Session G28: Statistical Mechanics of Social Systems

Characterizing production and consumption in Physics

We analyze the entire database of publications in the American Physical Society and generate longitudinal (50 years) citation networks at two different geographical levels. We define the knowledge diffusion proxy and Scientific Production Ranking algorithms to capture the complex nature of citation networks, and to provide a global view of spatial distributions of production and consumption of knowledge in Physics as well as its temporal evolution. Using the knowledge diffusion proxy we identify the key actors in producing and consuming knowledge in Physics as a function of time. The ranking results from the Scientific Production Ranking algorithm allow us to characterize the top countries/cities in the world for Physical sciences. Among all the results, we find that in 50 years major states and cities in US stably rank on the top and have been main knowledge producers, whereas the major European countries, Japan and Russia have greatly improved their their ranking since 1990. Interestingly, we notice that China and Spain as well as major cities in those countries have gradually become major knowledge consumers in the last two decades.   

The possible role of resource requirements and academic career-choice risk on gender differences in publication rate and impact

Many studies demonstrate that there is still a significant gender bias, especially at higher career levels, in many areas including science, technology, engineering, and mathematics (STEM). We investigated field-dependent, gender-specific effects of the selective pressures individuals experience as they pursue a career in academia within seven STEM disciplines. We built a unique database that comprises 437,787 publications authored by 4,292 faculty members at top United States research universities. Our analyses reveal that gender differences in publication rate and impact are discipline-specific. Our results also support two hypotheses. First, the widely-reported lower publication rates of female faculty are correlated with the amount of research resources typically needed in the discipline considered, and thus may be explained by the lower level of institutional support historically received by females. Second, in disciplines where pursuing an academic position incurs greater career risk, female faculty tend to have a greater fraction of higher impact publications than males. Our findings have significant, field-specific, policy implications for achieving diversity at the faculty level within the STEM disciplines.   

Analysis of Science and Technology Trend Based on Word Usage in Digitized Books

Throughout mankind's history, forecasting and predicting future has been a long-lasting interest to our society. Many fortune-tellers have tried to forecast the future by ``divine'' items. Sci-fi writers have also imagined what the future would look like. However most of them have been illogical and unscientific. Meanwhile, scientists have also attempted to discover future trend of science. Many researchers have used quantitative models to study how new ideas are used and spread. Besides the modeling works, in the early 21st century, the rise of data science has provided another prospect of forecasting future. However many studies have focused on very limited set of period or age, due to the limitations of dataset. Hence, many questions still remained unanswered. Fortunately, Google released a new dataset named ``Google N-Gram Dataset.'' This dataset provides us with 5 million words worth of literature dating from 1520 to 2008, and this is nearly 4\% of publications ever printed. With this new time-varying dataset, we studied the spread and development of technologies by searching ``Science and Technology'' related words from 1800 to 2000. By statistical analysis, some general scaling laws were discovered. And finally, we determined factors that strongly affect the lifecycle of a word.   

The Knowledge Economy

Although the study of scientific and citation networks is well developed, the way in which ideas and concepts flow between scientific groups scattered around the world is still an open problem. We take a first step in this direction by using the citation patterns over the course of decades to shed light on how areas and fields in the general area of physics have evolved both temporally and geographically. By geocoding the affiliations associated with each article published by the APS journals to the country level, and by borrowing concepts from the field of economics and international trade we can explore how ideas produced in one country are exported, through citations, to other countries. An objective way of ranking countries based on their contributions to the overall scientific effort is also proposed as well as a map of how the different subfields of Physics are related to each other.   

Power Law Distributions of Patents as Indicators of Innovation

The total number of patents produced by a country (or the number of patents produced per capita) is often used as an indicator for innovation. Such figures however give an overly simplistic measure of innovation within a country. Here we present evidence that the distribution of patents amongst applicants within many countries is well-fitted to a power law distribution with exponents that vary between 1.66 (Japan) and 2.37 (Poland). We suggest that this exponent is a useful new metric for studying innovation. Using simulations based on simple preferential attachment-type rules that generate power laws, we find we can explain some of the variation in exponents between countries, with countries that have larger numbers of patents per applicant generally exhibiting smaller exponents in both the simulated and actual data. Similarly we find that the exponents for most countries are inversely correlated with other indicators of innovation, such as research and development intensity or the ubiquity of export baskets. This suggests that in more advanced economies, which tend to have smaller values of the exponent, a greater proportion of the total number of patents are filed by large companies than in less advanced countries.   

