Models¶
By design, the specification of models is done in its own namespace: model
.
All model
classes are subclasses of model.model
. In the following
sections, I detail what each model can be used for and the mathematics behind
it.
SNooPy
now has two different parametrizations for the shape of the
light-curves. One is the traditional of
[Phillips 1999]_ and the newer ‘color-stretch’ introduced in
[Burns+2011], which more consistently deals with fast-declining events. Note
that the EBV_model
will only work with -based
templates.
EBV_model¶
This light-curve model is a variation of that given by [Prieto+2006]. It is most useful for determining a distance using the Phillips relation and correcting for dust extinction in the host. We compare the observed data in filters to the following model:
where is the observed magnitude in band ,
is the observed time relative to B maximum, is the
de-redshifted time relative to B maximum, is the decline
rate parameter [Phillips1999], is the absolute
magnitude in filter in the rest-frame of the supernova,
and are the reddening due to
galactic foreground and host galaxy, respectively, and
are the total-to-selective absorptions for filters and
, respectively, is the cross-band k-correction from
rest-frame to observed filter . Note that the k-corrections
depend on the epoch and can depend on the host and galaxy extinction (as
these modify the shape of the spectral template). In the fitting, one has the
choice of the spectral template of [Nugent+2002] or [Hsiao+2007]. The latter
is the default. You can choose to allow the K-corrections to vary during the
fit, keep them fixed, or not use them at all. See the parameters for the
fit()
function.
The template can be generated from
the code of citet{2006ApJ...647..501P} or citet{Burns2011}. In
the former case, you will be fitting to rest-frame
[1] while in the latter case, you can fit to from the CSP
data [Contreras+2010]. You can mix and match which template
generator you use: it is all a matter of which filter you choose for
the restbands
instance variable (see Doing the Fit).
In all, this model fits 4 parameters: Tmax
, dm15
,
EBVhost
, and DM
.
Note that to determine the host reddening, you need to fit at least two distinct restframe filters. For now, I’ve left the galactic and host reddening laws ( and ) as member variables rather than parameters to be fit [2]. This could change in the future.
EBV_model2¶
This is the same model as EBV_model, except that it can fit both
- and -based templates. And instead of the calibration
presented in [Folatelli+2010], the calibration from the
upcoming CSP analysis paper [Burns+2011] is used. When using
the choose_model
function to select this model, you can
specify stype='dm15'
or stype='st'
to select the
two different light-curve parameters.
max_model¶
An alternative to assuming some kind of relationship between the different
filters (i.e., some kind of reddening law as is done in the EBV_model
),
one can simply fit the maximum magnitude of each light-curve. This
is how the max_model
model works. It uses the same light-curve
templates as EBV_model
, but instead fits the following model:
where , the peak magnitude in filter is now a parameter.
Of course, depending on the number of filters you fit, you will have
that many parameters. Note that K-corrections and Milky-way
extinction are performed exactly as in the EBV_model
case.
Therefore, for filters, max_model will have parameters:
Tmax
, dm15
, and emph{f}``max``, where
emph{f} is the rest-band filter name (so if, for instance, you fit
an observed light-curve to restframe , there will be a Bmax
parameter). As with the EBV_model2
model, you can choose
which light-curve parameter you wish to use by specifying the stype
argument to choose_model()
.
NOTE: Please be aware that any non-linear fitter will only
find a emph{local} minimum in . It is up to you, the user,
to try different starting points in parameter space and see which
one gives the overall best-fit. After each fit, the member variable
rchisq
is updated with the reduced-. You can therefore
use this to discriminate between different solutions.
color_model¶
In this model, intrinsic colors from [Burns+2014] are used to infer the amout of extinction as well as the shape reddening law . Mathematically, the model being fit to the light-curves is:
where is the de-reddened and K-corrected maximum (treated as a free parameter) and is the intrinsic color, which is a function of . In [Burns+2014] it is modeled as a 2nd degree polynomial in . All other variables have the same meaning as in previous models. The model has 5 free parameters: , , , , and . Note that the distance modulus is not included in this model.
A major complication of this model is that two parameters,
and , appear as multiplicative factors in the same term.
At best, they will be highly covariant. At worst (for low values of
), the model becomse insensitive to . For this reason,
it may be necessary to impose priors on . Two such priors
were introduced in [Burns+2014]: a Gaussian mixture model that
applies to all SNeIa, and a binned prior, where a separate Gaussian
prior is applied to the SNIa depending on its value of .
Because SNooPy uses the LM least-squares algorithm by default, there
is no natural way to incorporate these priors using the standard fit()
routine. Instead, use the fitMCMC()
routine, which fits the
light-curves using Markov Chain Monte Carlo and allows priors to be
specified on parameters. In this case, use the keyword argument rvprior='mix'
for the Gaussian mixture model, or rvprior='bin'
for the
binned prior. See section Doing the Fit for more details.
[Phillips1999] | Phillips, M.M., AJ, 118, 1766 (1999) http://adsabs.harvard.edu/abs/1999AJ....118.1766P |
[Nugent+2002] | Nugent et al., PASP, 114, 803 (2002). http://adsabs.harvard.edu/abs/2002PASP..114..803N |
[Hsiao+2007] | Hsiao et al., ApJ, 663, 1187 (2007). http://adsabs.harvard.edu/abs/2007ApJ...663.1187H |
[Stritzinger+2005] | Stritzinger et al., PASP, 117, 810 (2005) http://adsabs.harvard.edu/abs/2005PASP..117..810S |
[Contreras+2010] | Contreras et al., AJ, 139, 519 (2010). http://adsabs.harvard.edu/abs/2010AJ....139..519C |
[Burns+2011] | (1, 2) Burns et al., AJ, 141, 19B (2011). http://adsabs.harvard.edu/abs/2011AJ....141...19B |
[Burns+2014] | (1, 2, 3) Burns et al., ApJ, 789, 32B (2014). http://adsabs.harvard.edu/abs/2014ApJ...789...32B |
[1] | The ‘s’ subscript refers to ‘standard’, which is to say the Bessel filters from [Stritzinger+2005] |
[2] | So far, data I’ve analyzed hasn’t been good enough to distinguish between reddening laws. |