Persistent collaboration as a structural signature of scientific resilience.

Scientific progress today is increasingly realized through collaboration. Recent global crises and shifting geopolitical and funding landscapes have shown that collaboration is not immune to shocks. These disturbances call for a deeper examination of the resilience of science, specifically how the scientific community sustains its functionality and collaborative output amid potential shocks and failures. In this study, we conceptualize resilience as a structural property of scientific collaboration and quantify it using an interpretable, network-theoretic measure. Using a large-scale bibliographic dataset, we construct coauthorship networks across multiple disciplines from 1985 to 2014 and perform stress-test experiments that simulate stylized perturbations to identify the structural elements that support system resilience. Our findings show that discipline-level resilience is positively associated with higher collaboration intensity and, more strongly, with greater variance in scientists' collaboration outcomes. Moreover, we find that persistent collaborations-stable and strong collaborations between scientists-play a central role in fostering resilient network structures across fields; they comprise only a small share of links yet are disproportionately held by scientists in the top decile of productivity. Overall, this study offers a network-based framework for understanding the structural resilience of scientific collaboration and implies the important role of persistent collaboration in supporting that resilience.