The offset fades at the window's floor
Worth reading first: The singlet the doublets never had · A precision is a spacing.
A four-parameter fit to a magnetic susceptibility curve has two nuisance parameters, an impurity fraction and a temperature-independent term, and at two exchange couplings they stop being correlated. Across every connected cluster of five and six spin-½ centres, after a band of seven clusters broke, the upper of those couplings is set, to within a factor of two, by the cluster’s spectrum, and the last refinement of that statement replaced a gap by a population: the coupling at which the states of spin above the ground’s hold a fraction of the Boltzmann population at 300 K, with fitted on the twenty five-centre doublets and carried unchanged to the twenty-one six-centre triplets.
The triplets kept a remainder, and it had a shape. Five of them have a ground level holding a triplet together with one or more singlets at exactly the same energy, and those five sat at or near the top when the triplets were ranked by the ratio of their upper separation to . Five spin-½ centres have no spin below one half, so the doublets that fixed could never have contained such a ground.
That essay named the obvious worry. The degeneracy is exact only because every coupling in the census is equal, and no real cluster has equal couplings. Give one edge a coupling a few per cent off and the singlet moves a little above the triplet or a little below it. If the offset survives until the splitting is a sizeable fraction of at 300 K, real clusters near such a degeneracy carry it. If it vanishes the moment the degeneracy breaks, it was an artefact.
It does neither. It fades smoothly, and the temperature it fades over is the window’s floor rather than its top.
Breaking a degeneracy with one coupling
Each of the five clusters is a graph of six centres with one antiferromagnetic coupling on each edge. Changing one edge’s coupling by a fraction leaves the graph and the rest of the spectrum nearly as they were, and splits the ground: the triplet and the singlet that were degenerate are now separated by an energy that grows with .
Which way the split goes depends on the sign. On one of cluster 35’s edges, lowering the coupling by five per cent leaves the triplet as the ground with the singlet 0.10 coupling units above; raising it by five per cent puts the singlet below and the triplet 0.10 above. That is the general pattern for the three clusters with one singlet in the ground. Every edge that lifts the degeneracy at all has one sign that keeps a triplet ground and one that makes the ground a singlet, at every change tried from 0.25 to 50 per cent. One of cluster 47’s edges lifts nothing: changing it leaves the triplet and singlet degenerate whichever way it moves, because its change shifts the triplet and the singlet by the same amount.
A cluster with a singlet ground is not part of the comparison at all. The census reads a cluster by the spin of its lowest level and compares only clusters with a spin in the ground; a singlet ground’s moment vanishes at low temperature instead of levelling off, and it was left out of the triplets for that reason, as it is here. So a real cluster near such a degeneracy is, half the time, not a triplet at all. The question the lead asked belongs to the other half.
How the offset falls
On the side that keeps a triplet ground, the split cluster is read exactly as the census reads any cluster: its susceptibility fit’s separations followed from 3 to 1,500 cm⁻¹, its upper separation the second of them, and its computed from its full split spectrum at the doublets’ . Its excess is its ratio less the ordinary triplets’ median, 0.299.
Unsplit, the three clusters’ excesses are 0.074, 0.059 and 0.085. Every edge’s curve falls smoothly as the splitting grows, with no step at the moment the degeneracy breaks: at a splitting of one kelvin the median excess is 0.066, at five kelvin 0.052, at twenty 0.030 and at fifty 0.011. On twelve of the thirteen edge classes it halves somewhere between 11.1 and 20.4 K, with a median of 14.9 K; the thirteenth is the one edge of cluster 47 whose largest change still splits the ground by only 11 K, and it has reached 53 per cent of its unsplit excess there. On three edges, followed far enough, the excess reaches zero, at 46, 81 and 109 K.
The splitting is measured in kelvin at the cluster’s own upper separation — the splitting in coupling units times the coupling at which the fit’s two nuisance parameters decouple — because that is the coupling at which the excess is measured. In wavenumbers the halving splittings run from 7.7 to 14.2 cm⁻¹.
The floor of the window, not its top
The lead framed the test against at the window’s top, 208 cm⁻¹ at 300 K, on the reasoning that a splitting small against the thermal energy is invisible to a measurement at that temperature. The offset halves at splittings about one twentieth of that. It halves, instead, at about at the window’s floor: 13.9 cm⁻¹ at 20 K.
