What the shape is for

The singlet the doublets never had

A susceptibility fit's upper separation, times the gap to the first multiplet of higher spin, came to a constant across forty-one spin clusters with a factor of about two left over. Replace the gap with the population of the raised spins, every state counted, and fit the one fraction it needs on the twenty doublets. The doublets close by a sixth under every design of measurement tried. The triplets close by much less, and what is left in them is a kind of ground no five-centre doublet can have: a triplet with a singlet at exactly the same energy.

Worth reading first: A precision is a spacing · The excitation that raises the spin.

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 the upper of those couplings, multiplied by the gap to the first multiplet of higher spin than the ground’s, comes to about one number. Then the number was followed through the window, the spacing of the temperatures and the precision of each reading. It moved with every one of them. What did not move was the spread of the forty-one clusters about it, a factor of about two under every design, and the last of those essays concluded that the remainder belonged to the clusters rather than to the measurement.

The candidate it named was the raising multiplet’s population. A gap is one number about a spectrum. A Boltzmann population knows how many states sit at that gap, and whether there are more raised states just above it. Two clusters with the same gap to their first quintet, one with five states there and one with ten, should not reach the same fraction of quintet population at the same coupling. If the upper separation is where the raised spins start to matter to the fit, a coupling defined by their population ought to collapse the clusters better than a coupling defined by their gap.

A coupling defined by a population

The refinement needs one number. For each cluster, at the window’s top temperature, find the coupling JpJ_p at which the states of spin greater than the ground’s hold a fraction pp of the population. Every level of the exchange spectrum enters, each with its full degeneracy: a quintet level of two multiplets counts ten states, and a ground of two degenerate doublets counts four. Then replace the product of the upper separation with the gap by the ratio of the upper separation to JpJ_p. The raising gap was a crude version of this. For a two-level spectrum, JpJ_p is the gap’s reciprocal times a constant.

The fraction pp is not known in advance. It was fitted on the twenty five-centre doublets, as the essay that raised it asked, by choosing the pp that makes their ratios most alike. The twenty-one six-centre triplets were then read at the same pp without being consulted.

One fraction of the population fitted on the doublets tightens them by a sixth and the triplets by an eighth. For an even 20–300 K design, the spread — largest over smallest — of the upper separation divided by the coupling at which the states of spin above the ground's hold a fraction p of the population at 300 K, against p. The doublets' spread is least at p = 0.0076, 1.251 against 1.496 for the raising gap alone. At the same p the triplets' is 1.889 against 2.146, and the least they reach at any p is 1.792.
Fig. 1 The spread of the upper separation over the population coupling against the population fraction, for the doublets and the triplets, on an even 20–300 K design.

For an even sweep of forty temperatures from 20 to 300 K the doublets’ spread, largest over smallest, is least at p=0.0076p=0.0076: 1.25 against 1.50 for the gap, a sixth of the spread removed. At the same fraction the triplets go from 2.15 to 1.89, an eighth. And if the triplets are allowed their own best fraction, around p=0.06p=0.06, they reach 1.79 and no lower. So the refinement transfers, and does less on the triplets than on the clusters it was fitted to. Their own optimum does not close the difference, so that is not a matter of where pp was set.

The fraction itself is small. A coupling at which seven or eight raised states in a thousand are occupied at 300 K is a coupling at which the raised spins are, in any ordinary sense, still frozen out. That is consistent with how the separation was found in the first place. It is a point at which two nuisance parameters decouple in a fit’s covariance, decided by the curvature at the hot end of the curve. It is not a point at which anything is visibly happening to the moment.

What the collapse looks like, cluster by cluster

A spread is a summary of two extreme clusters, and the refinement could improve it by moving those two and nothing else. It does not.

The population pulls the doublets together and leaves the triplets with a lump at the top. Every cluster's upper separation times its raising gap, and its upper separation divided by the population coupling, each scaled to the doublets' median of the same quantity, on an even 20–300 K design. The doublets close from 1.50 to 1.25. The triplets close from 2.15 to 1.89, and their highest values are the five grounds in which a triplet is degenerate with a singlet.
Fig. 2 Every cluster’s gap product and population ratio, each divided by the doublets’ median of the same quantity, on an even 20–300 K design.

