The Knots Were Right
In 1867, Kelvin and Tait built the wrong theory of everything from the correct mathematics. It took 150 years to find out.
In 1867, Peter Guthrie Tait blew a smoke ring at Lord Kelvin.
This is a more significant event in the history of science than it sounds.
Tait had built a smoke-ring generator from a wooden box with a hole in one end — a rubber-stretched drum on the other — and was demonstrating it in his Edinburgh lecture room. Kelvin watched as the rings floated across the room, held their shape, passed through each other, wobbled and recovered. Unlike ordinary puffs of smoke, which dissipate immediately, the rings were stable. Persistent. Self-organising. They behaved, Kelvin felt with sudden force, exactly like atoms ought to behave.
He wrote afterwards that “a magnificent display of smoke rings diminished by one the number of assumptions required to explain the properties of matter.” Then he sat down and began building a theory of everything.
The theory was wrong. The mathematics was not.
One: The Three Scots
To understand what happened next, you have to understand the world William Thomson (who would become Lord Kelvin), Peter Guthrie Tait, and James Clerk Maxwell inhabited together.
They were all Scots. They had overlapping educations — Edinburgh and Cambridge — and met at the nodes: British Association conferences, university appointments, the constant traffic of letters. Tait and Maxwell were schoolboys together at Edinburgh Academy, where Maxwell arrived at age ten already a published mathematician, making geometric discoveries about ovals that his father had to present to the Royal Society because the boy was too young to attend. Tait later recalled that Maxwell “at first seemed shy and rather dull,” but that this impression dissolved rapidly. They became fast friends for life.
Maxwell was the warmth at the centre of this whole story. He and Tait corresponded for decades in a register of elaborate private jokes — Maxwell signed postcards as “dp/dt” (which equals JCM, his initials, by thermodynamic notation), wrote letters backwards, peppered everything with Latin riddles. Kelvin and Tait’s great joint textbook on Natural Philosophy became “T and T’” in their letters; the Archbishops of York and Canterbury (who were at the time also called Thomson and Tait) were dubbed “the Archepiscopal Pair.” Members of the British Association for the Advancement of Science were “British Asses.” The three of them spent years exchanging jokes that would require a page of footnotes to unpack.
This intimacy matters because the vortex atom was not a solo theory. It was a collaborative bet, placed by close friends who trusted each other’s mathematical instincts completely, on a physical picture that Kelvin would not let go of for thirty years.
Two: The Programme
Kelvin’s vortex atom was elegant. Its central idea: atoms are stable knots in the aether, the frictionless fluid that everyone in 1867 believed filled all of space. Like Tait’s smoke rings, they would be persistent and self-organising. Different knots — different atoms. The trefoil knot is hydrogen. Some other knot is carbon. The periodic table is a knot table — the Scots’ knots, tying matter itself together.
Maxwell immediately saw both the beauty and the work this required. Someone had to catalogue all possible knots. “Someone” turned out to be Tait.
Between 1876 and 1885, Tait tabulated every knot up to seven crossings by hand. Seven papers in the Proceedings of the Royal Society of Edinburgh. He wrote to the British Association that the classification “promises absolutely endless work” and acknowledged that the labour “increases with extreme rapidity as the number of crossings is increased.” He was developing what we would now call combinatorial topology, without the benefit of combinatorial topology. He had no rigorous framework, no machinery, no algorithms. He had geometric intuition and extraordinary patience. He worked for nearly a decade in service of a theory of matter that would turn out to be wrong.
What he produced was the foundation of modern knot theory.
Maxwell died in 1879. Abdominal cancer. He was 48 — the same age and the same cause as his mother, who died when he was nine. Tait wrote his obituary for the Royal Society of Edinburgh and could not quite keep the grief out of it. The playfulness went out of the programme. Kelvin, now in his late fifties and increasingly authoritative, kept going.
Three: The Collapse
The end came from two directions at once.
In 1887, Heinrich Hertz set out to prove that James Clerk Maxwell was right about electromagnetic waves. He succeeded completely. The experiment demonstrated beyond doubt that electromagnetic radiation propagates through space as Maxwell had predicted — and in doing so, it made the aether almost unnecessary as a concept. If electromagnetic waves could travel through empty space, you didn’t need a fluid medium to carry them. And if you didn’t need the aether, you didn’t need Kelvin’s knotted vortex tubes.
