When the molecule does not stop

Two different defects rarely resonate

A net with half its sites low holds all thirty-five shapes of six adjacent low sites, and every pair measured so far was a shape and a copy of itself, whose shallowest levels start degenerate so that any coupling splits them. Two different shapes start apart. At twelve sites their levels differ by a median of 0.195, against a median coupling of a shape to its own copy of 3 × 10⁻⁴. Of the 595 pairs of different shapes, four share their states by more than a quarter, thirteen by more than a twentieth; at sixteen sites, two. A census of thirty-five shapes is thirty-five populations, with a handful of accidental twins.

Worth reading first: The run was one of thirty-five · The disc was a band with holes in it.

A square net with a fraction of its sites lowered in energy binds electrons on clusters of those sites, and when the clusters are close enough their levels couple and spread into a band. The run was one of thirty-five took the six-site clusters of such a net seriously: in a plane, six adjacent sites can take thirty-five shapes, every one of which a random net with half its sites low contains, and each binds its own levels at its own depths, the shallowest from 0.044 for a compact block to 1.10 for the deepest. It then measured how each shape’s shallowest level splits when a copy of the same shape sits a given distance away, and found decay lengths set by the level’s depth and directions set loosely by the shape.

Every pair in that census was a shape and itself. That is the natural first case and it has one property that makes it special: the two isolated levels are exactly degenerate, so any coupling at all mixes the two clusters’ states completely, half on each, and the splitting measures the coupling directly. Its continuation asked about the case a random net mostly holds instead. Two different shapes have different levels, and they can couple only through the difference. If that difference is usually larger than the coupling, the two clusters keep their states to themselves, and a census of thirty-five shapes is thirty-five separate populations rather than one.

It is, nearly.

Thirty-five depths

Thirty-five shapes, thirty-five depths: a handful of different shapes nearly share a level. The shallowest level of each of the thirty-five free hexominoes below the host band, on a logarithmic scale, from 0.044 to 1.102. Pairs of different shapes whose levels differ by less than 10⁻³ are marked: 10 and 19, 3.1×10⁻⁶; 6 and 30, 3.3×10⁻⁵; 4 and 16, 5.8×10⁻⁴; 8 and 30, 7.2×10⁻⁴; 6 and 8, 7.5×10⁻⁴; 8 and 11, 8.1×10⁻⁴; 1 and 32, 9.5×10⁻⁴.
Fig. 1 The shallowest level of each of the thirty-five free hexominoes below the host band, on a logarithmic scale, with the different shapes whose levels lie within 10⁻³ of each other marked.

The thirty-five shallowest levels lie between 0.044 and 1.10 host hopping units below the band edge, spread over more than a decade. Most are well separated from all the others: the median difference between two different shapes’ levels is 0.195. But they are not spread evenly, and a few sit almost on top of one another. Seven pairs of different shapes lie within 10⁻³ of each other, and they form four groups: four shapes near 0.099, two near 0.194, two near 0.394 and two near 0.505. The closest pair, shapes 10 and 19, differ by 3.1 × 10⁻⁶ in a level of 0.394, seven parts in a million.

No symmetry relates them. Shapes 10 and 19 are different trees — six sites joined by five bonds, with no square of four inside — one four sites wide and three high, the other three by three, and the net’s eight symmetries take neither into the other. All ten shapes in the near-degenerate pairs are trees, but so are 27 of the 35 hexominoes, so that says less than it seems to. These are accidental degeneracies: two clusters with different geometries happen to bind their shallowest levels at the same depth, to the precision the net’s Green’s function is computed to. A census of copies could not have found them, because it never set one shape’s level beside another’s.

Detuning against coupling

Two different shapes' levels are usually a thousand times further apart than any shape couples to itself at twelve sites. Top: the level difference of each of the 595 pairs of different shapes, median 0.195, smallest 3.1×10⁻⁶. Bottom: half the splitting of each shape's level with a copy of itself twelve sites away along the row, median 3.0×10⁻⁴, largest 3.8×10⁻³.
Fig. 2 The level difference of each pair of different shapes, against half the splitting of each shape’s level with a copy of itself twelve sites away.

