Seven questions mathematics couldn't answer in a hundred years, each with a one-million-dollar bounty. Here they are explained without scary formulas, with things to play with. One is already solved, and another had an announcement three days ago.
Every problem has a demo on the right. Look for the dashed box "How to play": it tells you what to drag and what to press.
per problem. The Clay Mathematics Institute, a private foundation in Massachusetts, chose them in 2000 as "the great unfinished business of the 20th century".
is how long a proof must wait after being published in a peer-reviewed journal, and be accepted by the community, before Clay will consider paying.
has been solved: Poincaré, by Grigori Perelman, who turned down the million in 2010. The other six remain officially open.
There are problems where checking a solution takes an instant, but finding one seems to require trying everything. Is that difference real, or have we just not found the shortcut yet?
Think of a combination lock. If someone tells you "it's 4-7-2-9", you check it in a second. But if nobody tells you, the only way we know is to try combination after combination. Mathematicians call P the problems that can be solved quickly, and NP the ones that can be checked quickly. Everything in P is in NP. The question is whether they're the same.
If P = NP, there would be a fast method for everything that's slow today: building perfect schedules, folding proteins, cracking any password. Almost everyone believes P ≠ NP, but nobody has been able to prove it. And "we believe" doesn't count.
All of internet security (your online banking, WhatsApp) rests on certain problems being hard to solve but easy to check. If P = NP with a practical algorithm, that collapses. It would also revolutionize science: optimizing almost anything would become trivial.
Formulated by Stephen Cook in 1971. Thousands of "NP-complete" problems are known to be equivalent: solve one quickly and you solve them all. Many known techniques have been proven unable to crack it, so a completely new idea is needed.
Each extra digit multiplies the search by 10 but changes nothing about checking. That asymmetry, scaled up to big problems, is what separates "one afternoon" from "longer than the age of the universe".
Primes seem to show up at random. Riemann found a function whose "zeros" control that randomness, and in 1859 he conjectured that all the important zeros line up on a single straight line.
Primes (2, 3, 5, 7, 11, 13...) are the atoms of numbers: every whole number is built by multiplying them. But nobody has a formula for which one comes next. Riemann discovered a function, the zeta, that "listens" to the primes like a sound wave. The points where that function equals zero are like the notes of that wave.
His hypothesis says all those notes sit on the same line (the "critical line", where the real part equals ½). More than ten trillion zeros have been verified and every one complies. But "all the ones we looked at" is not "all".
Hundreds of theorems begin with "if the Riemann Hypothesis is true, then...". Proving it confirms them all at once. It also pins down exactly how far primes behave "nicely", which underlies modern cryptography.
It's the oldest problem on the list and the only one that also appeared on Hilbert's famous 1900 list. There are partial results (it's known that "at least 40%" of the zeros are on the line), and every so often a claimed proof appears that doesn't survive review.
Top: the first real zeros of the zeta function, all on the ½ line (the numbers are their heights: 14.13 · 21.02 · 25.01…). Bottom: the primes up to the number you pick. If a single zero stepped off the line, the hypothesis would be false and the "rhythm" of the primes would have a stumble.
Physicists use Yang–Mills theory to describe the strong force, the one that glues quarks inside protons. It works in experiments. But nobody has built the theory with mathematical rigor, or proven why its lightest particles have mass.
Light (electromagnetism) can have any energy, as tiny as you like: there's no floor. The strong force is different: its lightest excitation has a minimum mass, a "gap" below which there is nothing. Experiments confirm it, simulations confirm it. The proof is missing.
The prize asks for two things: prove the theory exists as a well-defined mathematical object, and prove it has that mass gap.
It would be the first "realistic" quantum field theory in four dimensions built with full rigor. Today particle physics runs on mathematics that, strictly speaking, is not yet justified.
It's the most "physicist" problem on the list. In simpler theories (two or three dimensions) it has been achieved. In four dimensions, it isn't even clear what the correct definition of the object to study should be.
Left: electromagnetic field, any energy produces a wave, even a minuscule one. Right: Yang–Mills field, below the gap nothing happens; from the minimum up, a particle with mass appears. Move the slider and notice when each side "switches on".
