Source

Engine atom_fishbone_coupler · package package · assembled 2026-07-29 15:57 UTC.

Trustworthy atom-oriented FDTD observables¶

This is the fail-closed contract for the selected atom-fishbone finalist. It defines atom-oriented mode volume and physical-dipole emitter beta. It does not authorize a cloud task, and no emitter beta is currently claimed.

The executable, local-only foundation is split between:

observables_protocol.py, which owns candidate/run manifests, mode-volume normalization, convergence comparisons, and the emitter-beta evidence analyzer; and
observables_budget.py, which requires every future paid validation task to fit both the fishbone campaign ledger and the earlier 5-FC trustworthy- observables authorization.

Content-addressed candidate and run records¶

CandidateSpec freezes the geometry, material model, atom position, normalized dipole orientation, target wavelength, and provenance in canonical JSON. Its SHA-256 is the candidate identity.

RunSpec binds a stage to that candidate SHA-256 and separately hashes the realized grid and physical source. It also records the domain/PML, monitor definitions, automatic mesh resolution, mesh-override count, 16 ps maximum time, and automatic shutoff. The default production diagnostic is Auto-14, 16 ps, with no overrides. Convergence runs may increase the automatic mesh resolution but retain the same 16 ps ceiling and shutoff policy.

Downloaded evidence is promotable only when its candidate and run hashes match. A cached result from another geometry is a test fixture, not a substitute.

Atom-oriented mode volume¶

For a weak-loss, nonmagnetic material model, the stored energy is

U = 1/4 integral [
      epsilon_0 sum_i Re{d(omega epsilon_ii,r)/domega} |E_i|^2
      + mu_0 |H|^2
    ] dV.

The canonical oriented volume is

V_atom = 2 U
         / [epsilon_0 n_atom^2 |d_hat dot E(r_atom)|^2].

Here d_hat=Ey and the atom is in air. For a nondispersive model, the independent electric-energy construction is

V_atom,electric =
    integral epsilon_r |E|^2 dV
    / [n_atom^2 |d_hat dot E(r_atom)|^2].

The total-energy value is canonical; the electric value is a cross-check. Promotion requires:

colocated finite native-grid Yee components and positive physical cell volumes;
ordinary nondispersive electric/magnetic Maxwell energy balance within 5% (the dispersive material-energy term is included only in canonical U);
total-energy and electric forms agreeing within 5%;
at least two exact-pole meshes changing both forms by no more than 5%;
two matched source-free time windows changing V_atom and V_peak by no more than 5%;
pole wavelength change no more than 0.5%;
inner/outer boxes, signed faces, and Qflux/Qring each closing within 10%; and
exact-frequency detuning below 0.2 fitted linewidth.

The cached rejected M16 exact-v2 job is retained only as an optional analyzer regression fixture. Reanalysis gives 23.36294 um^3 from total energy and 23.50496 um^3 from the electric form, a 0.604% difference. It does not establish mesh convergence for a finalist.

Dispersive materials¶

For each colocated electric component, evaluate the same local material model on the exact Yee grid at f0*(1-delta) and f0*(1+delta). Central- difference f*epsilon(f); this equals d(omega*epsilon)/domega. Compare relative steps delta=1e-4 and 5e-5 and require the integrated Brillouin contribution to change by at most 1%.

The local epsilon(f0) must agree with the downloaded permittivity monitor within 1%. A large multifrequency volume monitor is added only if this local/server check fails. Magnetic dispersion is unsupported. If the maximum loss tangent exceeds 1e-3, the weak-loss Brillouin value is not promoted; an explicitly validated quasinormal-mode normalization is then required.

Physical-dipole emitter beta¶

The requested quantity is

beta_emitter = Gamma_cavity_pole / Gamma_total.

It is not an x/y/z cavity-loss fraction, a two-port fraction, a useful-mode collection ratio, or a ratio inferred from cavity quality factors. The analyzer has no path that converts those diagnostics into emitter beta.

The minimum evidence set is:

a full cavity with a physical Ey point dipole at the manifest atom;
a vacuum reference; and
at least two pole-removed structured references retaining the local dielectric environment, such as uniform-center-cell and detuned/no-defect fishbones.

Every run must have the same candidate SHA-256, realized full-cavity grid SHA-256, physical source-waveform SHA-256, atom position/orientation, domain/PML, symmetry, duration, and shutoff. Reference grids are copied from the realized full grid; independently regenerated automatic grids are not acceptable because the same-location point-dipole self field is grid sensitive.

The full and reference simulations record dense complex Ey(atom) spectra and nested closed-box emission spectra across the already fitted pole. Source normalization must be nonzero over the fit band. Matched structured-reference subtraction cancels the singular self field, then one complex pole plus a smooth background is fitted with its frequency and Q constrained by ringdown. The fit must permit Fano interference rather than force a positive Lorentzian residual.

Promotion requires all of the following:

positive Gamma_total, vacuum rate, and cavity-pole rate;
beta in [0, 1];
no more than 5 percentage points of beta spread across the two structured backgrounds and across fitting windows;
fitted-window and structured-background median beta values agreeing within 5 percentage points;
nested-box and closed-flux/source-work closure within 10%;
fitted frequency within 0.2 ringdown linewidth and fitted Q within 10%;
nonzero source spectrum across the fit; and
Gamma_cavity_pole/Gamma_vacuum within 10% of [3/(4 pi^2)] Q/(V_atom/lambda_air^3) for an exactly resonant, Ey-aligned emitter.

Any reported cavity-to-vacuum rate ratio must also equal the ratio recomputed from the primitive fitted rates; the analyzer does not trust a redundant precomputed value.

If any evidence is absent or any gate fails, the analyzer sets promoted_beta to null. It may retain a clearly labeled diagnostic fit value to aid debugging, but the public classification remains “no emitter beta is claimed.”

Cost gate and execution order¶

Future trustworthy-observables work is dual-accounted under independent authorization caps:

Ledger	Hard cap	Conservative maximum already started	Remaining
fishbone campaign	25 FC	15.511192631 FC	9.488807369 FC
earlier trust allocation	5 FC	2.019870981 FC	2.980129019 FC

The binding future headroom at this document revision is therefore 2.980129018900165 FC. It must be recomputed immediately before any reservation. One physical cloud task carries the same cross-ledger ID, candidate hash, run hash, task name, task ID, and maximum estimate in both ledgers. Mirroring the task is authorization accounting, not a second charge.

The paid sequence remains strictly staged:

freeze and locally gate the candidate, serialize all simulations, and calculate the complete maximum-cost envelope;
run a separated automatic-mesh broadband pair;
run a third mesh, then expanded domain/PML, only after earlier gates pass;
run exact-pole mode volume on two meshes, with dense driven point/box spectra included in the full-cavity exact job; and
only if V_atom is useful and converged, run vacuum and two structured beta references.

If mandatory maximum estimates exceed the binding cap, stop and request more authority. A reference, mesh, or uncertainty gate is never dropped merely to fit the budget.