The collection

Every essay — page 45

Page 45 of 45, continuing through the fields in the same order.

Orbitals Where the atoms go Bonding models What symmetry decides Beyond the octet What a spectrum settles When the molecule does not stop What the shape is for What is taught wrongly Series Named objects Orbitals Refutations Search

What is taught wrongly

The explanations that are confident, memorable and false — stated fairly and then tested against a calculation rather than an opinion.

The repulsion each bond needs, and the one Born and Mayer's law supplies. For each bond, the fraction of the ion pair's attraction a repulsion must reach before moving a whole electron stops paying (open circle), and the fraction a Born–Mayer repulsion of range 0.345 Å supplies at the measured length if the ion pair is to sit at its own minimum there, ρ/R (filled). Three bonds need none because they never transferred. Hydrogen fluoride needs 0.351 and is supplied 0.376; hydrogen chloride needs 0.116 and is supplied 0.271; the alkali halides need three quarters and are supplied a fifth or less. Every verdict comes out right — hydrogen fluoride's by a margin of 0.025.

A repulsion the proton does not have

Asked whether moving one whole electron pays, eight bonds with measured dipoles came out six right and two wrong: hydrogen fluoride and hydrogen chloride, both predicted ionic. The proposed fix was the repulsion between the two ions' closed shells. Add it the way an ion pair demands of itself, and the score becomes eight of eight — but the repulsion that does it is largest for the shortest bonds, and the shortest bonds are the ones whose cation is a bare proton with no shell at all. The term a proton really has, penetration into the anion's cloud, undoes hydrogen chloride and not hydrogen fluoride. And the measured bond energies demand a repulsion only for the three alkali halides.

6 figures
The covalent arrangement the inversion finds is Pauling's mean — except for hydrogen fluoride. For eight bonds, the covalent arrangement's binding found by solving the two-state problem exactly against the measured dipole and bond energy, against Pauling's covalent mean — the geometric mean of the two homonuclear bond energies — which the inversion never sees. The line is equality. HCl 0.97, HBr 0.98, HI 1.03, ClF 1.09 of the mean; HF 1.68. Open circles are the alkali halides, where the ionic weight is near 0.8 and the result is mostly the assumed repulsion.

Two measurements leave nothing to fit

A bond pictured as a mixture of its ionic and covalent arrangements has two unknowns — the coupling between them and the covalent arrangement's own energy — and a dipole fixes only one. Add the bond energy and the two-state problem inverts exactly, with no parameter left. The covalent arrangement it returns can then be held against something it never saw: Pauling's mean of the two homonuclear bond energies. For hydrogen chloride, bromide and iodide it lands on that mean to within three and a half per cent, and chlorine monofluoride within nine. Hydrogen fluoride lands 68 per cent above it. And the couplings rise and fall with length rather than falling — hydrogen fluoride's, on the shortest bond, is the smallest.

7 figures
The chain of four's strongest line is a hole's standing wave read against the singlet's ends. For the half-filled chain of four at infinite repulsion, each removal line's weight is a sum over the four sites of two factors. The first is the hole's standing wave squared on each site: the inner levels at ±0.62 put more of the hole on the end sites, the outer levels at ±1.62 more on the middle ones. The second belongs to the spin chain alone: removing an up electron at an end site leaves the stronger spin doublet with weight (2 + √3)/8 = 0.4665, at a middle site 1/4. Their sums are 0.4067 and 0.3098, which are 1/4 + √3/16 ± √15/80, and the limit of the contrast is their ratio, 1.312498.

The limit was two chains at once

At infinite repulsion every removal line of a half-filled chain is a hole's standing wave read against the spin chain the hole moves through, and every hole level carries exactly half an electron. On the chain of four that makes the contrast's limit (20 + 5√3 + √15)/(20 + 5√3 − √15) = 1.3124978 — not 21/16, which it misses in the sixth figure. On chains of eight and twelve, which no exact spectrum had reached, the limit is 1.0458 and 1.0195, and an energy that vanishes in the limit decides it.

6 figures
Where the correction comes from moves the worst case more than the start did. The polarisation difference's worst error ratio on the square's grid, with the correlation correction carried from eight references: the published one, three beside the band with no site energy, and the four systems where the two mean-field starts disagree. On the lowest mean field it runs from 9.7 to 213.2; the start alone moved the published reference's from 48.0 to 14.7. Rings are the conventional start; the dashed line is the census threshold of three.

The worst case measured the reference

Every score on the transferability square carries one correlation correction, fitted at a repulsion of eight and no site energy. Carried instead from seven other systems — three beside the band where the mean field's start misses, four on it — the polarisation's worst error ratio on the lowest mean field runs from 9.7 to 213, where fixing the start had moved it only from 48 to 14.7. Its median stays between 1.21 and 1.55. A worst ratio divides by the smallest comparable error it can find, and the reference decides where that is. And a reference fitted to the wrong solution passes the census at 2.05 while every system's error is above a half.

7 figures
One polarisation, every correlation energy: the two ends of the grid share a mean field's reading. The correlation energy each of the 81 systems misses, against its mean-field polarisation difference. The cloud is not a curve. Systems with the same polarisation run from no correlation at all to the largest on the grid: the system at repulsion 1, site energy 10 misses 0.000 and the one at 8, 8 misses 0.432, at polarisations of 0.000 and 0.000.

