Abstract: This study examines how research team knowledge background composition affects the knowledge innovation across eight major scientific fields by analyzing 32.63 million co-authored papers from the Crossref dataset spanning 1945-2023. Using Monte Carlo simulation and natural language processing, this study developed metrics to characterize team knowledge background composition, including internal knowledge similarity among team members and external knowledge similarity among teams. It was found that (1) Different knowledge background compositions exhibit significant differences in their capacity to generate knowledge innovation, with specific team knowledge background compositions demonstrating markedly superior innovation capabilities compared to others, indicating that team knowledge background composition has a decisive impact on knowledge innovation; (2) Each field has its own optimal team knowledge background composition: physics favors diverse teams with differentiated research directions, while social sciences perform better with specialized team compositions, suggesting that innovation policies should be tailored to specific disciplines; (3) Optimal team knowledge background compositions consistently represent only a minority across all fields, indicating structural misalignment in current research systems; meanwhile, despite exponential growth in publication volume, high-level knowledge innovation output remains constant across fields, revealing a "carrying capacity" phenomenon that highlights the disconnect between research scale expansion and knowledge innovation. These findings have important implications for scientific practice and policy development.

Key words: Team knowledge background composition, Knowledge innovation, Research collaboration, Interdisciplinary comparison, Big science, Science of Team Science