The ring's levels are its group's characters

The π energies of a ring of $n$ identical atoms are

$$E_k = \alpha + 2\beta \cos\left(\frac{2\pi k}{n}\right), \qquad k = 0, 1, \dots, n-1,$$

and they are usually obtained by building an adjacency matrix and handing it to an eigensolver. That works, it is checked against a closed form, and it is unnecessary. The expression is not the output of a diagonalisation; it is a statement about the ring’s own rotation group, and it can be written down without any linear algebra at all.

Why the group diagonalises the matrix #

A ring of $n$ identical atoms has the cyclic group $C_n$ acting on it. Its generator $R$ rotates every site into the next one. The Hückel matrix — the adjacency matrix, in units of $\beta$ — commutes with that rotation, because rotating the ring and then asking who is adjacent to whom gives the same answer as asking first and rotating afterwards.

Two commuting operators share eigenvectors. And the eigenvectors of a rotation of order $n$ are known without any calculation: they are the one-dimensional representations of $C_n$, and the $k$-th of them sends $R$ to

$$\omega^k = \exp\left(\frac{2\pi i k}{n}\right).$$

There are exactly $n$ of them, so they account for the whole space.

Now apply the Hückel matrix to the $k$-th of these vectors. Each site is connected to its two neighbours, which are the site rotated forward and the site rotated back, so the matrix acts as $R + R^{-1}$ — and on the $k$-th representation that multiplies by $\omega^k + \omega^{-k} = 2\cos(2\pi k/n)$.

That is the whole derivation. The sum $\omega^k + \omega^{-k}$ is the character of the $k$-th representation summed over the two neighbours, and the energy is twice its real part.

The pattern is a theorem #

Everything about the level structure follows from where those points sit on the circle, and the counting is arithmetic on the unit circle rather than a fact about molecules.

$k = 0$ is real and alone. Its character is 1, its level is $\alpha + 2\beta$, and there is no other representation with that real part. So the lowest level of every ring is non-degenerate.

Every other $k$ pairs with $n - k$. Cosine is even, so $\omega^k$ and $\omega^{n-k}$ are complex conjugates with equal real parts. Those two representations are degenerate, and they are degenerate for a reason with no energy in it: the Hamiltonian is real, so it cannot distinguish a representation from its conjugate.

If $n$ is even, $k = n/2$ is real and alone as well. Its character is $-1$ and its level is $\alpha - 2\beta$.

So the pattern is: one level, then pairs, then — for even rings — one more level at the top. Fill from the bottom two at a time and a shell closes after the single lowest level and some whole number of pairs: $2 + 4m$ electrons.

That is Hückel’s rule. It is a statement about the irreducible representations of a cyclic group and about nothing else, and every ring size obeys it because every cyclic group has one real representation at the identity end and conjugate pairs above.

Two routes that share nothing #

The agreement in that figure is worth more than it looks, because the two calculations have nothing in common.

One is a Jacobi eigensolver applied to an $n \times n$ integer matrix, rotating away off-diagonal elements until they vanish — several thousand floating-point operations, converging iteratively. The other is $2\cos(2\pi k/n)$, evaluated once per level.

An error in the adjacency matrix, in the solver, in the sorting, or in the closed form would show up. Nothing plausible leaves the two agreeing to a part in $10^9$ for every ring from three to twelve, in both the ordinary and the twisted case.

That is the same kind of check as the vanishing overlap integral, and for the same reason: a number computed two ways that share nothing is a much stronger statement than a number computed once and compared against a reference.

The orbitals are the representations too #

The energies are not the only thing the group supplies. The eigenvector belonging to level $k$ is the representation vector itself: the coefficient on site $j$ is $\omega^{jk}/\sqrt{n}$, a phase factor turning steadily round the ring.

Those are complex, and a real Hamiltonian’s eigenvectors can always be chosen real, so what is usually drawn is a pair of real combinations of the conjugate pair — a cosine and a sine of $2\pi jk/n$. Which pair is chosen is arbitrary, and this is the same freedom that appears whenever a degenerate shell is drawn: the projector fixes the subspace and not the basis in it, so the two pictures printed for benzene’s degenerate π levels are one choice out of a continuum.

The physical quantities are not affected, and that is checkable rather than a hope. A charge density and a bond order are sums over occupied orbitals, and within a completely filled degenerate shell any unitary mixing leaves both unchanged. Benzene’s uniform bond orders are therefore safe to quote even though its individual degenerate eigenvectors are not.