The doublets draw together as a body. The cluster with the largest gap product, number eight in the enumeration of connected five-vertex graphs, has eight raised states at its gap, two quartets, where most have four. It comes down from 1.29 times the doublets’ median to 1.11, which is exactly what counting states should do. The doublets with degenerate grounds, six of the twenty, made of two, three or five doublets at one energy, sit inside the body once their extra ground states are counted. They were already close to it with the gap alone.

The triplets do not draw together the same way. Their main group closes up much as the doublets did. But a group of five stays well above it, and under the population it separates more cleanly than it did under the gap. The colours in the figure are what the next section is about. Every one of those five has a ground that is not one triplet.

Under every weight

The obvious objection to any refinement of this constant is that the constant has turned out to depend on the design of the measurement at every step, and a refinement that works under one design might simply be tuned to it. That was the cheap test the earlier essay proposed: does the same refinement reduce the spread under every weight, or only under some?

Under every weight the population helps both sets, and the triplets keep half their spread. For ten designs — evenly, logarithmically and inversely spaced, with precision profiles and with constant absolute noise — the spread of the doublets and of the triplets under the raising gap alone and under the population, its fraction fitted on that design's doublets. Every one of the twenty arrows points down. The doublets end between 1.19 and 1.34; the triplets between 1.89 and 2.22.
Fig. 3 For ten designs, the spread of the doublets and of the triplets under the raising gap and under the population, its fraction fitted on that design’s own doublets.

It does under every one. The ten designs are those the precision essay compared: even, logarithmic and inverse spacings, an even sweep with trapezoid ends, an even sweep with precision growing as T\sqrt T, three levels of constant absolute noise, and two windows with noise. Under all of them all twenty arrows point down. The doublets end between 1.19 and 1.34, from gap spreads of 1.48 to 1.87. The triplets end between 1.89 and 2.22, from 2.14 to 2.66. The noisy designs, whose gap spreads were the worst, gain most, because their constant had drifted furthest from what one gap per cluster could follow.

What does not hold still is the fraction. Fitted on each design’s doublets it runs from 0.0062 for the even sweep with half-weight ends to 0.092 for the noisy wide window, a factor of fifteen. The earlier essays had already found the constant itself moving with the design, so a population fraction that moves too is not a surprise. It says that pp is a coordinate of the measurement, like the constant it refines, and not a property of spins. Anyone using this refinement on real data would have to fit pp under the design they actually used, and that needs a set of clusters with known spectra measured the same way.

A ground of mixed spin

Rank the triplets by their population ratio and the five that stand out are not random.

Five of the six triplets furthest from the doublets have a singlet in their ground. The twenty-one six-centre triplets ranked by upper separation over population coupling, on an even 20–300 K design at the doublets' fraction, with the doublets' range shaded; the numbers are the clusters' places in the enumeration of connected graphs. Five of the top six are grounds in which the triplet is degenerate with one or more singlets — a kind of ground no five-centre doublet can have, since ½ is the lowest spin five centres reach. The one ground of two triplets sits at the bottom.
Fig. 4 The twenty-one triplets ranked by upper separation over population coupling, coloured by what their ground level holds, with the doublets’ range shaded.

Five of the six highest are clusters whose ground level holds a triplet and one or more singlets at exactly the same energy. The census labels a ground by the largest spin it contains, which is why they were counted as triplets in the first place. The sixth of the top six, number two, is the only ordinary triplet up there, and it is the one triplet whose first excitation is itself the raising quintet, with two singlets at the same energy. At the very bottom is the only ground of two triplets.

Among the five mixed grounds there is an order with a reason. The three with one singlet beside the triplet sit at 0.359 to 0.384. The one with two singlets is at 0.397, and the one with three is at 0.482. The more of the ground’s population carries no moment, the higher the ratio. That is only five clusters and it is stated as an ordering, not a law. But it is in the direction that makes sense: a ground with a quarter or a half of its states non-magnetic has a low-temperature moment below a pure triplet’s, and a fit has to reconcile that plateau with an impurity term whose job is also to set the low-temperature moment.

