What symmetry decides

The theorem uses up an equation per species

The isotope product rule forced 86 per cent of ammonia's fitted misfit and, taken out, kept the bond-angle verdict just below nine tenths. Applied to every field fitted to two isotopologues it finds three different things. Water's misfit is entirely the theorem's; methane's is almost none of it, 0.8 per cent, and lives in how the T₂ species divides between its two modes, where the theorem is blind. And because the rule fixes one combination of lines per species for every field, it removes one equation per species from every fit: water has none to spare, ammonia one rather than three, methane one rather than four. Recounted, ammonia's band widens and its verdict crosses nine tenths again.

Worth reading first: The floor under ammonia's fit is a theorem · The worst case was the loosest constant.

A force field fitted to a molecule’s vibrational lines misses them by some amount, and that misfit has been used in these essays as a ruler: a field is as good as the best when its sum of squared misfits exceeds the best’s by no more than the best’s own sum divided by the degrees of freedom left over. The floor under ammonia’s fit found that 86 per cent of ammonia’s misfit is not the fit’s at all. Within each symmetry species the product of the ND₃/NH₃ frequency ratios is fixed by the masses alone — the Teller–Redlich product rule — so every harmonic field predicts the same two products, and the observed lines disagree with them. Taking that floor out of the noise estimate narrowed ammonia’s band of indistinguishable fields 2.6-fold and kept the bond-angle verdict below the nine-tenths line, by 0.0012.

That essay kept one thing fixed on purpose: the number of degrees of freedom, three, that the published band had used. It asked, at the end, what the same floor says about every other field fitted to two isotopologues. Asking it turns up the thing that was held fixed. A theorem that fixes one combination of lines per species is an equation the fit does not get to use, and the degrees of freedom have been counted as if it were.

Three fits, three kinds of misfit

Three fits, three floors: water's misfit is all theorem, ammonia's mostly, methane's almost none. For each field fitted to two isotopologues, its published residual (bar) and the least residual the product rule forces on any harmonic field (dot), on a logarithmic scale: H₂O / D₂O: residual 0.0073%, floor 0.0073%, 100.0% of the sum of squares; NH₃ / ND₃: residual 0.4632%, floor 0.4289%, 85.7% of the sum of squares; CH₄ / CD₄: residual 0.4010%, floor 0.0369%, 0.8% of the sum of squares.
Fig. 1 For each field fitted to two isotopologues, its published residual and the least residual the product rule forces on any harmonic field, on a logarithmic scale.

Three molecules in this collection are fitted to a parent and its fully deuterated isotopologue: water to H₂O and D₂O with four constants, ammonia to NH₃ and ND₃ with five, and methane to CH₄ and CD₄ with four. For each, every symmetry species gives a product of ratios that every harmonic field shares, and comparing it with the observed lines’ product gives the least misfit any field could reach.

The three come out completely differently. Water’s published residual is 0.0073 per cent, sixty times smaller than either of the others’, and the floor its products force is 0.0073 per cent: the whole of water’s misfit is the theorem’s. Ammonia’s is 0.463 per cent with a floor of 0.429, 85.7 per cent of its sum of squares, as found before. Methane’s is 0.401 per cent with a floor of 0.037, so the theorem forces 0.8 per cent of its sum of squares and the other 99 belongs to the field.

Species by species, how far each observed set's isotope product is from the one every harmonic field shares. For each symmetry species of each molecule, the logarithm of the observed D/H frequency product over the theorem's: H₂O / D₂O A₁ 0.036%; H₂O / D₂O B₂ 0.003%; NH₃ / ND₃ A₁ -1.292%; NH₃ / ND₃ E -1.892%; CH₄ / CD₄ T₂ 0.166%; CH₄ / CD₄ E -0.024%; CH₄ / CD₄ A₁ 0.081%.
Fig. 2 For each symmetry species of each molecule, how far the observed lines’ heavy-to-light product is from the product every harmonic field shares, in the logarithm.

Species by species the difference is in the data. Water’s two species are within 0.036 and 0.003 per cent of the theorem, which is what eight harmonic frequencies of one molecule look like. Methane’s three are within 0.17 per cent. Ammonia’s two are 1.29 and 1.89 per cent short, and those two numbers are the whole of ammonia’s floor.

Methane’s misfit is where the theorem cannot see

Methane’s residual is nearly as large as ammonia’s and almost none of it is forced, so it must be something the fields could change. The lines say what.

