Twelve defects, never eleven and never thirteen — and the arithmetic that decides which shells exist, which come in mirror pairs, and which cannot be built at all
Pour a box of magnetic spheres onto a table and they will find each other. Each ball settles against six neighbours, and the sheet they make lies perfectly flat. It will not curl. You can push it around, add to it, take from it, and it stays a plane. This is not stubbornness; it is arithmetic. Three hexagons meeting at a corner use $120° + 120° + 120° = 360°$, and a full turn is all there is. Nothing is left over to bend with.
To make that sheet into a ball you have to remove angle. Cut a $60°$ wedge out of the sheet and force the two cut edges together. The sheet cannot stay flat any more — it domes. At the point where you closed the cut, one ball now has five neighbours instead of six. That single operation is the only tool available, and this chapter is about how many times you have to use it.
The answer is twelve. Not approximately twelve, not usually twelve: exactly twelve, every time, for every shell of this kind that has ever been built or ever could be. Twelve for the soccer ball. Twelve for a shell of a hundred thousand cells. Twelve for the protein coat of a virus. You may add hexagons without limit and it costs nothing; you may never add or remove a pentagon.
Two further questions turn out to have equally sharp answers, and they are the reason this chapter is in a book about number theory rather than a book about geometry. Which shells can be built at all is decided by a norm form — a shell exists for a given size only when that size is a number of the shape $m^2+mn+n^2$. And which shells come in left- and right-handed pairs is decided by whether a prime splits in a particular ring of integers. The classic soccer ball is not mirror-symmetric by design; it is mirror-symmetric because $3$ ramifies.
The chapter closes by pointing at the thing this whole arc has been circling. The corpus has been writing contact forms for a hundred and sixty files. A contact form is Gibbs's object. That connection has never been made in these pages, and §21.8 makes it, carefully, marking what follows and what does not.
Let $\Lambda$ be the triangular lattice of sphere centres in the plane — the arrangement magnetic spheres adopt on their own, in which every site has six nearest neighbours at equal distance. Its dual is the hexagonal tiling. Identify the plane with $\mathbb{C}$ and $\Lambda$ with the Eisenstein integers $$\mathbb{Z}[\omega] = \{\,m + n\omega : m,n \in \mathbb{Z}\,\},\qquad \omega = e^{i\pi/3},$$ which is the ring of integers of $\mathbb{Q}(\sqrt{-3})$. Its norm is $$N(m+n\omega) = m^2 + mn + n^2 .$$
The flatness is a statement about angles that leaves no slack. At every vertex of the hexagonal tiling three faces meet, each contributing the interior angle of a regular hexagon, $120°$. The angular defect at that vertex — the amount by which the angles fall short of a full turn — is
$$\delta = 360° - 3\cdot 120° = 0 .$$Zero defect is zero Gaussian curvature. A surface built entirely of such vertices is flat, extends indefinitely, and closes on nothing. This is the first half of the commensurability picture developed in Chapter 12, seen from the other side: the hexagonal sheet is the perfectly commensurate case, and perfect commensurability buys you a tiling that goes on forever without ever returning.
A magnetic sphere lattice will not spontaneously form a shell, and this is not a limitation of the magnets. It is that a hexagonal sheet has exactly zero curvature budget, and a sphere requires $4\pi$ of it.
The one available operation is the Volterra construction: excise a wedge of angle $\theta$ with apex at a lattice site, then identify the two exposed edges. The result is a cone of deficit $\theta$, with all the curvature concentrated at the apex. Adding a wedge instead of removing one gives $\theta < 0$ and a saddle.
The wedge angle is not free. For the rejoined edges to match lattice to lattice, the rotation carrying one cut edge onto the other must be a symmetry of $\Lambda$. The rotations preserving $\mathbb{Z}[\omega]$ about a lattice point are exactly multiplication by the units,
$$\mathbb{Z}[\omega]^\times = \mu_6 = \{\pm 1, \pm\omega, \pm\omega^2\},\qquad |\mu_6| = 6,$$so the admissible wedge angles are multiples of $2\pi/6 = 60°$. This is the Frank angle, and its quantisation comes directly from the order of the unit group.
A positive elementary disclination in $\Lambda$ is a Volterra excision of angle $2\pi/6$. The apex site acquires five neighbours in place of six; in the dual tiling, one hexagon becomes a pentagon. A negative elementary disclination inserts the wedge; the apex acquires seven neighbours and the dual face becomes a heptagon.
