Bonding models

One number sets both prices

A ring current induced in one ring of a fused chain reaches another ring at a price per passage: one share kept across each linear ring, and across each angular ring a share that falls by a factor for every linear neighbour it has. From seven to thirteen rings the linear share rose by 1.21 and the factor fell by 1.22. Tied by one constant — the factor equal to 0.3821 divided by the linear share — the joint fit over 7,759 pairs loses one part in ten thousand of its accuracy, and at fifteen rings, computed afterwards on 104 more chains, the tie predicts the factor to 1.4 per cent where the factor's own trend misses by 3.6. Inside a run the company reaches further than one ring: an angular ring two rings from the nearest linear one keeps 0.33 of the response, one three or more rings deep keeps 0.45, against 0.57 for a linear ring.

Worth reading first: An angular ring is priced by its neighbours · The angular factor is a ratio of two prices.

A magnetic flux threaded through one ring of a chain of fused benzene rings changes the π current in every other ring — it does not divide between them the way a current divides between resistors — and how much depends on what lies between. In Hückel’s model the response of one ring to a flux through another is a second derivative of the π energy, computable exactly for any chain, and this sequence has been building a law for it. The essay that gave each passage its own price wrote the logarithm of a pair’s response as a sum: a constant, a log share for every linear ring between the two, a log share for every angular ring between them, and a factor for each angular ring the pair stands on. An angular ring is priced by its neighbours then split the angular share by the ring’s company: an angular ring between two linear ones keeps far less than one inside a run of angular rings, and the three classes fit as one share times a factor for each linear neighbour.

That model had two numbers that moved with the chain’s length. The linear share rose from 0.473 at seven rings to 0.571 at thirteen, as a longer linear stretch passes a current on more easily. The factor per linear neighbour fell, from 0.815 to 0.666. The essay noted that the rise was a factor of 1.21 and the fall a factor of 1.22, and asked whether the two were one. If the factor is a constant divided by the linear share, then a linear neighbour costs an angular ring exactly what the linear passage beside it gains, and the whole model has one length-dependent number.

It nearly does, and fifteen rings, computed afterwards, says how nearly.

The product of the two

The direct test of the tie is the product of the two numbers at each length. If the factor is KK divided by the linear share, the product is KK, the same at every length.

The linear price rises and the factor per linear neighbour falls, and their product barely moves from seven rings to fifteen. At each chain length the share of a pair's ring-current response kept across a linear ring, the factor an angular ring's price is multiplied by for each linear neighbour, and their product: 7 rings: 0.4732, 0.8148, 0.3856; 9 rings: 0.5129, 0.7525, 0.3860; 11 rings: 0.5447, 0.7048, 0.3839; 13 rings: 0.5712, 0.6664, 0.3807; 15 rings: 0.5941, 0.6342, 0.3768. Fifteen rings was computed after the tie was fitted on seven to thirteen; the tie's constant from those is 0.3821.
Fig. 1 The linear share, the factor per linear neighbour and their product at each chain length from seven to fifteen rings, with the tie’s constant fitted on seven to thirteen dashed.

At seven, nine, eleven and thirteen rings the product is 0.3856, 0.3860, 0.3839 and 0.3807, while the two numbers it is made of move by a fifth in opposite directions. Over the same lengths the linear share rises by 21 per cent and the factor falls by 22, and the product changes by 1.3 per cent. At fifteen rings — a design of 104 chains whose responses were computed only after the tie had been fitted, at about twenty seconds a chain — the linear share is 0.5941, the factor 0.6342 and the product 0.3768.

The product is not exactly constant. It sits within a per cent of 0.385 for the first three lengths and then drifts down, by 0.8 per cent to thirteen rings and a further 1.0 per cent to fifteen. A single number accounts for almost all of the two length dependences, and the remainder is a slow drift with its own sign.

A fit with one number less per length

A product read off separate fits is one test; a joint fit is a stricter one, because it asks whether the tie survives being imposed on every pair at once rather than read from the fitted numbers afterwards. The free joint fit has five numbers for each length — a constant, the linear share, the no-neighbour angular share, the factor per linear neighbour and the factor for standing on an angular ring — and is exactly the separate fits side by side. The tied fit replaces each length’s factor with one shared constant divided by that length’s linear share.