There are enormously complicated geometric shapes that can be understood by counting their "holes" and cycles. Hodge conjectured that, for shapes defined by polynomial equations, those cycles can always be assembled by gluing pieces that also come from equations.
Picture a donut. It has two essential loops: around the hole and around the tube. Those loops are topological information (about the shape, without measuring anything). Now imagine the donut is defined by an equation. Question: can those loops be drawn using only other equations? On the donut, yes. Hodge bet that, for a certain kind of "loop" (those of type (p,p)), it's always like that, in any dimension.
It's the most abstract of the seven. Many mathematicians admit they don't even know why it should be true; it's just that no counterexample has been found.
It's a bridge between two worlds: topology (soft shapes, holes) and algebra (equations, exactness). If true, algebraic geometry gains a tool for understanding shapes that are impossible to visualize.
William Hodge proposed it in 1950. It's proven in low dimensions (Lefschetz, 1924, for the simplest case). In higher dimensions there is evidence in both directions, and a more ambitious version has already been proven false.
Each cycle of the donut can be described by an equation (for example, slicing it with a plane). Hodge says that always happens, for objects with hundreds of dimensions nobody can draw. Tap the cycles to see what each is made of.
Elliptic curves are equations like y² = x³ + ax + b. Asking how many solutions they have using only fractions is extremely hard. The conjecture says a function (the L-function) tells you the answer just by looking at how it behaves at one point.
These curves have a trick: if you have two points with fractional coordinates, the line joining them meets the curve at a third point, which is also fractional. So from a few points you can generate infinitely many, or get stuck in a finite cycle. The number of independent "generators" you need is called the rank.
Birch and Swinnerton-Dyer, in the 1960s, used one of Cambridge's first computers and noticed a pattern: the rank seems to be exactly the "order of the zero" of the L-function at the point s = 1. It's a brazen numerical coincidence nobody knows how to explain.
Elliptic curves are in your phone's cryptography and were the key to proving Fermat's Last Theorem. BSD would give a practical way to know how many solutions an equation has, one of the oldest problems in mathematics (Diophantus, 3rd century).
It's proven when the rank is 0 or 1 (Kolyvagin, Gross–Zagier, 1980s). For rank 2 or higher almost nothing is known. Manjul Bhargava (Fields Medal 2014) proved it holds for "a large proportion" of curves, but not for all.
Move the sliders to change the shape. "Add P + Q" draws the line through two points, finds the third and reflects it: that's the curve's "addition". Repeating it generates more and more points; the BSD conjecture predicts whether that process gives infinitely many or closes on itself.
On a sphere, any loop you draw can slide around and shrink until it vanishes. On a donut, it can't: the loop gets caught on the hole. Poincaré asked in 1904 whether that property is enough to identify the sphere in three dimensions.
For two-dimensional surfaces (like the skin of a ball) this had been known for over a century. The problem was the three-dimensional version, the "3-sphere", which we can't see because it lives in four dimensions. The question sounds like common sense and resisted for almost a hundred years.
In 2002 and 2003, Grigori Perelman, a Russian mathematician, posted three papers online without sending them to any journal. He used a technique called "Ricci flow" (devised by Richard Hamilton): let the shape smooth itself out like dough stretching on its own, and show it always ends up a sphere. It took the community years to verify.
Perelman's proof actually established something much bigger (Thurston's geometrization): a complete classification of all possible three-dimensional shapes. It's one of the summits of 21st-century mathematics.
With a tool from another branch (analysis of differential equations) applied to a topology problem. It's the typical pattern of great breakthroughs: the problem doesn't yield from inside, it yields when someone brings an idea from outside.
Left: on the sphere, the loop contracts until it disappears. Right: on the donut, the loop catches on the tube and can't get any smaller. That difference is what Poincaré proposed as the sphere's "fingerprint".
This page reflects the status of the problems as of September 10, 2026. Sources: the Clay Mathematics Institute, Quanta Magazine and Nature on the Navier–Stokes announcements. The demos are cartoons to build intuition, not exact simulations.
Made by Agustina Cassi with Claude Fable 5.1