Two remarks, both of which matter later. First, the defect is topological: no continuous deformation of the sheet can remove it or change its charge, because the charge is a rotation class and rotation classes are discrete. Second, the sign is the whole difference between a ball and a doughnut. Positive disclinations buy positive curvature and build spheres. Negative disclinations buy negative curvature and build saddles, tubes, and the negatively curved carbon schwarzites. A shell of hexagons and heptagons is not a slightly different ball; it is not a ball at all.
Two proofs, each one line, arriving at the same integer from opposite directions.
Let $P$ be a convex polyhedron in which exactly three faces meet at every vertex and every face is a pentagon or a hexagon. Then $P$ has exactly twelve pentagonal faces, whatever its size.
Proof. Write $f_k$ for the number of $k$-gonal faces. Trivalence gives $3V = 2E$, and counting incidences gives $2E = \sum_k k f_k$. Substituting both into Euler's $V - E + F = 2$ and clearing denominators, $$\sum_k (6-k)\, f_k \;=\; 12 .$$ Hexagons contribute $0$ to the left-hand side. Hence $f_5 = 12$. $\;\square$
The identity $\sum_k (6-k) f_k = 12$ is worth keeping in that form, because it says something stronger than the theorem. It is a conservation law: hexagons are free, pentagons carry charge $+1$, squares $+2$, heptagons $-1$, and the total is pinned at $12$ for every closed trivalent surface of genus zero. A shell with one heptagon must carry thirteen pentagons to pay for it.
By Descartes' theorem on total angular defect, every convex polyhedron satisfies $\sum_v \delta_v = 4\pi$. In a shell of pentagons and hexagons every vertex is of type $(5,6,6)$, with defect $2\pi - (\tfrac{3\pi}{5} + \tfrac{2\pi}{3} + \tfrac{2\pi}{3}) = \pi/15$, i.e. $12°$; there are therefore $60$ vertices per unit of the smallest such shell, and the number of pentagons is $$\frac{4\pi}{2\pi/6} = 12,$$ the total curvature divided by the elementary disclination. $\;\square$
The denominator in Theorem 21.3 is $2\pi/|\mathbb{Z}[\omega]^\times|$. The number of defects in any closed hexagonal shell is the sphere's total curvature divided by the elementary rotation of the lattice — and that rotation is set by the order of the unit group of the Eisenstein integers. Twelve is not a fact about footballs. It is $4\pi \div (2\pi/6)$.
Fix an icosahedron and place one elementary disclination at each of its twelve vertices; the twelve are forced, and the icosahedron is the unique way to distribute them with full symmetry. What remains free is the hexagonal patch filling each of the twenty triangular faces, and a patch of the triangular lattice matching an equilateral triangle is specified by a single Eisenstein integer.
Let $z = m + n\omega$ with $m,n \ge 0$, not both zero, and let the icosahedral face correspond to the lattice triangle with vertices $0$, $z$, $\omega z$. Write
$$T \;=\; N(z) \;=\; m^2 + mn + n^2 .$$The shell determined by $z$ (the Goldberg polyhedron $\mathrm{GP}(m,n)$, equivalently the geodesic polyhedron dual to it) has
· $10T + 2$ sites — the sphere-magnet count,
· $30T$ contacts,
· $12$ five-coordinated sites and $10T - 10$ six-coordinated sites,
· $10T + 2$ panels in the dual: $12$ pentagons and $10T - 10$ hexagons.
These are Goldberg's counts (1937); the pentagon count is Theorem 21.2 and does not depend on $z$.
A shell of this family with parameter $T$ exists if and only if $T$ is a norm from $\mathbb{Z}[\omega]$ — equivalently, if and only if every prime $p \equiv 2 \pmod 3$ divides $T$ to an even power. These are the Löschian numbers, $$1,\;3,\;4,\;7,\;9,\;12,\;13,\;16,\;19,\;21,\;25,\;27,\;28,\;31,\;36,\;37,\;39,\;43,\;48,\;49,\dots$$ In particular $2$, $5$, $6$, $8$, $10$, $11$, $14$, $15$ are not available: there is no such shell of those sizes, and the obstruction is arithmetic, not engineering.
This is the point at which the object stops being a piece of solid geometry. The question "how many different balls of this kind are there, and of what sizes" is the question "which integers are represented by the quadratic form $m^2+mn+n^2$", which is a classical problem with a classical answer, and the answer is a statement about splitting behaviour in $\mathbb{Q}(\sqrt{-3})$.
