no. 12 · Tuesday 10 November 2026
Universality transitions in the logarithmic law of random determinants beyond the fourth moment
Abstract
For real random matrices $A_n=(a_{ij})_{1\le i,j\le n}$ with independent centered entries of unit variance, \citet{BaoPanZhou2015} established the classical logarithmic law under a uniform fourth-moment condition: \[ \frac{\log|\det A_n|-\frac12\log(n-1)!} {\sqrt{\frac12\log n}} \Rightarrow \mathcal N(0,1). \] We find that this Gaussian universality can be extended beyond the fourth-moment condition, with two distinct transitions in the center and fluctuation scale as the moment condition gradually weaken. The first critical condition is the uniform integrability of the family $\{|a_{ij}|^4/\sqrt{\log(e+|a_{ij}|)}\}_{n,i,j}$. The classical law remains valid under this weaker condition, with no change in center or scale. As the moment condition is weakened further, the classical center can fail. Replacing the classical center $\frac12\log(n-1)!$ by the actual mean $\mathbb E\log|\det A_n|$ preserves the standard Gaussian limit at the original scale and allows the moment condition to be relaxed to a second critical condition: uniform integrability of $\{|a_{ij}|^4/\log(e+|a_{ij}|)\}_{n,i,j}$. Under still weaker moment conditions, the classical scale $\sqrt{\frac12\log n}$ can also fail. Replacing this scale in turn by the actual standard deviation $\sqrt{\operatorname{Var}(\log|\det A_n|)}$ yields a standard Gaussian limit whenever the expectations of $|a_{ij}|^4/\{\log(e+|a_{ij}|)\}^α$ are uniformly bounded over $n,i,j$, for any fixed $α\ge0$. The results using the actual mean or standard deviation additionally assume uniformly bounded entry densities. We establish the sharpness of the first two critical conditions via counterexamples. We further obtain Berry–Esseen bounds for all three normalizations, with optimal rates under a uniform fourth-moment bound.
- comments
- 98 pages
- fields
- Probability