That has a reading. A susceptibility fit separates an impurity fraction from a temperature-independent term by the shape of the curve at low temperature, where the impurity’s Curie tail rises and everything else flattens. The mixed ground’s offset was traced to exactly that region: a ground with a non-magnetic state in it has a low-temperature moment below a pure triplet’s, which the fit must reconcile with an impurity term whose job is also to set the low-temperature moment. A singlet split off the ground by much less than at 20 K is still, at every temperature the fit reads, part of the ground’s population; one split off by much more is depopulated at the bottom of the window and becomes an excited state like any other. The offset therefore lives on the scale of the lowest temperature in the data. A window starting at 2 K would presumably have held it to a splitting ten times smaller, and that is a prediction this calculation has not tested.
Read as a ranking, the clusters move down through the ordinary triplets as their grounds split. Unsplit, their ratios of 0.359 to 0.384 put them among the five or six highest triplets in the census. At a twenty-kelvin splitting they spread from 0.315 to 0.343, in the upper middle of the ordinary triplets’ range of 0.255 to 0.420. At fifty kelvin, most sit within a few hundredths of the ordinary median. A splitting of a few tens of kelvin is enough to make a mixed ground an ordinary triplet as far as this fit is concerned, and a few kelvin is not.
How different a coupling has to be
A splitting in kelvin is the natural unit for the fit, and the wrong one for a chemist looking at a structure, who sees bridges and angles and asks how unequal the couplings are. The two are related edge by edge, and the relation is far from uniform.
A one per cent change in one coupling splits the ground by 0.3 to 3.6 K, depending on the edge; a five per cent change by 1.7 to 17.2 K. So the change needed to halve the offset runs from 4.4 per cent on the edges that move the ground most to 35 per cent on the edge that moves it least, with most edges between 10 and 16 per cent. In each of the three clusters the cheapest edge is the same kind: the single edge joining the two centres whose other neighbours are exactly the same pair. Each coupling on the graph stands for one superexchange path between two spins, and that is the path whose change separates the singlet from the triplet fastest; why it is that one, beyond the count, is not something the calculation shows.
That turns the scale into something checkable from a structure. A cluster whose couplings are equal by symmetry to within a few per cent keeps most of its offset whichever edge is unequal. One whose bridges differ by a tenth or more has lost half its offset on most edges and nearly all of it on the sensitive ones. The equal couplings of the census are the extreme of the first case. A census with every coupling equal finds every degeneracy a graph can have, which is what makes it a census; a real crystal would split any of those degeneracies its own symmetry does not protect, and a difference of a tenth between two bridges would split them by more than the window’s floor.
Two or three singlets cannot be split by one edge
The two clusters with more than one singlet beside the triplet behave differently, and the reason is a count.
A changed edge splits one combination of states from the rest. With one singlet that is enough: the singlet and the triplet separate and the ground is pure. With two or three, one edge can move one singlet away and leave the others where they were. In cluster 97, with two singlets, eight equivalent edges make the ground a singlet whichever way they move; the other two, lowered, leave a triplet and a singlet still degenerate, and raised, make a singlet ground. In cluster 98, with three singlets, the same: no edge, moved either way, leaves a triplet alone in the ground.
On the edges where a triplet stays in the ground, it stays with company, and the ratio falls only part of the way. Cluster 97 goes from 0.397 to 0.350 at a fifty per cent change; cluster 98 from 0.482 to 0.412 on one edge and 0.367 on another. Those are changes far larger than any real cluster’s disorder and they leave cluster 98 above most of the ordinary triplets. A ground with several singlets is not split by a single defect, and its offset survives one.
How the split clusters were read
Each split cluster’s spectrum is the isotropic exchange Hamiltonian on its graph, diagonalised in full and grouped into multiplets. Its susceptibility curve is fitted by the four-parameter model the census uses, a coupling, an impurity fraction, a temperature-independent term and a -factor, on forty readings evenly spaced from 20 to 300 K with flat weights; the correlation between the two nuisance parameters, from the fit’s own covariance, is followed on a 160-point logarithmic grid of couplings from 3 to 1,500 cm⁻¹ and each change of sign bisected thirty times. The upper separation is the second change of sign, kept only below 1,000 cm⁻¹ as the census does. is found by bisection from the split spectrum with every multiplet weighted by its degeneracy, at = 0.00761 and 300 K. Equivalent edges are grouped by the spectrum a five per cent change produces, and one edge of each class is followed through twenty changes from −50 to +50 per cent.
