This course develops the analytic tools used to study additive problems in number theory, with a focus on exponential sums, Fourier-analytic methods, and the circle method. The central goal is to understand how arithmetic structure in integers, primes, and polynomial sequences is reflected in oscillatory sums, and how that structure can be exploited to count representations, bound error terms, and prove quantitative distribution results.

The early chapters build the basic language: additive characters and finite Fourier analysis provide the framework for decomposing arithmetic functions into frequency components, while Weyl sums and polynomial phases introduce the main class of oscillatory sums that arise in Diophantine questions. From there, van der Corput estimates supply key cancellation techniques, and Vinogradov mean value estimates sharpen the control needed for higher-degree polynomial problems. These ideas culminate in the Hardy-Littlewood circle method, which turns Fourier analysis into a systematic machine for counting solutions to additive equations.

Later chapters apply this machinery to classical and modern problems. Waring’s problem and additive problems with prime variables show how the method handles representations by powers and primes, while additive energy, density, and Fourier control connect exponential-sum methods to combinatorial structure. The final chapter on fractional parts and Diophantine applications shows how the same analytic principles govern distribution modulo one and related approximation problems, tying together the course’s analytic, combinatorial, and Diophantine themes.

# Introduction

This course studies how finite Fourier analysis converts additive questions about integers into estimates for oscillatory sums. The guiding problem is to count solutions of equations such as

\begin{align*} x_1^k+\cdots+x_s^k=n, \end{align*}

or to understand when a structured set of integers has the expected number of additive representations. The main technical lesson is that cancellation in exponential sums is the mechanism that separates a main term from an error term.

The course sits after a first course in analytic number theory. We use congruences, elementary asymptotic notation, real and complex analysis, Fourier analysis on the circle, and basic measure and integration. Dirichlet series, sieve methods, and $L$-functions may appear as context, but the core method throughout is additive Fourier analysis.

## What Is Being Counted

How can an equation in integers be turned into something Fourier analytic? The common pattern is to encode a set or sequence by a generating exponential sum, multiply such sums to impose an additive relation, and then extract the desired coefficient by orthogonality. This viewpoint is useful because the size of an exponential sum measures how much arithmetic structure remains after oscillation.

The first kind of object is a representation function. It records how many ways an integer can be assembled from a prescribed list of summands, and it is the bridge between additive number theory and Fourier coefficients.

[definition: Representation Function] Let $A_1,\dots,A_s$ be finite subsets of $\mathbb Z$. The representation function associated to $(A_1,\dots,A_s)$ is the function $R_{A_1,\dots,A_s}:\mathbb Z\to\mathbb N\cup\{0\}$ defined by \begin{align*} R_{A_1,\dots,A_s}(n)=\#\{(a_1,\dots,a_s)\in A_1\times\cdots\times A_s:a_1+\cdots+a_s=n\}. \end{align*} [/definition]

This definition is deliberately finite: it lets us count exactly before taking limits. The later analytic work asks whether $R_{A_1,\dots,A_s}(n)$ is close to a predicted main term when the sets vary with a parameter such as $N$.

[example: Sum Of Two Intervals] Let $A=B=\{1,\dots,N\}$, with $N\ge 1$. For a fixed integer $n$, the value $R_{A,B}(n)$ is the number of integers $a$ such that $1\le a\le N$ and $1\le n-a\le N$. The second condition is equivalent to $n-N\le a\le n-1$, so the admissible values of $a$ are exactly the integers in the interval \begin{align*} \max(1,n-N)\le a\le \min(N,n-1). \end{align*} If $n<2$, then $a+b\ge 1+1=2$, so there are no pairs. If $n>2N$, then $a+b\le N+N=2N$, so again there are no pairs. Now assume $2\le n\le 2N$. If $2\le n\le N+1$, then $n-N\le 1$ and $n-1\le N$, so the admissible interval is $1\le a\le n-1$. Hence \begin{align*} R_{A,B}(n)=(n-1)-1+1=n-1. \end{align*} If $N+1\le n\le 2N$, then $n-N\ge 1$ and $n-1\ge N$, so the admissible interval is $n-N\le a\le N$. Hence \begin{align*} R_{A,B}(n)=N-(n-N)+1=2N+1-n. \end{align*} Combining the two ranges gives \begin{align*} R_{A,B}(n)=\min(n-1,2N+1-n) \end{align*} for $2\le n\le 2N$, and $R_{A,B}(n)=0$ otherwise. Thus the representation function rises by one at each step until $n=N+1$, then falls by one at each step, giving a triangular large-scale shape before any cancellation estimate enters. [/example]

The finite interval example is too regular to reveal the main difficulty. In problems such as Waring's problem, the summands are powers rather than consecutive integers, and the representation function is controlled by sums with polynomial phases.

## Exponential Sums As Measuring Devices

Where does cancellation enter the count? If the terms in a complex sum point in nearly the same direction, the sum is large; if they rotate sufficiently, the sum is smaller than its number of terms. Analytic number theory turns this geometric idea into quantitative bounds.

[definition: Exponential Notation] The exponential notation is the map \begin{align*} e:\mathbb R&\to\mathbb C, & x&\mapsto \exp(2\pi i x). \end{align*} [/definition]

The normalization makes $e(x)$ periodic with period $1$, so it is adapted to arithmetic modulo $q$ and to integration over $\mathbb R/\mathbb Z$. Once this notation is fixed, we can package a weighted arithmetic sequence together with an oscillating phase and ask whether the resulting complex sum cancels.

[definition: Exponential Sum] Let $(a_n)_{n\in I}$ be a finite complex sequence indexed by a finite set $I\subset\mathbb Z$, and let $\phi:I\to\mathbb R$ be a phase function. The associated exponential sum is \begin{align*} S=\sum_{n\in I}a_n e(\phi(n)). \end{align*} [/definition]