Loading field map
Analytic number theory is the branch of mathematics that uses the methods of real and complex analysis—limits, integrals, series, and the properties of functions of a complex variable—to answer questions about the integers, and especially about prime numbers. Its central preoccupation is the distribution of primes, but it also addresses the size of arithmetic functions, the solutions of Diophantine equations, and the behaviour of integer sequences that are too irregular to be fully described by algebraic methods.
The primes are the multiplicative atoms of the integers: every integer greater than 1 factors uniquely into primes. Yet their additive arrangement along the number line seems erratic. There is no simple formula that lists them, and their gaps are irregular. The fundamental problem of analytic number theory is to describe the statistical regularities hidden in this irregularity.
The first such regularity was glimpsed by Euclid: there are infinitely many primes. But a sharper question is: how many primes are less than \(x\)? The answer is given by the prime-counting function \(\pi(x)\). Around 1800, Gauss and Legendre independently conjectured, from numerical tables, that \(\pi(x)\) is approximately \(x/\log x\). This statement is the Prime Number Theorem, first proved in 1896 by Hadamard and de la Vallée Poussin.
The proof of the Prime Number Theorem marks the real birth of analytic number theory. Both proofs used the same central object: the Riemann zeta function
\[ \zeta(s) = \sum_{n=1}^{\infty} \frac{1}{n^s}, \]
defined for complex \(s\) with real part greater than 1. Euler had already noticed that this series has a multiplicative expansion over all primes,
\[ \zeta(s) = \prod_{p} \left(1 - p^{-s}\right)^{-1}, \]
which connects the additive world (the integers in a sum) to the multiplicative world (the primes in a product). Riemann, in a single celebrated 1859 paper, showed that the properties of \(\zeta(s)\) as a function of a complex variable control the distribution of primes. In particular, he revealed that \(\zeta(s)\), initially defined on a half-plane, can be extended to the whole complex plane (except for a pole at \(s=1\)), and that the locations of its zeros are deeply tied to the error term in \(\pi(x)\).
Hadamard and de la Vallée Poussin proved the Prime Number Theorem by showing that \(\zeta(s)\) has no zeros on the line \(\mathrm{Re}(s)=1\). This single fact forces \(\pi(x)\) to behave like \(x/\log x\). Everything finer, however, depends on where the other zeros lie. The Riemann Hypothesis asserts that all nontrivial zeros lie on the line \(\mathrm{Re}(s)=1/2\). If true, it gives the best possible error term for \(\pi(x)\) and implies extremely precise statements about the regularity of the primes. The hypothesis remains unproved and is one of the central open problems in all of mathematics. Analytic number theory is, to a large extent, the study of how much of the Riemann Hypothesis can be established, and how much can be deduced from partial information about it.
The zeta function is not only a tool for counting primes; it is also the gateway to understanding the structure of integers. By taking logarithms and using the Euler product, one obtains formulas for sums over primes in terms of integrals of the zeta function. For example, the statement that \(\zeta(s)\) behaves like \(1/(s-1)\) near its pole at \(s=1\) is equivalent to the statement that the sum of reciprocals of primes diverges—a stronger result than Euclid's theorem, and one that shows primes are relatively dense in a certain sense.
Beyond the zeta function itself, analytic number theory studies general Dirichlet series of the form
\[ \sum{n=1}^{\infty} \frac{an}{n^s}, \]
where \(an\) is some arithmetic sequence. The behaviour of the series in the complex \(s\)-plane encodes averages of the coefficients \(an\). This leads to the study of arithmetic functions: functions defined on the integers that capture their multiplicative or additive structure.
A few examples illustrate the scope:
The general moral is that the fine structure of integers—how they factor, how many divisors they have, whether they are squarefree—is governed by averages that can be expressed through the zeta function and its relatives. Analytic number theory provides the tools to compute these averages and to say how big the fluctuations around them can be.
Within analytic number theory, two major techniques dominate. They are not rival schools so much as complementary toolkits, though they often lead researchers to work on different problems and to value different kinds of results.
