Klaus M. Frahm

Google matrix analysis of directed networks

In past ten years, modern societies developed enormous communication and social networks. Their classification and information retrieval processing become a formidable task for the society. Due to the rapid growth of World Wide Web, social and communication networks, new mathematical methods have been invented to characterize the properties of these networks on a more detailed and precise level. Various search engines are essentially using such methods. It is highly important to develop new tools to classify and rank enormous amount of network information in a way adapted to internal network structures and characteristics. This review describes the Google matrix analysis of directed complex networks demonstrating its efficiency on various examples including World Wide Web, Wikipedia, software architecture, world trade, social and citation networks, brain neural networks, DNA sequences and Ulam networks. The analytical and numerical matrix methods used in this analysis originate from the fields of Markov chains, quantum chaos and Random Matrix theory.

Ulam method for the Chirikov standard map

AbstractWe introduce a generalized Ulam method andnapply it to symplectic dynamical maps with a divided phase space. Our extensive numerical studies based on the Arnoldi method show that the Ulam approximant of the Perron-Frobenius operator on a chaotic component converges to a continuous limit. Typically, in this regime the spectrum of relaxation modes is characterized by a power law decay for small relaxation rates. Our numerical data show that the exponent of this decay is approximately equal to the exponent of Poincaré recurrences in such systems. The eigenmodes show links with trajectories sticking around stability islands.

Spectral properties of Google matrix of Wikipedia and other networks

We study the properties of eigenvalues and eigenvectors of the Google matrix of the Wikipedia articles hyperlink network and other real networks. With the help of the Arnoldi method, we analyze the distribution of eigenvalues in the complex plane and show that eigenstates with significant eigenvalue modulus are located on well defined network communities. We also show that the correlator between PageRank and CheiRank vectors distinguishes different organizations of information flow on BBC and Le Monde web sites.

Time evolution of Wikipedia network ranking

AbstractnWe study the time evolution of ranking and spectral properties of the Google matrix ofnEnglish Wikipedia hyperlink network during years 2003–2011. The statistical properties ofnranking of Wikipedia articles via PageRank and CheiRank probabilities, as well as thenmatrix spectrum, are shown to be stabilized for 2007–2011. A special emphasis is done onnranking of Wikipedia personalities and universities. We show that PageRank selection isndominated by politicians while 2DRank, which combines PageRank and CheiRank, gives morenaccent on personalities of arts. The Wikipedia PageRank of universities recovers 80% ofntop universities of Shanghai ranking during the considered time period.n

Universal emergence of PageRank

The PageRank algorithm enables us to rank the nodes of a network through a specific eigenvector of the Google matrix, using a damping parameter α ∈ ]0, 1[. Using extensive numerical simulations of large web networks, with a special accent on British University networks, we determine numerically and analytically the universal features of the PageRank vector at its emergence when α → 1. The whole network can be divided into a core part and a group of invariant subspaces. For α → 1, PageRank converges to a universal power-law distribution on the invariant subspaces whose size distribution also follows a universal power law. The convergence of PageRank at α → 1 is controlled by eigenvalues of the core part of the Google matrix, which are extremely close to unity, leading to large relaxation times as, for example, in spin glasses.

Google matrix of Twitter

We construct the Google matrix of the entire Twitter network, dated by July 2009, and analyze its spectrum and eigenstate properties including the PageRank and CheiRank vectors and 2DRanking of all nodes. Our studies show much stronger inter-connectivity between top PageRank nodes for the Twitter network compared to the networks of Wikipedia and British Universities studied previously. Our analysis allows to locate the top Twitter users which control the information flow on the network. We argue that this small fraction of the whole number of users, which can be viewed as the social network elite, plays the dominant role in the process of opinion formation on the network.

Freed by interaction kinetic states in the Harper model

We study the problem of two interacting particles in a one-dimensional quasiperiodic lattice of the Harper model. We show that a short or long range interaction between particles leads to emergence of delocalized pairs in the non-interacting localized phase. The properties of these freed by interaction kinetic states (FIKS) are analyzed numerically including the advanced Arnoldi method. We find that the number of sites populated by FIKS pairs grows algebraically with the system size with the maximal exponent b = 1, up to a largest lattice size N = 10 946 reached in our numerical simulations, thus corresponding to a complete delocalization of pairs. For delocalized FIKS pairs the spectral properties of such quasiperiodic operators represent a deep mathematical problem. We argue that FIKS pairs can be detected in the framework of recent cold atom experiments [M. Schreiber et al., Science 349, 842 (2015)] by a simple setup modification. We also discuss possible implications of FIKS pairs for electron transport in the regime of charge-density wave and high Tc superconductivity.

PageRank of integers

We build up a directed network tracing links from a given integer to its divisors and analyze the properties of the Google matrix of this network. The PageRank vector of this matrix is computed numerically and it is shown that its probability is inversely proportional to the PageRank index thus being similar to the Zipf law and the dependence established for the World Wide Web. The spectrum of the Google matrix of integers is characterized by a large gap and a relatively small number of nonzero eigenvalues. A simple semi-analytical expression for the PageRank of integers is derived that allows to find this vector for matrices of billion size. This network provides a new PageRank order of integers.

Wikipedia mining of hidden links between political leaders

AbstractnWe describe a new method of reduced Google matrix which allows to establish direct andnhidden links between a subset of nodes of a large directed network. This approach usesnparallels with quantum scattering theory, developed for processes in nuclear andnmesoscopic physics and quantum chaos. The method is applied to the Wikipedia networks inndifferent language editions analyzing several groups of political leaders of USA, UK,nGermany, France, Russia and G20. We demonstrate that this approach allows to recovernreliably direct and hidden links among political leaders. We argue that the reduced Googlenmatrix method can form the mathematical basis for studies in social and political sciencesnanalyzing Leader-Members eXchange (LMX).n

Diffusion and localization for the Chirikov typical map.

We consider the classical and quantum properties of the Chirikov typical map, proposed by Boris Chirikov in 1969. This map is obtained from the well-known Chirikov standard map by introducing a finite-number T of random phase-shift angles. These angles induce a random behavior for small time-scales (t<T) and a T -periodic iterated map which is relevant for larger time-scales (t>T) . We identify the classical chaos border k(c) approximately T (-3/2)1 for the kick parameter k and two regimes with diffusive behavior on short and long time scales. The quantum dynamics is characterized by the effect of Chirikov localization (or dynamical localization). We find that the localization length depends in a subtle way on the two classical diffusion constants in the two time-scale regime.