optimize_solution

ell1fit.fitting.optimize_solution(observations, setup, nsteps=1000, minimize_first=False, outroots=('out',), reference_phases=None, sampler='emcee', nlive=1000, dlogz=0.1, workers=0)

Optimize and sample pulsar timing parameters for multiple event files.

Workflow: 1. Build a posterior from priors + profile likelihood. 2. Optionally run deterministic minimization for a starting point. 3. Run MCMC with safe_run_sampler(), or another backend – see sampler. 4. Produce diagnostic phaseogram comparisons and return summary fields.

Parameters are handled in local coordinates: for each fitted parameter, physical = local * factor + initial.

Parameters:

sampler{“emcee”, “nuts”, “nested”}, optional

Which posterior-exploration backend to run. "emcee" (the default) is safe_run_sampler(), unchanged from before this parameter existed. "nuts" is ell1fit.nuts_sampling.run_nuts(), needing pip install ell1fit[nuts]. "nested" is ell1fit.nested_sampling.run_nested(), needing pip install ell1fit[nested] – it is the only one of the three that reports log_evidence, which is what an eccentricity Bayes factor needs and neither of the other two can give.

nlive, dlogz, workersoptional

Only consulted when sampler="nested". Nested sampling can miss a narrow mode entirely and still report a confident, wrong evidence, so nlive matters more than it looks like it should – see ell1fit.nested_sampling.PEAK_SHORTFALL_GATE. workers – see ell1fit.nested_sampling.run_nested() – spreads likelihood evaluations across processes; 0 (the default) runs single-process.

reference_phaseslist of np.ndarray or None, optional

Phases to draw in the left-hand panel of the comparison phaseograms – the “before” of a before-and-after. Pass the phases of the solution the run started from.

This must be supplied by the caller rather than derived here. The obvious derivation, evaluating the posterior at local coordinates zero, silently means “whatever setup.baseline_values currently holds”, which is only the starting solution as long as nothing has re-centred the baseline in between. Iterative template refinement does exactly that, and the comparison then comes out as the refined solution against itself: two identical panels, and a diagnostic that always looks perfect. Defaults to the local-zero solution when omitted, which is correct only when no refinement has run.

Returns:

dict

Aggregated result dictionary containing posterior summaries, initial values, scaling factors, and copied model metadata.