 86, 05602 (202).

Motivation: chemical synapses and electrical gap junctions How does synchronization arises in this networks? Is a low dimensional approach still possible?

Mathematical formulation Two different coupling mechanisms: Stability?

Stability analysis Problem: In general, it is not possible to simultaneously diagonalize two matrices Question: Are there conditions under which these equations can be reduced in this form? The answer is yes in three cases.

Case I The matrices L A and L B commute Question: What are the graph properties for the Laplacians to commute?

Case II One of the two networks is fully connected and unweighted:

Case III One of the two networks is such that all the links originating from the same node have equal weights:

SBD approach What if none of these conditions (I-III) is satisfied? Danny Irving and I proposed an alternative approach based on simultaneous block-diagonalization (SBD) of matrices PROBLEM: Given the set of N-square real matrices = {L (),L (2),...,L (M) }, find the finest simultaneous block diagonalization (SBD) of. APPROACH: Find an invertible matrix P, such that P - L (i) P= j=,..n B j i and the dimensions of the blocks are minimal.

Method by Maehara & Murota (i) Let O (i) be the N 2 -matrix O (i) = I N L(i) L(i) I N. (ii) Construct the matrix S = Σ i O (i)t O (i) (iii) Let y be any N 2 -vector in the null subspace of the matrix S. Let y = [u T,u 2T,...,u NT ] T. (iv) Construct the matrix U=[u,u 2,,u N ]. (v) Output the matrix P whose columns are the eigenvectors of U. This procedure provides the finest simultaneous block diagonalization it provides the maximum reduction of the sync stability problem

Example I

Example II

Example III: undirected unweighted 3-motifs

Cluster synchronization in networks with symmetries L. Pecora, F. Sorrentino, A. Hagerstrom, T. Murphy, R. Roy Consider the following general equations: Where now the matrix A is completely arbitrary, e.g., non constant-row-sum QUESTION: Can synchronization be achieved? ANSWER: These equations are compatible with cluster synchronization: M synchronized motions {s,, s M }, where the clusters are determined by the network structure

Symmetries and Clusters Three -node random networks with the same number of edges These three networks are characterized by different numbers of symmetries (automorphisms) Equivalence relation partition into maximal disjoint sets of nodes

Sync solutions Graph symmetries Dynamical solutions: nodes that belong to the same cluster can synchronize, nodes that belong to different clusters cannot. The symmetries of the network form a group G. Each symmetry of the group can be described by a permutation matrix R g that re-orders the nodes in a way that leaves the dynamical equations unchanged (i.e., each R g commutes with A). Computation of the graph automorphisms enables us to find the possible cluster sync solutions corresponding to a given network structure - Are these solutions stable?

Sync solutions - stability Group theory provides a powerful way to transform the variational equations to a new coordinate system (the irreducible representation) in which the transformed coupling matrix B = TAT - has a block diagonal form that matches the cluster structure.

Experimental setup The dynamical oscillators that form the network are realized as square patches of pixels selected from a 32 x 32 tiling of the SLM array.

Stability results and Isolated Desynchronization

Application to power networks Geographical diagram of the Nepal Power Grid Network

Conclusions We are pushing the master stability function approach [Pecora and Carroll] to its limits In the case of networks with multiple type of connections, the stability problem can be reduced in a low-dimensional form by using SBD This guarantees that the problem is reduced in the lowest possible dimensional form In the case of arbitrary adjacency matrices, clusters may arise that correspond to a partition of the nodes based on the network symmetries Stability of each cluster can be studied independently from other clusters unless clusters are intertwined As a consequence we can experience isolated desynchronization of the cluster solutions Possible applications to networks of neurons and power grids