Two Geodesics and a Thermometer

Entry 047 ended with a question both programmes had been asking separately for months : is the modular flow the geodesic of entropic transport ? This entry answers it — with this programme’s own April machinery welded to the expedition’s transport maps, gates at machine precision, and one erratum issued within the hour, as the house demands. Three results came out. The two geodesics are genuinely different, and that difference turns out to be the thing that keeps this programme falsifiable. The bath’s thermometer — the one number every entry since 041 has leaned on — is no longer a postulate : it is a parabola, measured. And on the way, double-precision arithmetic was caught fabricating a result so plausible it would have been believed.

I. Two roads through the same states

Take two quantum states of the same system — Gaussian, so that everything is exact. Between them run two canonical paths. The transport geodesic moves probability the cheapest way : the Wasserstein interpolation of the Wigner functions, the expedition’s native road. The modular geodesic interpolates the modular Hamiltonians linearly — K_t = (1−t)K₀ + tK₁, the Gibbs road, built entirely from this programme’s April formulas (h_qq = X^−1/2F(C)X^−1/2, F(ν) = ν log((ν+½)/(ν−½))). If geometry is modular structure read at equilibrium, and transport is the cost of rearranging probability, the natural dream is that these two roads are one.

They are not. On a single mode they differ by 30 % of the covariance scale ; on a harmonic lattice (thermal interval, ten sites, mass quench, every gate at machine precision — the h→Γ inversion closes at 5×10⁻¹⁵, the endpoint relative entropy agrees across paths to 2×10⁻¹⁵) they differ by 37 %. And each road is virtuous in its own currency : along the modular path the first law δS = δ⟨K⟩ is betrayed as slowly as possible at the start ; along the transport path the kinetic action is least. Two metrics on the same manifold of states — Kubo–Mori and Wasserstein — and they genuinely disagree about what “straight” means.

One erratum, logged the same night : on the single mode the modular path also won the integrated first-law defect, and for an hour this entry’s draft contained a moral about the bath being “thermodynamically lazy.” The lattice inverted it — there the transport path holds the first law longer (∫D dt : 0.253 against 0.408 ; the soft long-wavelength modes punish the Gibbs road, D″(0) = 14.2 against 1.26). Which geodesic is gentler is a property of the states, not a law of nature. The draft moral lasted one hour. It is recorded here with its execution, because that is the method.

II. Why the disagreement is the good news

Entry 044 lists three ways to kill this framework. The third reads : a purely environmental or effective competitor reproduces the same 1/L law, the same prefactor, and the same shape dependence without any modular structure. For six months the most credible candidate for that competitor has been exactly the expedition’s formalism — entropic transport, which carries temperatures, costs, and geodesics of its own and never once mentions Tomita–Takesaki.

The 25–37 % is therefore not a disappointment. It is a distinguishability measurement. The transport world and the modular world make different predictions about the same family of states — about which path a relaxing system follows, about the profile of the first-law defect along the way. They can be told apart by experiments that live entirely at the level of density matrices. Death 3 requires the rival to be indistinguishable ; the rival is distinguishable by a third of the covariance. The programme survives by the width of the gap between itself and its nearest neighbour — and the gap is now a number.

What remains open, and stands now as the sharpest joint question : which road does nature take ? A levitated mode relaxing between two thermal states is, in principle, a referee. And the half-sided, Lorentzian modular flow — the one that carries causality (Borchers, Wiesbrock, Ceyhan–Faulkner) rather than the K-linear interpolation tested here — remains the true frontier, untouched by tonight’s answer.

III. The thermometer is measured — after the machine lied

Every quantitative claim this archive has made since entry 041 leans on one law : T_eff(L) = ℏc/(2πk_BL), the relational temperature of a region. Entry 043 admitted the honest gap : the law was imported as a continuum theorem (Bisognano–Wichmann, Hislop–Longo, Martinetti–Rovelli), never exhibited on anything one could hold. So the joint expedition went to hold it : compute the entanglement Hamiltonian of an interval in a critical chain — Peschel’s free fermions, H_E = ln((1−C)/C), nothing but a correlation matrix — and read the local inverse temperature off its couplings. The theorem says they must follow the parabola β(x) = π(R²−x²)/R : hottest at the centre, T_centre = ℏv/(πR k_B), diverging cold at the edges.

