dynphys_examples

Enhancement-75 — dynamic physics validation

The companion of Enhancement-57's physcheck_examples/: where that suite validated the static laws of compiled industry models (DC curves, autodiff Jacobians, noise), this one validates the charges — the reactive paths of the toolchain: ddt() lowering, the reactive autodiff Jacobian, the jω AC stamping, and the transient integrator, checked against physics that must hold across analyses.

The laws

1. One charge model, two code paths — PSP103's gate capacitance measured from AC (Im(i_g)/ω, the jω reactive Jacobian) equals the capacitance measured from a slow transient ramp (i_g/(dV_g/dt), the integrator) across the accumulation-to-inversion transition: worst relative difference 6×10⁻⁴ over a 4× capacitance swing.
2. Charge conservation — over a closed gate-bias loop the net gate charge integrates to 8×10⁻⁵ of the one-way charge: the charge model is conservative and the integrator preserves it.
3. Junction charge extraction — the leakage-subtracted transient integral of diode_cmc's reverse-ramp current equals ∫C(V)dV of the AC-measured capacitance (the physical charge-extraction technique) to 0.5%; C(V) is monotone under reverse bias. The leakage subtraction is itself the instructive part: the raw integral is 200× too large — reverse DC current dominates a 200 µs ramp.
4. Linear response — a PSP103 common-source stage driven with a 1 mV transient sine reproduces the .ac prediction to 2×10⁻⁶ in magnitude and 0.001° in phase (quadrature demodulation of the steady state) at 1 MHz and 10 MHz: the reactive matrix is the same matrix in both analyses.

Files

• verify_dynphys.py — the suite (9 checks); needs the VA_TEST/ corpus and skips gracefully without it.
• plot_dynphys.py — renders plots/*.png from the verify run's artifacts (run the verify first).