Hertz had no idea what he was ending. A student asked him what his discovery was good for. “It’s of no use whatsoever,” Hertz replied. “This is just an experiment that proves Maestro Maxwell was right — we just have these mysterious electromagnetic waves that we cannot see with the naked eye.” He died of blood poisoning in 1894, aged 36. Marconi would transmit across the Atlantic eight years later. Hertz never knew what he had started, or what he had finished.
The second blow came from inside. J.J. Thomson — who had won the Adams Prize in 1882 for work on the interaction of vortex rings, a product of Kelvin’s own programme — discovered the electron in 1897. Atoms have parts. Therefore they cannot be elementary vortices. The entire edifice collapsed.
Kelvin, by now 73, turned the wreckage into an epitaph with characteristic economy: “The vortex theory is only a dream. Itself unproven, it can prove nothing, and any speculations founded upon it are mere dreams about dreams.”
Three years later he gave his famous “Two Clouds” lecture at the Royal Institution — identifying two small dark patches on the otherwise clear horizon of classical physics. One was the failure of the Michelson-Morley experiment to detect the aether. The other was the problem of black-body radiation. Both clouds, he noted, were small. Both would turn out to be thunderstorms. One became special relativity. The other became quantum mechanics. Kelvin lived until 1907 and could follow into neither.
Tait had died in 1901, his knot tables completed. He never saw their vindication.
Four: The Mathematics Waits
Here is the thing about the mathematics Tait produced: it was not wrong. Knot theory did not depend on the aether for its validity. The tables he compiled for atoms that turned out not to exist were mathematically correct. They sat there, quietly correct, for fifty years while physics moved on.
Three things then happened in rapid succession, unconnected to each other and to Tait, that would eventually make his tables relevant again.
In 1931, Heinz Hopf discovered what is now called the Hopf fibration. He was working in Zurich, recently arrived to take Hermann Weyl’s chair at ETH, and he had found the first example of a map between spheres of different dimensions that could not be continuously deformed into a constant map. The three-sphere S³ — the surface of a four-dimensional ball — could be decomposed into circles, one through each point, stacked over a two-sphere in a precise and beautiful way. This was abstract topology, a branch of mathematics so pure that its practitioners prided themselves on its uselessness. Hopf spent the Nazi years in Switzerland in legal limbo — German citizen, Jewish grandfather, sheltering refugees, seeking Swiss citizenship only when told to return to Germany or lose his passport. He kept working. The fibration he discovered has since appeared in every corner of physics. He had no idea it would eventually touch the structure of atoms.
In 1935, Vladimir Fock published a paper in Soviet physics showing something remarkable about hydrogen. When you transform hydrogen’s bound-state wavefunctions from position space to momentum space and then project them onto a four-dimensional sphere, a hidden symmetry becomes visible: the energy levels of hydrogen, which depend only on n, are exactly what you would expect for the symmetry group of the four-sphere. The three-sphere S³ appears naturally as the momentum-space representation of hydrogen’s orbital structure. Fock was working in Leningrad, and he published this in 1935, which was the year Stalin’s terror was beginning to reach into Soviet science. He was arrested in 1937 on fabricated charges. His colleague Pyotr Kapitsa wrote directly to Stalin to secure his release. Fock returned to his office and kept working.
In 1936, Erwin Madelung published a handbook of mathematical methods for physicists. In it, almost as an aside, he noted that electrons fill atomic orbitals in order of increasing n+ℓ, where n is the principal quantum number and ℓ is the angular momentum. He offered no theoretical justification. He simply observed that it worked. The rule had actually been noticed before him — by the French engineer Charles Janet in 1929, and by Vladimir Karapetoff in 1930 — but Madelung’s name attached itself to it, the way names do. It sat in his handbook, without explanation, for nearly ninety years.
None of these three men knew they were handing a baton to each other.
Five: What S³ Actually Delivers
Here is the part that is solid, and it is not ours.