A pair’s two states mix when the coupling between the clusters is comparable with the difference between their levels, and the two quantities can be set side by side. At twelve sites between centroids along the row, a shape couples to a copy of itself with half-splittings from 4 × 10⁻⁷ to 3.8 × 10⁻³ and a median of 3.0 × 10⁻⁴. Two different shapes’ levels differ by a median of 0.195 and by less than 10⁻³ for only seven pairs. For a typical pair the detuning is six hundred times the coupling. Unless two shapes are accidental twins, their states have nothing to resonate with.

The share on the other shape

The comparison above uses each shape’s coupling to itself as a stand-in for the coupling to a different shape, and the mixing can be measured directly instead. For each of the 595 pairs of different shapes, the second is placed with its centroid twelve sites from the first’s along the row, and the combined twelve-site cluster’s levels nearest the two isolated levels are found from the same condition as for one cluster, det⁡(1+δG0(ε))=0\det(1 + \delta G_0(\varepsilon)) = 0. At each level the bound state’s amplitudes on the twelve low sites are the vector that makes that determinant vanish, so the share of the state on the other shape can be read off.

At twelve sites only four of 595 mixed pairs share their states, and every one is detuned by under 10⁻³. Each pair of different shapes at twelve sites along the row: the larger of its two states' shares on the shape it did not come from, against the difference of the two shapes' levels. 4 pairs exceed a quarter, 13 a twentieth and 23 a hundredth. A pair of identical shapes would sit at one half.
Fig. 3 Each pair of different shapes at twelve sites: the larger of its two states’ shares on the shape it did not come from, against the difference of the two shapes’ levels, with a share of a quarter dashed.

Four of the 595 put more than a quarter of a state on the other shape, thirteen more than a twentieth and twenty-three more than a hundredth. The four are the accidental twins, all detuned by less than 10⁻³. Below them the share falls roughly as the square of coupling over detuning, as two-level mixing should, down to shares of 10⁻⁸ for pairs a whole hopping unit apart. A pair of identical shapes, read the same way, sits at exactly one half at every separation, which is where the published census lived.

The six pairs of different shapes that share their states most at twelve sites: chiral trees with nearly equal levels. The six pairs of different hexominoes whose states mix most at twelve sites, drawn side by side, with the difference of their levels and the share on the other shape: shapes 10 and 19, 3.1×10⁻⁶, 0.484; shapes 6 and 30, 3.3×10⁻⁵, 0.481; shapes 6 and 8, 7.5×10⁻⁴, 0.448; shapes 1 and 32, 9.5×10⁻⁴, 0.394; shapes 8 and 11, 8.1×10⁻⁴, 0.214; shapes 14 and 26, 4.8×10⁻³, 0.177.
Fig. 4 The six pairs of different shapes whose states mix most at twelve sites, drawn side by side, with their level difference and the share on the other shape.

The six most strongly mixed pairs show the pattern. Shapes 10 and 19, three millionths apart, share 0.484; shapes 6 and 30, 3.3 × 10⁻⁵ apart, share 0.481; shapes 6 and 8, 7.5 × 10⁻⁴ apart, share 0.448. Further down the same group near 0.099, shapes 8 and 11 are 8.1 × 10⁻⁴ apart and share only 0.214: their detuning is similar to 6 and 8’s, and the difference is the coupling, which depends on how the two shapes’ amplitudes reach towards each other across twelve sites. Shapes 14 and 26, sixth on the list, are already 4.8 × 10⁻³ apart and share 0.177 only because they couple unusually strongly.

Why a small detuning is enough to stop it

A pair of clusters whose shallowest levels sit at εA\varepsilon_A and εB\varepsilon_B and couple with strength tt is, near those levels, a two-level problem. Its two states are mixtures of the clusters’ own states, and the share of each on the other cluster is

sin⁡2θ=12(1−ΔΔ2+4t2),Δ=∣εA−εB∣.\sin^2\theta = \tfrac12\left(1 - \frac{\Delta}{\sqrt{\Delta^2 + 4t^2}}\right), \qquad \Delta = |\varepsilon_A - \varepsilon_B| .

With Δ = 0 the share is one half whatever tt is, which is why every pair in the census of copies was fully mixed and why only the splitting could be read from it. With Δ much larger than tt the share is about (t/Δ)2(t/\Delta)^2, so a detuning ten times the coupling leaves one per cent of a state on the other cluster and a detuning a hundred times leaves one in ten thousand. The measured shares follow that law closely enough that the twelve-site scatter falls along a line of slope about −2 in its logarithms. The point is how little detuning it takes. The couplings at twelve sites are of order 10⁻⁴, so two shapes whose levels differ by one thousandth of a hopping unit — a twentieth of a per cent of the depth of either — already share only a few per cent.