A rational prime behaves in $\mathbb{Z}[\omega]$ in one of three ways, and each way corresponds to a visible property of the shells it generates.
| Prime | Behaviour in $\mathbb{Z}[\omega]$ | Consequence for the shell |
|---|---|---|
| $p = 3$ | ramifies, $3 \sim (1+\omega)^2$ | $z = 1+\omega$ has $m=n$; the shell is its own mirror image |
| $p \equiv 1 \pmod 3$ | splits, $p = \pi\bar\pi$ | $\pi$ and $\bar\pi$ give a left- and a right-handed shell |
| $p \equiv 2 \pmod 3$ | inert, $N(p) = p^2$ | no shell of size $p$ exists at all |
The shell $\mathrm{GP}(m,n)$ is achiral precisely when $n = 0$ (Goldberg class I) or $m = n$ (class II) — that is, precisely when $z$ is fixed by complex conjugation up to units. Otherwise $z$ and $\bar z$ generate distinct enantiomers (class III), and the mirror image of $\mathrm{GP}(m,n)$ is $\mathrm{GP}(n,m)$.
Three consequences worth stating plainly.
The soccer ball is achiral because $3$ ramifies. The 1970 Telstar pattern is $z = 1+\omega$, $T = 3$, thirty-two panels. It has $m = n$ because the ramified prime is the one fixed by conjugation. Its symmetry is not a design decision; it is the ramification of $3$ in $\mathbb{Q}(\sqrt{-3})$.
The smallest chiral shell is the seven-flower. Take $z = 2 + \omega$, so $T = 4 + 2 + 1 = 7$. The corresponding patch of the hexagonal lattice is one hexagon surrounded by six — the flower — which is exactly the fundamental domain of the index-$7$ sublattice $\mathbb{Z}[\omega]/(2+\omega) \cong \mathbb{F}_7$. Since $7 \equiv 1 \pmod 3$ splits, there are two conjugate primes above it, and therefore two shells, mirror images of each other, seventy-two sites each. This is the fullerene $\mathrm{C}_{140}$ and it is chiral for the same reason.
Seven enters as a norm, not as a heptagon. The pairing that closes a sphere is five-and-six. The number seven is nonetheless unavoidable here, but in the arithmetic rather than the geometry: it is the first split prime, hence the first handedness. A shell of hexagons and heptagons is the negatively curved object of §21.2 and does not close.
Caspar and Klug's triangulation number for icosahedral virus capsids is $T = h^2 + hk + k^2$ — the same norm form, arrived at independently in 1962 from electron micrographs. Capsids are catalogued at $T = 1, 3, 4, 7, 13, \dots$, and $T=7$ capsids are recorded with an explicit handedness label, laevo or dextro. The splitting of $7$ in $\mathbb{Z}[\omega]$ is observed in a protein shell.
This is why the chapter belongs beside Chapter 12 rather than beside the crystallography.
The commensurability ladder developed there runs on units in real quadratic fields. Baudhāyana's $\sqrt2 \approx 577/408$ is $\varepsilon_2^8$ where $\varepsilon_2 = 1+\sqrt2$; Archimedes' unexplained bound $\sqrt3 < 1351/780$ is $\varepsilon_3^6$ where $\varepsilon_3 = 2+\sqrt3$; and the golden ratio is $\varepsilon_5$, the smallest fundamental unit of any real quadratic field, which is exactly why $\varphi$ is the worst-approximable number and why phyllotaxis divergence angles built on it never repeat. Dirichlet's unit theorem gives these fields unit rank $r_1 + r_2 - 1 = 1$: an infinite cyclic ladder, and figures that approach without ever arriving.
The hexagonal field is the other case. $\mathbb{Q}(\sqrt{-3})$ has $r_1 = 0$, $r_2 = 1$, unit rank $0$ — no ladder at all, only the six roots of unity. There is nothing to iterate, and correspondingly nothing that fails to close. What the field supplies instead is its norm form, and the norm form does not approximate anything; it counts.
Unit rank one — real quadratic — gives an infinite multiplicative ladder, Diophantine approximation, and figures that never close. Unit rank zero — imaginary quadratic — gives a finite symmetry group, a norm form, and figures that do. Among all imaginary quadratic fields the largest finite unit group is $\mu_6$, belonging to $\mathbb{Q}(\sqrt{-3})$: the hexagonal lattice is the most symmetric closed case there is, which is precisely why it is the tiling that lies flat and why its elementary defect is $60°$.