In 1935 Vladimir Fock showed that hydrogen’s momentum-space wavefunctions are hyperspherical harmonics on S³, the three-sphere. This is the origin of the n² degeneracy and of the hidden SO(4) symmetry. It is a celebrated result.
And it explains something the textbook picture leaves as a coincidence. In ℝ³, the fact that 2s and 2p have the same energy is called an accidental degeneracy, and nothing accounts for it. On S³ it stops being accidental: the whole n = 2 shell is a single SO(4) irreducible representation, and ℓ and m are just labels for how that one object decomposes under rotations. The constraint ℓ < n, which the textbook derives from a differential equation, is on S³ the SO(4) → SO(3) branching rule: the irrep of dimension n² decomposes into exactly ℓ = 0, 1, …, n−1, giving n² = Σ(2ℓ+1). Pure representation theory.
So Kelvin’s instinct about the arena was better than anyone realised. Atoms really do have a natural home on a three-sphere. He was sixty-eight years early and looking at the wrong inhabitants.
Six: The Knots Are Not the Inhabitants
If orbitals live on S³, and S³ is full of torus knots, perhaps each orbital is a knot. We tried this. It does not work, and the way it fails is instructive.
The obvious map sends orbital (n, ℓ) to T(ℓ+1, n−ℓ). Under it two entire families collapse to the unknot — every ℓ = 0 and every ℓ = n−1 orbital — and through n = 5 exactly two orbitals get a genuine knot.
A better map exists, and it very nearly works:
T(n+ℓ+1, n−ℓ)
It knots 14 of the 21 orbitals up to n = 6, against 6 for the obvious map. Both indices mean something: n+ℓ+1 is the Madelung number plus one, and n−ℓ is the radial node count plus one. Better still, its unknots fall exactly on ℓ = n−1 — the circular Bohr orbits, the ones with zero radial nodes. A nodeless orbit getting an unknotted curve looks less like a failure than a result.
It even seems to impose a selection rule: p and q always come out with opposite parity, excluding half of all coprime pairs. If something physical forced that, it would be the first genuinely topological constraint in the programme.
Both hopes fail, and not because of anything specific to this map.
The parity rule is an artefact of arithmetic. For any map p = an+bℓ+c, q = dn+eℓ+f, the parity of p+q is fixed by c+f — the offsets. Map G adds 1 to p and 0 to q, so the rule records the +1. Drop it and the same orbitals obey the opposite rule.
And the deeper obstruction defeats every map at once. For a torus knot T(p, q), every standard invariant — genus, crossing number, Alexander and Jones polynomials, signature, braid index — is a function of (p, q) alone. If p and q come from n and ℓ, so does every invariant. The knot cannot tell you anything the quantum numbers do not already say.
There is also something the topology simply cannot express. n and ℓ are asymmetric — ℓ is bounded by n, n is unbounded — but T(p, q) and T(q, p) are the same unoriented knot. That asymmetry has no counterpart in the knot type, so ℓ < n cannot be encoded there by any map at all.
A knot assignment can still be a way of seeing orbital structure; the genus visualises the (n, ℓ) relationship rather neatly. It cannot be a source of predictions.
Seven: What the Knots Did Get Right
So were Kelvin and Tait simply wrong? No — and the honest accounting is more interesting than either verdict.
From topology alone, a closed curve on a torus in S³ gives you two integers, because H₁(T²) = ℤ². Two quantum numbers, falling out of the topology, before anyone knew quantum numbers existed. And it gives you discreteness — knot types do not vary continuously, so whatever they label comes in distinct species. In 1867, with no other candidate explanation for why matter comes in stable discrete kinds, that was a serious idea.
What knots do not give you is everything else:
| knots on S³ | harmonics on S³ | |
|---|---|---|
| discreteness | yes | yes |
| two integers | yes | yes |
| the degeneracy count n² | no | yes |
| the constraint ℓ < n | no | yes |
| energy ordering | no | yes |
| selection rules | no | yes |
Kelvin got the arena right, sixty-eight years early. Fock found the inhabitants. Knots were a reasonable guess about what lives on a three-sphere; harmonics were the right answer.