The first method, pioneered by Riemann and brought to maturity by Hadamard, de la Vallée Poussin, and later by Hardy, Littlewood, and others, treats the zeta function as an object in complex analysis. The central operation is contour integration: one expresses an arithmetic sum as an integral of a Dirichlet series over a vertical line in the complex plane, then shifts the contour, picking up contributions from poles and zeros that lie in the way.
This method yields the most precise results, but only when one knows—or can prove—something about the location of zeros. The Riemann Hypothesis would be the ultimate prize of this approach. In its absence, researchers work with zero-free regions: strips to the left of the line \(\mathrm{Re}(s)=1\) where the zeta function is known to have no zeros. The larger the zero-free region, the better the error term in the Prime Number Theorem. These regions have been extended incrementally over more than a century, each extension requiring deeper analysis of the functional equation of \(\zeta(s)\).
A key subsidiary object is the functional equation
\[ \zeta(s) = 2^s \pi^{s-1} \sin\left(\frac{\pi s}{2}\right) \Gamma(1-s) \zeta(1-s), \]
which relates \(\zeta(s)\) at \(s\) to its value at \(1-s\). This equation was discovered by Riemann and is the reason the zeta function has a rich structure on the whole plane. It also demonstrates a deep symmetry that persists in more general settings.
The second method, developed later and often called elementary (though not in the sense of "simple"), avoids complex analysis entirely. Its breakthrough came in 1949 when Selberg and Erdős independently proved the Prime Number Theorem without using the zeta function. Their proofs used combinatorial identities relating sums over primes to sums over integers, together with careful estimates for sums of arithmetic functions.
The elementary method does not usually give results as sharp as the best complex-analytic ones, but it has its own virtues. It often applies in situations where the analytic apparatus is unavailable or too cumbersome. Moreover, the identities used in elementary proofs have become powerful tools in their own right, particularly Selberg's sieve and related combinatorial techniques. The method also has a conceptual appeal: it shows that the Prime Number Theorem is not fundamentally a result about complex functions, but rather a genuine property of the integers.
The distinction between the two methods is not rigid. Many modern arguments mix them freely: an elementary identity might be used to set up a problem, and a complex integration might complete it. The choice of method often depends on the goal—sharpness of constants versus breadth of applicability.
A separate tradition within analytic number theory, closer in spirit to combinatorics, is the study of sieve methods. The oldest sieve is the Sieve of Eratosthenes for listing primes, but its modern descendants are far more refined. The basic idea is to estimate the number of integers in a set that survive after removing those divisible by a collection of primes, by inclusion–exclusion over that collection.
The most famous theorem proved by sieving is the Brun–Titchmarsh theorem and the work of Viggo Brun, who in 1919 used an early sieve to show that there are infinitely many integers that are the sum of two primes, and that the twin primes, if infinite in number, are sparse. Brun also proved that the sum of reciprocals of the twin primes converges—a result that remains true even though it is unknown whether there are infinitely many twin primes.
Sieves reached a high level of sophistication with the work of Selberg (the Selberg sieve, based on quadratic forms) and, later, of Bombieri, Friedlander, and Iwaniec, who developed versions that can be used in great generality. A central tool is the large sieve, which gives bounds on how many arithmetic progressions can contain many elements of a sparse set. The large sieve is closely related to the Bombieri–Vinogradov theorem, which states that, on average over moduli up to a certain size, the primes are distributed uniformly among reduced residue classes—a result that is often a usable substitute for the Generalized Riemann Hypothesis.
Sieve methods, however, have a structural limitation: they cannot distinguish between numbers with an even and an odd number of prime factors. This parity problem, first explicitly noted by Selberg, prevents standard sieves from proving that there are infinitely many primes with some given property; they can only prove results about almost-primes (numbers with at most \(r\) prime factors). The celebrated Chen's theorem (1966), that every sufficiently large even number is the sum of a prime and a number with at most two prime factors, is the sharpest known result in the direction of the Goldbach conjecture—and it is sharp because of the parity problem. It took the development of the Friedlander–Iwaniec and later Maynard methods to find special sets where a sieve could be combined with enough additional information to bypass the parity barrier in specific cases.
