Sharp endpoint asymptotics for entrywise powers of four-dimensional correlation matrices
We determine the largest first-order loss of positive semidefiniteness when the entries of a nonnegative real 4 by 4 correlation matrix are raised to the power 2-epsilon. The maximal negative part of the least eigenvalue equals c_0 epsilon + o(epsilon), where c_0 is specified by a unique scalar root and is approximately 0.0470564382481148. All first-order maximizing matrices form the permutation orbit of a rank-two Gram matrix with angles 0, alpha_0, pi/2-alpha_0, pi/2. A uniform kernel reduction on the full compact class and a two-by-two cross-Gram representation yield two strict reductions and the equality classification. Moving first-order near-maximizers are characterized by convergence to the maximizing orbit together with least eigenvalue of the entrywise square equal to o(epsilon), including sequences of ranks three and four.
Authors
- Zihuan Xu
Institutions
- Capital Normal University (CN)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-15
- DOI
- https://doi.org/10.5281/zenodo.24244988
- Primary Topic
- Matrix Theory and Algorithms
- Type
- preprint