The angular factor is a ratio of two prices
Worth reading first: The angular factor belongs to the chain · The scatter counts angular rings.
A magnetic flux through one ring of a fused chain of benzene rings changes the current in every other, and the size of the response between two rings falls with the number of fusions between them. The scatter counts angular rings found what most of the scatter at a fixed separation is made of: each angular ring between the pair multiplies the response by a factor, and each angular ring the pair stands on by another. The angular factor belongs to the chain then refitted that model on every chain of seven, nine and eleven rings and found that its form transfers exactly — the same 93.9 per cent of the variation explained at all three lengths — while its factors do not: the factor between a pair runs 0.656, 0.593 and 0.542.
That essay offered a reading of the drift and named the test it had not run. The model measures an angular ring against a free constant for each separation class, and those constants are set mostly by pairs joined through linear fusions. A straight chain’s response decays more slowly the longer the chain is, because its gap is closing, so a factor measured against linear passages would fall even if an angular ring’s own price never changed. The test: give the model one decay per linear passage and one per angular passage instead, and see which of the two carries the length.
It has now been run, at seven, nine, eleven and thirteen rings. The old factor is exactly the ratio of the two prices, and both of them move.
A price for each kind of passage
Between a pair of rings in a fused chain lie the rings strictly between them, and each is either linear — the chain continues straight through it — or angular. Call each a passage. The new model writes the logarithm of a pair’s response as one constant, plus a log price for every linear passage, plus a log price for every angular passage, plus a log factor for every angular ring the pair stands on. The separation is no longer a free class: it is simply the number of passages plus one. Four parameters replace the four, five or six separation constants of the old model.
The chains are the same design at every length: every chain with at most two angular rings, wherever they are, every run of three adjacent angular rings, the straight chain and the zigzag. That is 20, 35, 54 and 77 chains, and 300, 910, 2,160 and 4,389 pairs. Each chain is built from its list of angular rings, the angular rings are read back from the built geometry, and every pair is counted from that geometry rather than the list. Each pair’s response is the second cross-derivative of the Hückel π energy with respect to fluxes through its two rings, and pairs more than half the chain apart are left out, as in every earlier measurement.
The two-price model does the old model’s work. It explains 0.9375, 0.9382, 0.9378 and 0.9370 of the variation at the four lengths, against 0.9379, 0.9392, 0.9393 and 0.9391 for a free constant per separation. The difference is under a five-hundredth everywhere. A model with a single price for every passage, linear or angular alike, falls well short and gets worse with length: 0.883, 0.866, 0.853 and 0.842. So the distinction between the two kinds of passage is what carries the scatter, and the separation constants were never doing more than counting passages at an average price.
The factor is a ratio
Divide the angular price by the linear price at each length and the old factor comes back: 0.6561 against the fitted 0.6559 at seven rings, 0.5931 against 0.5929 at nine, 0.5421 against 0.5418 at eleven, 0.4985 against 0.4985 at thirteen. The factor per angular ring stood on is equally stable between the two models, 0.681, 0.629, 0.586 and 0.549 in both.
That is the reading the earlier essay gave, made exact. An angular ring between a pair replaces what would have been a linear passage, so in a model whose separation constants already charge every passage the linear price, the angular ring’s extra factor is what it costs on top of that: its own price divided by the linear one. A factor per angular ring was never a property of the angular ring alone. It is a relative price, and a relative price can move when either of its parts moves.
Both prices move
The linear price rises steadily: a linear passage keeps 0.465 of the response at seven rings, 0.498 at nine, 0.527 at eleven and 0.551 at thirteen. That is the half of the reading that holds. The straight chain fitted alone shows the same thing more strongly, from 0.558 to 0.642, and the reason is the one a reach that has no length found for a heteroatom’s reach along an acene: a response falls at a rate set by the gap, and the acenes’ gap closes as the chain lengthens.
