FAQ¶
Inputs¶
Do I pass magnitudes or fluxes?¶
Fluxes. brutus works in maggies, \(F = 10^{-0.4\,m}\), in each survey’s native system with no zero-point offset. inv_magnitude() converts magnitudes and their errors:
from brutus.utils import inv_magnitude
flux, flux_err = inv_magnitude(mag, mag_err)
What units do distances and parallaxes use?¶
Quantity |
Unit |
|---|---|
parallax (input) |
mas |
distance in |
kpc |
distance in |
pc |
\(A_V\) |
mag |
How many bands does a star need?¶
At least 4: each model has 3 free parameters (scale, \(A_V\), \(R_V\)), and the fit needs one more for a degree of freedom. With fewer than 4 valid bands, fit() warns and fits anyway. A star with none is not fitted: fit() writes it with model_idx = -99 and NaN values, warns once with the number of such stars, and fits the rest. A band is valid when its mask is true, its flux is positive, its error is positive and finite, and its error is below 0.25 mag (merr_max). Select the stars before fitting:
keep = (mask & np.isfinite(mag) & (mag_err < 0.25)).sum(axis=1) >= 4
Why does fit() fail to save my star IDs?¶
h5py cannot store unicode strings. Pass integers or byte strings as data_labels, for example np.array(ids, dtype="S").
Why does fit() ask for data_coords?¶
The default Galactic prior and the dust-map prior depend on the position on the sky. Pass data_coords, an array of \((\ell, b)\) in degrees with shape (Nstars, 2). Without coordinates, fit() applies no Galactic prior at all, and it accepts this only when you pass your own lngalprior (Priors).
Results¶
My fits look poor, but the stars are normal. Why?¶
Most often, the errors are too small. The models differ from the data by about 0.02 mag in the optical and 0.03 mag in the near-infrared, and brutus does not add this to your errors. Add it in quadrature (Photometric offsets). Judge a fit by its p-value, not by \(\chi^2 / N_{\rm bands}\) (Understanding results).
Why is the dust prior flat for some stars?¶
The Bayestar map covers only the sky north of declination −30°. Within that area, Bayestar also masks distances outside each sightline’s reliable range and pixels whose fit did not converge. Where the map is masked, the prior on \(A_V\) is uniform over avlim. Bayestar(apply_reliability_mask=False) returns the raw map values instead.
Why are my M dwarfs or massive stars fitted badly?¶
The default grid covers initial masses from 0.5 to 2.0 M☉ only. Build a grid that covers your stars (Model grids).
Why does a binary come out too close?¶
The default grid has single stars only. An unresolved binary is brighter than a single star of the same colour, so brutus places it closer. With a parallax, the fit is then often poor.
Running¶
Where are the data files?¶
The fetch_* functions in brutus.data download them from the Harvard Dataverse and return their paths. Installation lists each file and where it is stored.
How do I make fits faster?¶
Load only the bands you fit, and use a smaller grid: a narrower range of mass or metallicity, or include_postms=False for main-sequence stars. The time per star grows with the number of models (Model grids).
How do I get the same draws again?¶
Pass a seeded random generator: fit(..., rstate=np.random.default_rng(42)).