In 1867, William Thomson watched smoke rings collide. Peter Guthrie Tait had built a device in his Edinburgh laboratory that fired vortex rings across the room — fat, stable loops of smoke that bounced off each other, vibrated, and refused to dissipate. Thomson, who would become Lord Kelvin, saw in these rings something no one else saw: atoms.
His idea was gorgeous. Atoms are knots tied in the luminiferous ether — the invisible medium that Victorian physics believed filled all space. Different knot types produce different elements. The unknot is hydrogen. The trefoil is something heavier. The complexity of the knot determines the complexity of the atom. This explains, at a stroke, why matter comes in a small number of distinct types, each identical to every other instance of itself. A hydrogen atom in London is the same as one in Calcutta because both are the same knot.
The theory is wrong. Completely, beautifully wrong. Einstein killed the ether in 1905. Quantum mechanics provided a radically different and successful account of atomic structure. Kelvin’s vortex atoms belong to the graveyard of discarded physics, alongside caloric fluid and phlogiston.
But Tait, captivated by the vortex theory, had begun tabulating knots — classifying every possible knotted loop by the number of crossings in its simplest diagram. He was building a periodic table of atoms. He found 163 distinct knots with ten crossings or fewer, working entirely by hand. Along the way he formulated several conjectures about the structure of alternating knots — knots where the crossings alternate over-under-over-under as you trace the strand.
The physics died. The mathematics survived.
A knot, mathematically, is a closed loop embedded in three-dimensional space. Take a piece of rope, tangle it however you like, then fuse the two ends together. The fundamental question is deceptively simple: given two tangled loops, can one be deformed into the other without cutting? If yes, they are the same knot. If no, they are different. But how do you prove they are different? You cannot try every possible deformation — there are infinitely many. You need an invariant: a quantity you can compute from a knot diagram that is guaranteed to give the same answer no matter how you draw the diagram, as long as the underlying knot is the same.
In 1923, James Waddell Alexander found the first polynomial invariant. Given any knot, his method produces a polynomial — a mathematical expression that serves as a kind of fingerprint. If two knots have different Alexander polynomials, they are definitely different knots. If they have the same polynomial, they might be the same — or they might not. The fingerprint has resolution limits.
One limit in particular: the Alexander polynomial cannot tell a knot from its mirror image. The right-handed trefoil and the left-handed trefoil — mirror reflections of the simplest nontrivial knot — produce the same polynomial. This is like a tool for identifying hands that cannot distinguish left from right.
For sixty-one years, the Alexander polynomial was the only polynomial invariant anyone had. Sixty-one years is a long time for a field to have exactly one tool of a certain kind. Not for lack of trying. Topologists searched. They found other invariants — fundamental groups, colorings, Seifert matrices — but no new polynomial. The problem was not that the idea was exhausted. The problem was that the next polynomial was hiding in the last place anyone would look.
Vaughan Jones was not a topologist. Born in Gisborne, New Zealand, on the last day of 1952, he trained as an operator algebraist — a mathematician who studies infinite-dimensional spaces of bounded operators, the abstract structures that underlie quantum mechanics. His doctoral work, completed in Geneva in 1979, concerned actions of finite groups on a specific object called the hyperfinite type II1 factor. This is about as far from knots and tangles as mathematics gets.
In the early 1980s, at the University of Pennsylvania, Jones was studying subfactors — inclusions of one von Neumann algebra inside another. He developed an index theory for these inclusions and proved a remarkable rigidity result: below 4, the possible index values do not form a continuum but a discrete set, 4cos2(π/n) for integers n ≥ 3. The first few: 1, 2, ½(3 + √5) ≈ 2.618, 3, … The gaps between allowed values are themselves mathematical facts, rigid as crystal lattice spacings.
While computing with his subfactor indices, Jones noticed that the algebraic relations satisfied by his generators were identical to those of something called the Temperley-Lieb algebra — a structure from statistical mechanics. More crucially, he had constructed representations of the braid group without knowing it. Braids — collections of strands running from one bar to another, allowed to cross over each other — are the raw material from which knots are made. Close the top to the bottom, and a braid becomes a knot.
Jones did not realize this. He was an operator algebraist. He did not think in braids.
The chain of recognition began in the spring of 1984. Caroline Series, a mathematician on sabbatical at the Institute for Advanced Study in Princeton, happened to be there when Jones described the algebra he was working with. She recognized the braid group and told him: you must go and talk to Joan Birman about this.
Joan Birman, at Columbia University, was the world’s leading expert on braid groups. On May 14, 1984, Jones visited her office. He showed her his Hecke algebra representations. She explained two things: first, how knots and links can be formed from closed braids; second, a theorem by Andrei Markov that describes exactly when two different braids give the same link. Markov’s theorem is the bridge between algebra and topology. It tells you what algebraic conditions a quantity must satisfy to be a genuine knot invariant — not just a braid invariant.
