A count can be exact, parameter-free and able to return zero, and still forbid nothing — if it is computed over a measure the world does not sample. Here that demand is met completely, on the one object where it can be, and the answer makes this corpus’s own claim weaker and precise
Chapter 26 ended by adding a second way a ledger can fail. The first was easy to see: a count with a free parameter, which fits everything and therefore forbids nothing. The second is harder, and it is the one that matters here. A count can be exact, have no free parameter at all, and genuinely return zero on a great many inputs — and still forbid nothing that ever happens, because the inputs where it bites are not the inputs the world presents.
Worst-case complexity is the standing example. It is a real ledger by every criterion in §26.1, and its practical grip has all but gone. The bound is not wrong. It is true and has no purchase.
Which turns the diagnostic into a demand: name the measure, and show the count is nonzero on it. That demand is easy to state and hard to meet. This chapter meets it once, completely, on the smallest object in this corpus that can carry it — and the answer is not the flattering one.
Chapter 25 gave the closing field its entropy: $W(T) = d_1(T) - d_2(T)$, the number of distinct shells at size $T$, zero when none exists. Two things are known about that function, and they point in opposite directions.
It has a constant mean. Summing $W$ is counting lattice points of $\mathbb{Z}[\omega]$ inside a disc, and the fundamental domain has area $\sqrt3/2$, so
$$\sum_{T \le x} W(T) \;\sim\; \frac{\pi}{\sqrt{27}}\,x \;=\; 0.60460\ldots x .$$
It is zero almost everywhere. By the Landau–Ramanujan theorem in its Löschian form, the number of sizes that admit any shell at all is
$$\#\{T \le x : W(T) > 0\} \;\sim\; K\,\frac{x}{\sqrt{\log x}}, \qquad K = 0.6389\ldots$$
so the permitted sizes have density zero. Both statements are classical and neither is in dispute. Together they force something neither says on its own: if the average is constant while the support thins to nothing, the surviving values have to grow. The only question is how fast, and that question appears not to have been asked, because before Chapter 25 there was no reason to ask it — $W$ was a divisor function, not an entropy.
Conditioned on a size being realisable at all, the expected number of distinct shells grows as
$$\bigl\langle W(T) \,\bigm|\, W(T) > 0 \bigr\rangle \;\sim\; C\sqrt{\log T}, \qquad C = \frac{\pi/\sqrt{27}}{K} = 0.94617\ldots$$and therefore the closing field’s mean entropy at a realisable size is
$$\frac{\langle S(T)\rangle}{k} \;=\; \langle \log W \rangle \;\sim\; \tfrac{1}{2}\log\log T \;+\; \log C .$$Sieved to $T = 4\times10^6$: the summatory mean settles on $0.60460$ against $\pi/\sqrt{27} = 0.60460$; the permitted fraction is still falling, $0.230 \to 0.171$; and the conditional mean divided by $\sqrt{\log x}$ climbs $0.867, 0.887, 0.899, 0.902, 0.905$ — flattening, and below $C$, which is what a Landau–Ramanujan constant does on the way up. COMPUTED
Both inputs are classical: a lattice-point count and Landau–Ramanujan. The quotient is not, and the reason is prosaic — the conditional mean of a divisor function is an odd thing to want until the function is an entropy, which it became one chapter ago. So this is a new statement about a new object, assembled from old parts, and fully checkable by ch27-verify.py. That is the same standing as Proposition 26.2, and it is the honest description. Nothing here is a new theorem of number theory.
Chapter 25 said its entropy was a tie-breaker rather than a determinant: $k\log 2$ per shell, against an elastic cost that scales with the site count $10T+2$. It said an $O(1)$ quantity cannot outrun a quantity that scales, and left it there, because it had no growth rate to offer.
