What a spectrum settles

Deuterium breaks the pyramid's partner

Boron trifluoride's degenerate block carries two force fields that no symmetric isotope can tell apart, because its kinetic matrix has equal diagonals at every mass. Ammonia's does not, and the obvious guess is that the pyramid has no partner at all. It has one in each of its two blocks, built from the ratio of the block's two kinetic elements. What the pyramid lacks is a ratio that stays put: deuterium moves the E block's by 1.9 per cent and the A₁ block's by 12, which puts the partners 9 to 103 wavenumbers apart in ND₃ — far enough to reject one of them outright, and the other only by a factor of two.

Worth reading first: The partner field was a swap · The Coriolis constant was an isotope shift.

A harmonic force field is fitted to a molecule’s band centres, and the force field is not in the spectrum: more constants than bands leave a family of fields that fit equally well. Boron trifluoride’s degenerate block was the cleanest case. Three constants and two frequencies gave a closed loop of exact fits, a boron-10 band cut the loop to two fields, and no third symmetric isotope could choose between them. The partner field was a swap found why. In the right coordinates the block’s kinetic matrix has equal diagonal elements at every set of masses, so exchanging the stretching and rescaled bending constants commutes with it, and the two fields give the same frequencies for every isotopologue that keeps the symmetry.

That essay ended on the pyramid. Ammonia’s degenerate block pairs a stretch and a bend in the same way, but nothing forces its kinetic diagonals to be equal, and its fit had carried a wide band of fields rather than a closed pair. Either some other exchange exists there or its family has no such symmetry at all.

Neither, as it turns out. Ammonia has a partner field in each of its blocks, exact at its own masses, and deuterium breaks both — one decisively and the other barely.

Every two-by-two block has one

The argument needs one line of algebra and no property of boron. Write a block’s force field and kinetic matrix as

F=(xc yc yz),G=(g11g12g12g22),F=\begin{pmatrix} x & c\,y\\ c\,y & z\end{pmatrix},\qquad G=\begin{pmatrix} g_{11} & g_{12}\\ g_{12} & g_{22}\end{pmatrix},

with xx the stretch combination, zz the bend combination and yy the stretch–bend constant. The block’s two squared frequencies are the eigenvalues of GFGF, so they are fixed by its trace and determinant:

S=g11x+2g12c y+g22z,P=det⁡G (xz−c2y2).S=g_{11}x+2g_{12}c\,y+g_{22}z,\qquad P=\det G\,(xz-c^2y^2).

Now replace xx by (g22/g11) z(g_{22}/g_{11})\,z and zz by (g11/g22) x(g_{11}/g_{22})\,x, and keep yy. The trace is unchanged, because g11x′=g22zg_{11}x'=g_{22}z and g22z′=g11xg_{22}z'=g_{11}x. The product x′z′x'z' equals xzxz, so the determinant is unchanged too. The new field gives exactly the same two frequencies. At any one set of masses, every two-by-two block has a partner, and the only thing it needs to know is the ratio of the block’s two kinetic diagonals.

When the diagonals are equal, as in boron trifluoride’s rescaled coordinates, the ratio is one and the partner is the plain exchange. That is the case the earlier essay found, and its special property is not that the partner exists but that the ratio is one at every mass, so the same exchange works for every symmetric isotopologue at once. When the diagonals differ, the partner is a rescaled exchange, and the scale belongs to the masses it was built at.

Ammonia’s four fields

Ammonia’s published field has five constants — the N–H stretch, the HNH bend, the stretch–stretch, stretch–bend and bend–bend interactions — fitted to the band centres of NH₃ and ND₃ together. By symmetry they separate into two blocks of two coordinates each. The E block holds the degenerate stretch and bend, with stretch combination x=fr−frrx=f_r-f_{rr} and bend combination z=fθ−fθθz=f_\theta-f_{\theta\theta}. The A1A_1 block holds the symmetric stretch and the umbrella, with s=fr+2frrs=f_r+2f_{rr} and u=fθ+2fθθu=f_\theta+2f_{\theta\theta}. The single stretch–bend constant couples both.

At NH₃'s masses each of ammonia's two blocks has a second field that gives the same two bands. For the E block and the A₁ block, the published field's stretch and bend combinations and the partner field the rescaled exchange gives at NH₃'s masses. E: stretch 6.323 and bend 0.833 against 1.943 and 2.711. A₁: 5.974 and 0.963 against 1.359 and 4.231. The stretch–bend constant is the same in both, and the dashed curve is the set of fields with the same product of the two, along which the exchange moves.
Fig. 1 The published field and its partner at NH₃’s masses in the E block and in the A1A_1 block, with the curve of constant stretch-times-bend along which the exchange moves.