Discrete Lognormal Model as an Unbiased Quantitative Measure of Scientific Performance Based on Empirical Citation Data

Assessing the career performance of scientists has become essential to modern science. Bibliometric indicators, like the h-index are becoming more and more decisive in evaluating grants and approving publication of articles. However, many of the more used indicators can be manipulated or falsified by publishing with very prolific researchers or self-citing papers with a certain number of citations, for instance. Accounting for these factors is possible but it introduces unwanted complexity that drives us further from the purpose of the indicator: to represent in a clear way the prestige and importance of a given scientist. Here we try to overcome this challenge. We used Thompson Reuter's Web of Science database and analyzed all the papers published until 2000 by $\sim$1500 researchers in the top 30 departments of seven scientific fields. We find that over 97\% of them have a citation distribution that is consistent with a discrete lognormal model. This suggests that our model can be used to accurately predict the performance of a researcher. Furthermore, this predictor does not depend on the individual number of publications and is not easily ``gamed'' on.   

Information theory in econophysics: stock market and retirement funds

Information theory can help to recognize magnetic phase transitions, what can be seen as a way to recognize different regimes. This is achieved by means of zippers specifically designed to compact data in a meaningful way at is the case for compressor wlzip [1]. In the present contribution we first apply wlzip to the Chilean stock market interpreting the compression rates for the files storing the minute variation of the IPSA indicator. Agitated days yield poor compression rates while calm days yield high compressibility. We then correlate this behavior to the value of the five retirement funds related to the Chilean economy. It is found that the covariance between the profitability of the retirement funds and the compressibility of the IPSA values of previous day is high for those funds investing in risky stocks. Surprisingly, there seems to be no great difference among the three riskier funds contrary to what could be expected from the limitations on the portfolio composition established by the laws that regulate this market. \\[4pt] [1] E.E. Vogel, G. Saravia, L.V. Cortez, Physica A 391 (2012) 1591.   

DebtRank a centrality measure for financial systems and beyond

Use of network theory made possible to measure quantitatively many features of social and technological systems. In this spirit, inspired by traditional measures of centrality we introduce DebtRank a novel measure of systemic impact. We that we intend the risk of default of a large portion of the financial system, depends on the network of financial exposures among institutions. As an application, we analyse a new and unique dataset on the USD 1.2 trillion FED emergency loans program to global financial institutions during 2008--2010. We find that a group of 22 institutions, which received most of the funds, form a strongly connected graph where each of the nodes becomes systemically important at the peak of the crisis. Moreover, a systemic default could have been triggered even by small dispersed shocks. Other application to different systems are also presented.   

Cascading failures in Europe's transmission system

Cascading failures constitute an important vulnerability of infrastructure networks, hence understanding the origin and propagation of these failures is of great interest. To this end, we study cascades of overload failures within the framework of the cascade model introduced by Motter and Lai (2002) applied for flows that have a distributed character. We investigate numerically the properties of these failures in the high-voltage European electric power transmission system from 2002 (Zhou and Bialek, 2005). The network consists of 1254 nodes (substations) specified by geographical locations, and 1812 links (transmission lines) that are assumed to be undirected. We find that assigning excess capacities in proportion to initial loads does not significantly mitigate cascading failures. Moreover, increasing the fractional excess capacity does not yield monotonically increasing gains. Using a simple model of spatial network - random geometric graph - we investigate methods beyond the proportional excess capacity allocation in order to improve the gains in mitigating such failures.   

Cascade Failures in Power Grids with Distributed Generation

Power grids are nowadays experiencing a transformation due to the introduction of Distributed Generation based on Renewable Sources. At difference with classical Distributed Generation, where local power sources mitigate anomalous user consumption peaks, Renewable Sources introduce in the grid intrinsically erratic power inputs. By introducing a simple schematic (but realistic) model for power grids with stochastic distributed generation, we study the effects of erratic sources on the robustness of several IEEE power grid test networks with up to $2 \times 10^3$ buses. We find that increasing the penetration of erratic sources causes the grid to fail with a sharp transition. We compare such results with the case of failures caused by the natural increasing power demand.   