In the week that followed, Jones made the critical leap. He realized that by rescaling his trace function — a specific algebraic adjustment — both of Markov’s moves would affect the trace in the same way. The rescaling transformed an algebraic curiosity into a topological invariant. He called Birman, in great excitement.
On May 22 they met again. Jones announced: Look, I rescaled my representations, and now I have a polynomial invariant. But it must be the Alexander polynomial.
It was a reasonable assumption. The Alexander polynomial had been alone for sixty-one years. If you find a polynomial invariant for knots, the default hypothesis is that you have rediscovered the only one anyone knows.
Birman suggested they test it. Compute the polynomial for the right-handed trefoil and the left-handed trefoil — the two knots that the Alexander polynomial cannot tell apart. Within minutes, they had their answer. The two values were different.
It was not the Alexander polynomial. It was something entirely new.
Eight days. From first meeting to confirmed discovery: eight days. The drought had lasted 22,265.
Jones offered champagne. Birman declined, but insisted on adequate credit for her role — the Markov theorem insight was the key that made the trace into an invariant. She then made what she later called a snap decision: she chose not to pursue further work on the polynomial, having committed to a collaboration with Series on geodesics. I made a snap decision, and, yes, I had some small regrets, but basically I was OK with my decision.
The Jones polynomial, published in January 1985, detonated. Within a year, five independent groups generalized it — so many that the resulting invariant carries six names: HOMFLY-PT, for Hoste, Ocneanu, Millett, Freyd, Lickorish, Yetter, Przytycki, and Traczyk. In 1987, three mathematicians independently used the Jones polynomial to prove Tait’s conjectures about alternating knots — conjectures from the 1880s that had resisted proof for a century. The wrong physics from 1867 had produced the right conjectures, and the tool to prove them had arrived from operator algebras via a chain of accidental conversations.
In 1989, Edward Witten showed that the Jones polynomial emerges naturally from Chern-Simons gauge theory — a topological quantum field theory in 2+1 dimensions. The invariant is a partition function. It has physical meaning, though that meaning lives in a framework — quantum field theory — that is itself not fully rigorous. In 1990, both Jones and Witten received the Fields Medal. Jones for the polynomial, Witten partly for explaining where it comes from.
What the Jones polynomial can do: distinguish knots the Alexander polynomial cannot, detect chirality (tell a knot from its mirror image), and prove century-old conjectures about alternating knots. What it cannot do: distinguish all knots. Infinitely many non-equivalent knots share the same Jones polynomial. It is not even known whether the Jones polynomial can always detect the unknot — whether a knotted loop that happens to have the same Jones polynomial as a plain circle is necessarily a plain circle. This is an open problem.
We can tell many knots apart. We often cannot explain why they are different. The invariant is a detector, not an explanation. It assigns a number — or rather a polynomial — and when two numbers differ, we know the knots differ. But the number does not point to the geometric feature that makes them different. It does not say: here is where one knot twists in a way the other cannot.
This is the permanent tension of the field. Tait tabulated 163 knots by hand. By now, all prime knots up to 19 crossings have been classified — hundreds of millions of distinct knots. We can list them. We can tell them apart. We cannot fully describe the shape of the space they inhabit.
What I keep returning to is the accident. Not the minor kind — a slip that reveals something adjacent. A structural accident: the solution to a problem in topology was hiding inside a problem in operator algebras that had no topological content whatsoever. Jones was not looking for knots. He was not looking for braids. He was computing traces of operators in subfactors of von Neumann algebras, and the algebraic relations he found happened — for reasons no one fully understands even now — to encode the topology of knotted curves in three-dimensional space.
The human chain matters. Jones alone would not have recognized the braid group structure — he was not trained to see it. Series recognized it but did not have the topological expertise. Birman had the expertise but not the algebraic construction. Each person held one piece. The polynomial existed, in some abstract sense, from the moment Jones wrote down his subfactor representations. But it required three people, two chance encounters, and eight days to become visible.
Kelvin, watching Tait’s smoke rings in 1867, saw the wrong thing for the right reason. Atoms are not knots. But the intuition that the topology of tangled loops encodes something fundamental about physical reality turned out to be correct — not in the way he imagined, but in ways he could not have: DNA molecules are knotted, and the enzymes that manage them are topological machines. Topological quantum computation proposes to encode information in braids of exotic quasiparticles, protected from noise by the same topological rigidity that makes knots hard to untie. Jones himself, before his death in 2020, co-authored work on quantum algorithms for approximating his own polynomial.
The wrong theory produced the right mathematics. The right mathematics waited sixty-one years. The person who broke the drought was working on something else entirely. And the connection, when it came, required not insight but introduction — one mathematician telling another: you must go talk to Joan.
The deepest results do not come from working harder on the problem. They come from working honestly on a different problem until the two problems turn out to be the same.