The kinetic diagonals do not have to be derived by hand. Since SS is linear in the field, its derivative with respect to each combination is that combination’s kinetic element, up to one common factor, and the derivatives can be read off the site’s own normal-mode calculation by central differences, which are exact for a linear function. The second differences come out below 10−910^{-9} of SS, which is the check that the reading is right. For NH₃ the E block’s ratio g22/g11g_{22}/g_{11} is 2.3327 and the A1A_1 block’s 1.4121.

The partners follow. In the E block the published stretch combination is 6.323 mdyn/Å and the bend 0.833; the partner has a stretch of 1.943 and a bend of 2.711. In the A1A_1 block the published 5.974 and 0.963 become 1.359 and 4.231. Each partner reproduces NH₃’s bands in its block to 10−410^{-4} cm⁻¹ and leaves the other block’s bands exactly where they were, since the stretch–bend constant is untouched. With two blocks and a partner in each, NH₃ alone cannot distinguish four fields, and two of them make the umbrella or the degenerate stretch a mostly angular motion.

That is the ammonia version of what the swap did to boron trichloride: a stiff bond and a soft angle, or a soft bond and a stiff angle, chosen by nothing in the spectrum at those masses.

What the masses do to the ratio

Whether a second isotopologue separates partners is now a precise question: does it change g22/g11g_{22}/g_{11}?

What an isotope does to the ratio of a block's two kinetic diagonals: almost nothing, or an eighth. The change in the ratio of the bend's kinetic element to the stretch's, from NH₃ to two isotopologues, in each block. E: +0.14% for nitrogen-15, −1.87% for deuterium. A₁: −0.83% and +12.04%. The planar trihalide's ratio is one at every mass, and moves by nothing.
Fig. 2 The change in each block’s ratio of bend to stretch kinetic element from NH₃ to ¹⁵NH₃ and to ND₃, with the planar molecule’s ratio, which no symmetric isotope moves.

It does, by very different amounts in the two blocks. Deuterium moves the E block’s ratio by −1.87 per cent and the A1A_1 block’s by +12.0. Nitrogen-15 moves them by +0.14 and −0.83 per cent. The pattern has a plain reading. Each kinetic element is a sum of reciprocal masses weighted by geometry, and a ratio of two of them changes with a mass only as far as the two elements weight that mass differently. Replacing every hydrogen by deuterium multiplies the E block’s stretch and bend elements by 0.543 and 0.532, nearly the same factor, because hydrogen dominates both. It multiplies the A1A_1 block’s symmetric-stretch element by 0.516, close to the 0.500 that hydrogen alone would give, and the umbrella’s by 0.578, because the umbrella moves the nitrogen much more and a share of its element does not change at all. The two blocks’ ratios move by the gap between those factors.

In the planar trihalide neither isotope could do this, because both rescaled diagonals equal 1/mX+3/(2mB)1/m_{\mathrm X}+3/(2m_{\mathrm B}) whatever the masses are. That identity is what the pyramid lacks, and its absence is the whole difference between a partner that survives and one that does not.

ND₃ asked about the four fields

Asked about ND₃, the partners that NH₃ cannot tell apart give different spectra. ND₃'s four distinct bands: observed 749, 2564, 1191, 2660 cm⁻¹; under the published field 750.40, 2575.96, 1198.89, 2667.27; with the E partner 750.40, 2575.96, 1207.67, 2647.89; with the A₁ partner 721.58, 2678.86, 1198.89, 2667.27.
Fig. 3 ND₃’s four distinct band centres as observed, under the published field, with the E partner in place of the published E block, and with the A1A_1 partner in place of the published A1A_1 block.

Built at NH₃’s masses, the partners are no longer partners at ND₃’s. The E partner puts ND₃’s degenerate bend at 1207.7 cm⁻¹ and stretch at 2647.9, against the published field’s 1198.9 and 2667.3, differences of +8.8 and −19.4. The A1A_1 partner puts the umbrella at 721.6 and the symmetric stretch at 2678.9, against 750.4 and 2576.0, differences of −28.8 and +102.9. The symmetric stretch under the A1A_1 partner lands above the degenerate stretch, an inversion of the band order that the observed spectrum, 2564 against 2660, plainly does not show.

