## Markov Chain Methods For Analyzing Complex Transport Networks (2008)

### BibTeX

@MISC{Volchenkov08markovchain,

author = {D. Volchenkov and Ph. Blanchard},

title = {Markov Chain Methods For Analyzing Complex Transport Networks},

year = {2008}

}

### OpenURL

### Abstract

We have developed a steady state theory of complex transport networks used to model the flow of commodity, information, viruses, opinions, or traffic. Our approach is based on the use of the Markov chains defined on the graph representations of transport networks allowing for the effective network design, network performance evaluation, embedding, partitioning, and network fault tolerance analysis. Random walks embed graphs into Euclidean space in which distances and angles acquire a clear statistical interpretation. Being defined on the dual graph representations of transport networks random walks describe the equilibrium configurations of not random commodity flows on primary graphs. This theory unifies many network concepts into one framework and can also be elegantly extended to describe networks represented by directed graphs and multiple interacting networks.

### Citations

1282 |
Spline Models for Observational Data
- Wahba
- 1990
(Show Context)
Citation Context ... kernel (25) discriminate large eigenvalues that is preferable while performing estimation on the graph since it biases the estimate towards functions which vary little over the graph components [24]-=-=[25]-=-. The distribution of eigenvalues λk ∈ [0, 2[ of the self-adjoint Laplace operator (15) defined on the undirected graph G are characterized by the spectral density function, N∑ ρ(λ) = δ(λ − λk). (27) ... |

835 |
Applied Multiple Regression~Correlation Analysis for the Behavioral Sciences. Lawrence Erlbaum Associates
- Cohen, Cohen
(Show Context)
Citation Context ... (20) Covii Covjj instead of the covariance matrix Cov. We emphasize that the diagonal elements of (20) equal 1, while the off-diagonal elements are the Pearson’s coefficients of linear correlations, =-=[20]-=-. Let U be the orthonormal matrix which contains the eigenvectors {uk}, k = 1, . . . , n − 1, of the covariance (or correlation) matrix as the row vectors. These vectors form the orthogonal basis of t... |

495 | Semi-Supervised Learning Using Gaussian Fields and Harmonic Functions
- Zhu, Ghahramani, et al.
(Show Context)
Citation Context ...j), (25) and solve the diffusion equation defined on a connected, undirected graph G: ∫ ∞ ∫ ∞ G = Kβ dβ = exp(−βL) dβ = L −1 , (26) 0 0 9with initial conditions being a source at node i ∈ V at β = 0 =-=[23]-=-. The function (Kβ) ij describes the expected number of particles that would accumulate at vertex j after a given amount of time if they have been injected at vertex i and diffused through the graph a... |

375 | Reversible Markov chains and random walks on graphs, draft at http://www.stat.berkeley.edu/~aldous/RWG/book.html
- Aldous, Fill
- 2002
(Show Context)
Citation Context ...ere T ⊤ is the adjoint operator. The orthonormal ordered set of real eigenvectors ψi, i = 1 . . .N, of the symmetric operator ̂ T defines a basis in H. In the theory of random walks defined on graphs =-=[15, 14]-=- and in spectral graph theory [16], the properties of graphs are studied in connection with the eigenvalues and eigenvectors of self-adjoint operators defined on them. In particular, the symmetric tra... |

300 |
Probability theory
- Loève
- 1955
(Show Context)
Citation Context ...oduced in [36, 37]) for the transition operator: Pµ = N∑ s=1 Λsϕsiψsi ≡ N∑ Λs |ϕs〉〈ψs| , (51) where 0 ≤ Λ1 ≤ . . .ΛN and the functions ϕk ∈ H+ and ψk ∈ H− are related by the Karhunen-Loève dispersion =-=[38, 39]-=-, satisfying the orthogonality condition: s=1 Pµϕs = Λsψs (52) 〈ϕs, ϕs ′〉H+ = 〈ψs ′, ψs〉H− = δs ′ s. (53) Since the operators Pµ and P ⊤ µ act between different Hilbert spaces, it is insufficient to s... |

170 | Kernels and regularization on graphs
- Smola, Kondor
- 2003
(Show Context)
Citation Context ...of probability conservation considered in the theory of random walks on graphs are among them. Another example is given by the Laplace operators satisfying the mean value property (mass conservation) =-=[13]-=-. Markov’s operators on Hilbert space form the natural language of complex networks theory. Being defined on connected undirected graphs, a Markov transition operator T has a unique equilibrium state ... |

