Title: A Collision Probability Estimation Paradox and Two Solutions

Speaker: Cui Kaiqi (PMO)

Time: 12:15pm, November 28, 2025

Location: 3-302, PMO Xianlin Campus

Abstract: In traditional satellite conjunction analysis, collision risk is often expressed using epistemic probability. However, this approach suffers from two fundamental deficiencies — probability dilution and false confidence — that cause systematic underestimation of collision risk. As orbital uncertainty increases, the computed collision probability paradoxically first rises and then decreases, misleadingly implying improved safety with lower data quality. Based on the False Confidence Theorem, this paper demonstrates that the root cause lies in the additivity axiom of probability theory, which prevents proper representation of epistemic uncertainty. To overcome this limitation, the Martin–Liu validity criterion is introduced, requiring that the aleatory probability of assigning high belief to any false proposition must not exceed its belief level.