Delocalisation is stabilising, and other things that are false in general
The instinct is almost universal and it is stated in most first courses: electrons confined to a small region have a high kinetic energy, spreading them out lowers it, so delocalisation is stabilising.
Cyclobutadiene has four pi electrons spread over four carbons in a ring, delocalised in precisely the same sense benzene is, and it gains exactly nothing from the arrangement.
Zero, not merely small
The arithmetic takes one line. Four electrons occupying levels at give a total pi energy of
Two isolated ethenes give .
The difference is zero. Not small, not “less than benzene’s”, not “roughly comparable” — the ring is worth precisely nothing relative to two localised double bonds, and the statement is exact within the theory.
Benzene against three ethenes gives against : a delocalisation energy of . Same kind of ring, same kind of delocalisation, same theory, and two entirely different answers.
So the property doing the work cannot be delocalisation, because both molecules have it.
What the property actually is
It is the level pattern, and specifically whether the electron count closes a shell.
A monocyclic system’s levels are , and cosine takes the same value at and — so the pattern is one non-degenerate orbital at the bottom, then degenerate pairs, and for an even ring one more at the top. Filling that ladder, a closed shell takes 2 electrons and then 4 at a time.
Six electrons close a shell. Four do not: they put two electrons in the lowest orbital and two into a degenerate pair that lies at exactly the energy of an isolated p orbital, contributing nothing.
So the correct statement is not “delocalisation is stabilising”. It is:
A cyclic system’s level pattern is stabilising for particular electron counts and not for others.
That is a sharper claim, it predicts cyclobutadiene correctly, and it is why Hückel’s rule is a rule rather than a tendency.
The cases where it is worse than nothing
Zero is not the floor. Delocalisation can be actively destabilising, and the theory says which cases.
The comparison per carbon is the useful one, because a bigger ring has more bonds and would gain more from any general stabilisation. Benzene: over six carbons. Cyclooctatetraene: over eight. Cyclobutadiene: zero over four.
A tendency of delocalisation would give a monotone series. What appears instead is a series governed entirely by whether the count fits the pattern.
The experimental fact matches. Cyclobutadiene was not isolated until 1972, and then only trapped in an argon matrix at 8 kelvin; two molecules of it dimerise on contact. Cyclooctatetraene is a stable liquid that behaves like an ordinary polyene, because it declines the planar arrangement entirely and folds into a tub.
What the eigenvectors say about the non-bonding pair
The phrase “non-bonding orbital” is doing real work in the argument above, and it is worth seeing what such an orbital looks like rather than taking the label.
The two figures side by side settle the essay’s point without any energy arithmetic at all. Both molecules delocalise their electrons over the whole ring. In one, every occupied orbital accumulates density between nuclei; in the other, the last two electrons occupy orbitals that accumulate nothing anywhere.
Drag either slider up the ladder and the readout gives energy and node count together — and the node count is what distinguishes them. An orbital with nodes across half its bonds sits at α, and an orbital with nodes across fewer sits below it.
The prediction that is wrong, and why that is the good part
Hückel’s account of cyclobutadiene does not stop at “no stabilisation”. Hund’s rule puts one electron in each of the two degenerate non-bonding orbitals, so the prediction is a triplet ground state with two unpaired electrons.
That prediction is wrong. Cyclobutadiene is a singlet.
The reason is a distortion. A square cyclobutadiene has the degenerate pair the argument requires; a rectangular one does not, because the two Kekulé-like structures stop being equivalent, the degeneracy is lifted, and both electrons drop into the lower orbital. Real cyclobutadiene is rectangular, with alternating bond lengths of about 1.34 and 1.56 ångström.
That is a Jahn–Teller distortion — a system with a partly filled degenerate level lowers its energy by breaking the symmetry that produced the degeneracy — and it is invisible to a theory that discarded the geometry before the matrix was built.
Being wrong this specifically is what makes the theory worth having. A treatment predicting “cyclobutadiene will be unstable” could not have been caught out by anything. One predicting a triplet can be, and the mechanism of its being caught out is itself a result: the molecule escapes by moving, and the escape route was identified by the prediction failing.
What was computed, and how
Every delocalisation energy on this page is a computed pi energy minus a stated reference, and the reference is recorded in the system’s own definition as a count of isolated double bonds.
That is the repair for the thing this essay is really about. The experimental figure for benzene’s stabilisation ranges from about 120 to 180 kilojoules per mole across sources, and the range is not measurement uncertainty — it is four different comparisons being reported as though they were one. A delocalisation energy is a difference, and a difference without its reference is not a number.
The assertion the site’s gate runs is about the arithmetic of the comparison rather than about the answer: whatever the reference, the reported delocalisation must equal the computed pi energy minus it. A system reporting a delocalisation energy without naming a reference is refused, and a delocalisation claimed at the wrong value is refused.
The eigenvalues themselves are checked three ways — against the closed form for a ring, against the two trace relations which follow from the graph rather than from the solution, and against the pairing theorem asserted in both directions. And the shell-closure claim is asserted for every ring from three to ten: the closure found by filling must agree with the predicate, or the build stops.
Every energy here is in units of β, and β has no number. Fitting it would turn a clean statement about a graph into a calibration with a hidden choice, and values of β fitted to different observables differ by a factor of two.
The surprise: the instinct is not even wrong about the physics
The particle-in-a-box argument that motivates the instinct is correct as far as it goes. Confining a particle to a smaller region does raise its kinetic energy, and spreading a wavefunction out does lower it.
