Tail Bounds via Southwest Boundary

Abstract

We derive upper bounds for probabilities of the form $P(g(\mathbf{X})\geq t)$ using the southwest boundary (recently introduced in our previous work) $\partial_{\mathrm{SW}} Q(g^{-1}[t,\infty))$, where $Q$ is a reflection to the first quadrant. Under natural continuity, symmetry, and monotonicity assumptions on $g$, this yields explicit and computable bounds of the form $P(g(\mathbf{X})\ge t)\le ns_t$, where $s_t$ is the unique parameter at which the line $L(s)=(f_1^{-1}(s),\dots,f_n^{-1}(s))$ intersects the southwest boundary. In particular, when $g$ is a homogeneous polynomial of degree $k$ (plus a constant $C$) and all tail bounds on the random variables are identical, the bound proves to the closed-form expression $$ P(g(\mathbf{X})\ge t)\leq nf\bigg(\frac{(t-C)^{1/k}}{(\sum_i|a_i|)^{1/k}}\bigg) $$ where $a_i$ are the coefficients of the monomials in $g$. We then obtain an explicit tail bound for the trace of a Schur multiplier acting on random matrices with identical tail bounds on the random variables. No assumptions are made about independence or dependence.