Methane's misfits do not follow the rule's pattern: the field misses its lines where the theorem would not. Each observed line of CH₄ / CD₄ with its relative misfit under the published field (bar) beside the misfit the product rule forces on it (dot): CH₄ T₂ 1367.4: -0.34% against 0.04%; CD₄ T₂ 1035.9: 0.53% against -0.04%; CH₄ T₂ 3156.8: 0.52% against 0.04%; CD₄ T₂ 2336.1: -0.51% against -0.04%; CH₄ E 1582.7: -0.17% against -0.01%; CD₄ E 1119.3: -0.14% against 0.01%; CH₄ A₁ 3025.5: 0.04% against 0.04%; CD₄ A₁ 2141.9: -0.04% against -0.04%. The forced part is 0.8% of the published sum of squares.
Fig. 3 Each observed line of CH₄ and CD₄ with its relative misfit under the published field, beside the misfit the product rule forces on it.

The published field misses methane’s T2T_2 lines by half a per cent each — the bend 0.34 per cent low in CH₄ and 0.53 high in CD₄, the stretch 0.52 high in CH₄ and 0.51 low in CD₄ — while the theorem would force a few hundredths. The misfit has almost no overlap with the forced pattern: their alignment, as a normalised projection of one set of signed misfits on the other, is 0.09 for methane against 0.93 for ammonia and 1.00 for water. The E and A1A_1 lines are fitted to between 0.04 and 0.17 per cent.

Methane's T₂ misfits are large and cancel in the product: the field mis-divides the species between its two modes. In methane's T₂ species, the logarithmic D-minus-H misfit of the bend, 0.87%, and of the stretch, -1.03%, and their sum, -0.166% — the part the product rule sees.
Fig. 4 In methane’s T2T_2 species, the logarithmic heavy-minus-light misfit of the bend and of the stretch, and their sum, which is the only part the product rule sees.

The T2T_2 misfits are large and they cancel. The bend’s heavy-minus-light misfit is +0.87 per cent and the stretch’s −1.03; their sum, which is what the species’ product sees, is −0.17. So the field gets the T2T_2 species’ isotope product nearly right and divides it wrongly between its two modes: it puts too large an isotope shift on the bend and too small a one on the stretch. That division is set by how strongly the bend and the stretch are mixed in T2T_2 — the interaction between them — and it is exactly the kind of error a band along an interaction constant explores. It is also the error the product rule, which multiplies the species’ modes together, cannot see. Methane’s field was fitted without a bend–bend constant, which the redundancy of a tetrahedral centre’s angles forced out, and its misfit is the price of that form.

Two of CD₄’s lines need no constants at all

A species with a single mode is the limiting case, and methane has two of them. In A1A_1 and in E the product of ratios is a single ratio, so the theorem fixes the deuterated frequency from the parent’s outright: the field cannot change it. CD₄’s A1A_1 line is CH₄’s 3,025.5 cm⁻¹ times 0.707379, the square root of a ratio of kinetic determinants, which is 2,140.2 against the observed 2,141.9; its E line is 1,582.7 times the same factor, 1,119.6 against 1,119.3. Two of methane’s eight lines are predictions of the masses alone, to 0.08 and 0.02 per cent, before a single force constant is chosen. Water’s one-mode species, B2B_2, does the same for D₂O’s antisymmetric stretch: 3,942.5 times 0.732709 is 2,888.7 against 2,888.8.

So when methane’s field misses its E lines, both of them, by 0.17 per cent in CH₄ and 0.14 in CD₄, the same sign in both, it is not trading misfit between isotopologues: it has the E species’ frequency scale slightly too low in both, and the theorem, which is a statement about ratios, is indifferent to a scale. That is the second place in methane’s fit the theorem cannot see, and like the T2T_2 division it belongs to the field’s form, which ties the E species’ bend to the same angle-bending constant as the T2T_2 bend.

The equations a theorem uses up

The product rule does more than set a floor. It says that one combination of each species’ lines — the product of the heavy-to-light ratios — takes the same value for every field. A fit cannot adjust it and cannot be judged by it. Every distinct line is an equation, but that combination is an equation the masses solve, not the constants.

Each species spends one isotope equation on the theorem: water has none left over, ammonia and methane one each. For each molecule fitted to two isotopologues, its distinct observed lines, the one equation per symmetry species the product rule makes the same for every field, the constants fitted, and what is left: H₂O / D₂O: 6 lines, 2 species, 4 constants, 2 degrees of freedom as counted, 0 independent; NH₃ / ND₃: 8 lines, 2 species, 5 constants, 3 degrees of freedom as counted, 1 independent; CH₄ / CD₄: 8 lines, 3 species, 4 constants, 4 degrees of freedom as counted, 1 independent.
Fig. 5 For each molecule fitted to two isotopologues, one box per distinct observed line: the one per symmetry species fixed by the theorem, those spent on constants, and those left to test the fit.