A constant ties the polarisation

Every score on the transferability square moved with where the borrowed correction was fitted, and an absolute score was the natural repair. It repairs nothing, because a change of reference slides every error along one straight line. The only thing a diagnostic can be asked to predict without a reference is the correlation energy itself, and asked that, the polarisation fails more plainly than any ratio showed: twenty-seven systems read exactly zero, including both the system with the least correlation on the grid and the one with the most. No function of it can beat a single constant, and a second reference does not help.

7 figures
Refined twice, four of the six two-reading floors climb back towards what one reading or a constant gives, and two stay low. For six pairs of mean-field diagnostics, the largest error no function of the pair can avoid, on the square of 81 systems and on its refinements to 289 and 1,089. double occ. + highest occupied: 0.022, 0.025, 0.029; highest occupied + mean-field energy: 0.031, 0.034, 0.034; double occ. + mean-field energy: 0.025, 0.118, 0.123; polarisation + highest occupied: 0.025, 0.101, 0.136; polarisation + mean-field energy: 0.052, 0.202, 0.209; polarisation + double occ.: 0.212, 0.226, 0.294. The dashed lines are what one well-placed constant achieves on the coarsest and finest grids, 0.216 and 0.298.

Two readings separate the square only where they do not fold

Reading two cheap diagnostics together cut the floor that no composite can beat tenfold on eighty-one systems, and the warning was that two dimensions sampled that thinly leave most cells empty. Refined to 1,089 systems, four of the six pairs climb back to what one reading or a single constant gives. Two stay near 0.03, and halving the tolerance tells them apart from the rest: their floor shrinks with it, because the systems that set it are neighbours. The others fold the square, and no density removes a fold.

7 figures
Only a rule that knows every answer brings the worst miss down steadily, and even it stays above the median at 208. The composite's worst miss on the held-out systems against the number of references added to the coarse 81, for five placement rules, with the 17×17 grid's 208 added systems as a single point (0.130). at random: 4 0.166, 8 0.166, 16 0.166, 32 0.166, 64 0.166, 128 0.166, 208 0.164; far in the readings: 4 0.166, 8 0.166, 16 0.166, 32 0.130, 64 0.130, 128 0.130, 208 0.084; along ε = U: 4 0.166, 8 0.166, 16 0.173, 32 0.129, 64 0.089, 128 0.060, 208 0.109; where references disagree: 4 0.164, 8 0.164, 16 0.254, 32 0.254, 64 0.254, 128 0.129, 208 0.129; at the worst miss: 4 0.129, 8 0.125, 16 0.085, 32 0.054, 64 0.045, 128 0.031, 208 0.022. With none added, 0.1664.

A reference at the worst miss made the worst miss worse

A composite that borrows the correlation energy of the nearest reference system missed worst on a stretch of the transferability square where the correlation energy changes fast, and the hope was that a handful of references placed there would bring the worst miss down to the median's. The stretch is the line where the site energy equals the repulsion: all twenty-one misses above 0.08 lie near it. But no rule that can be run without knowing every answer gets the worst lower than 0.060, a rule that does know every answer needs 480 of the 1,008 systems to reach the median, and its very first placement raises the worst miss from 0.166 to 0.200.

7 figures
A plane through six neighbours halves the worst miss with the same references; a weighted average makes it worse. The worst and the median held-out miss of three lookups on the same references: the nearest reference, the inverse-square-distance average of three, and a least-squares plane through six, with the 81 coarse references, the 17×17 grid, and the coarse references plus 128 placed along ε = U. nearest reference: worst 0.166, 0.130, 0.060; median 0.0131, 0.0096, 0.0109; weighted, three nearest: worst 0.232, 0.138, 0.079; median 0.0073, 0.0034, 0.0063; local plane, six nearest: worst 0.088, 0.068, 0.052; median 0.0050, 0.0015, 0.0040.

The lookup was the problem, not the references

A composite that borrows the nearest reference system's correlation energy missed the transferability square's crossover line by up to 0.166, and no placement of extra references a calculation could run brought that below 0.060. Changing the lookup instead does better with no new references at all: a least-squares plane through the six nearest gives a worst miss of 0.088 and a median of 0.0050, against 0.166 and 0.013. A weighted average of three, the obvious smoother, makes the worst miss worse. Neither makes a reference harmless — the plane's own best placement raises its worst miss thirty times in 128 steps, twice to nearly four — but the spread among a system's three nearest references, known in advance, flags seventeen of the twenty-one worst misses among its top fifty.

6 figures
A plane that reads its own leverage loses both spikes and keeps every one of the ordinary rises. The worst held-out miss as each lookup places references one at a time at its own current worst miss, on a logarithmic scale, to 128: plane, unguarded: 30 rises, largest value 3.992, 0.0101 at 128; leverage guard, wider plane: 32 rises, largest value 0.117, 0.0101 at 128; leverage guard, nearest fallback: 32 rises, largest value 0.117, 0.0275 at 128. The guard falls back when the six-neighbour plane's leverage exceeds 2.

Leverage catches the spikes and none of the rises

A least-squares plane through a system's six nearest references was the best lookup the transferability square had found, and its own greedy placement twice extrapolated to misses of 3.99 and 2.05. Both are visible before any answer is known: the plane's leverage at the system, the variance factor of the value it returns, is 6,706 and 2,172 there and never above 1.88 on any held-out system of the coarse references. A guard anywhere between 2 and 100 removes both spikes and changes nothing else. It does not make the plane monotone. The guarded plane's worst miss still rises thirty-two times in 128 placements, and every one of those rises — and every one of the unguarded plane's thirty — happens at a system whose six neighbours have just gained the reference placed.

6 figures