Historical derivation — preserved in full
Source: design23_autonomous_optimizer_v1.0.0/design23_autonomous_optimizer/quan_loncar_series_phase25/NEXT_PHASE_QNM_OBSERVABLES.md
Snapshot: design23-v1.0.0. The body below is unabridged.
Next phase: one normalized pole for Q, mode volume, and feed coupling¶
1. QNM residue normalization¶
Let A(omega,p)x=0 and y^H A(omega,p)=0 at a simple pole. Define
N = y^H A_omega x.
For a source-to-field map B and a coefficient-to-field map F®, the pole part of the physical Green response has the form
G_pole(r,r';omega)
= F(r)x [y^H B(r')] / [N (omega-omega_tilde)].
The first acceptance test is scaling invariance under x -> alpha x and y -> y/alpha*, followed by a direct rank-one residue comparison with (omega-omega_tilde)G at several complex frequency offsets.
2. Complex effective mode volume¶
For a chosen emitter position r0 and unit polarization e, evaluate the QNM field amplitude E_e=e dot F(r0)x and use the same residue normalization N to define the complex effective volume in a convention consistent with the electric Green tensor. Constants must be fixed by the direct Green-residue test rather than by an arbitrary eigenvector norm.
Required checks:
- invariance to left/right eigenvector scaling;
- convergence with hole P1->P2 and thickness z0->z1 bases;
- convergence of E(r0) under the field-evaluation quadrature;
- agreement between the residue-derived Purcell factor and the pole response near resonance.
3. Feed-through waveguide coupling¶
Separate the guided-mode poles of the bare-waveguide beta Green function from the radiation continuum. Construct unit-power right/left port projectors at the two ends of the nanobeam. Their QNM residue amplitudes give the external decay rates kappa_L and kappa_R, while the remaining continuum gives kappa_rad.
Required checks:
-2 Im(omega_tilde) = kappa_L + kappa_R + kappa_rad
within the same frequency convention and within 10% initially, tightened with convergence. The resulting resonant residue is then inserted into the background-plus-pole through coefficient S21(omega) and compared with a direct driven solve.
4. Automatic-differentiation boundary¶
Once field and port functionals pass the above tests, provide JVP/VJP rules for A_p, F_p, and the port projectors. The pole pullback remains
d omega/dp = -(y^H A_p x)/N,
while mode-volume and coupling derivatives add the explicit derivatives of F, B, and the normalization N. No optimizer should differentiate through the Newton iteration or an eigensolver history.
Immediate implementation order¶
- Add coefficient-to-field and source-to-coefficient maps.
- Verify the full Green-function residue and eigenvector-scaling invariance.
- Define and converge complex mode volume.
- Extract and flux-normalize guided beta-pole port projectors.
- Verify the linewidth sum and direct S21 response.
- Replace centered geometry partials with analytic/AD JVPs and expose a multi-objective gradient API for Q, Veff, and feed coupling.