Where sieves study multiplicative structure, the circle method studies additive structure. It originates in work of Hardy and Ramanujan on the partition function and was developed systematically by Hardy and Littlewood in the 1920s. The idea is to represent the number of ways an integer \(N\) can be written as a sum of \(k\) terms from a set \(\mathcal{A}\) as an integral over the unit circle:
\[ rk(N) = \int0^1 S(\alpha)^k e^{-2\pi i N\alpha} \, d\alpha, \]
where \(S(\alpha) = \sum_{a \in \mathcal{A}} e^{2\pi i a\alpha}\). The integral is then approximated by splitting the circle into major arcs (near rationals with small denominator, where \(S(\alpha)\) is large and well understood) and minor arcs (everywhere else, where \(S(\alpha)\) is small and must be bounded by other means).
The circle method's greatest achievement is Vinogradov's theorem (1937): every sufficiently large odd integer is the sum of three primes. This is a strong partial answer to Goldbach's problem. The Hardy–Littlewood asymptotic formula for the number of representations gives a precise expected count, and the method works because the major arcs determine the main term and the minor arcs can be shown to contribute negligibly. The major difficulty is always the minor arc estimate, and Vinogradov's breakthrough was a bound for exponential sums over primes that was strong enough to control them.
The circle method has been extended far beyond primes. It has been used to study sums of squares of integers (where it recovers and refines classical results of Legendre and Jacobi), sums of powers, and representations of integers by general polynomial forms. In the hands of Bogolyubov and others, it has been transplanted into additive combinatorics, where it connects to the theory of Fourier analysis on finite groups and to the study of sets with additive structure.
The partition function \(p(n)\), which counts the number of ways to write \(n\) as a sum of positive integers, was one of Hardy and Ramanujan's earliest targets. Their asymptotic formula for \(p(n)\) was later refined by Rademacher into an exact convergent series. This is one of the few instances where an analytic method produces a formula that is, in principle, exact rather than merely asymptotic, and it remains a model of what the circle method can achieve when the generating function is modular.
A third central theme concerns primes in arithmetic progressions. The question is: given integers \(a\) and \(q\) with \(\gcd(a,q)=1\), how many primes are there congruent to \(a\) modulo \(q\)? Dirichlet proved in 1837 that there are infinitely many, and his proof introduced the key objects that govern the finer distribution: the Dirichlet characters and their associated L-functions.
A Dirichlet character \(\chi\) modulo \(q\) is a periodic function on the integers that is multiplicative, vanishes on integers sharing a common factor with \(q\), and takes values on the \(q\)-th roots of unity. The associated Dirichlet L-function is
\[ L(s,\chi) = \sum_{n=1}^{\infty} \frac{\chi(n)}{n^s}. \]
These functions satisfy a functional equation analogous to that of the zeta function, and their zeros control the distribution of primes among residue classes. The Generalized Riemann Hypothesis (GRH) states that all nontrivial zeros of all Dirichlet L-functions lie on the line \(\mathrm{Re}(s)=1/2\). It implies, among other things, that the primes are distributed as evenly as possible among the coprime residue classes modulo \(q\), with an error term nearly as small as \(O(\sqrt{x})\).
Without GRH, the best unconditional result is the Siegel–Walfisz theorem, which states that for each fixed \(A > 0\), the number of primes up to \(x\) in a residue class \(a\) mod \(q\) is
\[ \frac{\mathrm{Li}(x)}{\varphi(q)} + O\left(x e^{-c \sqrt{\log x}}\right), \] $\nu$niformly for \(q \le (\log x)^A\). The error term is worse than what GRH would give, but it holds for all fixed \(A\), which is often enough. The theorem has a curious flaw: the constant \(c\) is ineffective, because the proof relies on the possible existence of a certain exceptional zero (the Siegel zero) that would violate GRH in a mild way. This is a recurring theme in analytic number theory: one often proves strong results conditionally on the nonexistence of such zeros, and then must contend with the fact that the error terms are not computable in practice.
The distribution of primes in progressions is central to applications beyond the primes themselves, because many problems in additive number theory require knowing that primes are spread uniformly among residue classes. The Bombieri–Vinogradov theorem mentioned above is the key tool that allows this to be done without assuming GRH.
