Model file

The model file is a simple Python file that, at minimum, needs to contain the following information:

Names and prior distributions of the signal parameters.
Parametrized form of the signal, which can either be stochastic (and parametrized via its power-spectrum), or deterministic (and parametrized as a times eries).

In the following, we will explain how these two quantities are defined in the model file.

Priors¶

The priors for the signal parameters are defined via the parameters dictionary. The keys of this dictionary must be strings, which will be used as names for the model parameters. The values of this dictionary are [ENTERPRISE Parameter][enterprise.signals.parameter.Parameter] objects, the user can create these object via the prior helper function that can be imported from the ptarcade.models_utils module. The first argument that the user needs to pass to the prior function is a string with the name of the prior class they want to use for that parameter, the remaining arguments are used to set the attributes of the prior. By default, parameters are assumed to be common across all pulsars. If the user wants to define a pulsar-dependent parameter, this can be done by passing common=False as a keyword argument.

Priors Example

The parameters dictionary of a model described by the parameters \(a\) and \(b\), which are common among all the pulsars, will look as follows for different choices of the priors:

Uniform Priors1D Normal Priors2D Normal PriorsExponential Priors

parameters = {'a' : prior("Uniform", 0, 1), 'b' : prior("Uniform", 0, 1)} # (1)!

In this case, we have chosen uniform priors in the range [0,1] for both parameters.

parameters = {'a' : prior("Normal", mu=1, sigma=1), 'b' : prior("Normal", 1, 1)} # (1)!

In this case, we have chosen 1D normal priors with unit mean and variance for both parameters.

mu = [1, 1]
cov = [[1, 0.1],[0.1, 1]]

parameters = {'a_b' : prior("Normal", mu, cov, size=2)} # (1)!

In this case, we have chosen a joint 2D normal prior for the model parameters, which are grouped in a single two dimensional parameter called a_b.

parameters = {'a' : prior("LinearExp", 1, 1), 'b' : prior("LinearExp", 1, 1)} # (1)!

In this case, we have assumed

Constructing Priors

Notice, how we used both positional and keyword arguments: Both are allowed. These arguments correspond to the functions defined in either [enterprise.signals.parameter][] or ptarcade.models_utils. Below are links to all supported parameters:

[enterprise.signals.parameter.Normal][]
[enterprise.signals.parameter.Uniform][]
[enterprise.signals.parameter.TruncNormal][]
[enterprise.signals.parameter.LinearExp][]
[enterprise.signals.parameter.Constant][]
ptarcade.models_utils.Gamma.

Common Parameters vs. Pulsar-Dependent

Parameters are assumed to be common by default. If they are pulsar-dependent, you must pass common=False as a keyword argument to prior. For example, if we want to set the b parameters of previous examples to be pulsar-dependent, we can do that as follows

parameters = {'a' : prior("Uniform", 0, 1), 'b' : prior("Uniform", 0, 1, common=False)} # (1)!

In this example, b is a pulsar-dependent parameter. By default, the parameters are common to all pulsars in the PTA.

Stochastic Signals¶

Stochastic signals are defined via the spectrum function. The first parameter of this function should be named f and it is supposed to be a NumPy array containing the frequencies (in units of Hz) at which the spectrum will be evaluated. The names of the remaining parameters should match the keys of the parameters dictionary. The spectrum function should return a NumPy array containing the value of \(h^2\Omega_{\mathrm{GW}}\) at each of the frequencies in f.

Stochastic Signal Example

The spectrum function for a model with

\[ h^2\Omega_{\rm GW}(f) = \frac{a}{1+b/f} \]

is given by

def spectrum(f, a, b):

    return a * 1 / (1 + b/f)

Deterministic Signals¶

Deterministic signals are defined via the signal function. The first parameter of this function should be named toas and it is supposed to be a NumPy array containing the times of arrival (TOAs) (in units of seconds) at which the deterministic signal will be evaluated. The name of the remaining parameters should match the keys of the parameters dictionary. The signal function should return a NumPy array with the same dimensions as toas containing the value of the induced shift for each TOA contained in toas.

Deterministic Signal Example

For a deterministic signal,

\[ s(t) = a\sin(b t), \]

the signal is given by

def signal(toas, a, b):

    return a * numpy.sin(b * toas)

Model File Example

StochasticDeterministic

This is a model file for a stochastic signal with a broken power-law spectrum,

\[ h^2\Omega_{\rm GW}(f) = A_* \frac{f/f_*}{1+f^2/f_*^2}, \]

whose parameters, \(A_*\) and \(f_*\), are assumed to have a log-uniform prior between \([10^{-14},10^{-6}]\) and \([10^{-10},10^{-6}]\), respectively.

from ptarcade.models_utils import prior

parameters = {
            'log_A_star' : prior("Uniform", -14, -6),
            'log_f_star' : prior("Uniform", -10, -6)
            }

def S(x):
    return x / (1 + x**2)

def spectrum(f, log_A_star, log_f_star):
    A_star = 10**log_A_star
    f_star = 10**log_f_star

    return A_star * S(f/f_star)

This is a model file for a deterministic signal given by

\[ s(t) = A\sin(k t), \]

and assuming log-uniform priors between \([10^{-14},10^{-6}]\) and \([10^{-10},10^{-6}]\) for the two model parameters \(A\) and \(k\), respectively.

import numpy
from ptarcade.models_utils import prior


parameters = {
            'log_A' : prior("Uniform", -14, -6),
            'log_k' : prior("Uniform", -10, -6)
            }

def signal(toas, log_A, log_k):
    A = 10**log_A
    k= 10**log_k

    return A *  numpy.sin(k * toas)

Model File Flexibility

In defining the spectrum or signal functions in the model file, you have all the flexibility of a normal Python file. You can, for example, define auxiliary functions, import and interpolate tabulated data etc.

Additional Settings¶

The model file can also contain additional (optional) variables that can be used to control the new-physics signal in more detail. Specifically, you can control the following:

name: Default: "np_model" – This variable can be assigned to a string to specify the model name. This will be used to name the output directory.
smbhb: Default: False – If set to True, the expected signal from SMBHBs will be added to the user-specified signal.
orf: Experimental – This function can be used to specify the Overlap Reduction Function (ORF) of stochastic signals. The first parameter of this function should be named f and it is supposed to be a NumPy array containing the frequencies (in units of Hz) at which the ORF will be evaluated (this parameter should be present in the definition of the orf function even if the ORF that the user wants to specify does not depend on the frequency). The second and third parameters of the orf function should be named pos1 and pos2 respectively, and are supposed to represent the unit vectors pointing from Earth to a given pair of pulsars. The names of the remaining parameters should match the keys of the parameters dictionary (even if the ORF does not depend on them). The orf function should return a NumPy array containing the value of ORF for the given pulsar pair at each of the frequencies in f. If no orf is specified in the model file, the ORF for the GWB is assumed to be the Hellings & Downs (HD) correlation. Notice that an ORF different from HD can be used only in enterprise mode.

NG15 Model Files

The model files used in the NANOGrav 15-year new-physics search can be found here.