conatus.expert.ThresholdBisection

class conatus.expert.ThresholdBisection(fate, *, rtol, scale='linear', known=None, log=None)

Bases: object

Bisection on one amplitude (e.g. the seed energy E0 of a minimal seed) between a value that relaminarizes and one that transitions: the transition threshold as a bracket ]lo, hi], never as a single guessed number.

fate(a) -> -1 (relaminarizes) / +1 (transitions) / 0 (undecided) is the user’s declared verdict for amplitude a (e.g. one forward DNS of the seed rescaled to E0 = a, classified with declared thresholds); an undecided run stops the bisection (UndecidedTrajectory), it is never counted on either side. known gives fates already computed (e.g. by finished jobs): they are used first (no run is repeated, an end of [lo, hi] already settled by a known fate is not recomputed) and checked for consistency (a transitioning amplitude below a relaminarizing one is refused: the threshold would not be a single crossing on this family). Bisection is in the middle of the bracket (scale= "linear") or of its logarithm ("log") until (hi - lo) / hi <= rtol.

tb = ThresholdBisection(fate, rtol=1e-3, known={1.576778e-7: -1, 1.60e-7: +1}) lo, hi = tb.run(1.5e-7, 1.7e-7); print(tb.explain())

Parameters:

• fate (Callable)

• rtol (float)

• scale (str)

• known (dict | None)

run(lo, hi)

Bracket ]lo, hi] of the threshold in [lo, hi]: lo must relaminarize and hi transition. Known fates inside [lo, hi] narrow the bracket first; an end is run only when no known fate settles it (monotonicity: a relaminarizing amplitude >= lo settles lo).

Parameters:

• lo (float)

• hi (float)

Return type:

tuple[float, float]

explain()

Return type:

str