lemonergy/ wiki

Worked examples

Synthetic pulse parameter recovery

A separately authored pulse experiment distinguishes forward-model checking from parameter recovery and independent holdout testing.

Document created Updated
Draft explanationSource record reviewed

Separate fitting from forward solving

The prepared two-RC graph solves a fixed forward model. It does not estimate its own resistance or time constants.

The original calibration sequence is 12 s rest, 6 s at 2 A, 36 s rest, 96 s at 6 A, 120 s rest, 24 s at 4 A and 96 s rest. The independently shaped holdout is 18 s rest, 72 s at −3 A, 96 s rest, 48 s at 4 A and 120 s rest. The holdout includes charging and is not an amplitude-scaled copy of the calibration waveform.

Both branches use the newly chosen synthetic input contract. The browser-safe source evaluates closed-form responses by superposition of all current edges. End-of-step fitting samples use the left limit, while event-aware forward checks use the right limit.

The earlier externally seeded fitting benchmark and its numerical payloads have been retired. The native forward report on this site is not evidence of optimizer recovery or measured accuracy. A fitting implementation must separately retain its admission checks, initial guesses, recovered parameters, sensitivities and fixed holdout result.

Two-RC identification literature provides scientific context; it is not the source of these chosen numbers.

Related papers and references