Logarithmic asymptotics for multidimensional extremes under non-linear scalings

Abstract: Let $\boldsymbol W={\boldsymbol W_n:n\in\mathbb N}$ be a sequence of random vectors in $\mathbb R^d$, $d\ge 1$. This paper considers the logarithmic asymptotics of the extremes of $\boldsymbol W$, that is, for any vector $\boldsymbol q>\boldsymbol 0$ in $\mathbb R^d$, we find $$\log\mathbb P\left(\exists{n\in\mathbb N}:\boldsymbol W_n> u \boldsymbol q\right), \quad\text{as} u\to\infty.$$ We follow the approach of the restricted large deviation principle introduced in Duffy et al. \textit{Logarithmic asymptotics for the supremum of a stochastic process} (Ann. Appl. Probab., 13:430--445, 2003). That is, we assume that, for every $\boldsymbol q\ge\boldsymbol 0$, and some scalings ${a_n},{v_n}$, $\frac{1}{v_n}\log\mathbb P\left(\boldsymbol W_n/a_n\ge u \boldsymbol q\right)$ has a, continuous in $\boldsymbol q$, limit $J_{\boldsymbol W}(\boldsymbol q)$. We allow the scalings ${a_n}$ and ${v_n}$ to be regularly varying with a positive index. This approach is general enough to incorporate sequences $\boldsymbol W$, such that the probability law of $\boldsymbol W_n/a_n$ satisfies the large deviation principle with continuous, not necessarily convex, rate functions. The formula for these asymptotics agrees with the seminal papers on this topic.