The paper studies a mathematical question about how large certain kinds of oscillating sums can get. Specifically, it looks at families of functions that repeat periodically and asks: if you add up the values of such a function step by step, how big can the running total get at its peak? The functions studied include well-known objects in number theory called Kloosterman sums and Birch sums. The central goal is to understand not just whether these peaks can be large, but precisely how likely large peaks are across many such functions.
The first main result improves on earlier work by giving a much sharper formula for how the probability of seeing a very large peak drops off as the size of that peak grows. Previous work gave rough bounds on this probability; the new result pins down the precise rate of decay in a wide and uniform range of parameters. This kind of precise asymptotic information is valuable because it tells you both that large peaks are rare and exactly how rare they are.
The second main result is a structural finding that is somewhat surprising. It shows that whenever one of these sums achieves a large peak, the peak almost always occurs near the halfway point of the summation range, and the sum at that point is almost entirely imaginary rather than real. This is a universal feature across the whole class of functions studied. Interestingly, this behavior is the opposite of what happens for a related but different class of functions called Dirichlet character sums, where large peaks behave quite differently. The contrast highlights that despite surface similarities, these two families have fundamentally different internal structure.