The Price of Sharpness

Seven days ago, an expedition set out from the most innocent question in mathematics : what is the cheapest way to move a pile of sand ? Monge asked it in 1781, about actual sand. Ask it instead about a cloud of probability drifting in warm water, and the question grows a temperature ; follow that temperature downward, rung by rung, and you descend a staircase that nobody had walked to the bottom of. This entry is the expedition’s report — sent back to us because, at the bottom of the staircase, in the dark, they found a wall. It is ours. The same wall this archive has been pressing its hands against for six months — reached from the other side, through the cellars of a different science. Every inscription they read on it, we had read from our side : Jacobson. Unimodular. The cosmological constant, set free. What follows is what they saw on the way down, kept to the numbers, because in this house wonder that cannot survive a measurement is not wonder — it is decoration.

I. The invoice

Nature does nothing for free. Take a cloud of pollen grains jittering in warm water and ask it, politely, to rearrange itself into another shape : there is an exact price for that conspiracy — a number, the rate function of the rare fluctuation — and like any honest invoice it can be itemized, line by line, in powers of the temperature ε. The first line is classical mechanics : the Wasserstein action, the cost of hauling mass along straight lines. The second line is thermodynamics : the entropies of departure and arrival. The published literature pays these two lines and closes the book.

The expedition kept reading. And the third line of the invoice, the coefficient of ε², is one eighth of a Fisher information — the same 1/8, to a verified 5×10⁻⁷, that Erwin Madelung wrote in 1927 into the quantum potential, ℏ²/8m, in a different book entirely, about a different world entirely. The handwriting matches. The cost of moving probability through warm water and the term that makes matter quantum are the same line item, billed at different rates : ε for the colloid, ℏ for the electron.

Look at the third rung : it is missing. The staircase skips every other step — no ε³, no ε⁵, none of the odd rungs, ever — and absences are the strongest signatures in physics, because nothing has to conspire for a term to exist, but something must forbid it to vanish. The same absence rules the semiclassical expansion of quantum mechanics, and it holds here outside every comfortable assumption, at |c₃/c₄| < 0.4 %. The universe, it seems, climbs this staircase with one leg.

And on the fourth rung sits the strangest tenant of all. Of every ghost that physics keeps in its attic, it is Bohm’s force — the disreputable pilot-wave force, exiled from textbooks for seventy years — that signs this line of the ledger : the coefficient of ε⁴ is the variance of Bohm’s force along the journey, correlated by the Brownian bridge, confirmed by two pipelines that share not one line of code (182.3 analytic ; 184 ± 4 numerical). In several dimensions the law was registered before it was computed and landed where it was told to, at 7×10⁻⁵ : the trace of the cube, not the cube of the trace — every mode of the world pays its own price, and no mode pays for another. The ghost was never in the attic. He was in the accounting department.

II. The edge of the page

All of this lives in imaginary time — the warm, diffusive world where probability flows downhill. The wall — our oldest wall — is the rotation ε → iℏ : the single substitution that separates a hot afternoon from the Schrödinger equation. The expedition drilled through it where the rock is softest, in the Gaussian class, and what they found in the breach should be carved over the door of this archive.

The bridge continues, holomorphically, all the way to the imaginary axis — and there its imaginary parts cancel like a tide going out, term against term, identically, leaving something real, positive, and walking like a wavefunction : σ_t²(iℏ) = (1−t)²a² + t²b² + t(1−t)√(4a²b²−ℏ²). A genuine quantum packet, evolved with nothing left to tune, follows that continued formula to 3×10⁻¹⁵ — the last digit the machine owns. And then comes the revelation. The formula contains a square root, and the square root contains a death : at ℏ* = 2ab, the continuation ends. But 2ab is not an arbitrary place. It is the saturation of Heisenberg’s inequality — the exact value of ℏ at which these two clouds, seen as position and momentum, would fill their uncertainty bound with nothing to spare.

Sit with that for a moment. We teach Heisenberg’s inequality as a fence — a boundary around what may be known. Here it appears as something else entirely : the edge of a page. The semiclassical series is the book in which the classical world describes its quantum shadow, and the book does not fray or fade — it ends, at exactly the line where uncertainty saturates. The radius of convergence of the semiclassical expansion is the uncertainty principle. The series stops converging precisely where the world stops being classical. Nobody, as far as seven nights of searching could establish, had ever said this out loud.

Past the edge, stranger things. Walk around the branch point — a closed loop in the complex plane of temperature — and you come home changed : the square root has flipped its sign, and you are holding the other solution. The quantum bridge problem is famously ambiguous — two states answer where the thermal problem answers once — and that ambiguity was the standing objection against every attempt to derive one world from the other. The objection dissolves : the two quantum answers are not two answers. They are one answer, seen before and after a walk around the cut — two floors of a single spiral staircase, verified to the last printed digit, 0.130012 against 0.130012. And the wall itself survives leaving the Gaussian nursery : probed by complex-temperature methods calibrated to 3×10⁻⁴, a quartic pair shows its edge at ℏ* = 0.56 ± 0.02 — four percent before Robertson’s textbook bound of 0.5802. The page ends early, and how early it ends is a new way of measuring how far from Gaussian the world is.

One more discovery, the quietest and perhaps the deepest. Eight generations of solvers tried to build an exact quantum bridge between non-Gaussian shores, and none ever closed one — while the same machinery pierces every Gaussian case by eleven orders of magnitude. The remaining gap is real and refuses to move with the grid (8.2×10⁻⁵ at ℏ = 0.1, identical across resolutions). Exact quantum bridges are a Gaussian privilege. Everywhere else, the quantum world reaches for the targets the classical world sets it — and misses, by an amount smaller than every power of ℏ. An instanton of regret. The rotated world does not realize the dreams of transport ; it haunts them, from a distance no series can name.