The near-degenerate pairs are therefore not merely close; they are close to the precision the couplings set, which at sixteen sites is a few millionths. Shapes 10 and 19 are three millionths apart, which is why they are the last pair to stop resonating as the separation grows.

What a resonant share has to count

The chain essays measured a defect band by the share of runs that sit in a resonant pair — a run close enough to another run of the same length to share a state with it — and found that the share grows slowly with the chain, so that the commonest runs reach a band only in a grain of matter. In a chain every run of six sites is the same shape, and every pair of runs of one length is a possible resonance.

In a plane that stops being true. A cluster of six sites has thirty-four other shapes it could find nearby, and against all but a few of them it cannot resonate at any distance a net would put it at. The resonant share in a plane has to be counted shape by shape: the chance that a cluster has a copy of its own shape, or one of its accidental twins, close enough. Since each shape is only a fraction of all six-site clusters, that chance is smaller than a count over all clusters would suggest, by roughly the shape’s own share of the population — and for a shape with no twin, exactly that.

Distance decides which twins resonate

The mixed pairs that share their states thin out with distance, because a coupling falls and a detuning does not. The number of the 595 pairs of different shapes whose states put more than a quarter, a twentieth and a hundredth of themselves on the other shape, at 10, 12, 14 and 16 sites between centroids: share over a hundredth 49, 23, 10, 8; share over a twentieth 25, 13, 7, 4; share over a quarter 10, 4, 3, 2.
Fig. 5 The number of the 595 pairs of different shapes whose states put more than a quarter, a twentieth and a hundredth of themselves on the other shape, at ten, twelve, fourteen and sixteen sites between centroids.

A coupling falls with distance — exponentially, with the decay length the reach of each shape was measured to have — and a detuning does not, so the resonant pairs thin out as the clusters move apart. At ten sites, ten pairs share more than a quarter and 49 more than a hundredth; at sixteen, two and eight. Each near-degenerate pair has its own distance at which its coupling falls below its detuning and it stops resonating.

The most strongly mixed pairs at every separation. For the pairs of different shapes that mix most at any separation, their level difference and the share on the other shape at 10, 12, 14 and 16 sites. 10-19: 0.499, 0.484, 0.430, 0.263; 6-8: 0.492, 0.448, 0.346, 0.201; 6-30: 0.480, 0.481, 0.482, 0.469; 1-32: 0.458, 0.394, 0.241, 0.075; 8-11: 0.396, 0.214, 0.088, 0.028; 4-16: 0.377, 0.110, 0.008, 0.000; 14-26: 0.299, 0.177, 0.076, 0.027.
Fig. 6 For the pairs of different shapes that mix most at any separation, their level difference and the share on the other shape at ten to sixteen sites.

Shapes 1 and 32, 9.5 × 10⁻⁴ apart, share 0.458 at ten sites, 0.394 at twelve, 0.241 at fourteen and 0.075 at sixteen. Shapes 4 and 16, 5.8 × 10⁻⁴ apart, share 0.377 at ten and less than a thousandth at sixteen. Shapes 6 and 30, a thirtieth as far apart, still share 0.469 at sixteen, and shapes 10 and 19, ten times closer still, are falling only because, at sixteen sites, even their tiny detuning has begun to exceed the coupling. In a dilute net, where clusters are far apart, only the closest twins mix; in a dense one, the groups of near-degenerate shapes become small bands of their own.

How the pairs were solved

The net is a square lattice with every site at energy 4 except the low ones at 0, and unit hopping between neighbours, the setting of the disc essay and its successors. A cluster’s bound levels below the band are the energies ε at which the matrix of the host’s Green’s function on the cluster’s sites, G0(ε)G_0(\varepsilon), has an eigenvalue of exactly −1/δ-1/\delta with δ = 4, found by regula falsi on the logarithm of the eigenvalue condition; the Green’s function is the host’s exact one, computed from its Bessel-function representation. A pair is two shapes, the second’s centroid placed a whole number of sites from the first’s along the row, and every level of the twelve-site cluster is found the same way. The two that matter are the levels nearest each isolated shallowest level — chosen by nearness rather than by order, because one shape’s second level can lie shallower than another shape’s first. The share of a state on a shape is the sum of squares of its amplitude vector on that shape’s six sites over the sum on all twelve.