So the degrees of freedom are the distinct lines less the species less the constants. Degenerate lines count once, because a degenerate pair is one frequency; and the species count is the number of symmetry species that appear in both isotopologues’ spectra, each of which carries one product. Water has six distinct lines in two species and four constants: no equations left. Its fit was counted as having two degrees of freedom and has none; the four constants are determined exactly by the four combinations of lines the theorem does not fix, and the residual of 0.0073 per cent is entirely the data’s own departure from the theorem, which no field could remove. Ammonia has eight lines, two species and five constants: one left, not three. Methane has eight lines, three species — T2T_2, E and A1A_1 — and four constants: one left, not four.

That makes water’s band, twenty-two times narrower than ammonia’s, a different object than it looked. Water’s four constants are fixed exactly by the four combinations of its lines the theorem leaves free, and holding any constant away from its value costs misfit at once, which is why the band was narrow. Its width came from a tolerance of half the published misfit, and all of that misfit is the floor every field carries. With the floor taken out there is no tolerance left, and the band shrinks to the published field itself: water’s field is not the best of a family, it is the only one.

Ammonia’s verdict, recounted

Counted after the theorem's equations, ammonia's band crosses nine tenths again. The share of a least-cost bond-angle distortion that goes into ammonia's degenerate bend, along the held stretch–bend constant, with three bands: all misfit as noise with ν = 3 (-1.000 to -0.634); the product rule's floor taken out with ν = 3 (-0.899 to -0.760, verdict 0.867 to 0.899); and the floor taken out with ν = 1, the degrees of freedom left once the theorem's equations are removed (-0.946 to -0.705, verdict 0.855 to 0.910).
Fig. 6 The share of a least-cost bond-angle distortion that goes into ammonia’s degenerate bend along its held stretch–bend constant, with the band read three ways: all misfit as noise with three degrees of freedom, the floor taken out with three, and the floor taken out with the one that remains.

With the floor taken out and one degree of freedom rather than three, the criterion allows each field twice the excess over the floor rather than four-thirds of it, and ammonia’s band widens from −0.899 to −0.760 to −0.946 to −0.705 mdyn/rad. That is still narrower than the published band, −1.000 to −0.634, by a factor of 1.5. And the bond-angle verdict across it runs from 0.855 to 0.910. It crosses nine tenths again. The reading that put the verdict below the line by 0.0012 depended on three degrees of freedom the theorem says the fit does not have.

The honest statement about ammonia’s bond angle is therefore the one the census began with, restated with both corrections in it. With the misfit split into the part the masses force and the part fields can trade, and the equations counted after the masses have used theirs, the fields ammonia’s two isotopologues cannot tell apart put the bond-angle verdict on both sides of nine tenths — over a band two-thirds as wide as the one first drawn.

The product rule on every field fitted to two isotopologues. For water, ammonia and methane: the published residual, the floor the product rule forces and its share of the sum of squares; each species' product under every harmonic field and from the observed lines; and how well the published misfits line up with the forced pattern (1 is exactly, 0 not at all). H₂O / D₂O: residual · floor · share forced: 0.0073% · 0.0073% · 100.0%; A₁: theorem · observed product: 0.527601 · 0.527790 (0.036%); B₂: theorem · observed product: 0.732709 · 0.732733 (0.003%); alignment of the misfits with the forced pattern: 1.000; NH₃ / ND₃: residual · floor · share forced: 0.4632% · 0.4289% · 85.7%; A₁: theorem · observed product: 0.542935 · 0.535966 (-1.292%); E: theorem · observed product: 0.533762 · 0.523758 (-1.892%); alignment of the misfits with the forced pattern: 0.926; CH₄ / CD₄: residual · floor · share forced: 0.4010% · 0.0369% · 0.8%; T₂: theorem · observed product: 0.559688 · 0.560617 (0.166%); E: theorem · observed product: 0.707379 · 0.707209 (-0.024%); A₁: theorem · observed product: 0.707379 · 0.707949 (0.081%); alignment of the misfits with the forced pattern: 0.092.
Fig. 7 For water, ammonia and methane: the published residual, the theorem’s floor and its share, each species’ product under every field and from the lines, and how the published misfits align with the forced pattern.

What a smaller count does to every verdict read against it

The census of verdicts that began with a label that prices nothing read every symmetry label’s verdict off a fitted field, and the essay that walked ammonia’s field asked how far those verdicts move across the fields a fit cannot tell apart. Both questions assumed the fit had the degrees of freedom its lines and constants suggested. A smaller count does two things at once. It makes the criterion more generous, since a fit with one spare equation cannot be expected to resolve fields as finely as one with three; and it makes the published residual a worse estimate of noise, since one spare equation is a sample of one.

For ammonia the two effects pull the same way, towards a wider band, and the verdict that the floor alone had pushed below the line crosses it again. For methane, whose one remaining equation is the only test of a field whose misfit is all form, a band along an interaction constant would mostly measure how the T2T_2 division changes, which is the thing its misfit already says is wrong. For water there is nothing to read: the field is the data’s. The partner field that deuterium was asked to rule out was ruled out, or not, by misfits whose noise level the same count now changes.

