A spectrum counts environments, not atoms

Count the fluorines in phosphorus pentafluoride and the answer is five. Count the environments and the answer is two. Every resolving spectroscopy reports the second number, and the difference between the two counts is a fact about the group rather than about the chemistry.

Two atoms are in the same environment when some operation of the molecule’s symmetry group carries one onto the other. That is a strong condition: it does not mean the two are similar, it means there is an exact symmetry exchanging them, so no measurement of any kind can distinguish them. The sets it produces are the orbits of the group acting on the atoms, and they are computable from the coordinates.

The orbits, and the relation that checks them #

The computation is short. The group’s operations each realise a permutation of the atoms — that permutation is exact, being a fact about labels rather than about numbers, and character tables and reduction explains why the site keys its operations on it. Take the union of the cycles those permutations produce and what is left is the orbits.

Each orbit comes with a check. The stabiliser of an atom is the set of operations that leave it where it is, and the orbit–stabiliser relation says

$$|\text{orbit}| \times |\text{stabiliser}| = |G|.$$

Phosphorus pentafluoride’s group has twelve operations. Its axial orbit has two members, so each axial fluorine must be fixed by six of them — the threefold rotations, the identity and the three vertical mirrors. Its equatorial orbit has three members, so each equatorial fluorine is fixed by four. Two times six and three times four both come to twelve, and the relation is checked for every orbit of every molecule drawn this way.

That relation is what makes an orbit count trustworthy. A miscounted orbit gives a stabiliser order that does not divide the group order, which is impossible for a subgroup, and the figure refuses to draw.

What the count is worth #

This is the sharpest structural inference available from a resolving spectrum, and it runs in a useful direction: the number of signals is a lower bound on the number of environments, and the number of environments is fixed by the group. A structure that predicts two signals and shows three is wrong.

Three cases from this collection show the range:

Benzene: two environments from twelve atoms. One ¹³C signal, one ¹H signal.

Ethene: two environments from six atoms — the two carbons in one orbit, the four hydrogens in another. All four hydrogens are equivalent despite two of them being cis to each other and two trans, because the operations of D2h carry every one onto every other.

Bromochlorofluoromethane: five environments from five atoms. Its group is C₁ — the trivial group, one operation — so every atom is its own orbit, and nothing in the molecule is equivalent to anything else. Chirality is a symmetry statement is about the same molecule and the same group, from the other side.

Where the prediction fails, and why the failure is interesting #

Phosphorus pentafluoride has two fluorine environments and its fluorine magnetic resonance spectrum shows one signal.

The structure is not in doubt. Its vibrational spectrum, its electron diffraction pattern and its symmetry are all those of a trigonal bipyramid with two distinct kinds of fluorine, and five sites are not alike computes why five points on a sphere cannot be equivalent: every other common coordination number gives an arrangement with one or two distinct angles, and five gives three, because two positions sit on an axis and three around an equator.

What happens is that the molecule exchanges them. The axial and equatorial sites interconvert by a low-energy pathway — the two axial ligands bending together while two equatorial ones straighten, so that a different pair ends up axial — and it does so millions of times a second at room temperature. A magnetic resonance measurement takes milliseconds to distinguish two frequencies a few hundred hertz apart; a molecule that has been in both environments a million times during that interval gives one signal at the average.

The count from symmetry is a count for the static structure. Whether an experiment sees it depends on how the experiment’s timescale compares with any exchange available, and different experiments answer differently about the same molecule:

• Vibrational spectroscopy samples a molecule in about 10⁻¹³ seconds, far faster than any exchange, and sees the trigonal bipyramid with its two fluorine environments. Site symmetry, and what it constrains computes twelve vibrations in four species for exactly that structure.
• Magnetic resonance samples in about 10⁻³ seconds and sees the average.
• Cooling slows the exchange, and at low enough temperature the two signals separate. That the separation can be produced by cooling is the proof that the single signal is an averaging effect rather than a symmetry.

The same counting, on a vibrational spectrum #

The orbit count is not confined to magnetic resonance. It sets how many of anything a molecule can have, and a vibrational analysis is the same arithmetic with a different basis on the atoms.

The relation between the two counts is worth stating because it is not the obvious one. One orbit of six atoms does not give one band; it gives as many bands as the reduction produces species. Six equivalent fluorines give three symmetry species of stretching motion, so a molecule with a single fluorine environment still has three distinct stretching frequencies, of which some are active and some are not.

