1: \begin{abstract}

2:   A refinement of Bennett's inequality is introduced which is strictly

3:   tighter than the classical bound. The new bound establishes the

4:   convergence of the average of independent random variables to its

5:   expected value. It also carefully exploits information about the

6:   potentially heterogeneous mean, variance, and ceiling of each random

7:   variable. The bound is strictly sharper in the homogeneous setting

8:   and very often significantly sharper in the heterogeneous

9:   setting. The improved convergence rates are obtained by leveraging

10:   Lambert's ${\cal W}$ function. We apply the new bound in a portfolio

11:   optimization setting to allocate a budget across investments with

12:   heterogeneous returns.

13: \end{abstract}