And the vortex instinct itself — that topology can confer stability with no dynamical mechanism — was not wrong either. It simply found its home elsewhere: in quantised vortex lines in superfluid helium, in the knotted vortex tubes Kleckner and Irvine made and photographed in 2013, in magnetic helicity in plasmas, and in the Faddeev–Skyrme hopfion, a genuinely knotted field configuration classified by an invariant of maps S³ → S². Kelvin would have recognised the hopfion immediately.
It did not find its home in atoms.
What the Chemists Were Actually Doing
It is tempting to read the periodic table’s structure — shells, periods, the transition-metal block, the lanthanide/actinide split — as knot topology in disguise. It isn’t, and the reason is worth stating plainly.
Every standard invariant of a torus knot T(p, q) is a function of p and q alone. If p and q are themselves determined by n and ℓ, then every knot invariant is a function of n and ℓ. The knot can never tell you more than the quantum numbers you built it from. Anywhere the topology seems to explain a chemical fact, it is the quantum numbers doing the work in a costume.
And the costume fits badly. The unknots include 1s, 2s and 2p — so any story of the form “unknotted, therefore chemically quiet” has to explain carbon. Two- component links include both 5f and 3p, and phosphorus is not baroque.
What the chemists were actually doing is less romantic and more impressive. Bohr, Sommerfeld, Slater, Hartree and Fock mapped the structure of quantum-number space using real-space tools, and got it right. The filling order itself — Madelung’s n+ℓ rule — was later derived group-theoretically by Barut, by Rumer and Fet, and by Kibler, from the SO(4,2) spectrum-generating group. That explanation exists, it works, and it is not topological. Chromium’s [Ar] 3d⁵4s¹ really does come down to exchange energy.
Knots offered a shortcut past all of that. There isn’t one.
The Honest Frontier
There is still an open question here, but it is not the one this essay was originally written to pose.
Fock’s S³ picture is standard physics and essentially unknown to chemists. A quantum chemistry course covers the ℝ³ solution and mentions SO(4) as a remark about the Runge–Lenz vector, if at all; the twistor description is less known still. So the open work is not a theorem but a translation. What does a chemist gain from knowing that the 2s/2p degeneracy is a single SO(4) irrep rather than an accident?
Possibly nothing. Possibly a better way to decide which orbitals a correlated calculation must treat exactly — which is a real, expensive, unsolved problem. That is a question about the value of an existing method, and it deserves to be asked plainly rather than dressed up as a discovery.
Coda: What Tait’s Knots Were For
Tait worked for nine years tabulating knots for atoms that do not exist. His tables — published between 1876 and 1885, compiled by hand with geometric intuition and no algebraic machinery — are the direct ancestors of the knot tables now used in low-dimensional topology, DNA biology, quantum field theory, and, as it turns out, the assignment of orbitals in the periodic table.
The mathematics was right all along. The aether was wrong. The knots were right. They were in the wrong medium — not a Victorian fluid filling all of space, but the quantum-mechanical state space of a hydrogen atom, the Fock sphere in momentum space, the Hopf-fibred three-sphere that Kelvin and Tait could not have imagined in 1867 and that the field needed another sixty years to find.
Kelvin’s famous confession — “I can never satisfy myself until I can make a mechanical model of a thing” — was the temperament of a man born fifty years too early for his own best idea. He needed to be able to see the vortex, to touch it, to watch it spin. The actual knot is not in the fluid. It is in the wavefunction. You cannot build a mechanical model of it. You can only compute it.
When Tait blew that smoke ring in 1867, something real passed between him and Kelvin. Not an atom. Something harder to name — the intuition that topology was the right language for matter, that the stability and distinctness of knots was the right kind of stability and distinctness for the building blocks of everything. That intuition was correct. The medium was wrong. The mathematics waited.
It has now been waiting long enough.
A note on sources. This essay rests on Fock’s 1935 result and the SO(4) → SO(3) branching rule, both standard and neither mine. An earlier version cited three papers from my own research programme in support of stronger claims; those claims did not survive checking, and the citations have been removed. The Madelung rule’s group-theoretic explanation belongs to Barut (1971), to Rumer and Fet (1971), and to Kibler.
Further reading: Bander & Itzykson’s review of the hydrogen atom’s hidden symmetry; Kleckner & Irvine, Nature Physics 9, 253 (2013), on knotted vortices made in a laboratory.