Tying the factor to the linear price costs the joint fit almost nothing: three parts in ten thousand of its error. Joint fits of the log ring-current responses over every pair at each set of lengths: free, with five numbers per length, and tied, with the factor per linear neighbour set to one constant divided by that length's linear price. seven to thirteen: 7759 pairs, R² 0.96283 free and 0.96283 tied, error 0.011 per cent larger, constant 0.3821; seven to fifteen: 15767 pairs, R² 0.96502 free and 0.96501 tied, error 0.026 per cent larger, constant 0.3792.
Fig. 2 Joint fits of every pair’s log response over seven to thirteen rings and over seven to fifteen, free and tied.

Over seven to thirteen rings the tied fit’s R2R^2 is 0.96283, the free fit’s 0.96283: the sum of squared residuals is larger by one part in ten thousand. Over seven to fifteen, with 8,008 more pairs, it is larger by three parts in ten thousand, and the shared constant moves from 0.3821 to 0.3792. A model with one length-dependent number fits nearly 16,000 pairs essentially as well as a model with two. The three parts in ten thousand are the drift seen above, and they grow when fifteen rings joins the fit, as a drift should.

The tie also moves very little else. Fitted with one shared constant over seven to fifteen rings, each length’s linear share comes back within half a per cent of its free value — 0.4712 against 0.4732 at seven rings, 0.5942 against 0.5941 at fifteen — and each factor within one and a half per cent: 0.8046 against 0.8148 at seven, 0.6381 against 0.6342 at fifteen. The largest adjustments are at the two ends of the range, as they must be when a constant is asked to describe a quantity drifting slowly through it: the tie pulls the shortest chain’s factor down and the longest chain’s up, by about a per cent each.

That small price is worth setting against what the tie buys. The model now has a constant per length, a linear share per length, an angular share per length, a stand-on factor per length, and one number for the whole family that turns the linear share into the company factor. The quantity the scatter first counted as a single factor per angular ring, and that one turned fusion was found to change throughout a chain, is now a derived quantity rather than a fitted one.

A prediction made before the measurement

A tie fitted on four lengths could describe those four and nothing else. Fifteen rings was the test it was built to fail.

Fitted on seven to thirteen rings, the tie predicts fifteen rings' factor to a per cent and a half; the factor's own trend misses by more than three. The factor per linear neighbour at seven to thirteen rings, and at fifteen: measured 0.6342; predicted by the tie from fifteen rings' own linear price (0.5941) as 0.382 ÷ 0.5941 = 0.6432, 1.4 per cent off; predicted by continuing the straight-line trend in length through the four fitted lengths, 0.6115, -3.6 per cent off.
Fig. 3 The factor per linear neighbour at seven to thirteen rings, and at fifteen: measured, predicted by the tie from fifteen rings’ own linear share, and predicted by continuing the factor’s straight-line trend.

The tie’s constant from seven to thirteen rings is 0.3821. Fifteen rings’ own linear share is 0.5941, so the tie predicts its factor as 0.3821 ÷ 0.5941 = 0.6432. The measured factor is 0.6342: the prediction is 1.4 per cent high. The plainest rival prediction uses no physics at all — continue the straight line the factor followed from seven to thirteen rings — and gives 0.6115, 3.6 per cent low. The tie, which needs fifteen rings’ linear share and nothing else, is two and a half times closer than the trend, and on the other side of the measurement.

That is a modest success and should be read as one. A per cent and a half is the size of the drift in the product, and the prediction is high because the product has fallen. What the tie captures is that the factor’s fall is not its own: it is the linear share’s rise, read through one constant.

Why the two should be one

The reading the earlier essay offered for the length dependence gives a reason for the tie as well. A reach that has no length found the acenes’ ring-current decay set by their gap, which closes as a linear stretch lengthens; the linear share rises with length for that reason. An angular ring between two linear ones is a break in the chain’s character, and the more easily the linear stretch on either side carries a current, the sharper that break is in relative terms. If the cost of a linear neighbour to an angular ring is exactly the ease of the linear passage it interrupts, the factor is the reciprocal of the linear share up to a constant.

The constant itself, 0.38 or so, is then the share an angular ring would keep if its linear neighbours passed a current perfectly — a number about the angular ring alone. Whether that reading is more than a story the numbers permit is not something the fit can settle, and the drift says it is not exact.