Here is why the doublets could not have taught this. Five spin-½ centres have no spin below one half. A ground of five spin-½ centres can be several doublets at once, and six of the twenty are, but it can never hold a doublet together with a state of lower spin, because there is none. In the six-centre census such grounds are not rare: five of twenty-one triplets have one. A refinement fitted on the doublets met degenerate grounds only in the form a population handles correctly, several multiplets of one spin, and never in the form it cannot.

Two clusters with one difference

The cleanest comparison the census offers holds everything the refinement can see and varies only the ground.

Same raising gap, same raised states, and a singlet in one ground: the two separations differ by a fifth. The lowest levels of two six-centre clusters, in units of the coupling, by spin and number of states. Both have their first quintets four coupling units up and ten raised states there; one has a triplet ground and the other a triplet degenerate with a singlet. Their upper separations over population coupling are 0.320 and 0.384; the raising gap alone gives them products of 401 and 455 cm⁻¹.
Fig. 5 The lowest levels of two six-centre clusters with the same raising gap and the same raised states, one with a triplet ground and one with a triplet and a singlet.

Clusters thirteen and forty-seven both have their first quintets four coupling units up, two quintets each, ten raised states. Their excited structure below that differs, singlets in one and triplets with singlets in the other, but nothing in that structure is a raised spin, so the population coupling counts them both by the same rule. The difference the refinement does see is the ground: three states in one, four in the other. That makes the mixed cluster’s raised fraction slightly smaller at any coupling, its population coupling smaller, and its ratio larger.

The result is that their population ratios, 0.320 and 0.384, are 20 per cent apart, where their gap products, 401 and 455 cm⁻¹, were 13 per cent apart. So counting the extra ground state moved them in the wrong direction. A population that dilutes the raised fraction with a state carrying no moment is counting something the fit treats quite differently, because to a susceptibility a singlet is not one more state of the ground but a piece of the sample that does not respond.

Is it the singlets above the ground?

Every triplet in the census has singlets somewhere in its spectrum, and in most of them the first excitation is a singlet. So the remainder might belong to the spin-lowering states as a class, and the mixed grounds would just be the extreme case, singlets at zero energy rather than a fraction of a coupling unit up. That is testable, because a population of lowered spins can be added to the refinement with its own weight: find the coupling at which the raised fraction minus ww times the lowered fraction reaches pp, and fit ww and pp together on the triplets themselves.

The answer is that the best weight is zero. No mixture of a lowered population helps the triplets, even fitted on the triplets, which is the most generous test there could be. The singlets above the ground are not what separates the clusters. The singlets in the ground are, and a population with a weight cannot express “in the ground” as anything but a lowered state at energy zero, which it then weighs like all the others.

The numbers, design by design

Every design: the fitted fraction and the spreads it leaves. For each design, the fraction of the population fitted on its doublets, the doublets' and the triplets' spread under the raising gap and under the population, and the triplets split by their ground: the fifteen with one triplet and the five with a singlet beside it. The fraction moves with the design from 0.0062 to 0.092. The five mixed grounds among themselves spread by 1.29 to 1.39, about as little as the doublets.
Fig. 6 For each design, the fitted fraction, the doublets’ and triplets’ spreads under the gap and under the population, and the triplets split by their ground.

Two columns of the table are worth reading together. The fifteen triplets with a single triplet ground spread by 1.51 to 2.00 under the population, depending on the design: more than the doublets, but most of their excess is the one cluster, number two, whose raising quintet is its first excitation. The doublet with the same property, number two of the five-centre graphs, is also at the top of its group. The five mixed grounds, among themselves, spread by only 1.29 to 1.39 under every design, about as little as the doublets do. They are a tight group of their own with a different constant. So the triplets are better described as two populations than as one population with a broad spread, and the line between them is a fact about each cluster’s ground that can be read before any fit.

What a reader of a fit should take from it

The constant was always meant to be used backwards. A paper reports a susceptibility fit with its impurity fraction and temperature-independent term, a reader sees the two nuisance parameters decouple near some coupling, and the constant turns that coupling into an estimate of the gap to the next spin. This essay changes what such an estimate needs. With the gap alone the cluster’s spectrum entered as one number. With the population it enters as a count of raised states, which is usually known from the cluster’s spin and symmetry, and the estimate tightens by a sixth on doublets.