136 |
The Social Logic of Space
- Hillier, Hanson
- 1984
(Show Context)
Citation Context ...extensively investigated within the concept of space syntax, a theory developed in the late 1970s, that seeks to reveal the mutual effects of complex spatial urban networks on society and vice versa, =-=[10, 11]-=-. Spatial perception that shapes peoples understanding of how a place is organized determines eventually the pattern of local movement which is predicted by the space syntax method with surprising acc... |

128 | Diffusion kernels on graphs and other discrete structures
- Kondor, Lafferty
- 2002
(Show Context)
Citation Context ... role of the Green functions describing long-range interactions between different diffusion modes induced by the graph structure (the main graph components). The Green function recently introduced in =-=[22]-=- is specified by the parameter β > 0, (Kβ) ij = exp(−βLi,j), (25) and solve the diffusion equation defined on a connected, undirected graph G: ∫ ∞ ∫ ∞ G = Kβ dβ = exp(−βL) dβ = L −1 , (26) 0 0 9with ... |

87 |
Über ein Paradoxon aus der Verkehrsplanung
- Braess
- 1969
(Show Context)
Citation Context ...is to determine the structure and properties of transport networks that are important for the emergence of complex flow patterns of vehicles (or people) through the network such as the Braess paradox =-=[1]-=-. This counter-intuitive phenomenon occurs when adding more resources to a transportation network (say, a new road or a bridge) deteriorates the quality of traffic by creating worse delays for the dri... |

69 |
Space is the Machine. A configurational theory of architecture
- Hillier
- 1996
(Show Context)
Citation Context ...extensively investigated within the concept of space syntax, a theory developed in the late 1970s, that seeks to reveal the mutual effects of complex spatial urban networks on society and vice versa, =-=[10, 11]-=-. Spatial perception that shapes peoples understanding of how a place is organized determines eventually the pattern of local movement which is predicted by the space syntax method with surprising acc... |

61 | Diffusion maps, spectral clustering and reaction coordinate of dynamical systems
- Nadler, Lafon, et al.
- 2006
(Show Context)
Citation Context ...ll agents participating in transport processes. Being defined on a graph, the time evolution of such a system can be described by just a few dynamically relevant variables called reaction coordinates =-=[21]-=-. Identification of slow variables and dynamically meaningful reaction coordinates that capture the long time evolution of transport systems is among the most important problems of transport network a... |

59 |
Principal Component Analysis, 2nd edition
- Jolliffe
- 2002
(Show Context)
Citation Context ...he reducing size of a data set. It is achieved by the optimal linear transformation retaining the subspace that has largest variance (a lower-order principal component) and ignoring higher-order ones =-=[18, 19]-=-. Given an operator S self-adjoint with respect to the measure m defined on a connected undirected graph G, it is well known that the eigenvectors of the symmetric matrix S form an ordered orthonormal... |

14 |
Space Syntax and Spatial Cognition or Why the Axial Line? Environment and Behavior
- Penn
- 2003
(Show Context)
Citation Context ...ial perception that shapes peoples understanding of how a place is organized determines eventually the pattern of local movement which is predicted by the space syntax method with surprising accuracy =-=[12]-=-. 2.2 Euclidean space of undirected graphs associated to random walks Any graph representation naturally arises as the outcome of a categorization, when we abstract a real world system by eliminating ... |

12 |
Theory of Reproducing Kernels and its
- Saitoh
- 1988
(Show Context)
Citation Context ...usion kernel (25) discriminate large eigenvalues that is preferable while performing estimation on the graph since it biases the estimate towards functions which vary little over the graph components =-=[24]-=--[25]. The distribution of eigenvalues λk ∈ [0, 2[ of the self-adjoint Laplace operator (15) defined on the undirected graph G are characterized by the spectral density function, N∑ ρ(λ) = δ(λ − λk). ... |

11 |
Lecture notes on spectral graph theory
- Chung
- 1997
(Show Context)
Citation Context ...rthonormal ordered set of real eigenvectors ψi, i = 1 . . .N, of the symmetric operator ̂ T defines a basis in H. In the theory of random walks defined on graphs [15, 14] and in spectral graph theory =-=[16]-=-, the properties of graphs are studied in connection with the eigenvalues and eigenvectors of self-adjoint operators defined on them. In particular, the symmetric transition operator of the random wal... |

8 | Minimal Volume Entropy on Graphs
- Lim
- 2005
(Show Context)
Citation Context ...ile the volume of G is defined as the sum Vol(G) = 1 2 ∑ ℓ(e). The volume entropy h is defined to be the exponential growth of the balls in a universal covering tree for G with the lifted metric, [3]-=-=[6]-=-. e∈E 2The degree of a node v ∈ V is the number of its neighbors in G, deg(v) = kv. It has been shown in [6] that among all undirected connected graphs of normalized volume, Vol(G) = 1, which are not... |