Company deeper than one ring

The second open question was the run. The no-neighbour class lumped together every angular ring whose two neighbours are angular, and at thirteen rings most of those were the zigzag’s interior rings, which sit several rings deep inside an angular run. Runs of three, the only other source, contribute one ring each, whose angular neighbours themselves have linear neighbours. Runs of four and five angular rings at every position were added to the designs at eleven and thirteen rings — 11 and 15 more chains, 65 and 92 in all — and each no-neighbour angular ring was classed by its depth: its distance along the chain to the nearest linear ring.

Deep in a run, an angular ring keeps nearly as much as a linear one; one ring in from the run's edge it keeps a third less. The share of a pair's response kept across each kind of passage at eleven and thirteen rings, with runs of four and five angular rings added to the design and the no-neighbour angular class split by the ring's distance to the nearest linear ring: 11 rings: linear 0.537, depth three or more 0.451, depth two 0.333, one linear neighbour 0.314, two 0.209; 13 rings: linear 0.565, depth three or more 0.454, depth two 0.335, one linear neighbour 0.307, two 0.193.
Fig. 4 The share of a pair’s response kept across each kind of passage at eleven and thirteen rings, with the no-neighbour angular class split by depth.

The two depths are far apart. At thirteen rings an angular ring at depth two — both neighbours angular, but each of them next to a linear ring — keeps 0.335 of the response; one at depth three or more keeps 0.454. The linear share is 0.565, one linear neighbour gives 0.307 and two give 0.193. Deep in a run an angular ring keeps four fifths of what a linear ring keeps; one ring in from the run’s edge it keeps less than three fifths. At eleven rings the numbers are within a per cent of those.

The deep class is not the zigzag in disguise. At thirteen rings the zigzag contributes most deep passages, from its seven interior rings at depth three or more, but 97 of the 194 deep passages between pairs lie in other chains — every one of them the middle ring of a run of five, since a run of four has no ring three deep. At eleven rings it is 48 of 96. A price fitted half on chains that are angular everywhere and half on chains that are mostly linear, and still coming out at four fifths of the linear share, is a property of depth rather than of one molecule.

So company reaches past the nearest ring. An angular ring’s price depends on its neighbours’ neighbours, and the no-neighbour price the earlier model fitted — 0.443 at thirteen rings, 0.384 once the runs are in the design — was an average over depths, weighted towards the zigzag’s deep rings. The prices form a scale in depth, from two linear neighbours through one, depth two and depth three, with each step keeping more.

What the depth scale means for a whole passage is easiest to see on a run of five between two linear stretches at thirteen rings. Its five rings are, in order, an angular ring with one linear neighbour, one at depth two, one at depth three, one at depth two and one with one linear neighbour. A pair on either side of it keeps, across the run, 0.307×0.335×0.454×0.335×0.307=0.00480.307 \times 0.335 \times 0.454 \times 0.335 \times 0.307 = 0.0048 of its response, against 0.5655=0.0580.565^5 = 0.058 across five linear rings: a twelvefold difference, of which the run’s middle ring accounts for less than its edges. The single no-neighbour price would have charged the three inner rings alike, 0.384 each, and put the run at 0.0053 — ten per cent too generous, the error concentrated in the rings one step in from the edges.

The gain from splitting by depth belongs to which rings are deep: three shuffles of the same counts buy almost none of it. The gain in R² of the log responses from splitting the no-neighbour angular class by depth, and the gain from the same two counts dealt to the wrong pairs among pairs with the same no-neighbour count, under three shuffles: 11 rings: 0.00163 against 0.000025, 0.000027, 0.000019; 13 rings: 0.00140 against 0.000000, 0.000012, 0.000001.
Fig. 5 The gain in fit from splitting the no-neighbour class by depth, against the gain from the same depth counts shuffled among pairs with the same no-neighbour count.

The split buys 0.0016 of R2R^2 at eleven rings and 0.0014 at thirteen, out of the four per cent the model leaves unexplained. Three shuffles of the depth counts among pairs that share their total of no-neighbour rings buy under two per cent of that gain, so it belongs to which rings are deep and not to an extra parameter.