The node count is the neatest consequence. A phase factor turning $k$ times round the ring crosses zero $2k$ times, so the $k$-th level has $2k$ nodes — which is why the lowest orbital has none, the first pair has two, and the top of an even ring has $n$. The energy ordering and the node ordering are the same ordering because both are read off the same integer.

The odd ring, and why five is different #

An odd ring has no representation at $k = n/2$, so it has one non-degenerate level and $(n-1)/2$ pairs. Filling it completely takes $2 + 4m$ electrons where $m = (n-1)/2$ — which is $n + 1$, an even number for odd $n$.

Cyclopentadienyl is the case everyone knows: five carbons, and the closed shell needs six electrons, so the anion is the aromatic species — and the same ring at three different charges is the cleanest demonstration that the count rather than the framework decides. Tropylium is seven carbons and six electrons, so the cation is. Neither of those is a special fact about carbon; they are the arithmetic of an odd cyclic group with a neutral ring supplying $n$ electrons.

The twist, as a projective representation #

A Möbius ring is a cyclic system with one resonance integral reversed in sign, and it is usually treated as such: change one matrix element and diagonalise again. The group-theoretical account is better, and it is the same account with one alteration.

Going once round a twisted ring returns the basis function to minus itself rather than to itself. So a function on the ring does not carry a representation of $C_n$ at all — it carries a projective one, in which the generator has eigenvalue $\omega^{k + 1/2}$ instead of $\omega^k$. The half-integer is the sign.

The levels are then

$$E_k = \alpha + 2\beta \cos\left(\frac{(2k+1)\pi}{n}\right),$$

and everything about the pattern changes in one place. No representation lands on the real axis when $n$ is even, because a half-integer multiple of $2\pi/n$ never comes out at zero or $\pi$. So the lowest level is a pair, the shells hold four electrons each, and closure is at $4n$ electrons rather than $4n + 2$.

The two rules are therefore not two rules. They are one statement about which representations a function on the ring can carry, and the twist changes the boundary condition rather than the physics.

Two rules because a real matrix admits only ±1 #

Reading the twist as a half-integer in the exponent invites the obvious question: why a half, and not any other fraction? Nothing in the derivation required the generator’s eigenvalue to be a root of unity. Allowing an arbitrary phase $\varphi$ picked up on one circuit of the ring gives

$$E_k(\varphi) = \alpha + 2\beta \cos\!\left(\frac{2\pi k + \varphi}{n}\right),$$

with $\varphi = 0$ the ordinary ring and $\varphi = \pi$ the twisted one. The two cases usually computed are the two ends of a continuum, and the continuum is worth a paragraph because it says why there are exactly two.

Filling a ring at a fixed electron count and sweeping the phase angle gives a monotone curve between the ends:

Neither end is a local feature. The π energy of a $4n+2$ ring falls all the way from $\varphi = 0$ to $\varphi = \pi$ and the π energy of a $4n$ ring rises all the way, without an interior extremum in either case. So each family has its best phase at one end of the range and its worst at the other, and the two families disagree about which end.

That reframes the counting rule into a single sentence. A ring’s electron count decides which boundary phase minimises its π energy, and Hückel’s rule is the statement that the answer is $0$ for $4n+2$ and $\pi$ for $4n$. There is no third rule because there is no third answer.

And there is no third answer for a reason with nothing to do with chemistry. A phase angle other than $0$ or $\pi$ makes the generator’s eigenvalue complex, which makes the Hückel matrix complex, which a system with no magnetic field cannot have — a real Hamiltonian admits exactly two ways of carrying a basis function once round a loop, back to itself or back to minus itself. The two counting rules are two because $\pm 1$ are the only real square roots of one.

Which also says what would be needed to reach the middle of the table. A magnetic flux through the ring supplies exactly such an angle, continuously, and the columns above are then the ground-state energy against the flux rather than a formal interpolation. Nothing here computes a field, so the interior columns are arithmetic on a formula rather than a prediction about a molecule — but they establish that the two rules in use are endpoints of one expression rather than two separate results that happen to look alike.

Antiaromatic, and what the pattern says about it #

Four electrons in a four-ring is the standard antiaromatic case, and the group picture says exactly what goes wrong with it.

The four representations of $C_4$ sit at $1$, $i$, $-1$, $-i$, so the levels are $\alpha + 2\beta$, $\alpha$, $\alpha$, $\alpha - 2\beta$: one below, two exactly at the reference, one above. Four electrons fill the lowest and put one in each of the degenerate pair, leaving an open shell.