8 |
Mixing of Random Walks and Other Diffusions on a Graph. Surveys in combinatorics
- Lovász, Winkler
- 1995
(Show Context)
Citation Context ...s before it had originated from a node j. The matrix element (40) is zero, provided there is no directed path from j to i in ⃗ G. If ⃗ G is strongly connected and aperiodic, the random walk converges =-=[32, 31, 33, 34]-=- to the only stationary distribution π given by the Perron vector, πP = π. If the graph ⃗ G is periodic, then the transition probability matrix P can have more than one eigenvalue with absolute value ... |

7 |
What if they closed the 42 nd Street and nobody noticed?”, The
- Kolata
(Show Context)
Citation Context ...iorates the quality of traffic by creating worse delays for the drivers, rather than alleviate it. The Braess paradox has been observed in the street vehicular traffic of New York City and Stuttgart, =-=[2]-=-. The paper is partitioned into three sections. In Sec. 2, we demonstrate that Markov chains arise naturally in the problem of network equilibriums. Random walks embed connected undirected graphs into... |

4 |
Environment and Behavior
- Kuipers
- 1982
(Show Context)
Citation Context ...Euclidean one, but are rather related to the perceiving of the vista spaces (streets and squares) as single units and of the understanding of the topological relationships between these vista spaces, =-=[9]-=-. Decomposition of city space into a complete set of intersecting vista spaces produces a spatial network which we call the dual graph representation of a city. Therein, the relations between streets ... |

3 |
Introduction to Statistical
- Fukunaga
(Show Context)
Citation Context ...he reducing size of a data set. It is achieved by the optimal linear transformation retaining the subspace that has largest variance (a lower-order principal component) and ignoring higher-order ones =-=[18, 19]-=-. Given an operator S self-adjoint with respect to the measure m defined on a connected undirected graph G, it is well known that the eigenvectors of the symmetric matrix S form an ordered orthonormal... |

2 |
Local structure of directed networks,” eprint arXiv:0707.4084 [physics
- Bianconi, Gulbahce, et al.
- 2007
(Show Context)
Citation Context ...the diameters of directed networks can essentially exceeds that the one corresponding to the same networks regarded as undirected. A recent investigation into the local structure of directed networks =-=[29]-=- shows that directed networks often have very few short loops as compared to random models usual in contemporary theory 14of complex systems. In undirected networks, the high density of short loops (... |

2 |
renewal performance in complex networks Case studies
- Haffner, Elsinga, et al.
- 2007
(Show Context)
Citation Context ...ions. From Euler’s time, urban design and townscape studies were the sources of inspiration for the network analysis and graph theory. It is common now that networks are the reality of urban renewals =-=[41]-=-. Flows of pedestrians and vehicles through a city are dependent on one another and that requests for organizing them in a network setting. The networking is structurally contagious. In order to be ab... |

1 |
Nature 393(6684
- Watts, Strogatz
- 1998
(Show Context)
Citation Context ...ontemporary theory 14of complex systems. In undirected networks, the high density of short loops (high clustering coefficient) together with small graph diameter gives rise to the small-world effect =-=[30]-=-. In directed networks, the correlation between number of incoming and outgoing edges modulates the expected number of short loops. In particular, it has been demonstrated in [29] that if the values k... |

1 |
Electronic
- Butler
- 2007
(Show Context)
Citation Context ...1/2) , (41) ij where π is a diagonal matrix with entries πi. The matrix (41) is symmetric and has real non-negative eigenvalues 0 = λ1 < λ2 ≤ . . . ≤ λN and real eigenvectors. It can be easily proven =-=[35]-=- that the Laplace operator (41) defined on the aperiodic strongly connected graph ⃗ G is equivalent to the Laplace operator defined on a symmetric undirected weighted graph Gw on the same vertex set w... |

1 |
Spatio-temporal symmetries, Preprint CPT-93/P.2923, Centre de Physique Theorique
- Aubry, Lima
- 1993
(Show Context)
Citation Context ... = 〈ψs ′, ψs〉H− = δs ′ s. (53) Since the operators Pµ and P ⊤ µ act between different Hilbert spaces, it is insufficient to solve just one equation in order to determine their eigenvectors ϕs and ψs, =-=[40]-=-. Instead, two equations have to be solved, { Pµϕ = Λψ, or, equivalently, ( 0 Pµ P⊤ µ 0 P⊤ µ ψ = Λϕ, ) ( ϕ ψ ) ( ϕ = Λ ψ (54) ) . (55) The latter equation allows for a graph-theoretical interpretation... |