Now it has one, and the news is bad in the direction that makes the earlier claim stronger.
| $T$ | sites $10T+2$ | $\langle S\rangle/k$ | ratio |
|---|---|---|---|
| $10^{1}$ | 102 | 0.362 | $3.5\times10^{-3}$ |
| $10^{3}$ | 10,002 | 0.911 | $9.1\times10^{-5}$ |
| $10^{6}$ | 10,000,002 | 1.258 | $1.3\times10^{-7}$ |
| $10^{12}$ | $10^{13}$ | 1.604 | $1.6\times10^{-13}$ |
The entropy is not merely bounded against the elastic cost. It is doubly logarithmic against linear. Going from a hundred sites to ten trillion multiplies the elastic bill by $10^{11}$ and the entropy by four. A quantity growing like $\log\log T$ is, for every purpose a physical argument could have, a constant with pretensions.
So the ledger does bind — the count is genuinely nonzero on the sizes that occur, and we can say by how much — and what it binds is almost nothing. That is a real answer to §26.2’s demand, and it is a negative one about our own chapter. Meeting the demand is what let us find that out. A framework that had only asked whether the count could return zero would have stopped one chapter earlier, satisfied.
The closing field is a toy: one arithmetic function, one measure, a sieve and an afternoon. The demand is the same in complexity theory and there it is a research programme. Three answers exist and each names a measure rather than complaining about the absence of one.
Average-case complexity (Levin, 1986) makes hardness relative to a distribution and asks which problems are hard on average, giving distributional NP its own completeness theory. Smoothed analysis (Spielman and Teng, JACM 51(3), 2004; Gödel Prize 2008) does the sharpest version: it proves the simplex method runs in polynomial time under small random perturbation of its input, which is Trefethen’s observation that simplex is fast in practice, converted into a theorem by naming the measure. Phase transitions take the third route: random $k$-SAT has a sharp satisfiability threshold, and the instances that are hard concentrate near it. The threshold’s existence is proved for sufficiently large $k$ (Ding, Sly and Sun); for $k = 3$ the location is numerically near $\alpha \approx 4.267$ and the proof is still open. CHECKED
A sharp threshold at which a problem changes character and its cost peaks is a $\kappa^*$. That is the shape this corpus has been describing since Book 3 — the $K$ operator, a critical parameter, a transition — appearing in a discrete computational setting rather than a fluid or a lattice. Whether that is more than a shared shape is exactly the question WP-29 exists to refuse until a mechanism is produced, and no mechanism is produced here. It is noted as a place to look, not a bridge to cross.
Nothing here bears on whether P equals NP. Average-case hardness and worst-case hardness are different questions; smoothed analysis does not collapse the classes; the SAT threshold concerns random instances, not all instances. This chapter takes a diagnostic from that literature and repays it with an example small enough to finish. It does not approach the problem.
Result 27.1 is about one object. It is the closing field’s entropy and nothing else. The analogous question for $\mathbb{Z}[i]$ should have the same shape with a different constant, and is not done here. OPEN
And the κ\* remark in §27.4 is a remark. A shared shape between a satisfiability threshold and this corpus’s critical parameter is exactly the kind of resonance WP-29 was written to refuse. It is recorded because refusing to record it would be a different dishonesty, and marked as unspent.
What the chapter does establish is one thing, and it is the thing §26.2 asked for: here is a ledger, here is the measure the system actually samples, here is the count on that measure, and here is how little it forbids. The demand can be met. Meeting it is how you find out what your own claim was worth.
27.1 Show that a function which is zero on a set of density one but has positive constant mean must have unbounded values, and that the growth rate is forced by the density of its support.
27.2 Carry out §27.2 for $\mathbb{Z}[i]$: the analogue of $\pi/\sqrt{27}$ is the reciprocal of the unit cell area, and the analogue of $K$ is the original Landau–Ramanujan constant. What is $C$?
27.3 At what $T$ does the mean number of shells first exceed ten? Comment on whether the answer is physically reachable.
27.4 §26.2's second failure mode asks whether a count returns zero on the measure that occurs. State the corresponding question for Chapter 21's twelve defects, and say whether it has an answer.
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.