And then the two halves meet on the shell. The twelve defect sites sit at the vertices of an icosahedron, whose rotation group $A_5$ has character field $\mathbb{Q}(\sqrt5)$ — the field of $\varphi$, the field of the ladder that never closes. The hexagons between them are counted by the norm form of $\mathbb{Q}(\sqrt{-3})$. A closed hexagonal shell is the surface on which those two fields are obliged to share a skin: a $\mathbb{Q}(\sqrt{-3})$ lattice draped on a $\mathbb{Q}(\sqrt5)$ skeleton. The defect is carried by the number that never closes; the field that does the closing is the one with no units to spend.
Counts are combinatorial and exact. Panel geometry is not, and it decides whether a shell is any good as an object. The table below reports the spherical Voronoi tessellation of the $10T+2$ sites on the unit sphere: one panel per site, areas computed by angular excess. ball3.py generates the site set from the Eisenstein patch on each consistently oriented icosahedral face, and the pentagon count of twelve in every row is a check on the construction rather than an input to it.
| shell | $z$ | $T$ | sites | 5 / 6 | hex : pent area | max / min | CV | roundness |
|---|---|---|---|---|---|---|---|---|
| GP(1,0) | $1$ | 1 | 12 | 12 / 0 | — | 1.0000 | 0.00 % | 0.7547 |
| GP(1,1) | $1+\omega$ | 3 | 32 | 12 / 20 | 1.0623 | 1.0623 | 2.90 % | 0.9058 |
| GP(2,0) | $2$ | 4 | 42 | 12 / 30 | 1.1296 | 1.1296 | 5.36 % | 0.9359 |
| GP(2,1) | $2+\omega$ | 7 | 72 | 12 / 60 | 1.2826 | 1.2826 | 8.52 % | 0.9623 |
| GP(3,0) | $3$ | 9 | 92 | 12 / 80 | 1.2679 | 1.3393 | 7.95 % | 0.9693 |
| GP(2,2) | $2+2\omega$ | 12 | 122 | 12 / 110 | 1.4196 | 1.5342 | 10.88 % | 0.9767 |
| GP(3,1) | $3+\omega$ | 13 | 132 | 12 / 120 | 1.3978 | 1.4872 | 10.48 % | 0.9786 |
| GP(4,0) | $4$ | 16 | 162 | 12 / 150 | 1.3563 | 1.4397 | 9.72 % | 0.9820 |
| GP(3,2) | $3+2\omega$ | 19 | 192 | 12 / 180 | 1.5045 | 1.6249 | 11.61 % | 0.9849 |
| GP(4,1) | $4+\omega$ | 21 | 212 | 12 / 200 | 1.4666 | 1.6571 | 11.38 % | 0.9863 |
| GP(3,3) | $3+3\omega$ | 27 | 272 | 12 / 260 | 1.5757 | 1.8021 | 11.81 % | 0.9892 |
| GP(4,4) | $4+4\omega$ | 48 | 482 | 12 / 470 | 1.6640 | 1.9459 | 12.34 % | 0.9938 |
| GP(5,5) | $5+5\omega$ | 75 | 752 | 12 / 740 | 1.7208 | 2.0343 | 12.64 % | 0.9960 |
Roundness is the isoperimetric quotient $36\pi V^2 / S^3$ of the convex hull, equal to $1$ for a sphere. CV is the coefficient of variation of panel area. Chiral rows are GP(2,1), GP(3,1), GP(3,2), GP(4,1).
Roundness rises with $T$ and panel uniformity falls, without exception in this range: $0.9058 \to 0.9960$ against $2.90\% \to 12.64\%$. The classic thirty-two-panel ball has the most uniform panels of any shell in the family that has hexagons at all. Every rounder shell pays for its roundness in panel unevenness, and the unevenness concentrates in the same twelve places — the pentagons shrink relative to their neighbours, from $1{:}1.06$ at $T=3$ to $1{:}1.72$ at $T=75$. This is the same statement fullerene chemistry makes about strain: the pentagons carry it. It is also why the modern match ball abandoned this family entirely and went to small numbers of equal-area non-polygonal panels, keeping the twelve defects in the curvature rather than as faces.
