The run was one of thirty-five
Worth reading first: The disc was a band with holes in it · A plane spends the same reach on a disc.
A run of low sites in a host is a box for an electron, and one such box is a level in the gap. Two boxes near each other split their levels, and many make a band once the splittings are larger than the spacing of the levels they join. The chain’s version of that story was told in runs, because a run is the only shape six adjacent sites can take on a line. When the same story was told in a plane, the plane spent the chain’s reach on a disc, and then the disc turned out to be a band with holes in it: computed from the plane’s own Green’s function, two runs of six couple with a direction and a parity, thirty times more strongly along their row than stacked above it at the same distance.
That essay ended by asking whether a plane’s defects should be runs at all. In a square net with half its sites low, six adjacent low sites are as likely to form an L, a T or a block as a straight line. The level-and-coupling calculation accepts any set of low sites, so a census of the shapes is the same calculation with a different list of sites.
Six low sites in a plane are thirty-five defects, and the run is a middling one. Its level is in the middle of the range, its reach is in the middle of the range, and its strong preference for its own row is not a property of defects in a plane but of being six sites long and one wide.
Thirty-five shapes
A shape of six cells joined edge to edge is a hexomino. Counted as distinct up to the eight rotations and reflections of the square, there are thirty-five, a number known since polyominoes were first catalogued; counted in fixed orientations there are 216. They are generated here by growing every shape of five cells by one cell in every possible place and reducing the results under the square’s symmetries, and the count is checked against the known thirty-five before anything else is computed.
Each shape becomes a defect exactly as the run was: its six sites at energy zero in a host of sites at energy four, every site joined to its four neighbours by a unit hopping. The host’s Green’s function is known exactly for an infinite square net, as a single integral over two Bessel functions, and a set of perturbed sites binds a state at an energy exactly when the Green’s function restricted to those sites has an eigenvalue of , with the contrast. For six sites that is a six-by-six problem at each trial energy, with no finite net, no boundary and nothing large to diagonalise. For the run it reproduces the published levels to .
All but two of the thirty-five bind four levels below the host band, as the run does. The two that bind only three have the deepest shallowest levels of all, which is to say their fourth level has already crossed into the band. The level the coupling is about is the shallowest one, the one nearest the band edge, because its tail reaches furthest.
The block binds least deeply
Ordered by that depth, the shapes run from 0.044 to 1.102 below the band edge, a factor of twenty-five. The run sits at 0.239, eighteenth of thirty-five, almost exactly in the middle. The shallowest of all is the two-by-three block, and the next few are compact too: a three-by-three outline at 0.050, two four-by-two shapes at 0.070 and 0.098.
That ordering runs against a natural expectation. A compact cluster gives each of its sites more low neighbours, so one might expect it to hold its electron more tightly and its levels to sit deeper. Its deepest level does: the block’s lowest bound state is at 2.717 below the edge, against the run’s 2.298. Its shallowest level does the opposite, and the shallowest level is the one that couples.
Nothing simple predicts the depth
It is tempting to turn the block into a rule — compact clusters bind their top level shallowly, extended ones deeply — and the census does not support it. Across the thirty-five shapes, the rank correlation between the depth of the shallowest level and the number of bonds inside the cluster is 0.13; with the cluster’s radius of gyration it is −0.15; with the area of its bounding box, 0.07. None of those is a relation. The twenty-seven shapes with five internal bonds, the fewest a connected six-site cluster can have, have shallowest levels anywhere from 0.050 to 0.818. The seven with six internal bonds range from 0.070 to 1.102. The block, with seven, is alone at 0.044.
The deepest levels, by contrast, are bunched: every shape’s lowest bound state lies between 2.298 and 2.717 below the edge, a range of a fifth, while the shallowest levels span a factor of twenty-five. A plausible reading is that the lowest state of a box is nodeless and depends mostly on how many low sites there are, which is six for every shape. The shallowest bound state is the most nodal one, and where its nodes can go depends on the shape’s whole connectivity in a way no single count captures. Two shapes with the same number of internal bonds and the same bounding box can put their top level at very different depths; presumably one admits a nodal pattern the other does not, though that is a reading rather than something computed here.
