Synthetic Photometry

starkit.base.operations.imager.Photometry converts a spectrum to magnitudes in a set of filters. The filters are curves from the SVO Filter Profile Service, curves read from text files, or curves given as arrays. The magnitudes are in the Vega or the AB system. The flux is taken to be \(f_\lambda\) in erg s-1 cm-2 Å-1 at the observer; read Grid Flux Units before comparing the magnitudes of a grid with observed ones.

No package beyond the StarKit requirements is needed; the Vega spectrum for Vega magnitudes is included with StarKit.

Quick start

Filters are named by their SVO filter IDs, as listed at http://svo2.cab.inta-csic.es/theory/fps/:

from starkit.gridkit import load_grid
from starkit.base.operations.imager import Photometry

grid = load_grid('phoenix_t4000_10000_w3000_9000_r3000.h5')
model = grid | Photometry.from_grid(
    grid, filter_set=['SLOAN/SDSS.g', 'SLOAN/SDSS.r', 'Generic/Johnson.V'])
model.teff_0, model.logg_0, model.mh_0 = 6000, 4.5, 0.0
magnitudes = model()

assemble_model adds the same operation when given filter_set and, optionally, mag_type:

from starkit import assemble_model

model = assemble_model(grid, filter_set=['SLOAN/SDSS.g', 'SLOAN/SDSS.r'],
                       mag_type='ab', vrad=0.)

Filter curves

filter_set is a list whose items are SVO filter IDs, Filter objects, or both. starkit.filters.Filter holds a curve made in one of three ways:

from astropy import units as u
from starkit.filters import Filter

# From the SVO Filter Profile Service
ks = Filter.from_svo('2MASS/2MASS.Ks')

# From a text file with two columns, wavelength and transmission;
# lines starting with # are skipped
nb217 = Filter.read('nb217.txt', wavelength_unit=u.um)

# From arrays; wavelengths without a unit are taken as Angstrom
box = Filter([4000., 4010., 5990., 6000.], [0., 1., 1., 0.], name='box')

photometry = Photometry(filter_set=[ks, nb217, 'SLOAN/SDSS.r'])

Only the shape of a curve matters, so a transmission in percent gives the same magnitudes as one in fractions. The wavelengths must increase. Between its points a curve is interpolated linearly, and it is zero outside them.

Filter.from_svo downloads the curve on first use and keeps it in the astropy download cache, so later uses need no network connection. For example, SVO has 11 Keck/NIRC2 curves, taken from the Keck NIRC2 filter page: Keck/NIRC2.J, .H, .K, .Kp, .Ks, .Lp and .Ms, and the narrow-band .Hcont, .Fell (Fe II), .Brgamma and .Kcont. Other NIRC2 filters can be read with Filter.read from a tabulated curve. IDs in the form used by wsynphot, such as '2MASS/2MASS/Ks', are also accepted. wsynphot FilterSet and FilterCurve objects are accepted in filter_set, without wsynphot being installed.

Detector type

A response curve describes either the relative number of photons detected at each wavelength or the relative energy detected. The mean flux density in the filter is

\[\langle f_\lambda \rangle = \frac{\int f_\lambda S(\lambda) \lambda\,d\lambda} {\int S(\lambda) \lambda\,d\lambda}\]

for a photon-counting response \(S\), and

\[\langle f_\lambda \rangle = \frac{\int f_\lambda S'(\lambda)\,d\lambda} {\int S'(\lambda)\,d\lambda}\]

for an energy response \(S'\). The two describe the same system when \(S' = S \lambda\) (Bessell & Murphy 2012, PASP 124, 140, Appendix A). CCDs and infrared arrays count photons, so their response curves are photon-counting. A filter transmission is the fraction of the light transmitted at each wavelength, the same for photons and for energy, so a measured transmission used with such a detector is also a photon-counting response. For a photomultiplier read by integrating its current, the system response is the filter transmission multiplied by the radiant response of the photocathode, in mA/W, which is an energy response; Bessell & Murphy (2012, Section 5.4) note that the response curves generally published before CCDs were of this kind.

The detector parameter of the Filter constructor and of Filter.read is 'photon' by default and 'energy' for an energy response. Filter.from_svo takes it from the DetectorType that SVO gives for each curve, unless detector is given. SVO lists the generic Johnson U, B, V, R and I curves (Generic/Johnson.*) and the GCPD V curves measured with photomultipliers as energy responses, and the V curves of CCD instruments that include the detector response, such as Paranal/OmegaCAM.V, as photon-counting. For curves of the filter alone the SVO labels are not consistent: OSN/Circ.Johnson_V is listed as an energy response and OSN/Johnson.V as photon-counting. A measured filter transmission used with a CCD or an infrared array is photon-counting. Using the generic Johnson V curve with the wrong integral changes the magnitude of a 6000 K blackbody by 0.011 mag.

Curves that are all called Johnson V differ by more than this. For a 3500 K blackbody, the curves CAHA/Astralux.V_noatm and GCPD/Johnson.V_Landolt give magnitudes 0.103 and 0.089 mag fainter than Generic/Johnson.V; at 6000 K the differences are 0.043 and 0.036 mag. To compare with observed magnitudes, use the curve of the system the observations are calibrated to.