The ratio is monotone in $T$ within class I and within class II separately; across classes it is not (GP(4,0) at $T=16$ sits below GP(2,2) at $T=12$), and class III is not monotone even internally (GP(4,1) below GP(3,2)). Uniformity is a property of the lattice class, not of $T$ alone.
An earlier table computed for this chapter reported the $T=3$ shell at CV $\approx 15\%$ and a hexagon-to-pentagon area ratio of $1.39$, and returned twenty-seven pentagons for GP(2,1) instead of twelve. Both errors had one cause: the icosahedral faces were taken from a convex-hull routine without enforcing a consistent outward orientation, so the chiral lattice patch was applied in mirrored handedness on roughly half the faces and the seams did not match. Achiral shells were unaffected, which is why the fault survived a first check. The table above is the corrected computation; the pentagon count is now twelve in every row, as Theorem 21.2 requires.
Everything above is static. A shell either closes or it does not, and nothing in §§21.1–21.7 says what drives a sheet to fold or what it costs. That question has been standing open across this arc, and the honest first step is to report where the corpus currently stands on it.
| term | files in geometry/ | reading |
|---|---|---|
contact form | 164 | the structure is everywhere |
Reeb | 164 | |
Legendrian | 33 | |
| $\alpha \wedge d\alpha$ | 24 | |
entropy | 80 | the vocabulary is everywhere |
temperature | 65 | |
heat | 62 | |
Gibbs | 1 | one subordinate clause, in gcm-framework.html |
Carnot | 1 | a history section, 1824–1865 |
Legendre transform | 0 | — |
disclination | 0 | — (before this chapter) |
So the position is precise. The corpus has been writing contact forms for a hundred and sixty-four files and has never once said what a contact form is. It is Gibbs's object. The first law $dU = T\,dS - p\,dV$ is a contact form on the odd-dimensional thermodynamic phase space; equilibrium states are its Legendrian submanifolds; changing thermodynamic potential is a Legendre transformation. This is Hermann (1973), Mrugała (1978), and Arnold (1990), who titled the point "the geometrical method of Gibbs's thermodynamics" and observed that every contact geometer has been doing thermodynamics without noticing.
Read the corpus's own form in that light:
$$\alpha \;=\; dz - r^2\,d\theta ,\qquad \alpha \wedge d\alpha \neq 0 .$$The non-integrability condition is a physical prohibition. There is no surface everywhere tangent to $\ker\alpha$, which is to say: you cannot gain $z$ without turning. Height is not available on its own; it is available only in exchange for circulation, and $r^2\,d\theta$ is the action — angular momentum times angle. The Reeb field is the direction of pure gain, transverse to every trade.
A lift needs heat, and the form as the corpus writes it has none. Set it beside the Gibbs form for a system that can rotate. The first law for such a system is $dU = T\,dS + \Omega\,dJ$, with $\Omega$ the angular velocity and $J$ the angular momentum, so its contact form on the extended phase space is
$$\alpha_{\mathrm{G}} \;=\; dU - T\,dS - \Omega\,dJ .$$Match this term by term against $\alpha = dz - r^2\,d\theta$. The potential $z$ is the internal energy $U$; the extensive variable $\theta$ is the angular coordinate whose conjugate momentum is $J$; and $r^2$ sits where $\Omega$ sits. One term is present. One term is absent.
Setting $dS = 0$ in $\alpha_{\mathrm{G}}$ gives exactly $dU - \Omega\,dJ$, which is $dz - r^2\,d\theta$. The contact form this corpus has been using for a hundred and sixty-four files is the Gibbs form with the heat term deleted — thermodynamics restricted to constant entropy. Its Legendrian submanifolds are adiabats. This is why nothing ever appeared to be spent: on an adiabat, nothing is.
Restore the term and evaluate $\alpha_{\mathrm{G}}$ along an actual process. Using the first law in the form $dU = \delta Q + \Omega\,dJ$,
$$\alpha_{\mathrm{G}}(\dot\gamma) \;=\; \delta Q - T\,dS .$$That quantity has a name. Clausius' inequality says $\delta Q \le T\,dS$ for every physical process, with equality if and only if the process is reversible. So
$\alpha_{\mathrm{G}}(\dot\gamma) \le 0$ along every physical path, and $\alpha_{\mathrm{G}}(\dot\gamma) = 0$ exactly when the path lies in $\ker\alpha_{\mathrm{G}}$. Hence the Legendrian submanifolds are precisely the reversible, quasi-static processes, and the failure of a path to be Legendrian is, up to the factor $T$, the entropy it produces: $$dS - \frac{\delta Q}{T} \;=\; -\frac{1}{T}\,\alpha_{\mathrm{G}}(\dot\gamma) \;\ge\; 0 .$$
Verified symbolically in gibbs-check.py, together with Observation 21.7 and the Legendre invariance below.