First attempt, double precision : the profile refused the parabola, and the centre temperature came out identical — 0.0245 — for intervals of 50, 100 and 200 sites. A constant. No 1/R law at all. A lesser night would have published the anomaly. The gates said otherwise : the entanglement spectrum of a critical interval is doubly exponential, the eigenvalues of C crowd against 0 and 1 beyond the reach of sixty-four bits, and ln((1−w)/w) silently saturates near ±32. The “constant temperature” was the ceiling of the floating-point format, wearing the costume of a discovery. float64 does not fail loudly on entanglement Hamiltonians ; it fabricates plausible physics.

At eighty decimal digits the truth surfaced whole :

The Hislop–Longo parabola at σ = 1.7 % across the whole interval ; the centre coefficient within 5–7 % with the slow logarithmic convergence the literature promises. The thermometer of entries 041–046 is no longer an import. It is a measurement. One identification remains, and it is named rather than hidden : that the bath reads regions through their causal diamonds — the only Lorentzian structure on offer, but an identification still. And one piece of housekeeping follows from the lie : any entanglement-Hamiltonian numerics ever run in double precision beyond ℓ ≈ 20 — including parts of this programme’s April Sage scans — are hereby under audit.

IV. The fiction grows an equation

Entry 041 dreamed a laboratory in 2038 measuring the ratio of decoherence to dissipation on a levitated crystal and finding a temperature that belonged to no cryostat. With the thermometer now measured, that fiction acquires a closed form. For a single mechanical mode of frequency ω in a trap of causal size L, the substrate inflicts an anomalous damping γ ∝ λ² and an anomalous noise S ∝ λ² — and their ratio sheds the unknown coupling entirely, exactly as Einstein’s D = k_BT/6πηa shed the molecular details in 1905 :

One reconciliation, owed to the archive’s own bookkeeping : entry 046’s Prediction B reads Γ/γ = c/(πL), without the ω. Both are correct : 046 states the integrated, low-frequency limit of the same coth ; this entry’s R is the spectral ratio at the mode frequency. One law, two readouts — the entry that measures one should know which it is measuring. The chain behind the relation is now load-bearing at every link but one : gravitational drag is measured (PSR B1913+16, 0.2 %, forty years) ; quantum consistency makes its noise partner mandatory ; the thermometer is derived and measured (§III) ; only the diamond identification stands between the bath hypothesis and a zero-parameter laboratory number. The energy channel stays hopeless — Dyson recomputed : 10⁻³⁰ quanta per second for a one-ton bar — but the silent rotational interferometer of this site’s Experiment page was never an energy measurement. It was always a ratio measurement waiting for its equation. It has one now, with the silent sphere (Q_ℓ = 0) as its built-in control.

V. In honesty

• The geodesic comparison is Gaussian and K-linear. The unitary modular flow e^iKs, and above all the half-sided Lorentzian flow that carries causality, are untouched. The frontier stands where 042 left it.
• The thermometer theorem is free/conformal. The critical boson polluted its own test through its zero mode (first attempt rejected at the gates) ; interactions remain open. The 5–7 % centre coefficient converges logarithmically — this is a verified law, not yet a precision measurement.
• Which geodesic honours the first law longer is state-dependent — the single-mode moral died on the lattice within the hour. Recorded above.
• A full reread of this archive (47 entries, four parallel readers) flagged two entries for status banners : 030, whose α⁻¹ = 4π³+π²+π is presented without the reservation that entry 031 honestly applies to its own 6π⁵, and 020, which wears a “Verified” header on a speculative vote. The 031 standard should apply uniformly. The CMB execution of entry 047, for the record, killed the saturated dark-energy family — not the area law ; entry 024’s derivation stands.

VI. The code, so you can break it yourself