This is a fixed-text formatted version of a Jupyter notebook.

Try online You can also contribute with your own notebooks in this GitHub repository. Source files: sed_fitting_gammacat_fermi.ipynb | sed_fitting_gammacat_fermi.py

Flux point fitting in Gammapy¶

Introduction¶

In this tutorial we’re going to learn how to fit spectral models to combined Fermi-LAT and IACT flux points.

The central class we’re going to use for this example analysis is:

In addition we will work with the following data classes:

And the following spectral model classes:

Setup¶

In [1]:

%matplotlib inline
import numpy as np
import matplotlib.pyplot as plt

In [2]:

from astropy import units as u
from astropy.table import vstack
from gammapy.spectrum.models import PowerLaw, ExponentialCutoffPowerLaw, LogParabola
from gammapy.spectrum import FluxPointFitter, FluxPoints
from gammapy.catalog import SourceCatalog3FGL, SourceCatalogGammaCat, SourceCatalog3FHL


For this analysis we choose to work with the source ‘HESS J1507-622’ and the associated Fermi-LAT sources ‘3FGL J1506.6-6219’ and ‘3FHL J1507.9-6228e’. We load the source catalogs, and then access source of interest by name:

In [3]:

fermi_3fgl = SourceCatalog3FGL()
fermi_3fhl = SourceCatalog3FHL()
gammacat = SourceCatalogGammaCat()

In [4]:

source_gammacat =  gammacat['HESS J1507-622']
source_fermi_3fgl = fermi_3fgl['3FGL J1506.6-6219']
source_fermi_3fhl = fermi_3fhl['3FHL J1507.9-6228e']


The corresponding flux points data can be accessed with .flux_points attribute:

In [5]:

flux_points_gammacat = source_gammacat.flux_points
flux_points_gammacat.table

Out[5]:

Table length=6
e_refdndednde_errndnde_errp
TeV1 / (cm2 s TeV)1 / (cm2 s TeV)1 / (cm2 s TeV)
float32float32float32float32
0.86092.29119e-128.70543e-138.95502e-13
1.561516.98172e-132.20354e-132.30407e-13
2.763751.69062e-136.7587e-147.18838e-14
4.89167.72925e-142.40132e-142.60749e-14
9.988581.03253e-145.06315e-155.64195e-15
27.04037.44987e-165.72089e-167.25999e-16

In the Fermi-LAT catalogs, integral flux points are given. Currently the flux point fitter only works with differential flux points, so we apply the conversion here.

In [6]:

flux_points_3fgl = source_fermi_3fgl.flux_points.to_sed_type(
sed_type='dnde',
model=source_fermi_3fgl.spectral_model,
)
flux_points_3fhl = source_fermi_3fhl.flux_points.to_sed_type(
sed_type='dnde',
model=source_fermi_3fhl.spectral_model,
)


Finally we stack the flux points into a single FluxPoints object and drop the upper limit values, because currently we can’t handle them in the fit:

In [7]:

# stack flux point tables
flux_points = FluxPoints.stack([
flux_points_gammacat,
flux_points_3fhl,
flux_points_3fgl
])

# drop the flux upper limit values
flux_points = flux_points.drop_ul()


Fitter Setup¶

We initialze the fitter object with the 'chi2assym' statistic, because we have assymmetric errors on the flux points. As optimizer we choose the 'simplex' algorithm and to estimate the errors we use 'covar' method:

In [8]:

fitter = FluxPointFitter(
stat='chi2assym',
optimizer='simplex',
error_estimator='covar',
)


Power Law Fit¶

In [9]:

pwl = PowerLaw(
index=2. * u.Unit(''),
amplitude=1e-12 * u.Unit('cm-2 s-1 TeV-1'),
reference=1. * u.TeV
)


After creating the model we run the fit by passing the 'flux_points' and 'pwl' objects:

In [10]:

result_pwl = fitter.run(flux_points, pwl)


And print the result:

In [11]:

print(result_pwl['best-fit-model'])

PowerLaw

Parameters:

name     value     error         unit      min max frozen
--------- --------- --------- --------------- --- --- ------
index 1.950e+00 2.656e-02                 nan nan  False
amplitude 1.248e-12 1.599e-13 1 / (cm2 s TeV) nan nan  False
reference 1.000e+00 0.000e+00             TeV nan nan   True

Covariance:

name/name   index   amplitude
--------- --------- ---------
index  0.000706 -2.25e-15
amplitude -2.25e-15  2.56e-26


As a quick check we print the value of the fit statistics per degrees of freedom as well:

In [12]:

print(result_pwl['statval/dof'])

2.5038842921602655


Finally we plot the data points and the best fit model:

In [13]:

ax = flux_points.plot(energy_power=2)
result_pwl['best-fit-model'].plot(energy_range=[1e-4, 1e2] * u.TeV, ax=ax, energy_power=2)
result_pwl['best-fit-model'].plot_error(energy_range=[1e-4, 1e2] * u.TeV, ax=ax, energy_power=2)
ax.set_ylim(1e-13, 1e-11)

Out[13]:

(1e-13, 1e-11)


Exponential Cut-Off Powerlaw Fit¶

Next we fit an exponential cut-off power law to the data.