What the argument leaves out is that the electrons have to go somewhere in the level pattern, and the pattern is not a single lowered level. Spreading six electrons over benzene’s ring puts all six into orbitals below the isolated-ethene energy. Spreading four over cyclobutadiene’s puts two below and two exactly at it.
So the instinct is right about the mechanism and wrong about the bookkeeping. Delocalisation lowers the lowest orbital in every case — cyclobutadiene’s lowest level is at α+2β, deeper than ethene’s α+β — and the question is whether the electrons available fill the levels that went down or the ones that did not.
That reframing is worth having because it rescues the intuition rather than discarding it. A reader who understands why the lowest orbital drops has understood the physics. What they need in addition is a count.
What it costs
Nothing arithmetically: a four-by-four or eight-by-eight eigenvalue problem.
What the false version costs is worth pricing, because it is a teaching cost paid repeatedly.
It makes cyclobutadiene inexplicable. A student holding “delocalisation is stabilising” meets a delocalised molecule that is spectacularly unstable and has to file it as an exception. There are no exceptions here; there is a count.
It makes the bond orders unaccountable. A reader expecting delocalisation to be uniformly good expects benzene’s uniform bonds to be the general case, and naphthalene’s three different bond orders then look like a complication rather than a consequence.
It makes aromaticity mysterious. If delocalisation were intrinsically good, benzene would need no special explanation and would be an arbitrary extra rule. Getting the general statement right makes the rule a consequence rather than an addition.
And it obscures the reference-state problem. “Stabilising” is a comparative with the comparison suppressed, so the habit of saying it trains a reader not to ask what the comparison is — which is the same habit that lets a 120-to-180 range be quoted as a single number.
Where the model stops
Four limits.
Hückel has no electron repulsion at all. The singlet–triplet question above is exactly the kind of question a theory without repulsion cannot answer, and the fact that its answer is wrong for cyclobutadiene has two causes rather than one — the missing distortion and the missing repulsion.
No geometry. A theory handed a connectivity returns the levels of that connectivity, and cannot report that the connectivity would prefer to be a different shape. Every escape described on this page — the rectangle, the tub — is invisible to it.
Energies carry an unfitted parameter, so nothing here converts to kilojoules without a fit.
And “stability” means several things. A delocalisation energy is a ground-state energy difference. That cyclobutadiene dimerises on contact is a statement about a reaction barrier, and no barrier is computed anywhere on this site. The two usually run together and they are not the same claim.
What the bond orders say when the count is wrong
The eigenvectors give a second reading of the same conclusion, and it is worth having because it is a reading in terms of bonds rather than of energies.
A half is exactly what two localised double bonds averaged over four positions would give, which is the bond-order form of the delocalisation energy coming out zero. The ring has redistributed its π character and gained nothing by the redistribution.
Benzene’s two thirds is the excess, and it is the same excess the delocalisation energy measures, arrived at from the coefficients rather than from the eigenvalues. Two quantities computed from different parts of the same diagonalisation, agreeing about which ring is worth having.
That agreement is not automatic. Bond order sums over occupied eigenvectors and delocalisation energy sums over occupied eigenvalues, and a solver that had converged to the wrong place could easily have made one look reasonable while the other did not. Both are checked, and both are checked against systems whose closed forms are known.
Who found it, and when
Hückel’s rule dates from 1931 and the antiaromatic half of it was ignored for two decades along with the rest.
The word “antiaromatic” is Breslow’s, from 1967, and it was coined precisely to name the thing this essay is about — that a system is not merely un-aromatic but destabilised relative to a localised reference. That the term needed inventing thirty-six years after the theory predicted the phenomenon says something about how long the instinct held.
Cyclobutadiene’s history is a long chain of failed syntheses. Attempts began in the 1900s and continued for seventy years, with several claimed preparations withdrawn. Pettit’s iron tricarbonyl complex in 1965 stabilised it by coordination; Chapman and Lin’s matrix isolation in 1972 produced the free molecule at 8 kelvin. Its rectangular structure was settled shortly after, confirming a distortion that had been predicted on symmetry grounds.
The Jahn–Teller theorem itself dates from 1937 and is a general statement about any non-linear molecule in a degenerate electronic state: such a state is always unstable with respect to some distortion. Applied here it says the square arrangement cannot survive, and it says so without any reference to carbon.
Where the ladder goes next
The general claim this essay corrects is delocalisation.
The count that decides the outcome is aromaticity as a shell closure.
The theory throughout is Hückel theory, with its checks.
And the chain version, where a length decides instead of a count, is conjugation and its limits.
The repair is a small edit to a sentence and a large change to what a reader can predict. “Delocalisation is stabilising” covers benzene and fails on cyclobutadiene, with no way to tell in advance which case is which. “A cyclic system’s level pattern is stabilising for particular electron counts” covers both, predicts the ring sizes that will work, and puts the count where a reader can check it. Two extra clauses, and a rule that can be wrong instead of a slogan that cannot.
It is worth adding that nothing in the correction makes delocalisation less interesting. What it does is make the interesting quantity explicit — a level pattern and an electron count, both of which are computable and checkable — instead of leaving it inside an adjective that sounds like an explanation and is not one.
A count, a level pattern and a stated reference. All three are cheap, all three are checkable, and between them they replace an adjective that was doing work it could not do.
What the pictures here cannot show. Every figure on this page is a ladder of energies in units of β, so no two of them can be compared for absolute energy — only the pattern, the degeneracies and the closure are meaningful. Nothing here can show cyclobutadiene’s rectangular distortion, because the geometry was discarded before the calculation began; nothing can show that it dimerises on contact, because that is a rate; and nothing can show a kilojoule, because β has no number.