That has a practical consequence for any census that wants to treat clusters by type. A cluster’s reach is set by the depth of its shallowest level, as the next section shows, and that depth has to be computed for the shape; it cannot be read from how compact the shape looks. For six sites the computation is a six-by-six problem and costs nothing, so the honest course is simply to compute it for every shape, which is what the figure at the top does.
The depth sets the rate
How far a level’s coupling reaches is a question about its tail, and the tail of a bound state outside the cluster is the host’s Green’s function at the level’s energy. In a square net near its band edge, the host’s dispersion is quadratic, so a state at depth below the edge decays as at long range. That predicts a decay length of for every cluster, whatever its shape.
It holds. The splitting of two copies of each shape is computed at separations of 10, 16, 22 and 28 sites along the row and along the column, and a straight line through its logarithm gives a decay length. Every one of the seventy fitted lengths is between 0.72 and 1.12 of , and most sit a little below it. That is expected: in two dimensions the Green’s function at long range is a modified Bessel function, , which falls as , and the square root in the denominator makes a straight-line fit over a finite window read a decay length slightly shorter than the exponential’s own. The shallowest shapes, whose windows span the fewest decay lengths, show it most. The block’s decay lengths are 4.05 and 4.32 sites against a predicted 4.75; the run’s are 1.88 and 1.83 against 2.05; the most deeply bound shape’s 0.95 and 0.96 against 0.95.
So the thirty-five shapes do not need thirty-five coupling laws. The rate at which a cluster’s coupling falls is set by one number, the depth of its shallowest level, and the shape enters only through that depth and through a prefactor.
Ten decades at twenty-eight sites
A rate set by depth turns modest differences in depth into large differences in reach. The block loses about a decade of coupling for every ten sites of separation; the most deeply bound shape loses a decade every two. Between ten sites and twenty-eight the block falls from to , the run along its row from to , and the deepest shape from to . At ten sites the run along its row is the stronger of the first two, because its prefactor is larger; by sixteen the block’s slower decay has carried it past, and the gap then widens by a decade every six or seven sites.
At twenty-eight sites the thirty-five clusters’ couplings span ten decades, from to . The block is at the top in both directions. Nothing about the run’s own position is special: its level at 0.239 below the edge gives a decay length of 2.05 sites by the same law, and its couplings fall at that rate in both directions. The run is in the middle of the cloud along its row and among the weaker shapes across it, at and . A census that counted every six-site cluster as a run would therefore have given every cluster the run’s middling reach. At twenty-eight sites seven of the thirty-five reach more than ten times further than the run along its row in their stronger direction, the block more than a hundred times, and fourteen reach less than a hundredth as far.
Direction is elongation
The finding that made the plane’s runs look strange was their direction: a twin in the same row coupled thirty times as strongly as one stacked directly above at the same distance. The census says where that comes from.
The direction follows the shape’s elongation, loosely. The run, six wide and one high, couples forty-three times as strongly along its row as across it at twenty-eight sites. The five shapes two high and five wide prefer their long axis by factors from 1.1 to 48, so elongation allows a strong preference without forcing one; shapes as wide as they are high couple within a factor of a few either way, and a shape symmetric under reflection in a diagonal couples exactly equally, since that reflection exchanges the row and the column. Turned through a right angle a shape swaps its two numbers. A run’s strong direction is therefore not a property of the plane — the host’s Green’s function decays at the same rate in every direction at long range, since its dispersion near the band edge is round — but of the shape’s shallowest level and where its amplitude sits; the census does not decompose that further.