The runs, at eleven and thirteen rings. For the design with runs of four and five angular rings added: 11 rings: chains · pairs 65 · 2,600, passages: depth two · three or more 278 · 96, deep passages outside the zigzag 48, no-neighbour price, unsplit 0.376, depth two · three or more 0.333 · 0.451, linear price 0.537, R², three classes · four 0.95571 · 0.95734; 13 rings: chains · pairs 92 · 5,244, passages: depth two · three or more 533 · 194, deep passages outside the zigzag 97, no-neighbour price, unsplit 0.384, depth two · three or more 0.335 · 0.454, linear price 0.565, R², three classes · four 0.95861 · 0.96001.
Fig. 6 The designs with runs added: chains and pairs, passages at each depth, the share at each depth and the fits.

How the prices were fitted

Every response is the second cross-derivative of the Hückel π energy with respect to magnetic fluxes through two rings, at half filling with uniform hopping, on chains built from their lists of angular rings with the angular rings read back from the geometry. The designs are those of the earlier essays — every chain with at most two angular rings anywhere, every run of three, the straight chain and the zigzag — at seven to fifteen rings, 104 chains and 8,008 pairs at fifteen; the runs of four and five add 26 chains at eleven and thirteen. The log responses are fitted by ordinary least squares on the counts of each kind of passage and of angular rings stood on, and each share is the exponential of its coefficient.

The free joint fit reproduces each length’s published linear share and factor to 10−910^{-9}. The tied fit is nested in it, so its error is no smaller. The refusal was set before fifteen rings was computed: the tie fitted on seven to thirteen must predict fifteen rings’ factor to two per cent and more closely than the factor’s own trend. It does both.

What a least-squares price on a Hückel chain cannot say

Equal hopping everywhere. The responses are those of a chain with no bond alternation, and in a real phenacene the angular rings’ outer bonds are short and their inner bonds long. That alternation would change an angular ring’s passage in the direction its company does, and the separation of company from alternation is not one this model can make.

A fitted price is an average. Each share is fitted to logarithms with equal weight on every pair, so pairs far apart, whose responses are tiny, count as much as near ones; the depth split and the tie are both statements about that weighting.

Fifteen is not a long lever. The tie was tested one length beyond its fit, and its constant moved by 0.8 per cent when that length joined. A drift that continued at its present rate would put the product near 0.37 at twenty rings; whether it continues, levels off or turns is a question for seventeen and nineteen rings, which at forty seconds and more a chain are not computed here.

Runs only at two lengths, and depth only to three. The runs of four and five were added at eleven and thirteen rings, not at seven or nine, where they barely fit, nor at fifteen, where the design was already the expensive one. So the depth scale has no length dependence of its own measured yet, and whether the deep price rises with length like the old no-neighbour price or holds still is open. Depth beyond three needs runs of seven or more, which were not built; the zigzag’s interior supplies deeper rings, but only in one molecule, so the deep class here lumps depth three with everything beyond it.

The length was the linear stretch’s

The angular factor belongs to the chain found the old angular factor falling with length and could only report it. Two prices for a passage showed it was a ratio of two prices; the company essay split one of them by its neighbours; and now the length dependence of that split turns out to be borrowed from the other. What changes with a chain’s length is how easily its linear stretches pass a current, and everything else follows it through one constant — up to a drift of a per cent per two rings that this calculation can measure and cannot yet explain.

Still open: the drift, and a run with linear rings on one side only

The obvious open question is the drift. The product of the factor and the linear share falls by about a per cent for every two rings from eleven onward, and a constant tie cannot carry it. Two readings predict different things at seventeen and nineteen rings: if the drift is the end effect — the same company priced differently near a chain’s end — it should shrink as the chain grows and the ends matter less; if it belongs to the linear share’s own approach to its long-chain limit, it should track the linear share’s increments, which are shrinking too, but at a different rate.

The nearer question is the depth scale’s direction. Depth here was the distance to the nearest linear ring on either side, so a ring at the end of a long run next to the chain’s terminal ring and one in the middle of a run between two linear stretches can have the same depth. Splitting depth into the distances on each side would say whether a run’s company is felt from both ends at once or only from the nearer one.

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.

AromaticityBand gapHückel theoryLeast-squaresRing current