Add Jiles-Atherton B-H hysteresis to inductor

[esaruoho]

Summary

• Adds real B-H hysteresis to inductors using a normalized Jiles-Atherton model — captures coercivity, remanence, and minor-loop memory, not just saturation.
• Extends the saturable-core infrastructure introduced in Add saturable core model for inductors and transformers #308 (same element, same dump type, same degradation pattern) so there is no new element and no new UI surface beyond two fields.
• Fully backwards-compatible: coerciveCurrent = 0 degrades to Add saturable core model for inductors and transformers #308 saturation; saturationCurrent = 0 as well degrades to the original linear inductor.

Why

Critics have pointed to "linear-biased, can't do real B-H hysteresis cleanly" as a gap. #308 closed half the gap (single-valued saturation). This closes the other half — history-dependent magnetization, which is what you need to see real hysteresis loops, coercivity-driven demagnetization, and realistic flyback/Joule-Thief core loss.

How it works

Internal dimensionless Jiles-Atherton state:

• h = I / Ic (normalized field)
• Man(he) = coth(he/a) - a/he (Langevin, Ms = 1)
• he = h + alpha*m
• dMirr/dh = (Man - Mirr) / (delta - alpha*(Man - Mirr)) with delta = sign(dh) and k normalized to 1
• m = Mirr + c*(Man - Mirr)

Fixed internal constants a = 1/3, alpha = 1e-3 are tuned so the virgin small-signal inductance equals the user-specified L0. The user only exposes:

• Coercive Current (A) — 0 disables hysteresis (default)
• Hysteresis Reversibility (0-1) — JA c, default 0.1

State is advanced once per timestep in startIteration() (not per Newton iteration), following the capacitor-charge pattern — Newton sees a linearized inductor per step. Companion resistance is derived from the incremental slope dm/dh, with a [0.01, 3.0] floor/cap and a |denom| >= 1e-3 clamp on the JA denominator to avoid singularities at loop tips.

Backwards compatibility

• Text dump: new tokens appended at the end, read inside the existing try/catch block. Old circuit files load unchanged.
• XML: ich and hrev attributes are only emitted when hysteresis is active. Old XML files round-trip unchanged.
• All existing Inductor.setup() callers (RelayCoilElm, RelayElm, DCMotorElm) use the 3-arg overload which is preserved; the 6-arg overload is new.

Files changed

• Inductor.java — JA state, advanceHysteresis(), extended startIteration() / stamp() / nonLinear() / resetTo(), new 6-arg setup()
• InductorElm.java — two new EditInfo slots, dump/undump tail tokens, XML attrs, info-panel additions
• war/circuits/hysteresis-loop.txt — minimal demo (AC source, inductor, 1-ohm sense)

Scope

In: InductorElm B-H hysteresis, 2 UI params, backwards-compat dump, demo circuit.

Deferred (future PRs):

• TransformerElm hysteresis (coupled-state bookkeeping warrants its own review)
• Core-loss energy as power output
• Temperature dependence of Ms

Happy to simplify the model (fewer params, different formulation) if you prefer a lighter touch — JA is the SPICE-standard but the framework here can host a simpler Tellinen-style atan() model just as easily.

Test plan

• Linear check: Ic = 0, Isat = 0 -> identical behavior to current inductor
• Saturation check: Ic = 0, Isat > 0 -> identical to Add saturable core model for inductors and transformers #308
• Hysteresis check: AC-driven inductor shows asymmetric I vs V on sine excitation; loop width ~ 2*Ic
• Reset clears M to 0 (virgin curve on next excitation)
• XML and text round-trip preserve the two new parameters
• Joule Thief / flyback topologies still converge (degrades to saturation when Ic = 0)