How the floors and counts were computed

Each molecule’s frequencies are the harmonic normal modes of its valence force field at the parent’s and the deuterated isotopologue’s masses, with the published constants; species are taken from the fit’s own assignment and degenerate lines grouped. A species’ product is the product of its distinct lines’ heavy-to-light ratios. The floor in each species is the least sum of squared relative misfits, weighted as the fit weights them with each degenerate line counted as often as it appears, subject to the species’ product being the field’s: the deficit is spread equally over the species’ lines with opposite signs on the two isotopologues. The alignment is the cosine between the published field’s signed misfits and that forced pattern, with the same weights. Ammonia’s band is read along the same refitted family of fields as before, with the limit set by the floor plus the excess over the floor times one plus one over the degrees of freedom, and its edges found by regula falsi on the refitted residual.

The checks run where these figures are drawn. In every species of every molecule, the published field and the same field with every constant scaled by a tenth — different constants, different frequencies — give the same product to 10−710^{-7}. No floor exceeds its fit’s residual. Ammonia’s forced share comes out at the 85.7 per cent found by the earlier essay’s separate route. And the refusal: a target set made of a field’s own frequencies, with no observation in it, must have no floor at all, to 10−1210^{-12}, or the floor would be measuring the bookkeeping rather than the data.

What a count of equations cannot say

It is a count for harmonic fields. The product rule is exact for harmonic frequencies, and the equations it uses up are used up only for a harmonic model. An anharmonic model fitted to fundamentals has no such identity, and its degrees of freedom are the ordinary count — but it has more constants, and the question of which lines test it begins again.

One parameter’s band. Ammonia’s band is read along one held constant, the direction its frequencies see least. With one degree of freedom left, the fit can barely test the form at all, and fields off that line, like the partner that exchanges two definitions, are bounded by nothing the residual can say.

Three molecules and full deuteration. Every fit here pairs a parent with its fully deuterated isotopologue, which keeps the symmetry and so keeps the species. A partly deuterated molecule lowers the symmetry, merges species, and carries a product rule of its own over the merged block; a fit that added NH₂D or CH₃D would gain lines and lose fewer equations to the theorem than it gained, which is one reason partly substituted molecules are worth their trouble.

Methane’s remaining equation is not a test of its form. Methane’s one free equation is spent the moment a constant is added; the bend–bend constant the redundancy removed would use it, and the fit would then be exactly determined, as water’s is. Whether its T2T_2 misfit would vanish or merely move is a fit this does not run.

A residual has three parts

Read through the product rule, a fit’s residual separates into the part the masses force, which no field can change and every field shares; the part the field’s form cannot reach, which a better form would remove; and the part that distinguishes one field of the form from another, which is all a band of fields is about. Water’s residual, small as it is, is all of the first. Methane’s, nearly as large as ammonia’s, is almost all of the second. Ammonia’s is mostly the first and a little of the third. Nothing in a single residual says which mixture a fit has; it takes the species’ products to separate them, and a count of the equations they use up to know how much of the remainder is evidence. It was the ammonia band that every verdict in the census of symmetry labels was eventually read against.

It is symmetry that makes the separation exact. Without species there would be one product for the whole molecule and one equation used up; with them there are as many as there are species, and the isotope shift is arithmetic species by species. The same factorisation of the secular equation that lets a label assign a distortion to a vibration lets a theorem take an equation away from every fit that ignores it.

Still open: data that obey the theorem, and a fit with nothing left over

The obvious open question is ammonia’s data. A deficit of 1.3 and 1.9 per cent in its isotope products is far larger than water’s or methane’s, and its sign is not the one anharmonicity in both isotopologues would give. Replacing NH₃’s and ND₃’s eight numbers with harmonic frequencies derived consistently for both — from fundamentals and measured anharmonic constants — would remove most of the floor at its source, and then the one remaining degree of freedom would be a real test of the five-constant form rather than a test of the data’s consistency.

The nearer question is methane with its bend–bend constant restored under a constraint that removes the redundancy, which would use its last equation. An exactly determined methane field would say whether the T2T_2 species’ mis-division between bend and stretch is the missing constant or the valence form itself, and its mode shapes would give the T2T_2 labels’ verdicts a field that every line, rather than all but one, has fixed.

What links here

Computed from the collection rather than written here: the essays that point at this one.

Reads more easily once this is understood

Essays that name this one as worth reading first.

Shares its objects with

Essays naming at least two of the same things, that neither author linked.

Named objects

A dashed tag is an object no other essay names yet.

Force constantHarmonic approximationIrreducible representationsIsotope substitutionLeast-squaresNormal mode