The State of the Art

You have arrived at the end of the course — and the edge of the map. Behind you lies a whole machinery for interrogating the primes: the Prime Number Theorem and its zero-free regions, the circle method, sieves, L-functions, and the astonishing bridge from zeta zeros to random matrices. This final page is not a new tool. It is a map of the whole territory — a survey of what has been proved, what remains stubbornly open, and where the living frontier of prime-number research runs today.

The primes are the atoms of arithmetic, and for two and a half millennia they have refused to give up their pattern. What is remarkable is how much we have wrestled from them — and how sharply the questions that remain still resist. Let us take stock, honestly, of both.

SOLVED — what we actually know

A surprising amount of what once looked hopeless is now theorem. These are not conjectures with strong evidence; they are proved, and they anchor the whole subject.

The Prime Number Theorem (Hadamard & de la Vallée Poussin, 1896): \pi(x)\sim x/\ln x. The sharpest unconditional error term, from the Vinogradov–Korobov zero-free region, gives \pi(x)=\operatorname{Li}(x)+O\!\big(x\,e^{-c(\ln x)^{3/5}(\ln\ln x)^{-1/5}}\big).
Dirichlet's theorem (1837) and PNT in arithmetic progressions: every progression a, a+q, a+2q,\dots with \gcd(a,q)=1 contains infinitely many primes — and they share out evenly, \pi(x;q,a)\sim \tfrac{1}{\varphi(q)}\operatorname{Li}(x).
Ternary Goldbach (Helfgott, 2013): every odd number \ge 7 is a sum of three primes — now fully proved, unconditionally.
Bounded gaps between primes (Zhang 2013; Maynard & the Polymath project 2014): infinitely many prime pairs differ by at most a fixed constant — pushed down to \le 246.
Green–Tao (2004): the primes contain arbitrarily long arithmetic progressions.
Sato–Tate (for non-CM elliptic curves, 2008–11) and the Modularity Theorem (Wiles, Taylor–Wiles, and the full proof by Breuil–Conrad–Diamond–Taylor, 2001): every rational elliptic curve is modular — the engine behind Fermat's Last Theorem.
AKS (Agrawal–Kayal–Saxena, 2002): primality testing is in P — a deterministic polynomial-time algorithm, no unproven hypothesis required.

OPEN — the questions that still defeat us

Against that ledger of victories stands a roll-call of problems that are, by any honest measure, unsolved — some of them centuries old, some phrased simply enough for a child, all of them beyond current reach.

The Riemann Hypothesis and its generalisation GRH — the crown jewel. Every non-trivial zero of \zeta (and of every Dirichlet L-function) on the line \Re(s)=\tfrac12. A Clay Millennium Problem.
The twin prime conjecture: infinitely many primes with gap exactly 2. Bounded gaps got us to 246; the last step to 2 is blocked by the parity barrier.
Binary (strong) Goldbach: every even number \ge 4 is a sum of two primes. The ternary case fell in 2013; the binary case stands.
The Lindelöf hypothesis: \zeta(\tfrac12+it)=O(t^{\varepsilon}) — weaker than RH, still open.
Cramér's conjecture on prime gaps: g_n=O\!\big((\ln p_n)^2\big).
The Elliott–Halberstam conjecture on the distribution of primes in progressions — which, if true, would push bounded gaps down to 6.
Landau's problem: are there infinitely many primes of the form n^2+1? Not one polynomial of degree \ge 2 is known to produce infinitely many primes.
Birch and Swinnerton-Dyer (BSD) — another Millennium Problem — linking the rank of an elliptic curve to the order of vanishing of its L-function at s=1.

The ledger — solved vs open at a glance

Here is the centrepiece: the major problems of prime-number analysis, side by side, with dates and status. Read it as the state of the art in one table.

Problem	Statement (informally)	Status	Date / bound
Prime Number Theorem	\pi(x)\sim x/\ln x	SOLVED	1896
Dirichlet / PNT in APs	primes spread evenly over coprime residues	SOLVED	1837 / 1896
Ternary Goldbach	odd \ge 7 = sum of 3 primes	SOLVED	2013 (Helfgott)
Bounded prime gaps	gap \le C infinitely often	SOLVED	2013–14, \le 246
Green–Tao	arbitrarily long prime APs	SOLVED	2004
Sato–Tate (elliptic curves)	distribution of a_p	SOLVED	2008–11
Modularity theorem	rational elliptic curves are modular	SOLVED	1995 / 2001
AKS primality	PRIMES \in P	SOLVED	2002
Riemann Hypothesis / GRH	zeros on \Re(s)=\tfrac12	OPEN	since 1859
Twin primes	infinitely many gaps of 2	OPEN	—
Binary Goldbach	even \ge 4 = sum of 2 primes	OPEN	since 1742
Lindelöf hypothesis	\zeta(\tfrac12+it)=O(t^{\varepsilon})	OPEN	—
Cramér's conjecture	g_n=O((\ln p_n)^2)	OPEN	—
Elliott–Halberstam	primes in APs, level \to 1	OPEN	—
Primes n^2+1 (Landau)	infinitely many?	OPEN	—
Birch–Swinnerton-Dyer	rank = order of vanishing	OPEN	—

Notice the shape of the ledger: nearly everything about counting and averaging primes is solved, while nearly everything about individual, close, or specially-shaped primes is open. That is not a coincidence — it is the fingerprint of a single obstruction, which we meet next.