The objection that opened this thread was that thermodynamics needs a direction and a cost. Both are here, and neither had to be added by hand. The direction is the sign of $\alpha$, fixed by Clausius and not by convention. The cost is $-\alpha(\dot\gamma)/T$, the entropy produced. A contact manifold with a distinguished sign on its form is an arrow of time; the corpus had the manifold and, by deleting $T\,dS$, had thrown away the arrow.
Trade $J$ for its conjugate by setting $G = U - \Omega J$, so that $dG = dU - \Omega\,dJ - J\,d\Omega$ and $dG = T\,dS - J\,d\Omega$. The contact form built on the new potential is
$$\alpha' \;=\; dG - T\,dS + J\,d\Omega \;=\; dU - T\,dS - \Omega\,dJ \;=\; \alpha_{\mathrm{G}} .$$The two are not merely equivalent; they are the same form. Changing thermodynamic potential is a contact transformation that leaves $\alpha$ fixed and moves only the coordinates in which it is written — which is Arnold's observation, and the reason a Legendre transform is a change of chart rather than a change of physics. The corpus has used contact transformations throughout without once calling one by this name.
A vortex takes in heat at a warm base, discharges it at a cold top, and produces lift — and it can only produce that lift by rotating, which is $\alpha \wedge d\alpha \neq 0$ said in air. Emanuel's (1986) theory of the tropical cyclone models exactly this as a Carnot engine running between sea-surface temperature $T_s$ and outflow temperature $T_o$ at efficiency $\eta = (T_s - T_o)/T_s$. The petals lifted by a spring gust are tracing the Reeb flow: they cannot rise except by going round. An engine is precisely a device whose cycle is not Legendrian: if it were, $\alpha$ would vanish on it, no entropy would be produced, and no work would be extracted. Emanuel's Carnot efficiency $(T_s-T_o)/T_s$ is the bound on how much of $\oint\alpha$ the storm can convert into $\oint\Omega\,dJ$. This is the geometry the corpus has been using as a formal device, with the heat put back in.
And it closes the loop with §21.2, because a vortex and a disclination are the same species of object. Both are topological defects; both carry a charge valued in a discrete group (a winding number for the flow, a Frank angle in $\mu_6$ for the lattice); neither can be removed by any continuous deformation; and in both cases the charge must sum to a value fixed by the topology of the surface — $4\pi$ of curvature for a sphere, total circulation for a closed flow. The pentagon in a magnet shell and the funnel in a storm are counted by the same kind of theorem.
21.A · answered, and it changes the question. The corpus's $\alpha$ is a Gibbs form: the potential is the internal energy, the conjugate pair is $(\Omega, J)$, and Legendre transforms act on it as contact transformations that leave the form invariant. But it is the adiabatic Gibbs form, missing $T\,dS$. So the open question is no longer whether a potential exists. It is this: what is the corpus's entropy? Until a function $S$ is named on the corpus’s own phase space — not borrowed from an analogy — the $T\,dS$ term cannot be restored, and Observations 21.7–21.8 describe a thermodynamics the corpus is adjacent to rather than one it has.
21.B The phase-space volume rate computed elsewhere in this arc, $c/\pi = \tfrac{1}{2\pi}\log(t/2\pi)$, is a bookkeeping identity. It is not established as a Second Law statement, and nothing here licenses a thermodynamic argument about $\zeta$. Observation 21.8 makes this stricter, not looser: the inequality $\alpha(\dot\gamma)\le 0$ is Clausius, and Clausius is a statement about physical processes with a real entropy. The critical strip supplies no such process. A rate that happens to be positive is not entropy production.