Zero points

mag_type='vega' gives magnitudes relative to the spectrum of Vega, with Vega at 0 mag in every filter. The spectrum is the CALSPEC spectrum alpha_lyr_stis_011.fits (Bohlin, Gordon & Tremblay 2014, PASP 126, 711), from 900 Å to 300 µm, which is included with StarKit; it is the default Vega spectrum of synphot 1.9.0. Its mean flux density is computed with the same integral as for the model spectrum. StarKit keeps the spectrum for the rest of the Python session; to use another CALSPEC spectrum, set starkit.filters.VEGA_FILE to its path or URL and call starkit.filters.vega_spectrum.cache_clear().

mag_type='ab' gives \(m = -2.5 \log_{10} (\langle f_\nu \rangle / 3631\ {\rm Jy})\), with \(\langle f_\nu \rangle = \langle f_\lambda \rangle \lambda_p^2 / c\) and the pivot wavelength

\[\lambda_p^2 = \frac{\int S \lambda\,d\lambda}{\int S / \lambda\,d\lambda} = \frac{\int S'\,d\lambda}{\int S' / \lambda^2\,d\lambda}\]

(Bessell & Murphy 2012, equation A16). Filter.zero_point(mag_type) returns the zero point of a filter in erg s-1 cm-2 Å-1.

Wavelength coverage

The spectrum must cover the filter. If more than 1% of the response of a filter, \(\int S \lambda\,d\lambda\) or \(\int S'\,d\lambda\), lies outside the wavelengths of the spectrum, Photometry stops with a ValueError that names the filter. Photometry.from_grid checks this when it is created, against the wavelengths of the grid.

Photometry.from_grid interpolates the curves onto the grid wavelengths once and uses them for every spectrum. A wavelength shift applied by an earlier operation, such as DopplerShift, therefore does not move the spectrum relative to the filters. Photometry created without grid wavelengths interpolates the curves onto the wavelengths it receives at each evaluation.

Observed magnitudes in a fit

PhotometryColorLikelihood compares the colors between consecutive filters, which do not depend on the flux scale of the grid, using the covariance of adjacent colors, which share a magnitude. RelativePhotometryLikelihood compares the magnitudes themselves. Both take the observed magnitudes and their uncertainties, in the order of the filters:

from starkit.fitkit.likelihoods import PhotometryColorLikelihood

likelihood = model | PhotometryColorLikelihood(
    [17.64, 17.22, 17.03], [0.035, 0.053, 0.066])

An object with the attributes magnitudes and magnitude_uncertainties, such as a wsynphot MagnitudeSet, is also accepted.

Accuracy

The magnitudes from Photometry were compared with those of Observation.effstim in synphot 1.9.0, using the same SVO curves and Vega spectrum, for two spectra: the spectrum of the example grid phoenix_t4000_10000_w3000_9000_r3000.h5 at 6000 K, log g = 4.5 and [M/H] = 0, and a 3000 K blackbody tabulated from 0.9 to 6 µm.

Filters

Detector

Vega (mag)

AB (mag)

SLOAN/SDSS.g, SLOAN/SDSS.r (6000 K grid)

photon

< 0.000001

0.000071

Paranal/OmegaCAM.V (6000 K grid)

photon

< 0.000001

0.000071

Generic/Johnson.B, .V (6000 K grid)

energy

0.000005

0.000066

11 Keck/NIRC2 curves (3000 K blackbody)

photon

0.000035

0.000066

The columns give the largest absolute difference between StarKit and synphot. Of the AB differences, 0.000066 mag is the difference between the AB zero point of 3631 Jy used by StarKit and the 3630.78 Jy (AB magnitude defined by the constant -48.60) used by synphot through astropy. StarKit’s Vega zero points for the NIRC2 curves, converted to \(f_\nu\) at the pivot wavelength, agree with the zero points listed by SVO to 0.069%.

Two settings were needed on the synphot side. synphot holds the end values of a curve that does not end at zero constant at all wavelengths beyond it; StarKit sets the curve to zero there. The curves were therefore given to synphot with fill_value=0; with synphot’s default, its magnitudes of the 3000 K blackbody in the narrow-band NIRC2 filters change by up to 0.645 mag. synphot also treats every curve as photon-counting, so the energy responses were interpolated onto a fine sampling and then divided by the wavelength. Dividing the Johnson curves, which are sampled every 100 to 200 Å, before interpolating them changes the synphot magnitudes by up to 0.0013 mag.

Changes from StarKit 0.8 and wsynphot

StarKit 0.8 and earlier computed the magnitudes with wsynphot. The changes that affect results are:

  • The Vega spectrum. wsynphot used the CALSPEC model alpha_lyr_mod_002.fits. Vega magnitudes computed with it differ from the present ones by -0.010 mag in Generic/Johnson.V, -0.004 mag in SLOAN/SDSS.r and +0.020 mag in 2MASS/2MASS.Ks (wsynphot minus present). Setting starkit.filters.VEGA_FILE to https://archive.stsci.edu/hlsps/reference-atlases/cdbs/calspec/alpha_lyr_mod_002.fits and calling starkit.filters.vega_spectrum.cache_clear() reproduces the wsynphot zero points.

  • The detector type. wsynphot from October 2017 to January 2022 used the photon-counting integral for every curve, including the curves SVO lists as energy responses, such as the generic Johnson filters; earlier versions used the energy integral for every curve. Since February 2022 it used the SVO detector type, as StarKit does now.

  • Observed magnitudes. The likelihoods take arrays of magnitudes and uncertainties instead of a wsynphot MagnitudeSet, which is still accepted.