How each cluster was solved
The host is an infinite square net of sites at energy four with unit hopping to each neighbour; its lattice Green’s function at energy below the band edge is an integral over a product of two modified Bessel functions, evaluated on a quadrature in the logarithm of the time variable, which is exponentially accurate for an integrand this smooth. A cluster’s sites are set to zero, and its levels are the energies at which the Green’s function restricted to its six sites, times the contrast, has an eigenvalue of : each is found by scanning for a bracket and closing it by regula falsi on the logarithm. A pair of clusters is the same problem on twelve sites, and the splitting of the shallowest level is the difference between the two roots that come from it.
The checks run where the figures are drawn. The enumeration must produce thirty-five free shapes. The run must bind exactly the levels of the run in the plane, to . And the refusal is a symmetry: a second copy displaced by a vector and a second copy displaced by its negative are the same pair of clusters, so their splittings must agree to for every shape tested. A census that read a direction into a mirror image would be reading its own arithmetic, and this one does not.
A clean host is not a half-low net
The host is clean. Every coupling here passes through sites at the host’s energy, and in a net where half the sites are low there is no clean host between clusters; the next cluster’s tail passes through other clusters on the way. That was the first question the disc essay left open and it is still open: a disordered host has no closed-form Green’s function, though the same restricted problem can be solved with one computed numerically.
Thirty-five shapes, one size. Every cluster here has six sites. A real half-low net has clusters of every size, and the census of runs counted runs of every length. The shapes of each size could be counted the same way — the number of each hexomino per site in a random net is a known polynomial in the low fraction — but that census has not been run here.
One orientation per pair. Both copies of a shape point the same way. In a real net a copy turned through a right angle, or reflected, is as likely, and while its levels are the same its coupling to the first copy is not; for an elongated shape the cross-oriented pair should sit between the row and column numbers above, but it has not been computed.
One contrast. The depths, and so the reaches, depend on the contrast of four. A weaker contrast would push more shapes’ shallowest levels into the band and leave fewer bound at all; which shapes lose a level first has not been computed.
A shape is a depth and a direction
The plane’s runs had looked as if they needed a coupling law with a direction and a parity in it, and the earlier essays built that law carefully. The census separates the two things the law was carrying. The rate at which coupling falls with distance belongs to the depth of a level, and is the host’s, the same in every direction. The direction belongs to the shape, and is a statement about where a level’s amplitude sits inside the cluster. A run happens to be the extreme case of the second and an ordinary case of the first, which is why it made direction look like the main thing.
The more general lesson is the one the share that was read as a line has made before in another form: a quantity measured on one member of a family and carried to the rest carries that member’s peculiarities with it. Here the member was the only one a chain could offer, and the plane has thirty-four others.
Still open: how many of each, and the reach of the census
The obvious open question is the count. In a net with a fraction of its sites low, each fixed hexomino occurs as a maximal cluster with a probability that is a polynomial in , set by the shape’s size and the number of sites on its boundary. Combining those probabilities with each shape’s coupling would give the plane’s defect band for clusters of six as a whole, and since the shallowest shapes are the compact ones, which have the fewest boundary sites to be high, they may be the commoner ones too — which would move the half-way length of the disc essay by more than the run’s own correction did.
The nearer question is mixed pairs. Every pair here is a shape and a copy of itself, whose shallowest levels are degenerate. Two different shapes have different levels and couple only through the difference, which at the depths above is often larger than the coupling itself; whether such pairs resonate at all, or only the matched ones do, decides whether a census of thirty-five shapes is one population or thirty-five.
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.
- A band that is a hundred and seventy decades of nothing — both name closed form, defect state, disorder, model limit, tight-binding models
- A count rather than an average — both name closed form, defect state, disorder, model limit, tight-binding models
- The length at which levels become a band — both name closed form, defect state, disorder, model limit, tight-binding models
- A particle in a box the alloy made — both name closed form, disorder, model limit, tight-binding models
- A band becomes a bell curve — both name closed form, model limit, tight-binding models
- A decay that keeps slowing down — both name closed form, model limit, tight-binding models
Named objects
A dashed tag is an object no other essay names yet.
Closed formDefect stateDimensionalityDisorderModel limitTight-binding models