In [14]:

ecpl = ExponentialCutoffPowerLaw(
index=2. * u.Unit(''),
amplitude=1e-12 * u.Unit('cm-2 s-1 TeV-1'),
reference=1. * u.TeV,
lambda_=0. / u.TeV
)


We run the fitter again by passing the flux points and the ecpl model instance:

In [15]:

result_ecpl = fitter.run(flux_points, ecpl)
print(result_ecpl['best-fit-model'])

ExponentialCutoffPowerLaw

Parameters:

name     value     error         unit      min max frozen
--------- --------- --------- --------------- --- --- ------
index 1.876e+00 4.388e-02                 nan nan  False
amplitude 1.932e-12 3.873e-13 1 / (cm2 s TeV) nan nan  False
reference 1.000e+00 0.000e+00             TeV nan nan   True
lambda_ 6.147e-02 5.795e-02         1 / TeV nan nan  False

Covariance:

name/name   index   amplitude lambda_
--------- --------- --------- --------
index   0.00193 -1.35e-14 -0.00178
amplitude -1.35e-14   1.5e-25 1.73e-14
lambda_  -0.00178  1.73e-14  0.00336

In [16]:

print(result_ecpl['statval/dof'])

2.001353939388082


We plot the data and best fit model:

In [17]:

ax = flux_points.plot(energy_power=2)
result_ecpl['best-fit-model'].plot(energy_range=[1e-4, 1e2] * u.TeV, ax=ax, energy_power=2)
result_ecpl['best-fit-model'].plot_error(energy_range=[1e-4, 1e2] * u.TeV, ax=ax, energy_power=2)
ax.set_ylim(1e-13, 1e-11)

Out[17]:

(1e-13, 1e-11)


Log-Parabola Fit¶

Finally we try to fit a log-parabola model:

In [18]:

log_parabola = LogParabola(
alpha=2. * u.Unit(''),
amplitude=1e-12 * u.Unit('cm-2 s-1 TeV-1'),
reference=1. * u.TeV,
beta=0. * u.Unit('')
)

In [19]:

result_log_parabola = fitter.run(flux_points, log_parabola)
print(result_log_parabola['best-fit-model'])

LogParabola

Parameters:

name     value     error         unit      min max frozen
--------- --------- --------- --------------- --- --- ------
amplitude 1.954e-12 2.797e-13 1 / (cm2 s TeV) nan nan  False
reference 1.000e+00 0.000e+00             TeV nan nan   True
alpha 2.140e+00 7.377e-02                 nan nan  False
beta 4.947e-02 1.757e-02                 nan nan  False

Covariance:

name/name amplitude  alpha     beta
--------- --------- -------- --------
amplitude  7.82e-26 1.79e-15 2.19e-15
alpha  1.79e-15  0.00544  0.00112
beta  2.19e-15  0.00112 0.000309

In [20]:

print(result_log_parabola['statval/dof'])

1.5798902196277185

In [21]:

ax = flux_points.plot(energy_power=2)
result_log_parabola['best-fit-model'].plot(energy_range=[1e-4, 1e2] * u.TeV, ax=ax, energy_power=2)
result_log_parabola['best-fit-model'].plot_error(energy_range=[1e-4, 1e2] * u.TeV, ax=ax, energy_power=2)
ax.set_ylim(1e-13, 1e-11)

Out[21]:

(1e-13, 1e-11)


Exercises¶

• Fit a PowerLaw2 and ExponentialCutoffPowerLaw3FGL to the same data.
• Fit a ExponentialCutoffPowerLaw model to Vela X (‘HESS J0835-455’) only and check if the best fit values correspond to the values given in the Gammacat catalog

What next?¶

This was an introduction to SED fitting in Gammapy.

• If you would like to learn how to perform a full Poisson maximum likelihood spectral fit, please check out the spectrum pipe tutorial.
• If you are interested in simulation of spectral data in the context of CTA, please check out the spectrum simulation cta notebook.
• To learn more about other parts of Gammapy (e.g. Fermi-LAT and TeV data analysis), check out the other tutorial notebooks.
• To see what’s available in Gammapy, browse the Gammapy docs or use the full-text search.
• If you have any questions, ask on the mailing list .