The angular price does not stay put. It falls from 0.305 at seven rings to 0.296, 0.285 and 0.275 at thirteen: an angular passage keeps a tenth less of the response in the longest chains than in the shortest. That half of the reading fails. The zigzag fitted alone, where every passage is angular, goes the other way, from 0.318 to 0.337, and its gap settles near 0.82. So an angular passage’s price inside a mixed chain is not the zigzag’s price; it depends on how much linear chain surrounds it, and it depends on it in the opposite sense from the linear passage’s own.
Because the factor is a ratio, its logarithm is a difference, and the drift splits exactly. From seven rings to thirteen the factor’s logarithm falls by 0.274. The linear price’s rise accounts for 0.171 of that, and the angular price’s fall for 0.104: three-fifths of the drift belongs to the chain’s linear passages, and two-fifths to the angular rings themselves. The earlier reading had the larger part right and missed the smaller one.
A linear passage’s price depends on how far it reaches
One price per linear passage is itself an average, and the straight chain shows over what. On thirteen rings, going from a pair one fusion apart to two keeps 0.537 of the response; from two to three, 0.611; then 0.653, 0.677 and 0.689. Each further fusion costs less than the one before, by 0.074, 0.043, 0.024 and 0.011 — a series that is closing in on a limit near 0.70. And at each separation the ratio is larger on a longer chain: the first step is 0.526 on seven rings and 0.537 on thirteen, the second 0.573 and 0.611. So the linear price the two-price model reports is a mixture of short, expensive passages and long, cheap ones, and as the chain lengthens the long ones are both cheaper and more numerous. That is how a closing gap reaches the fit.
Not settled at thirteen
The first question the earlier essay left was whether the factor settles. Thirteen rings say: not yet. It falls by 0.063 from seven rings to nine, 0.051 from nine to eleven and 0.043 from eleven to thirteen. The steps are shrinking by about a sixth each time, so if they continued to shrink that way the factor would settle somewhere near 0.30 — but that is an extrapolation over four points, and the acenes’ gap, which drives the linear price, closes towards zero rather than settling, so there is no guarantee the steps keep shrinking. What the decomposition adds is that even a settled linear price would leave the factor moving, because the angular price is falling too.
What the drift costs, carried to thirteen rings
The practical meaning of a drifting factor is the error in carrying it from one length to another, and thirteen rings make that error larger than it looked at eleven. Carried from nine rings, the factor 0.593 overstates a thirteen-ring pair spanning one angular ring by 19 per cent, a pair spanning three by 68 per cent and a pair spanning five by a factor of 2.4, because the factor enters as a power. The earlier refit put the same transfer from nine rings to eleven at 9, 31 and 57 per cent; two rings more nearly double each of them.
The two-price model does not remove that error, but it says where to look for a correction. Of the 0.274 by which the factor’s logarithm falls between seven rings and thirteen, the 0.171 that belongs to the linear passage is a property of the straight chain’s gap, which is known in closed form for a Hückel acene at any length. The 0.104 that belongs to the angular passage has no such handle, and it is the part a transfer rule would have to model. A factor quoted with its chain length is still the honest form, as the earlier essay concluded; a pair of prices quoted with their chain length is a better one, because one of the two can then be carried by a formula rather than by a refit.
The other factor, and the ends
The factor for each angular ring a pair stands on drifts too, from 0.681 at seven rings to 0.549 at thirteen, and in the two-price model it is a free parameter rather than a ratio, so nothing above explains it. It falls by almost exactly as much, in logarithm, as the factor between a pair: 0.216 against 0.274. One reading is that a pair standing on an angular ring has one fewer linear passage leading into it than a pair that does not, and so also pays the relative price; the two-price model does not test that, and the matching sizes may be a coincidence of four points.
The end of the chain, which the end effect with two signs found to matter for pairs that touch it, is worth almost nothing once passages are priced. Adding a factor for pairs touching an end changes the explained share by at most 0.0002 at any length, and the factor itself is between 0.958 and 1.004. In the class model the end term had been the supposed leading correction to the separation constants and was already shrinking with length; with prices in place of constants it has nothing left to correct.
Taken together, the four parameters of the two-price model account for the scatter as well as the five to eight of the class model, whose separation classes grow from three at seven rings to six at thirteen, and the single number the class model was built around — a factor per angular ring between a pair — is recovered as a quotient of two of them. What the extra separation constants were buying, at every length, was a way of not deciding how much a linear passage costs.
