[Paper Review] Stratification and averaging for exponential sums: bilinear forms with generalized Kloosterman sums

Emmanuel Kowalski, Ph. Michel+1 more|arXiv (Cornell University)|Feb 27, 2018

Analytic Number Theory Research11 references4 citations

TL;DR

This paper introduces a novel cohomological comparison principle using stratification and vanishing cycles to bound bilinear exponential sums involving generalized Kloosterman sums over finite fields. By showing that cohomology groups associated with two exponential sums are isomorphic, the authors achieve non-trivial cancellation without explicit computation, leading to improved bounds below the Pólya-Vinogradov range—specifically, $ MN \geqslant q^{7/8 + \delta} $ for hyper-Kloosterman sums—enabling new estimates for the first moment of degree-3 $L$-functions.

ABSTRACT

We prove non-trivial bounds for bilinear forms with hyper-Kloosterman sums with characters modulo a prime $q$ which, for both variables of length $M$, are non-trivial as soon as $M\geq q^{3/8+δ}$ for any $δ>0$. This range, which matches Burgess's range, is identical with the best results previously known only for simpler exponentials of monomials. The proof combines refinements of the analytic tools from our previous paper and new geometric methods. The key geometric idea is a comparison statement that shows that even when the "sum-product" sheaves that appear in the analysis fail to be irreducible, their decomposition reflects that of the "input" sheaves, except for parameters in a high-codimension subset. This property is proved by a subtle interplay between étale cohomology in its algebraic and diophantine incarnations. We prove a first application concerning the first moment of a family of $L$-functions of degree $3$.

Motivation & Objective

• To develop a new comparison principle for exponential sums over finite fields, particularly for bilinear forms with generalized Kloosterman sums.
• To establish non-trivial bounds for bilinear sums below the classical Pólya-Vinogradov threshold of $ q^{1/2} $.
• To apply these bounds to estimate the first moment of a family of $ L $-functions of degree 3.
• To unify and extend previous results on hyper-Kloosterman sums using étale cohomology and monodromy.
• To demonstrate that cancellation in exponential sums arises from cohomological isomorphism rather than explicit term-by-term evaluation.

Proposed method

• Introduce a stratification of the parameter space to analyze the behavior of exponential sums across different geometric loci.
• Use vanishing cycles to show that if cohomological comparison fails on a stratum, it fails generically, reducing the problem to a generic statement.
• Apply a variant of Katz’s diophantine criterion of irreducibility to control monodromy and ensure sheaf irreducibility.
• Establish isomorphism between Frobenius-equivariant endomorphism spaces of sheaves $ \mathcal{K}_{\mathbf{b}} $ and $ \mathcal{R}^*_{\mathbf{b}} $, implying trace equality.
• Leverage the Grothendieck–Lefschetz trace formula to relate sums of trace functions to cohomological traces, enabling bounds via Deligne’s Riemann Hypothesis.
• Control Betti numbers using Bombieri–Katz bounds, ensuring uniformity in implied constants across the parameter space.

Experimental results

Research questions

• RQ1 Can a new cohomological comparison principle be developed to bound bilinear exponential sums with generalized Kloosterman sums below the Pólya-Vinogradov range?
• RQ2 Does the cancellation in such sums arise from isomorphism of cohomology groups rather than explicit evaluation of terms?
• RQ3 Can this method yield non-trivial bounds for $ M,N \geqslant q^{\delta} $ and $ MN \geqslant q^{7/8 + \delta} $ for hyper-Kloosterman sums?
• RQ4 How does the stratification of the parameter space and the use of vanishing cycles facilitate the proof of cohomological comparison?
• RQ5 What is the impact of this method on the first moment of degree-3 $ L $-functions?

Key findings

• The authors establish a non-trivial bound for bilinear sums involving hyper-Kloosterman sums when $ M,N \geqslant q^{\delta} $ and $ MN \geqslant q^{7/8 + \delta} $, improving upon the previous $ 7/8 $ threshold.
• The key innovation lies in showing that the cohomology groups of the two exponential sums are isomorphic, leading to cancellation without explicit computation.
• For $ \mathbf{b} \notin \mathcal{W}(\mathbf{F}_q) $, the sum $ \Sigma_I(\mathbf{b}) = \sum_r \mathbf{R}(r,\mathbf{b}) \ll q $, which is a significant saving over the trivial $ \ll q^2 $.
• For $ \mathbf{b} \in \mathcal{W}(\mathbf{F}_q) \setminus \mathcal{V}^\Delta(\mathbf{F}_q) $, the sum $ \Sigma_I(\mathbf{b}) \ll q^{3/2} $, again improving on the trivial bound.
• The complete sum $ \Sigma_{II}(\mathbf{b}) \ll q^2 $ when $ \mathbf{b} \in \mathcal{W}(\mathbf{F}_q) \setminus \mathcal{V}^\Delta(\mathbf{F}_q) $, and $ \ll q^{3/2} $ otherwise, with the latter relying on cohomological isomorphism.
• The implied constants in all bounds depend only on Betti numbers, which are controlled via Bombieri–Katz bounds, ensuring uniformity across the parameter space.