Seeing the frontier move: the bounded-gap chase

The frontier is not static — it lurches. The single most dramatic recent example is the collapse of the bounded-gap constant. In April 2013, Yitang Zhang announced the first finite bound — an astonishing 70{,}000{,}000. Within a year, the Polymath8 collaboration and then James Maynard's independent method hammered it down by more than five orders of magnitude, to 246. The plot below traces that fall (on a log scale).

The last stretch to 2 (twin primes) is not just more of the same grinding — it runs headlong into a genuine theoretical wall, the same wall that stops us proving binary Goldbach. That wall has a name.

FRONTIER — where the research lives

Beyond the ledger of named problems lies the frontier of methods — the ideas people are actually pushing on. Four strands dominate.

The parity barrier

Classical sieve theory has a built-in blind spot, made precise by Selberg: a sieve alone cannot tell a number with an even number of prime factors from one with an odd number. This parity problem is exactly why sieves get you to "P_2" (a prime or a product of two primes) but never to a genuine prime — and why twin primes and binary Goldbach remain out of reach. The great recent advances chip at the barrier rather than break it: Zhang and Maynard exploit new estimates for primes in arithmetic progressions to bound gaps; and Friedlander–Iwaniec (1998) achieved a landmark parity-breaking result, proving infinitely many primes of the form a^2+b^4 by importing extra "bilinear" arithmetic information a sieve alone can't see.

The Langlands program

The deepest organising vision in modern number theory. In one breath: the Langlands program is a vast web of conjectures predicting a dictionary between automorphic forms (analytic objects — generalised modular forms, and their L-functions) and Galois representations (algebraic symmetries of number fields). It reinterprets an L-function as coming from an automorphic representation, so that its analytic continuation, functional equation, and (conjecturally) its zeros are all forced by symmetry. The modularity theorem and Sato–Tate — both now proved — are the first fruits of this program; almost everything on the open side of the ledger would follow from its grand conjectures.

Additive combinatorics

The Green–Tao theorem grew from here: transferring tools built for dense sets (Szemerédi's theorem, Gowers norms, the regularity method) onto the sparse set of primes via a "transference principle." This is now a field of its own, feeding back into questions about additive structure in the primes.

Analytic number theory meets random matrices

Perhaps the most tantalising frontier of all: the statistics of zeta zeros match, to extraordinary precision, the eigenvalue statistics of large random unitary matrices (the Montgomery–Odlyzko phenomenon). The random-matrix philosophy now predicts moments of \zeta, distributions of L-function values, and more — a physicist's toolkit turned onto the primes, still largely conjectural but stunningly accurate.

This is the single most important habit of mind to carry out of the course. The Riemann Hypothesis has been verified for more than 10^{13} zeros; binary Goldbach has been checked for every even number up to about 4\times 10^{18}; twin primes have been found with hundreds of thousands of digits. None of this is a proof. Mathematics draws a hard line between "true for every case anyone has computed" and "true for every case, forever" — and there are always infinitely many cases left unchecked.

History has punished the complacent. The inequality \pi(x)<\operatorname{Li}(x) holds for every x anyone could compute — yet Littlewood proved in 1914 that it flips infinitely often, the first flip hiding somewhere near 10^{316} (Skewes' number), far beyond any search. Overwhelming numerical evidence is a reason to believe, never a licence to claim. RH and Goldbach are open — magnificently, stubbornly open — no matter how high the computers climb.

It feels like 246 \to 2 should be a formality. It is not. Every gap improvement so far comes from feeding the sieve better information about how primes distribute in arithmetic progressions (the "level of distribution"). But even granting the strongest such conjecture — Elliott–Halberstam — the machinery bottoms out at gap 6, not 2. The final step from 6 (or any even number) down to a guaranteed prime pair is precisely a parity-barrier crossing, and no known method crosses it in this setting. Twin primes will need a genuinely new idea, not a sharper constant.

The music of the integers

Why do the primes hold us like this? Because they sit exactly at the meeting point of the two great halves of mathematics — the discrete and the continuous. A prime is the most rigid, combinatorial object imaginable, and yet the deepest facts about primes are theorems of analysis: about smooth functions, contour integrals, and the zeros of \zeta. The explicit formula literally writes the fluctuations of the primes as a sum of waves, one for each zeta zero — which is why the zeros are so often called the "music of the integers". Prove the Riemann Hypothesis and you are saying every note in that music is perfectly in tune.

The honest state of the art is this: we can count the primes beautifully, and we can prove deep structural facts about them on average — but the primes as individuals, close together or specially shaped, still guard their secrets. The frontier is not a wall so much as a shoreline, and it is advancing.

Where you go next

If this course has done its job, you now have the vocabulary to read the frontier, not just admire it. Natural next steps:

The Langlands program — the unifying horizon; start with modular forms and automorphic L-functions.
Iwaniec–Kowalski, Analytic Number Theory — the field's standard graduate reference, and where every method in this course is pushed to its research edge.
Elliptic curves and BSD — the arithmetic-geometry doorway, via L-functions you already understand.
Sieve theory and additive combinatorics — where the bounded-gaps drama is still unfolding, and where a new idea for the parity barrier will one day come from.

Two and a half thousand years after Euclid proved there are infinitely many primes, we still cannot say whether infinitely many of them come in twos. That gap between what is obvious and what is provable is not a failure of the subject — it is the subject. Welcome to the frontier.