The checks run where these figures are drawn. Every pair binds both levels at every separation. From twelve sites on, every pair is a two-level problem: the state nearest the first shape’s level has a share on the first shape within a tenth of one minus the other state’s. At ten sites a few pairs bring a third level in, and they are counted in the figures rather than removed. Off resonance, each state stays mostly on its own shape. A matched pair, the compact block with a copy of itself, read the same way, puts exactly half of each state on each copy. And the refusal: a chiral shape and its own mirror image are different placements of one free shape with the same levels, and they must come out sharing their states by more than three-tenths — which they do. If they had not, the share would be measuring something about placement rather than resonance.

What a pair of clusters cannot say about a net

Two clusters, not a net. A random net holds many clusters of every shape at once, and a cluster’s state can mix with a third cluster even when it does not mix with the second. The pair is the first step to a band; whether the accidental twins in a real net form small bands of their own, or are swamped by their many detuned neighbours, needs the clusters’ levels in a net with their real abundances.

Along the row only, at whole-number separations. Every pair is placed with one orientation of each shape and with the second shape’s centroid along the row. Turning either shape changes the coupling, often by a large factor as the published census found for copies, and would move individual pairs across the resonance line. The count of resonant pairs is a count for this placement.

Twelve low sites, and no host disorder. Each pair is computed in an otherwise perfect host, so the only difference between the two clusters’ surroundings is each other. In a real net the host sites between two clusters are themselves disordered, and a cluster’s level shifts with its surroundings by amounts that, for the shallow levels, could easily exceed the millionths that make two shapes twins.

The shallowest level only. Each shape binds up to four levels, and a shape’s deeper levels can be nearly degenerate with another shape’s shallower ones. At ten sites that is what brings a third level into a few pairs, and in a net with clusters of different sizes, deeper levels of large clusters will meet shallow levels of small ones.

A census is a list of populations

The sequence of essays that built this picture of a defect band — the length at which levels become a band, a count rather than an average, a plane spending its reach on a disc — treated the band as built from copies: clusters of a given size, coupling to their neighbours of the same kind. The thirty-five shapes made that assumption visible. Their levels differ by much more than any of them couples to anything, so a band formed from a random net’s six-site clusters is thirty-five sub-bands, one per shape, each formed only by copies of its own shape. A cluster of a different shape nearby is, to a cluster’s shallowest state, an obstacle rather than a partner.

It also sharpens a distinction the chain essays drew between where a band begins and how wide it is. A band’s width is set by the typical coupling, not by the closest pair, and in a plane of mixed shapes a cluster’s typical neighbour is a different shape, whose coupling to it is effectively divided by a detuning a few hundred times larger. The typical coupling that sets a mixed net’s band width is therefore not the copy census’s median of 3 × 10⁻⁴ but something closer to its square over 0.2 — a band narrower, for a given density of clusters, by the factor that separates resonance from its absence.

The exceptions are where the census earns its keep. Seven pairs of shapes are near enough in energy to resonate, unrelated by symmetry, and they would merge their sub-bands in a dense enough net. That is not something a census of copies could find; it needs every pair.

Still open: the abundances, and the twins in a net

The obvious open question is how many of each shape a random net holds. Each shape’s probability of occurring as a maximal cluster is a polynomial in the fraction of low sites, set by its size and the number of sites on its boundary, and those abundances decide which of the thirty-five sub-bands is wide enough to matter. With them, the defect band of a net at any filling becomes a sum of sub-bands with known weights, and the accidental twins become a correction whose size can be stated.

The nearer question is the twins themselves. Four groups of near-degenerate shapes, near 0.099, 0.194, 0.394 and 0.505, would each form a merged sub-band in a dense net. A net seeded with clusters of only the four shapes of the 0.099 group, at their natural abundances, would say whether their accidental degeneracy survives being surrounded by each other — or whether each cluster’s level shifts, in a crowd, by more than the few ten-thousandths that made them twins.

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.

Defect stateDimensionalityDisorderResonanceTight-binding models