Multi-band light curves

When you have the same star observed in several filters (e.g. g and r), you can model all bands jointly instead of running each separately. A joint model shares the period across bands while letting each band have its own amplitude and offset, which boosts sensitivity when no single band is well-sampled.

Multi-band is supported by GLS, BLS, and MHAOV. (The other methods take a single band.)

Building a MultiBandLightCurve

From a {band: LightCurve} mapping:

from cuperiod import LightCurve, MultiBandLightCurve

lc_g = LightCurve.from_arrays(t_g, mag_g, err_g)
lc_r = LightCurve.from_arrays(t_r, mag_r, err_r)

mb = MultiBandLightCurve.from_light_curves({"g": lc_g, "r": lc_r})
print(mb.band_names, mb.n_bands)       # ('g', 'r'), 2

Or split one long-format table on a band/filter column:

mb = MultiBandLightCurve.from_dataframe(df, band_column="filter")

The band column is auto-detected (band, filter, passband, fid, …) when you don’t name it, and the time/value/error columns resolve exactly as for a single light curve (with ColumnMap if needed).

Running a multi-band periodogram

Pass the MultiBandLightCurve to periodogram() like any other input:

pg = cup.periodogram(mb, "GLS")        # VanderPlas & Ivezić shared-phase model
print(pg.best_period())

The result is an ordinary Periodogram, so peak finding, the raw spectrum, and serialization all work as in Reading results.

pg = cup.periodogram(mb, "MHAOV")      # pooled F-statistic across bands
pg = cup.periodogram(mb, "BLS")        # joint box search across bands

Asking a single-band method for a multi-band run raises a clear error, so you’ll know immediately if a method doesn’t support it.

A note on the default grid

For multi-band input the default grid is built from the stacked baseline of all bands (their combined time span), so the frequency resolution reflects all your data. You can still override it with grid= (Tuning: settings & grids).

---

Next: Batch processing — scaling to many light curves.