What was fitted, and how it was checked
Each fit is ordinary least squares on the natural logarithm of every pair’s response. The two-price model has a constant, a count of linear passages, a count of angular passages and a count of angular rings stood on; the class model has a constant for each separation class, the count of angular rings between and the count stood on, exactly as before. Each uniform chain is also fitted alone, the straight chain with its linear passages only and the zigzag with its angular passages and rings stood on. The thirteen-ring chains are the expensive part, about a quarter of an hour of flux derivatives, and they are computed once and kept.
The checks, run where the figures are drawn: the class model refitted here reproduces the published factors at seven, nine and eleven rings to , on the same pairs; on every pair the passages counted add up to the separation; the angular factor equals the ratio of the two prices to half a per cent at every length; and both parts of the drift are positive. The refusal is the uniform pair: fitted on the straight chain and the zigzag together, where each chain has only one kind of passage, the two-price model must return the straight chain’s own linear decay to within a few per cent, and does. A model that mixed the two kinds would fail there first.
Hückel prices
One electron. The prices are Hückel prices, quoted per passage as the share of the response kept, and they have no meaning outside the model that defines the response. The responses are second derivatives of a Hückel π energy, not computed magnetic shieldings, and the acenes’ closing gap is itself a Hückel feature that bond alternation and correlation soften. Aromaticity as a shell closure is the reminder of how far an electron-count picture is from a measurement. The structure found here — a factor that is a ratio of two prices, each of which depends on the chain — is what is likely to survive a better model, not the prices themselves.
Chains only, and few angular rings. Every chain is a path of rings with at most three angular rings. The angular price is measured mostly on chains where angular passages are rare, which is presumably why it differs from the zigzag’s own. A design with more angular rings per chain would sample the angular price in more crowded company, and the zigzag, the one chain in the design that is all angular, already suggests that the answer there is different.
An average over reach. The linear price is a single number standing for a decay that weakens with distance, as the straight chain shows; a model with a price per linear passage that depended on its distance from the pair’s nearer ring would be more faithful, and would have more parameters to fit.
A relative price is not a property of either good
The angular factor belonged to the chain because a factor fitted against free separation constants inherits whatever moves those constants. Writing the model with a price for each kind of passage removes the constants and shows what the factor always was: the price of an angular passage measured in units of a linear one. Both units drift, and the drift was never one thing. The same lesson ran through the end effect with two signs and the scatter that counts angular rings in other forms: a number that summarises a comparison carries both sides of the comparison with it, and changing the model so that each side has its own number is the cheapest way to find out which side moved.
Still open: a price that depends on reach, and the angular ring’s company
The obvious open question is the linear price’s own structure. The straight chain shows each further fusion costing less than the one before, converging near 0.70 on thirteen rings. A model giving a linear passage a price that depends on how many passages lie between it and the pair’s nearer ring — one parameter for the near limit and one for the far — would say whether the whole length dependence of the linear price is that convergence, in which case it would stop once chains are long enough for most passages to be far ones.
The nearer question is why the angular price falls in mixed chains while the zigzag’s rises. The two differ in how much linear chain surrounds each angular ring. Fitting the angular price separately for angular rings with linear neighbours on both sides, on one side and on neither would say whether an angular passage is cheaper or dearer according to its company, and whether the zigzag’s rising price and the mixed chains’ falling one are the two ends of a single dependence.
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.
- Neither of the two separations — both name aromaticity, band gap, hückel theory, least-squares, model limit, ring current
- One integer, and everything it changes — both name aromaticity, hückel theory, least-squares, model limit, ring current
- The floor was in the bookkeeping — both name band gap, hückel theory, least-squares, model limit
- The reach is the molecule's — both name band gap, hückel theory, least-squares, model limit
- A decay that keeps slowing down — both name band gap, least-squares, model limit
- A parameter that never finds a value — both name hückel theory, least-squares, model limit
Named objects
A dashed tag is an object no other essay names yet.
AromaticityBand gapHückel theoryLeast-squaresModel limitRing current