20b.C Is the Voronoi panelisation of §21.7 the minimiser of panel-area variance among all twelve-defect shells at fixed site count? Not known. The table measures one construction, not an optimum.
| claim | status | source |
|---|---|---|
| $\sum_k (6-k) f_k = 12$; hence exactly 12 pentagons | PROVED | Euler; Thm 21.2, one line |
| Total angular defect $= 4\pi$ | CLASSICAL | Descartes |
| Frank angle quantised in multiples of $2\pi/6$ | PROVED | $|\mathbb{Z}[\omega]^\times| = 6$; §21.2 |
| Counts $10T{+}2$, $30T$, $10T{-}10$ | CLASSICAL | Goldberg 1937 |
| Shell exists iff $T$ is a norm from $\mathbb{Z}[\omega]$ | CLASSICAL | Löschian numbers, A003136 |
| Chirality $\Leftrightarrow$ $z \not\sim \bar z$ $\Leftrightarrow$ $p$ splits | CLASSICAL | restated here; Prop 21.6 |
| Unit rank $1$ vs $0$ dichotomy | CLASSICAL | Dirichlet; §21.6 |
| Panel geometry table | MEASURED | ball3.py, reproducible |
| Caspar–Klug $T = h^2+hk+k^2$; $T{=}7$ handedness observed | CITED | Caspar & Klug 1962 |
| Cyclone as Carnot engine, $\eta = (T_s-T_o)/T_s$ | CITED | Emanuel 1986 |
| Contact form $=$ Gibbs form | CITED | Hermann 1973; Mrugała 1978; Arnold 1990 |
| Corpus $\alpha$ is the adiabatic Gibbs form ($dS=0$) | PROVED | Obs 21.7; gibbs-check.py |
| $\alpha(\dot\gamma) = \delta Q - T\,dS$; Legendrian $\Leftrightarrow$ reversible | PROVED | Obs 21.8; Clausius |
| Legendre transform leaves $\alpha$ invariant | PROVED | gibbs-check.py (2) |
| An entropy function on the corpus's own phase space | OPEN | 21.A, restated |
| Second-Law reading of the $c/\pi$ rate | NOT ESTABLISHED | 21.B |
| Optimality of the Voronoi panelisation | OPEN | 20b.C |
Nothing in this chapter is claimed as new mathematics. Theorems 21.2–21.3 and Propositions 21.4–21.6 are classical results assembled in one place; what is offered as the chapter's own contribution is the assembly — that the existence of a shell, its size, and its handedness are three readings of the same arithmetic in $\mathbb{Z}[\omega]$, placed beside the unit-theoretic ladder of Chapter 12 as its complementary half — together with the measured panel table and the correction recorded in §21.7.
Chapter 16 built the lattice and put a hexagonal colony in it. Chapter 20 asked what the lattice does where it fails, and found that failure is a different structure rather than an absence of one. This chapter takes the same defect and asks what happens when there are only twelve of them and they are made to close a surface. The answer moves the discussion out of crystallography and into $\mathbb{Q}(\sqrt{-3})$.
The companion reading is Chapter 12, whose commensurability ladder runs on units in real quadratic fields. That ladder describes what never closes; this chapter describes what does, and §21.6 states the dichotomy as a consequence of Dirichlet's unit theorem. Readers coming from Chapter 20 should note that its Burgers-vector defects are translational and this chapter's are rotational — dislocations and disclinations respectively — and that the quantisation arguments are parallel but not the same argument.
§21.8 opens a thread rather than closing one, and does so deliberately. It is the first place in the corpus where the contact form is identified as Gibbs's, and the three items in the open box are the price of saying so honestly.
21.1 Show that a closed trivalent shell built from hexagons and exactly one heptagon must contain thirteen pentagons. Then show that no closed trivalent shell can be built from hexagons alone.
21.2 Verify that $2+\omega$ has norm $7$ and that the seven-hexagon flower is a fundamental domain for $\mathbb{Z}[\omega]/(2+\omega)$. Deduce that the flower tiles the plane, and identify the self-similar map that sends the flower to a flower of forty-nine.
21.3 A standard set contains $216$ spheres. Find every shell in the family that can be built from at most $216$, and identify which of them are chiral. (Answer: $T \in \{1,3,4,7,9,12,13,16,19,21\}$; the largest is $\mathrm{GP}(4,1)$ at $212$ sites with four left over, and it is chiral because $21 = 3 \cdot 7$ carries the split prime $7$.)
21.4 Explain, without computing anything, why no shell in this family has exactly $10 \cdot 2 + 2 = 22$ or $10 \cdot 5 + 2 = 52$ sites.
sorryAx. A clean axiom report is not a reading of the statement: per R20, a theorem can assume its conclusion and still report clean. Follow the link before citing one as evidence.