Scientific discovery and topological transitions in collaboration networks

Abstract

Abstract:We analyze the advent and development of eight scientific fields from their inception to maturity and map the evolution of their networks of collaboration over time, measured in terms of co-authorship of scientific papers. We show that as a field develops it undergoes a topological transition in its collaboration structure between a small disconnected graph to a much larger network where a giant connected component of collaboration appears. As a result, the number of edges and nodes in the largest component undergoes a transition between a small fraction of the total to a majority of all occurrences. These results relate to many qualitative observations of the evolution of technology and discussions of the "structure of scientific revolutions". We analyze this qualitative change in network topology in terms of several quantitative graph theoretical measures, such as density, diameter, and relative size of the network's largest component.To analyze examples of scientific discovery we build databases of scientific publications based on keyword and citation searches, for eight fields, spanning experimental and theoretical science, across areas as diverse as physics, biomedical sciences, and materials science. Each of the databases is vetted by field experts and is the result of a bibliometric search constructed to maximize coverage, while minimizing the occurrence of spurious records. In this way we built databases of publications and authors for superstring theory, cosmic strings and other topological defects, cosmological inflation, carbon nanotubes, quantum computing and computation, prions and scrapie, and H5N1 influenza. We also build a database for a classical example of "pathological" science, namely cold fusion. All these fields also vary in size and in their temporal patterns of development, with some showing explosive growth from an original identifiable discovery (e.g. carbon nanotubes) while others are characterized by a slow process of development (e.g. quantum computers and computation).We show that regardless of the detailed nature of their developmental paths, the process of scientific discovery and the rearrangement of the collaboration structure of emergent fields is characterized by a number of universal features, suggesting that the process of discovery and initial formation of a scientific field, characterized by the moments of discovery, invention and subsequent transition into "normal science" may be understood in general terms, as a process of cognitive and social unification out of many 2 initially separate efforts. Pathological fields, seemingly, never undergo this transition, despite hundreds of publications and the involvement of many authors.

General Critical Properties of the Dynamics of Scientific Discovery

“…(1), the average degree, k , should obey a simple scaling relation with network size, N , of the form Regardless of the details of the temporal evolution of the field, the number of collaboration (co-authorship) links increases in a scaleinvariant way with the number of authors in the network, as shown in C for nanotechnology and D for sustainability science. These patterns are similar to those we have found in [17] for a variety of smaller scientific fields.…”

“…Second, in our previous study of collaboration networks [17], we found a simple scaling relationship between the number of co-authorship links (or edges) per node, E(t), and the number of authors (nodes), N (t),…”

“…The fields vary greatly in size and composition: from relatively modest-sized communities in theoretical physics such as cosmic strings or cosmological inflation, in which authors have similar training; to benchtop biomedical topics like research on scrapie and prions, which incorporate co-authors of varied expertise who work together on a narrowly-defined problem; to huge interdisciplinary fields like nanotechnology and sustainability science, which feature authors from a wide range of specialties. Whereas we have examined other aspects of the growth of several of these fields in previous work [7,17], the two largest fields in our sample -sustainability science and nanotechnology -represent brand-new datasets. Table 1.…”

“…Fields with α > 1, in other words, grow more dense as they evolve [29]. Empirically, virtually every field we studied had an exponent, α > 1, with most falling in the range α = [1.05, 1.38]; only cold fusion displayed a scaling exponent consistent with 1 [17]. In our latest studies, we have found the same behavior for much larger scientific fields: both for sustainability science, a field that has included approximately 37,000 authors over the past few decades and has a densification exponent of α = 1.27; and for nanotechnology, a field that has included more than 330,000 authors over the past twenty years and has a densification exponent of α = 1.…”

“…There are certainly a variety of emerging methodologies that unify the study of science. These include population dynamical models [1,2,3,4,5,6,7], networks of cocitation [8,9] and collaboration [10,11,12,13,14,15,16,17], disciplinary maps of science [18,19,20,21] and phylogenetic term analyses [22,23], among others. But a more profound question is whether studies using these methodologies and others capture common dynamics across fields.…”