The Twin Prime Conjecture proposes that there are infinitely many pairs of prime numbers differing by exactly 2 — like (3,5), (11,13), (17,19), and (29,31). Open since Alphonse de Polignac's general 1849 formulation, it saw a stunning 2013 breakthrough when a virtually unknown lecturer, Yitang Zhang, proved a bounded gap between infinitely many primes exists at all — a result the mathematical community rapidly refined, though the specific gap of exactly 2 remains unproven.
As numbers grow larger, prime numbers become progressively rarer and more spread out — yet remarkably, pairs of primes separated by just 2 keep appearing, seemingly forever: (3,5), (5,7), (11,13), (17,19), (29,31), (41,43)... all the way up to enormous pairs discovered only recently by dedicated computer searches. The Twin Prime Conjecture proposes that this pattern genuinely never stops — that there are infinitely many such twin prime pairs, no matter how far out along the number line you search.
Zhang's own original bound of 70 million was, by his own account, deliberately not optimised — he prioritised getting a correct, rigorously verifiable proof out quickly rather than spending additional years trying to minimise the bound himself. Once published, an enormous, rapid, highly collaborative online mathematical effort (the Polymath8 project, coordinated in part by Terence Tao) progressively refined the techniques, driving the proven gap down to just 246 within about a year — though even 246 remains far short of the twin prime conjecture's specific target of exactly 2.
Zhang's breakthrough became one of the most widely reported pure mathematics stories of the 2010s specifically because of his unusual personal background — years working outside academia (including, according to some accounts, at a Subway sandwich shop) before eventually returning to serious mathematical research — combined with the sheer scale of the achievement on a problem that had resisted dramatic progress for well over a century.
The Twin Prime Conjecture states: there exist infinitely many primes p such that p+2 is also prime. It's a special case of a more general 1849 conjecture by Alphonse de Polignac, which proposed that for every even number k, there are infinitely many prime pairs differing by exactly k (twin primes being the specific case k=2).
Yitang Zhang proved in 2013 that there exists SOME specific number N (he showed N could be taken as 70,000,000) such that infinitely many pairs of primes differ by at most N. This was the first unconditional proof that primes don't spread out indefinitely without bound — some finite gap size recurs infinitely often — even though which specific gap sizes recur (and whether 2 itself is among them) remained unresolved by Zhang's argument alone.
Terence Tao organised an open, massively collaborative online research project (Polymath8) immediately following Zhang's announcement, bringing together numerous mathematicians worldwide to optimise the technical details of Zhang's method. This rapidly reduced the proven bound from 70 million to 4,680, and shortly after, James Maynard and (independently) Terence Tao developed a substantially different, more efficient sieve-theoretic approach that further reduced the bound to 246 (as of 2014) — the current best unconditional bound, still standing as of 2026. Under certain additional, widely-believed but unproven conjectures (specifically the Elliott-Halberstam conjecture), the bound can be further reduced to 6, tantalisingly close to but still not reaching the twin prime target of exactly 2.
Both Zhang's original proof and the subsequent refinements rely on sieve theory — a family of techniques for estimating the number of integers in a range satisfying certain divisibility conditions, tracing back to the ancient Sieve of Eratosthenes for finding primes, but developed into an enormously sophisticated modern analytic toolkit (particularly via the Goldston-Pintz-Yıldırım method, 2005, which supplied crucial technical groundwork that Zhang and Maynard both built directly upon).
Daniel Goldston, János Pintz, and Cem Yıldırım (2005) developed a refined sieve-theoretic method that came remarkably close to proving bounded prime gaps — falling short only because a technical input (an improved "level of distribution" for primes in arithmetic progressions, related to the Bombieri-Vinogradov theorem) wasn't yet available at a strong enough level. Zhang's key technical contribution in 2013 was precisely supplying an unconditional improvement to exactly this needed technical ingredient, for a restricted but sufficient class of arithmetic progressions — finally allowing the GPY framework to yield an unconditional, if not yet optimal, bounded-gap result.
James Maynard and (independently, within days) Terence Tao developed, in late 2013, a substantially generalised and more flexible multidimensional sieve method that both simplified Zhang's argument considerably and pushed the achievable bound down dramatically to 246, without requiring Zhang's specific technical breakthrough on the level of distribution at all — an independent and more broadly applicable route to essentially the same style of bounded-gap result, and now the standard modern approach to the entire family of such problems.
Current sieve-theoretic techniques suffer from a well-understood, rigorously identified fundamental limitation called the parity problem in sieve theory: these methods are, in a precise technical sense, unable to distinguish between numbers with an even number of prime factors and those with an odd number, which provably prevents current sieve techniques alone from ever reaching all the way down to a gap of exactly 2 without some entirely new additional mathematical idea beyond currently known sieve methods.
A vastly more general 1923 conjecture by Hardy and Littlewood predicts the precise asymptotic density of prime constellations of essentially any specified pattern (not just pairs differing by 2), including a specific predicted quantitative density for twin primes themselves. This conjecture, if proven, would immediately imply the Twin Prime Conjecture as a special case — but the general Hardy-Littlewood conjecture remains just as unproven as the twin-prime special case it would resolve, illustrating how this entire cluster of additive prime-distribution questions (also connecting closely to the Goldbach Conjecture — see that entry) remains substantially unified by shared open technical obstacles.