How far apart each isotope puts a pair of fields that NH₃ cannot separate. For each of ammonia's four distinct bands, the difference between the partner field's prediction and the published field's, at nitrogen-15 and at deuterium, on a logarithmic axis. E lower, ¹⁵NH₃ −0.93 cm⁻¹; E upper, ¹⁵NH₃ +1.97 cm⁻¹; E lower, ND₃ +8.77 cm⁻¹; E upper, ND₃ −19.38 cm⁻¹; A₁ lower, ¹⁵NH₃ +3.06 cm⁻¹; A₁ upper, ¹⁵NH₃ −10.44 cm⁻¹; A₁ lower, ND₃ −28.82 cm⁻¹; A₁ upper, ND₃ +102.90 cm⁻¹.
Fig. 4 For each of ammonia’s four distinct bands, how far the partner field’s prediction sits from the published field’s, at ¹⁵NH₃ and at ND₃, on a logarithmic axis.

Nitrogen-15 separates the partners too, but by much less: 0.9 and 2.0 cm⁻¹ in the E block, 3.1 and 10.4 in the A1A_1 block. That ordering follows the ratios exactly. Taking the larger band of each block, the separation per per cent of ratio shift is 10.3 cm⁻¹ for the E block under deuterium and 14.5 under nitrogen-15, 8.6 for the A1A_1 block under deuterium and 12.6 under nitrogen-15. The separation is close to proportional to the shift, within a factor of 1.7 across a ninety-fold range of shifts, which is what a first-order change in one ratio should give. It also means the separation can be estimated for any isotopologue that keeps the symmetry from its kinetic ratio alone, before any frequency is computed. The partner’s prediction at a new isotopologue differs from the published field’s by an amount set by how far that isotopologue moves the ratio, and the A1A_1 block under deuterium is the largest move available.

The contrast with the planar molecule is worth a number. In boron trichloride the only thing that separated the two fields was a chlorine isotope that broke the molecule’s symmetry, and the bands differed by 0.06 cm⁻¹, at second order in a mass difference. Ammonia’s partners differ by tens of wavenumbers under a substitution that keeps every symmetry the molecule has. The pyramid needs no symmetry-breaking isotope at all, because it never had the identity that made the planar fields inseparable.

Ruled out, and only disfavoured

A difference in prediction is not yet a verdict; it has to be compared with how well the published field itself does on ND₃.

Measured against ND₃'s bands, the A₁ partner is ruled out by a factor of eleven and the E partner only by two. The root-mean-square relative miss of each block's two ND₃ bands, under the published field and under the partner built at NH₃'s masses: E 0.507% against 1.041%, A₁ 0.355% against 4.091%. The published field's own miss is the size of what a harmonic field gets wrong anyway.
Fig. 5 The root-mean-square relative miss of each block’s two ND₃ bands under the published field and under the partner built at NH₃’s masses.

The published field misses ND₃’s A1A_1 bands by 0.36 per cent and the A1A_1 partner by 4.09: a factor of eleven, and an inverted band order. That partner is ruled out. The E block is different. The published field misses ND₃’s degenerate bands by 0.51 per cent and the partner by 1.04, only a factor of two, and the published field’s own miss is the size of what a five-constant harmonic field gets wrong about real band centres anyway. A harmonic field fitted to anharmonic band centres carries errors of that order, and the umbrella’s tunnelling and the anharmonic shifts of the stretches are not in it. A partner that does twice as badly as a field already wrong by half a per cent is disfavoured, but a change of anharmonic correction of the same size could in principle move the verdict.

The same question from the other side

The construction does not prefer NH₃. Built with ND₃’s own kinetic ratios, each block has a partner that keeps ND₃’s bands exactly, and that partner can be asked about NH₃ instead. The E partner built at ND₃’s masses has a stretch combination of 1.906 and a bend of 2.762, close to the NH₃-built partner’s 1.943 and 2.711, and it puts NH₃’s degenerate bands at 1666.8 and 3594.4 cm⁻¹ against the published field’s 1679.7 and 3566.7. The A1A_1 partner built at ND₃’s masses, with a symmetric-stretch combination of 1.523 and an umbrella of 3.776, puts NH₃’s umbrella at 1060.7 and its symmetric stretch at 3356.5, against 1020.1 and 3490.3 — a shift of 134 cm⁻¹ in the band that the stretch carries.

So the separation is symmetric in the way the algebra says it must be. Each molecule has its own pair of exact fields in each block, both pairs contain the published field, since both partners were built from it, and their other members lie apart by an amount set by how far the ratio moved. Neither molecule is privileged as the one the partner “belongs” to.

That also says something about the published field itself. It reproduces neither molecule exactly: NH₃’s degenerate bands come out at 1679.7 and 3566.7 against an observed 1691 and 3577, and ND₃’s at 1198.9 and 2667.3 against 1191 and 2660. It is a compromise between the two molecules’ exact fields, and each partner above is a partner of that compromise. A fit that gave the two molecules different weights would sit at a different compromise with different partners, though the sizes of the separations come mostly from the ratios rather than from the field and should change much less.