The zeta function and Dirichlet L-functions are the first members of a much larger family: the automorphic L-functions. These arise from modular forms, which are complex functions on the upper half-plane with strong transformation laws under the modular group. The connection to number theory was discovered by Ramanujan in his study of the tau function, defined by
\[ \Delta(q) = q \prod{n=1}^{\infty} (1-q^n)^{24} = \sum{n=1}^{\infty} \tau(n) q^n, \]
where \(e^{2\pi i z} = q\). Ramanujan conjectured that \(\tau(n)\) is multiplicative, that it satisfies certain congruences, and that \(|\tau(n)| \le n^{11/2} d(n)\). The multiplicativity was proved by Mordell, and the bound—the Ramanujan–Petersson conjecture—was proved in 1974 by Deligne as a consequence of the Weil conjectures. Between these two results lies one of the deepest threads in modern arithmetic.
The central object attached to a modular form is its L-function, which encodes the Fourier coefficients. The Langlands philosophy predicts deep relationships among these L-functions: all arise from automorphic representations, and their properties are interconnected. For analytic number theory, the importance is practical: these L-functions provide a vast testing ground for the same questions about zeros, averages, and error terms that one asks about \(\zeta(s)\). The Generalized Riemann Hypothesis extends to all automorphic L-functions, and progress on it—or even on the weaker statement that there are no zeros very close to the line \(\mathrm{Re}(s)=1\)—yields results about the coefficients of modular forms, which in turn count solutions to Diophantine equations.
A major achievement in this direction is the proof, by Iwaniec, Luo, and Sarnak, of the best known bounds for the size of Fourier coefficients of Maass forms on \(\mathrm{GL}(n)\), which partially resolves the Ramanujan–Petersson conjecture in the automorphic setting. The subconvexity problem—bounding L-functions on the critical line better than the trivial convexity bound—remains open in general but has been solved in many cases. Subconvexity bounds imply, among other things, the equidistribution of solutions to certain Diophantine equations.
The analytic study of automorphic forms has become one of the most active fronts of the field. It draws on representation theory and harmonic analysis in addition to classical complex analysis, and it connects analytic number theory to arithmetic geometry via the Langlands program. While the core of the subject remains the distribution of primes, this modern branch treats the primes as the simplest case of a much broader phenomenon: the behaviour of families of arithmetic objects as captured by their associated analytic functions.
Contemporary analytic number theory is a highly interconnected discipline. The classical core—the zeta function, Dirichlet L-functions, the Prime Number Theorem and its refinements—remains foundational, but the field has expanded in several directions.
One development is the rise of probabilistic number theory. The idea, pioneered by Erdős and Kac, is that many arithmetic functions behave like sums of independent random variables. Their theorem on the normal distribution of the number of prime factors of a random integer is a model for this viewpoint. It has been extended enormously: the Erdős–Kac theorem now has many proofs and generalizations, and probabilistic heuristics guide conjectures about primes, such as the distribution of prime gaps and the prime tuples conjecture.
A second development is the use of higher-order Fourier analysis, which began with Gowers's work in additive combinatorics and was brought into number theory by Green and Tao. Their theorem that the primes contain arbitrarily long arithmetic progressions is perhaps the most famous result of the modern era. The proof is a delicate combination of the circle method, sieve theory, and new structural results about functions on finite groups. It shows that the primes, despite their irregularity, contain the same additive patterns one would expect from a random set of density \(1/\log x\).
A third is the moments of L-functions problem. The conjecture of Keating and Snaith, based on random matrix theory, predicts the size of mixed moments of L-functions at the critical point. Progress has been slow but steady, and the connection between L-functions and the eigenvalues of random matrices remains one of the most mysterious and promising bridges between number theory and mathematical physics.
Throughout these developments, the unresolved questions remain the same ones Riemann posed: Where are the zeros of the zeta function? How regular are the primes? Each partial answer has required new methods, and each new method has opened new questions. The field is not a collection of solved problems but a web of connected conjectures, with the Riemann Hypothesis at its center. The most striking feature of analytic number theory is that, despite more than a century of work, the gap between conjecture and proof remains enormous—and yet the partial results have proven extraordinarily useful, both within mathematics and beyond it, in areas ranging from cryptography to the theory of random walks on groups.