Curvature test¶

The curvature() method tests for spectral curvature (deviation from a power-law energy spectrum) for a given source via likelihood ratio test.

The likelihood is maximized under three different spectral hypotheses for the source in question:

PowerLaw

LogParabola, and

PLSuperExpCutoff4.

For the first two models, all parameters except for the pivot energy are fit. For the power law with super-exponential cutoff, the parameter Index2 (also referred to as b) is fixed to 0.6667 (the recommended value for pulsars from 4FGL-DR3) by default. The user may supply a different value of Index2 and/or allow its value to float during the likelihood fit. The latter is only recommended for sources with high detection significance.

The likelihood ratios are calculated with the PowerLaw fit as the baseline, e.g. \(TS_{LP} = -2 \left( \mathrm{ln}\left(L_{PL}\right) - \mathrm{ln}\left(L_{LP}\right) \right)\). The LogParabola and PLSuperExpCutoff4 model with fixed Index2 have three free parameters each compared to two for the baseline model, and both models contain the baseline model as a special case. In the absence of spectral curvature, the likelihood ratios defined as above should thus follow a chi2 distribution with one degree of freedom.

The PLSuperExpCutoff4 model with free Index2 has four free parameters and its associated likelihood ratio should thus follow a chi2 distribution with two degrees of freedom in the absence of curvature.

Note that the PLSuperExpCutoff4 model with free Index2 contains both the LogParabola and PLSuperExpCutoff4 model with fixed Index2 as special cases. PLSuperExpCutoff4 with Index2=0 is equivalent to LogParabola.

Warning

The likelihood fits within the curvature() function sometimes fail to find the global minima. This can lead to an over- or under-estimate of the curvature TS. If in doubt, please perform dedicated fits with your own starting values and check fit quality.

Usage¶

The cuvature() method is executed by passing the name of a source in the ROI as a single argument. Additional keyword arguments can also be provided to override the default configuration of the method:

# Run curvature test with default settings
sed = gta.curvature('sourceA')

# Override the value for Index2 and free said parameter
sed = gta.sed('sourceA', Index2=0.2, free_Index2=True)

The return value of curvature() is a dictionary with the results of the analysis. The contents of the output dictionary are documented in Curvature Dictionary.

Curvature Dictionary¶

The following table describes the contents of the curvature() output dictionary:

*curvature* Output Dictionary¶
Key	Type	Description
`lp_ts_curv`	`float`	Likelihood ratio for `LogParabola` fit.
`ple_ts_curv`	`float`	Likelihood ratio for `PLSuperExpCutoff4` fit with fixed `Index2`.
`ple_free_ts_curv`	`float`	Likelihood ratio for `PLSuperExpCutoff4` fit with free `Index2`.
`loglike_pl`	`float`	Likelihood ratio for `PowerLaw` fit.
`loglike_lp`	`float`	Likelihood ratio for `LogParabola` fit.
`loglike_ple`	`float`	Likelihood value for `PLSuperExpCutoff4` fit with fixed `Index2`.
`loglike_ple_free`	`float`	Likelihood value for `PLSuperExpCutoff4` fit with free `Index2`.
`Index2`	`float`	(starting) value of `Index2` for `PLSuperExpCutoff4` fit.

Configuration¶

The default configuration of the method is controlled with the Curvature section of the configuration file. The default configuration can be overriden by passing the option as a kwargs argument to the method.

*curvature* Options¶
Option	Default	Description
`Index2`	0.6667	Index2 parameter for PLSuperExpCutoff4 fit.
`free_Index2`	False	Whether or not to perform curvature test with PLSuperExpCutoff4 model with free Index2`` parameter

Reference/API¶

GTAnalysis.curvature(name, **kwargs)[source]

Test whether a source shows spectral curvature by comparing the likelihood ratio of PowerLaw and LogParabola spectral models.

Parameters:

name (str) – Source name.
Index2 (double) – Exponent for super-exponential cutoff PL test. Default: 2/3
free_Index2 (bool) – Test also super-exponential cutoff PL with free index. Only recommended for sources with high TS.