So there are two quite different counting questions, and they have different answers:

• How many environments? The number of orbits. Answered by the permutations alone.
• How many bands? The number of species the relevant basis spans, filtered by activity. Answered by the reduction.

A resolving spectrum like magnetic resonance measures the first; a vibrational spectrum measures the second. Both are computed from the same group, and confusing them is the same error as confusing modes with frequencies in how many frequencies, not how many modes.

Timescale, made concrete #

The failure case above deserves a number rather than a hand-wave, because “faster than the measurement” is doing all the work.

A magnetic resonance experiment distinguishes two environments by the difference in their resonance frequencies. If two fluorine environments differ by, say, 500 hertz, then the measurement needs about a millisecond to tell them apart — the reciprocal of that difference. A process that swaps the two nuclei many times within a millisecond averages them, and one signal appears at the mean.

Cooling slows the exchange. At the temperature where the exchange rate falls to about the frequency difference, the single line broadens and splits; below it, two sharp lines appear. That progression is the standard experimental signature and it is what proves the averaging: a symmetry cannot be switched off by cooling and an exchange can.

The comparison across techniques then follows from their timescales rather than from anything about the molecule:

• A vibrational measurement samples in about 10⁻¹³ seconds and sees the static structure.
• A magnetic resonance measurement samples in about 10⁻³ seconds and sees the average.
• A diffraction measurement sees a time and space average of a different kind again, and reports disorder rather than an averaged position.

None of them is wrong. Each reports the structure averaged over its own window, and the disagreements between them are measurements of what the molecule does between windows.

Two counts that are often confused #

It is worth separating the two things this essay’s title distinguishes, because the confusion has a standard form.

How many atoms of an element there are is chemistry: five fluorines, six hydrogens.

How many environments they occupy is group theory: two, one.

A spectrum reports the second, and the ratio of intensities within it reports the sizes of the orbits — a two-to-three ratio for phosphorus pentafluoride’s fluorines if the exchange is stopped, a one-to-one for benzene’s if there were two carbon environments to compare. So a resolving spectrum gives the orbit decomposition directly: the number of signals is the number of orbits and their relative intensities are the orbit sizes.

That is a complete statement of what such a spectrum measures about structure, and it uses no bonding argument at all. It is the same kind of statement as symmetry forbids a dipole — a property fixed by the point group with no reference to how the electrons are arranged.

Orbits are the same object twice over #

The orbit arithmetic used here is not new; it is the same arithmetic three other essays already run, and noticing that is worth a paragraph because it is the kind of connection worth making.

A vibration’s localisation is an orbit question. Group frequencies, and where they stop computes that a normal mode can belong to one internal coordinate only when no symmetry operation carries that coordinate into another — which is to say, only when the coordinate is alone in its orbit. Four of twenty-one distinct frequencies in this collection qualify.

An atom’s site symmetry is its stabiliser. Site symmetry, and what it constrains is about the subgroup fixing a position, which is the other half of the relation used above, and the orbit sizes here are the group order divided by those subgroup orders.

A chiral molecule is one whose group has no improper operation — chirality is a symmetry statement — and the extreme case of that, C₁, is the extreme case here too: every atom its own orbit, nothing equivalent to anything.

So one computation, run on different objects, answers four questions: how many signals, which modes can be localised, what constrains an atom’s local environment, and whether the molecule can be chiral. The arithmetic is a group acting on a set, and the set is chosen by the question.

The general statement, which holds for every resolving experiment: the number of signals is the number of orbits, and their relative intensities are the orbit sizes. That is a complete description of what such a spectrum measures about structure, it uses no bonding argument at all, and it is computable from the coordinates in a few lines.

What symmetry cannot count #

Chemical shift is not computed here. Symmetry says how many signals and in what ratio, and says nothing whatever about where they appear. Two inequivalent environments can have shifts so close that no instrument separates them, which is an absence of the kind what an absence proves is about: accidental coincidence rather than symmetry-required equivalence, destroyed by changing the solvent or the field.

Coupling is not computed here either. The multiplet structure of a magnetic resonance spectrum comes from interactions between nuclei, which are not computed here and not claimed.

