Hidden memory and stochastic fluctuations in science

2025-03-04

AI Summary

Citation networks exhibit a paradoxical coexistence of log-normal and high-end power-law in-degree distributions, lacking a unified mechanistic explanation. Method: We discover that the logarithmic variance of citations grows as a power law (∝t^H) over time and propose a hidden-memory stochastic model based on fractional Brownian motion, where the Hurst exponent H quantifies the memory of scientific attention: H<0.5 induces anti-persistent dynamics (yielding log-normality), while H>0.5 drives persistent dynamics (generating power-law tails). The model integrates cumulative advantage with latent-variable mechanisms. Contribution/Results: Numerical simulations and empirical validation on arXiv data confirm the model’s ability to interpolate and predict citation distributions across all scales. Empirical estimation yields H≈0.13 for arXiv, indicating strong anti-persistence dominates real-world citation evolution—providing the first mechanistically grounded, unified framework resolving the long-standing distributional paradox in scientometrics.

Technology Category

• Reasoning under Uncertainty: Causality
• Cognitive Modeling & Cognitive Systems: Simulating Human Behavior
• Application Domains: Humanities & Computational Social Science

Application Category

• Web Mining and Content Analysis: Models for Web evolution
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for heterogeneous, signed, attributed, multi-relational, temporal, higher-order, and annotated Web-related graphs
• Social Networks and Social Media: Computational social science

Abstract

Understanding the statistical laws governing citation dynamics remains a fundamental challenge in network theory and the science of science. Citation networks typically exhibit in-degree distributions well approximated by log-normal distributions, yet they also display power-law behaviour in the high-citation regime, presenting an apparent contradiction that lacks a unified explanation. Here, we identify a previously unrecognised phenomenon: the variance of the logarithm of citation counts per unit time follows a power law with respect to time since publication, scaling as $t^{H}$. This discovery introduces a new challenge while simultaneously offering a crucial clue to resolving this discrepancy. We develop a stochastic model in which latent attention to publications evolves through a memory-driven process incorporating cumulative advantage. This process is characterised by the Hurst parameter $H$, derived from fractional Brownian motion, and volatility. Our framework reconciles this contradiction by demonstrating that anti-persistent fluctuations ($H< frac{1}{2}$) give rise to log-normal citation distributions, whereas persistent dynamics ($H> frac{1}{2}$) favour heavy-tailed power laws. Numerical simulations confirm our model's explanatory and predictive power, interpolating between log-normal and power-law distributions while reproducing the $t^{H}$ law. Empirical analysis of arXiv e-prints further supports our theory, revealing an intrinsically anti-persistent nature with an upper bound of approximately $H=0.13$. By linking memory effects and stochastic fluctuations to broader network dynamics, our findings provide a unifying framework for understanding the evolution of collective attention in science and other attention-driven processes.

Problem

• Understanding citation dynamics in science networks
• Resolving contradiction between log-normal and power-law distributions
• Modeling memory-driven attention evolution in publications

Innovation

• Stochastic model with memory-driven attention evolution
• Incorporates Hurst parameter from fractional Brownian motion
• Reconciles log-normal and power-law citation distributions