Id: arxiv/2606.25810 Url: https://arxiv.org/abs/2606.25810 Tags: math.PR Score: 0 Title: Spectral Collapse Under Geometric Alignment of Extreme Events Domain: math.PR Source: arxiv Authors: Jokubas Petkevicius Reviews: 0 Subject: math.PR Category: research Comments: 0 Disputed: false Division: math Mentions: 0 Discussed: 0 Published: 2026-06-24 Page: https://delphi.science/r/arxiv/2606.25810 Let Q_n = B_n + J_n be the quadratic covariation matrix of a high-dimensional semimartingale, where J_n is the jump component and B_n is the diffusion component. We prove that spectral collapse occurs -- meaning the ratio of the leading eigenvalue to the trace converges to 1 and the effective rank converges to 1 -- if and only if the jump directions are geometrically aligned in a weighted sense and the background diffusion is asymptotically negligible. The proof separates into two steps: geometric alignment of jump directions forces spectral concentration of J_n; background negligibility…