So the answer to the question the planar case left is in two parts. Ammonia’s A1A_1 block has a partner that its own deuterated form throws out. Its E block has a partner that deuterium only leans against, and the reason is that hydrogen dominates both of that block’s kinetic elements nearly equally, which is the pyramid’s nearest approach to the planar molecule’s identity.

The four fields NH₃ allows, block by block, and what each says about ND₃. For each block, the published field and its partner at NH₃'s masses: the stretch and bend combinations, the shared stretch–bend constant, the two ND₃ bands each predicts and the relative miss against the observed bands.
Fig. 6 The four fields NH₃ allows, block by block, with the constants of each and what each predicts and misses for ND₃.

How the ratio was read

The published field is the site’s five-constant valence field for ammonia, fitted to NH₃’s and ND₃’s band centres; its frequencies at any isotopologue come from the full Cartesian normal-mode calculation, with each block’s bands identified by degeneracy — the E bands as the two degenerate pairs, the A1A_1 bands as the two singles — rather than by their order, since a partner can move an A1A_1 band past an E pair. A field is rewritten in block combinations with the A1A_1 sums held when the E differences move and the other way round, so that each partner touches one block only. The kinetic elements are derivatives of the block’s sum of squared frequencies with respect to the stretch and bend combinations at a step of 0.05, central differences, at each isotopologue.

Three things are checked where the figures are drawn: the sum is linear in the field, its second differences below 10−910^{-9} of itself; each partner keeps NH₃’s bands in its block to 10−410^{-4} cm⁻¹; and it leaves the other block’s bands alone to the same tolerance. The refusal runs the construction the other way: a partner built with ND₃’s own ratio must reproduce ND₃’s bands in its turn, to 10−210^{-2} cm⁻¹, which the numerical splitting of a degenerate pair at those masses allows, and it does. Whatever separates the NH₃-built partner at ND₃ is then the change of ratio and nothing else.

A harmonic field is not the molecule

The partners are partners of a harmonic field. A real band centre carries anharmonic shifts of the order of a hundred wavenumbers in the N–H stretches, and ammonia’s umbrella is a tunnelling doublet rather than a harmonic level. Every verdict above compares two harmonic predictions with each other and with observed centres, and the E block’s factor of two is close enough to the published field’s own misfit that a better model of the anharmonicity could change it.

The proportionality is an observation on four points. Two isotopologues and two blocks give four separations, and a first-order argument explains their rough agreement; it is not a law that has been tested on a molecule with a third symmetric substitution, such as ¹⁵ND₃, where both masses change at once.

Only the stretch–bend constant is shared, and there is only one. Ammonia’s field has a single stretch–bend interaction coupling both blocks, and a field with separate E and A1A_1 interactions would have more freedom in each block than the partner construction uses. The construction itself does not care: it holds the block’s coupling fixed whatever it is made of.

Two isotopologues, not a natural sample. Partly deuterated ammonia breaks the three-fold symmetry and mixes the blocks, which is exactly the situation the chlorine isotope created for the trihalide; it is a further test that has not been run here.

A partner is a statement about one set of masses

Read together with the planar case, the pattern is that a two-by-two block always has two exact fields at one set of masses, and the question that matters is how much a second set of masses moves one ratio. For boron trifluoride the answer is nothing, by an identity, and the partner survives every symmetric isotope. For ammonia the answer is two per cent in one block and twelve in the other, and the partners fail in proportion. The Coriolis constant that turned out to be an isotope shift was the planar version of the same fact: a quantity that depends on mode shapes still cannot separate fields whose kinetic matrix the masses never distinguish.

The surprising part is which block is the hard one. The degenerate pair is the block every earlier argument about ammonia’s field worried over, and it is the block deuterium barely helps with. The totally symmetric block, which the planar molecule does not even have as a pair, is where a single substitution settles everything.

Still open: the partly deuterated molecules, and a field with two couplings

The obvious open question is NH₂D and NHD₂. Without the three-fold axis the E and A1A_1 blocks mix into one four-by-four problem, and a partner built in either block at NH₃’s masses has no reason to keep any of the mixed molecule’s bands. Their band centres are measured, and the size of the miss would say whether one partly deuterated molecule does better than ND₃ at separating the E partners, which is the block that still needs it.

The nearer question is the field’s form. Every valence field here uses one stretch–bend constant for both blocks. Giving the E and A1A_1 blocks separate interactions adds a constant and leaves each block’s partner construction unchanged, but it changes which fields the fit to NH₃ and ND₃ together can reach. Whether the E partner then becomes as decisively wrong as the A1A_1 partner, or the fit spreads along a new family instead, is a fit with one more parameter.

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 approximationIsotope substitutionModel limitNormal modeValence force field