The exchange rate is not computed. That phosphorus pentafluoride’s ligands exchange is quoted; how fast, and by what pathway, is a question about a potential energy surface with a barrier on it, which is beyond a geometric repulsion model. The pseudorotation mechanism is Berry’s, from 1960, and the fact that five-coordinate species are stereochemically non-rigid is one of the standard results of inorganic chemistry.

The counts assume the free molecule. In a crystal an atom’s environment is set by its site in the lattice rather than by the molecule’s own group, and inequivalences appear that the molecule does not have. That is the periodic version of the same orbit arithmetic and belongs to crystallography rather than here.

One consequence for reading a structure paper. When a spectrum is quoted as evidence for a structure, the load-bearing claim is nearly always a count — how many signals, in what ratio — rather than any individual position. Positions need models; counts need only the group, and the group needs only the coordinates. That is why counting arguments outlive the models built around them, and it is why the essays here lean on them wherever both are available.

The failure to resolve is a measurement of a rate #

One signal where two were expected is reported here as the motion outrunning the measurement, and the comparison between the two timescales is not merely qualitative — it is the basis of a standard method for measuring a barrier.

The two signals are separated by some frequency difference. The molecule exchanges the two environments at some rate. Which of the two wins is decided by their ratio, and the ratio can be changed at will by changing the temperature.

Cool the sample and the rate falls. The averaged single line broadens, then flattens, then splits, and at low enough temperature the two separate signals appear at their own positions — the spectrum the static structure predicts.

The temperature at which the line is broadest is the coalescence point, and there the exchange rate is fixed by the separation alone: it is the separation times a constant of order unity, with no other input. So a temperature and a frequency difference give a rate, directly.

A rate at a temperature gives an activation barrier, and the barrier is the quantity of chemical interest — how much energy it costs the fluorines to exchange places.

That turns the finding from a caution into a technique. A spectrum that fails to resolve two environments is measuring how fast they interconvert, and the failure is the signal rather than the noise.

It also explains why the same molecule gives different answers to different instruments. A magnetic resonance measurement is slow — the separations it works with are hundreds or thousands of hertz — so almost any intramolecular rearrangement outruns it. A vibrational measurement is fast, with separations of hundreds of wavenumbers, which is a frequency ten orders of magnitude higher; almost nothing outruns it, and the same molecule shows its two environments there.

So phosphorus pentafluoride has one fluorine environment to a magnetic resonance experiment and two to a vibrational one, at the same temperature, in the same sample. Neither is wrong, and the ratio of the two timescales is the reason.

That comparison also settles which technique to reach for, and the answer is the opposite of the intuition. A structural question — how many environments does this molecule have — is best asked of the fast technique, because a fast technique sees the molecule frozen. A dynamic question — how quickly do they exchange — is best asked of the slow one, because a technique that outruns nothing measures no rates.

So a vibrational spectrum is the better structural instrument and a magnetic resonance spectrum is the better kinetic one, which is not how either is usually introduced. And the choice is not about resolution or convenience: it is about where the molecule’s own rates sit relative to the two, and a molecule with no motion faster than either would give the same count to both.

Who worked it out #

The orbit–stabiliser relation is elementary group theory and is older than any of its chemical applications. Its use for counting spectroscopic environments arrived with high-resolution nuclear magnetic resonance in the 1950s, when the question “how many signals should this structure give?” became the first thing anybody asked of a spectrum.

Phosphorus pentafluoride’s single fluorine signal was reported by Gutowsky and co-workers in 1953 and was a genuine puzzle: the structure was known to be a trigonal bipyramid and the spectrum said the fluorines were equivalent. Berry’s 1960 pseudorotation mechanism resolved it, and the general lesson — that a spectrum measures a structure averaged over whatever the molecule does during the measurement — has been part of the subject since.

One last framing, because it is the sentence this essay exists for. A molecule has atoms; a spectrum has environments; and the map between them is a group acting on a set. Everything else — which technique, which nucleus, which timescale — decides only whether the map is visible.

Where to read on #

Counting can go no further here. The next questions measure rather than count: the rotational spectrum is a moment of inertia starts from a molecule’s mass distribution and produces a spectrum whose whole structure is one number, and a bond length out of a spectrum runs that arithmetic backwards to recover a geometry from two measured line spacings.