Distribution of Betweenness in Networks Failing by Overload

We study the Motter and Lai [1] model of cascading failures based on the betweenness centrality of the nodes, for a random regular network. After removing a fraction of the nodes, we study the size of the giant component at the end of the cascade of failures, as a function of the fraction of the nodes that survived the initial attack. We find that the type of transition through which the network disintegrates changes from first order to second order as the maximum capacity of the nodes increases. We examine the distribution of the betweenness of the nodes in the vicinity of the critical fraction of initial surviving nodes, and we look at the distribution at different stages of the cascade. We explore the disintegration of the network when the size of the initial attack approaches the percolation threshold of the network. We compare our results with an analytical ansatz of the role of subcomponents of the network nearly isolated from the giant component.\\[4pt] [1] A. Motter, Y. Lai, ``Cascade-based attacks on complex networks,'' \textit{Phys. Rev. E} \textbf{66}, 065102(R) (2002)   

Cascading Failures in Networks with Proximate Dependent Nodes

We study a system composed of two identical, random regular, interdependent networks. When a fraction of nodes in the first network are eliminated by failure or attack, further nodes that become isolated or lose their dependent node fail in turn, initiating a process of cascading failures. In contrast to previous models, these networks are constructed such that interdependent nodes are no more than a set distance away, with the distance defined by the number of intervening nodes. We find that as the maximum distance and the degree of connectivity increase, the disintegration of the system shifts from being similar to that of a single network to resembling the failure found in other models of interdependent networks. As the distance and degree increase, the collapse at the critical threshold changes from a second-order transition to a first-order one. The critical threshold monotonically increases as the distance increases.   

Comparative advantage between traditional and smart navigation systems

The smart navigation system that refers to real-time traffic data is believed to be superior to traditional navigation systems. To verify this belief, we created an agent-based traffic model and examined the effect of changing market share of the traditional shortest-travel-time algorithm based navigation and the smart navigation system. We tested our model on the grid and actual metropolitan road network structures. The result reveals that the traditional navigation system have better performance than the smart one as the market share of the smart navigation system exceeds a critical value, which is contrary to conventional expectation. We suggest that the superiority inversion between agent groups is strongly related to the traffic weight function form, and is general. We also found that the relationship of market share, traffic flow density and travel time is determined by the combination of congestion avoidance behavior of the smartly navigated agents and the inefficiency of shortest-travel-time based navigated agents. Our results can be interpreted with the minority game and extended to the diverse topics of opinion dynamics.   

The Cancho i Ferrer - Sol\'e model does not explain Zipf's Law

We examine the cost-minimization problem posed by Ferrer i Cancho and Sol\'e in their information-theory based communication model [1], proposed in efforts to explain Zipf's Law (that is a power-law frequency-rank relation for words in written texts). Using a simple inequality, we obtain the exact minimum-cost solution as a function of the parameter $\lambda$, as obtained previously via other methods [2-4]. ($\lambda$ defines the relative weights of speaker's and listener's costs.) We show that at the phase transition, the minimum-cost solutions do not correspond to a power law except for a vanishingly small subset, even if we impose the additional condition of equal costs to speaker and listener. Finally we consider the model at finite temperature using mean-field theory and entropic Monte Carlo simulation, and find a line of {\it discontinuous} phase transitions in the $\lambda$-$T$ plane. The simulations yield no evidence of a power-law frequency-rank distribution. \noindent 1. R. Ferrer i Cancho and R.~V. Sol\'e, PNAS {\bf 100}, 788 (2003).\\ 2. R. Ferrer i Cancho and A. D\'{\i}az-Guilera, J. Stat. Mech.: Theory Exp. (2007) P06009.\\ 3. A. Trosso, Master's thesis, 2008, University of Turin, Italy.\\ 4. M. Prokopenko et al., J. Stat. Mech. (2010) P11025.