Limiting distribution under equicorrelation covariance structures

Characterize the limiting distribution of the relevant high-dimensional covariance statistic when the covariance matrix has the equicorrelation form $(1-\rho)I_d+\rho 1_d1_d^\top$ with $\rho\in(0,1)$ and the fourth-trace condition used in prior work fails.

Background

In discussing related covariance-testing results, the paper considers covariance structures of the form (1ρ)Id+ρ1d1d(1-\rho)I_d+\rho 1_d1_d^\top. For this structure, the ratio tr(Γn4)/tr2(Γn2)\operatorname{tr}(\Gamma_n^4)/\operatorname{tr}^2(\Gamma_n^2) does not vanish, so a condition imposed in earlier work is unavailable. The paper explicitly notes that the limiting distribution in this setting is unresolved; the present paper provides Gaussian approximation results for its own U-statistic norm framework but does not characterize that previously unclear limiting distribution.

References

But for this choice of $\Gamma_n$, $\tr(\Gamma_n4)/\tr2(\Gamma_n2) = C$ and thus the condition in is not satisfied, and therefore the limiting distribution remains unclear.

Berry--Esseen bounds and bootstrap approximations for the Hilbert-space norm of $U$-statistics  (2608.25463 - Chakraborty et al., 26 Aug 2026) in Section 2, discussion following Proposition 2.2