The delocalisation energy comes out at exactly zero — $2(\alpha + 2\beta) + 2\alpha$ against two isolated double bonds at $2(\alpha + \beta)$ each — and the zero is a consequence of the character sum rather than a coincidence. The characters of any group’s representations sum to zero over the non-identity elements, and here that forces the four levels to sum to $4\alpha$; with the extreme pair symmetric about $\alpha$, the middle two have to be at $\alpha$ exactly.

Being at exactly zero is a stronger statement than being small, and it is the statement the antiaromaticity essay is built on: a ring can be perfectly delocalised and gain nothing whatever by it.

What this does not explain #

Three limits, and the first is the important one.

This is not an explanation of aromaticity. It is an explanation of a level pattern, and the level pattern is what the counting rule counts. Whether a closed shell of π electrons makes a molecule stable, planar, or magnetically distinctive is a separate question, and the answers to it are not all favourable — delocalisation is not always stabilising, and the hexagonal geometry of benzene is the σ frame’s doing rather than the π system’s.

The model is one electron per site with no repulsion. The degeneracies above are exact within it and would be modified by any treatment of electron repulsion, which for the half-filled degenerate shells of an antiaromatic ring is not a small correction: cyclobutadiene’s predicted triplet ground state is wrong for exactly that reason, and it is wrong in a way that makes the Jahn-Teller distortion discussable.

The full point group is larger. A real benzene has D₆ₕ symmetry, with reflections and a centre as well as the rotations. The cyclic subgroup is what commutes with the Hückel matrix and therefore what diagonalises it; the rest of the group sorts the resulting levels into the g and u, primed and double-primed species a spectroscopist uses. The argument above needs only the rotations, which is why it works for a Möbius ring, which has no reflections at all.

The same argument, one dimension up #

It is worth noticing what the derivation actually used, because it applies far beyond rings.

The only ingredients were: a Hamiltonian that commutes with a symmetry operation, and a symmetry operation whose eigenvectors are known. That is all. The eigenvectors of a cyclic rotation are known because $C_n$ is abelian and all its representations are one-dimensional, so every function on the ring is an eigenfunction of the rotation.

The same argument diagonalises a linear chain with periodic ends, a two-dimensional lattice, and a three-dimensional one — where the representations of the translation group are labelled by a wavevector rather than by an integer $k$, and the level expression becomes a band. The band limit, reached by taking rings to large $n$, is exactly this statement in its limiting form, and the reason the levels there come out as a cosine is the reason they come out as a cosine here.

What is not available in the molecular case is any way to avoid the finite group. A ring of six has six representations and six levels, and there is nothing continuous about it. That is the difference between a molecule and a solid stated in the sharpest possible way: not a matter of size, but of how many representations the symmetry group has.

What had to be excluded from the count #

One detail of the counting rule needed care, and it is the kind that makes a check useful rather than decorative.

Asking which electron counts close a shell, and requiring every answer to be $4m + 2$, is refused by the four-ring at eight electrons. Eight electrons in a four-ring fills every level, which is a closed shell by any definition, and eight is $4m$ rather than $4m + 2$.

That is not a counterexample; it is the case the rule does not speak about. Hückel’s rule concerns shells below the top of the manifold, where a closed shell means a gap above it. A completely filled π system has no gap above it because there is nothing above it, and for an even ring the total count is $2n$, which is a multiple of four whenever $n$ is even.

Restricting the count to $e < 2n$ makes the rule exact for every ring size tested, and stating the restriction is more useful than quietly choosing a range that happens to work. It also marks the boundary of the claim: the rule is about a gap, and a molecule with every π level occupied has no π chemistry left to have.

Still open: the heteroatomic ring, and where the counting rule fails #

The immediate open question is the heteroatomic ring, and the news there is negative in an instructive way. Substituting one carbon by a nitrogen breaks the cyclic symmetry entirely: the rotation no longer commutes with the Hamiltonian, the representations no longer diagonalise it, and the levels have to be computed. What survives is a perturbation argument — the degenerate pairs split by an amount proportional to the difference in site energy weighted by the coefficients — and the pattern is recognisably the ring’s, with the degeneracies lifted.

The other is the one all of this circles. The counting rule is exact within a model that has no electron repulsion in it, and the cases where the rule is known to fail — antiaromatic rings that distort, rings large enough for the levels to crowd — are precisely the cases where a half-filled degenerate shell makes the one-electron picture untenable. The group theorem is exact and the model it is a theorem about is not, which is the ordinary condition of a symmetry argument and worth restating each time it is used.