What the triplets add is a condition on the ground. A cluster whose lowest level might hold a singlet beside its triplet, which in practice means any cluster whose couplings are equal or nearly so by symmetry, is outside the group the refinement was fitted on. The offset is about a fifth, in the direction of a larger separation, so a raising gap read from such a cluster with the doublets’ constant would come out about a sixth too small. This is the same lesson the essay on what is fitted drew from a chain read with a pair’s expression: a fit returns numbers that look sound and are not, unless what the model assumes about the spectrum matches the sample. And as the product a curve measures found for the free direction, it is a combination of parameters that is determined, never one parameter alone. Here the combination is a separation and a population, and the population has to be counted from the right ground.

How the comparison was made

The census is the one the earlier essays read: every connected graph on five and six vertices with one antiferromagnetic coupling on each edge, diagonalised exactly, the twenty with a doublet ground and the twenty-one with a triplet ground. Each design’s upper separations are the second sign change of the correlation between the impurity fraction and the temperature-independent term, as before, with the census’s rule that a product under half the median is set aside (one cluster in two of the noisy designs). Each level of the spectrum was split into its spin multiplets by counting states of each ∣Sz∣|S_z|, and the raised fraction at the window’s top temperature is a Boltzmann sum over all of them. JpJ_p is a bisection in the logarithm of the coupling. The fraction was fitted on a grid of sixty-one values in the logarithm and refined by golden section. The refusal is the doublets’ degenerate grounds: were their states miscounted, the gap, which ignores them, would beat the population on the doublets, and under no design does it.

Where the model stops

One coupling on every edge. Every degeneracy in the ground of these clusters is exact because every coupling is equal. In a real cluster with couplings that differ by a few per cent the triplet and singlet in a mixed ground split, and the singlet becomes a low-lying excitation. Whether a small splitting restores the ordinary behaviour, or whether the cluster carries the mixed ground’s offset until the splitting reaches some fraction of kTkT, is exactly the thing this census cannot say.

The fraction is fitted, not derived. It moves with the design by a factor of fifteen, and nothing here explains its value. A derivation would start from where the fit’s covariance changes sign and work back to a population. This is a refinement that works, not an account of why.

Forty-one clusters, two spins. Five mixed grounds make a group and an ordering by their number of singlets. They do not make a law.

A noiseless fit. As throughout, the separations are read from the covariance at an exact fit, with no noise added.

Who said what

Fitting magnetic susceptibilities to the Heisenberg spectrum of a cluster, with impurity and temperature-independent terms, is long-standing practice in molecular magnetism, and the Van Vleck formula is what turns a spectrum into a curve. Equal-coupling clusters are full of accidental degeneracies — a frustrated tetrahedron has two singlets at its ground — and the sign of the free combination already turned out to follow a count of spins rather than the geometry that seemed to explain it. How many parameters a curve is worth and the moment a fit invents are where the fit’s two nuisance parameters first became the subject. What is measured here is the population refinement’s transfer from doublets to triplets under ten designs, its failure on mixed grounds, and the null result for a lowered-spin term.

Still open: grounds that split, and a third spin

The obvious open question is what a split ground does. Give one edge of a mixed-ground cluster a coupling a few per cent different from the others, so that its singlet moves a little above or below its triplet, and follow the upper separation as the splitting grows. If the mixed grounds’ offset survives until the splitting is a sizeable fraction of kTkT at the window’s top, then real clusters near such a degeneracy carry it too, and the warning applies to them. If it disappears as soon as the degeneracy is broken, it is an artefact of equal couplings.

The nearer question is a third spin. Seven spin-½ centres can have quartet grounds, and a quartet ground can hold a doublet beside it at the same energy, which is the seven-centre version of a mixed ground. A refinement fitted on six-centre triplets that knew about mixed grounds could be tested there, on clusters it has not seen, and that would turn the ordering among five clusters here into a prediction.

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.

DegeneracyExchange couplingLeast-squaresMagnetic susceptibilityModel limitSpin stateTemperatureUnderdetermination