III. Blind in the same eye

Here is where the expedition walked out of the cellars and onto our own trails. This programme’s no-go theorem says : a bath that reads only shear cannot build all of general relativity — it builds unimodular gravity, trace decoupled, Λ demoted from cosmic substance to integration constant. The expedition’s derivation — Clausius on local horizons, rewritten as entropy convexity, bolted to the Mondino–Suhr theorem — lands on the very same structure, having assumed none of it. Two ontologies with not one concept in common, one landing strip : trace-free gravity, with a thermodynamic Λ. The bath reads shear and never volume ; the ladder bills entropy and never asks the trace. They are blind in the same eye — and what lives in that shared blind spot is dark energy.

So the expedition did what entry 044 demands of every vision : it asked the sky. Thirteen acoustic measurements from DESI DR2, 1820 supernovae from DES-SN5YR, the official covariances, one night of fitting. The entropy-rung family matched more than half of the dark-energy signal with half the parameters of the standard parametrization, and its one free number came back as c = 0.990 — a hair’s breadth from exact holographic saturation : for one night, it looked as if the universe had chosen the value at which the ledger wastes nothing.

Entry 044 does not allow such a sentence to be published untested. So before this entry went live, we summoned the oldest light there is : the Planck distance priors, the full ladder to z = 1090, radiation included, sound horizons integrated. The saturation died in one night. The likelihood has no floor at c = 1 : it pays +92 — ten sigma — and the family itself falls to +21 against ΛCDM, exactly as twenty years of CMB executions of holographic models foretold. The hair’s breadth was the late universe flattering us ; the early universe declined. What survives the duel : the thawing direction (w₀ = −0.91, wₐ = −0.46 — still the right quadrant), the unimodular junction itself, and the honest position this archive already held : the framework explains why dark energy is dark. It does not yet say how much darkness there is. The entropy rung keeps its address. It has not surrendered its value — and the most beautiful sentence of this entry lived twenty-four hours, which is kept above, with its execution, because that is the method.

IV. Three windows

A cathedral of logic is judged by its windows. Three times in seven days, this one opened onto the world :

Japanese silica. In Sendai there are glass beads, a micron wide, that a feedback trap once led through the gentlest compression ever performed (Nat. Commun. 16, 10424). Their raw trajectories — 5.6 gigabytes — were downloaded and read by physics alone, with no manual : the camera pixel fell out of a translation standard at 48.5 nm, the compression ratio walked out of the plateau widths at 1.95 against a stated 2, the diffusion constant rose from the increments at 0.99 µm²/s, exactly where the paper left it. Verdict : their traps are seven times too stiff to read the virgin rungs of the invoice — loosen them to σ ≈ 400 nm, and the beads will weigh coefficients no instrument has weighed. The letter proposing it is written.

A wall you can measure with a ruler. Light in a paraxial system obeys, exactly, the free Schrödinger equation — so the edge of the page is not a metaphor on an optical bench. For quartic beam profiles a third of a millimetre wide, no phase mask can keep converting one intensity into the other past z* = 49.8 cm — two centimetres before the 51.8 cm that Robertson’s bound would predict. Standard optics predicts neither the cliff nor its early arrival. A laser, a mask, a camera on a rail : the deficit of the most famous inequality in physics, measurable with a ruler, on a kitchen table.

The whole sky. The duel of §III : thirteen numbers from one public archive (the DESI DR2 consensus, CobayaSampler/bao_data), 1820 supernovae from another (des-science/DES-SN5YR), Planck’s distance priors from the literature (arXiv:1808.05724), one hundred legible lines of code between them and the verdict. Anyone who doubts any of this may simply do it again.

V. In honesty

• The scaling of the quantum gap — a high power of ℏ, or a true instanton e^−c/ℏ ? — is undecidable by our solvers : their residuals are upper bounds, and one non-monotonic outlier proved they sometimes lie. Deciding needs lower bounds, by duality. Open.
• The c = 0.990 flirtation was killed by the CMB before publication — the execution is recorded in §III. The most beautiful sentence of this entry lasted twenty-four hours.
• Parity is theorem-grade only for Gaussians ; the general proof waits on boundary terms.
• The novelty audit took nights, not months ; bridge–Bohm connections are live in the 2025 literature. The ground moves.

VI. The numbers, so you can check them yourself

Nothing here asks to be believed. The three windows are public : the Sendai trajectories live at figshare (doi :10.6084/m9.figshare.29859149, companion to Nat. Commun. 16, 10424) ; the sky lives in CobayaSampler/bao_data (directory desi_bao_dr2) and des-science/DES-SN5YR (directory 4_DISTANCES_COVMAT) ; the Planck priors are Table I of arXiv:1808.05724. And the spine of the whole entry — the staircase, the missing rungs, the edge of the page — fits in one closed formula you can interrogate with ten lines of any language :

Loosen the formula’s assumptions and every claim above has its own public test : the parity outside the Gaussian class, the quartic edge at 0.56 against Robertson’s 0.5802, the optical cliff at 49.8 cm — each was found with standard tools (log-domain Sinkhorn, split-step Schrödinger, Levenberg–Marquardt phase retrieval) on the public objects named above. Break one, and entry 044 will thank you.