The checks run where these figures are drawn. Unsplit, every cluster’s separations reproduce the census’s upper separation to a part in a million. With one singlet in the ground, every edge that lifts the degeneracy keeps a triplet one way and makes a singlet the other; with two or three, no edge leaves a pure triplet. On every edge with a triplet side, the excess at the largest splitting is below its value at the smallest. And the refusal: four ordinary triplets, with no singlet in their ground, are given the same one per cent change on one edge. Their ratios move by at most 0.0011, symmetrically in the sign of the change, against a median one-sided move of 0.0091 for the mixed clusters. If the ordinary clusters had moved as far, the fall would have been the reading’s sensitivity to any change of coupling rather than the ground’s.
What one changed edge cannot represent
Real disorder changes every coupling. A cluster in a crystal has all its couplings slightly different, not one, and the ground’s splitting is then a sum of small contributions. The finding that the offset depends on the splitting’s size rather than on which edge produced it — the twelve halving points fall within a factor of two of one another across three clusters and twelve different edges — suggests the splitting is what matters, but a cluster with every coupling perturbed has not been read.
Only isotropic exchange. A real triplet ground is split further by anisotropy, zero-field splittings of a wavenumber or less in many copper clusters, which is below the scale found here but not far below it. A triplet and a singlet a few wavenumbers apart, with the triplet itself split by one, is a three-level problem this does not touch.
The singlet side is not read. Half of every split produces a singlet ground with a triplet a few kelvin above it, and those clusters were set aside because the census sets them aside. A fit to such a cluster’s curve still returns two nuisance parameters and still has couplings at which they decouple; where those fall, and whether a near-degenerate singlet ground carries an offset of its own in the other direction, is a separate census with a separate reference set.
The fit’s window is fixed. Every reading here is on a 20–300 K design. The conclusion that the offset lives at the window’s floor is drawn from one floor; a second window would test it directly.
A degeneracy is a scale, not a switch
The doublets could not have taught this, and neither, it turns out, could equal couplings alone. It is the second time a property of these clusters has turned out to be a matter of counting rather than of the physics it was first credited to, after the exponent’s sign that was put down to frustration. An exact degeneracy looks like a property a cluster either has or lacks, and a refinement fitted on clusters without it seems to need a separate rule for clusters with it. What the split clusters show is that the rule is continuous: a singlet in the ground counts fully, a singlet at 50 K counts nearly not at all, and in between it counts in proportion to how much of the window it is populated over. That is what a precision turned out to be for the readings’ weights and what the window was for the constant: something measured against the temperatures in the data rather than an intrinsic label.
It also says what to do with a real cluster. A fitted susceptibility from a cluster whose spectrum puts a singlet within about fifteen kelvin of a triplet ground, at the coupling the fit returns, should be read with the mixed grounds’ offset; one with the nearest singlet fifty kelvin or more away should be read as an ordinary triplet. Between them, the excess interpolates, and the parameters a curve is worth do not include a separate one for it.
Still open: a low-temperature window, and every coupling at once
The obvious open question is the window’s floor. If the offset lives on the scale of at the lowest temperature in the data, then the same five clusters fitted on a 2–300 K window should keep it to splittings ten times smaller, and on a 50–300 K window lose it at splittings two and a half times larger. That is a direct test of the reading above, at the cost of re-reading the census’s separations under two more designs, which the spacing essays already know how to read.
The nearer question is disorder in every coupling. Drawing each of a cluster’s couplings from a narrow distribution, a per cent or two wide, and reading the excess against the resulting ground splitting would say whether the halving scale of 11 to 20 K survives when the splitting is the sum of many small changes, and what a two-singlet ground does when every edge can move one of its singlets away at once.
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.
- The model is what is fitted — both name exchange coupling, least-squares, spin state, temperature
- The moment a fit invents — both name exchange coupling, least-squares, spin state, temperature
- The sign a frustrated ring changes — both name degeneracy, exchange coupling, least-squares, magnetic susceptibility
- A moment between two integers — both name degeneracy, spin state, temperature
- Purity renames the poles — both name exchange coupling, magnetic susceptibility, spin state
- The frustrated cluster with an even count — both name degeneracy, exchange coupling, magnetic susceptibility
Named objects
A dashed tag is an object no other essay names yet.
DegeneracyExchange couplingLeast-squaresMagnetic susceptibilitySpin stateTemperature