Quasi-periodic plus derivative kernel#
The kernel employed in the gp_quasiperiodic_derivative model is a combination of the quasi-periodic kernel and its first derivative, not different from using the multivariate approachon a single dataset. If we consider the correlated noise of a given dataset \(\mathrm{D}\) as the combination of a Gaussian process and its first derivative:
Then the final covariance between observations of G and its derivative at times \(t_i\) and \(t_j\) is given by:
Where \(\gamma_{\rm GP}^{(G,G)} (t_i, t_j)\) is the quasi-periodic kernel and \( \gamma^{(dG,dG)}(t_i, t_j) = \left.\left.\frac{\partial}{\partial t'} \frac{\partial}{\partial t} \gamma^{(G,G)}(t, t') \right|_{t=t_i} \right|_{t'=t_j} \). Whit a single dataset, the terms \( \gamma^{(G,dG)}(t_i, t_j)\) and \( \gamma^{(dG,G)}(t_i, t_j)\) cancel each other. See Section 3.3 of Rajpaul et al. 2015 for more details.
This model can be useful when a single dataset needs a more flexible stellar-activity covariance than the standard quasi-periodic kernel, without switching to a multidimensional GP where the first derivative is usually employed.
Model definition and requirements#
The fastest implementation relies on tinyGP, but it requires a few extra tricks in the configuration file and execution (see Caveats on the use of tinyGP). If you use this model, cite the Zenodo repository.
An independent implementation that relies only on basic packages is maintained for legacy reasons; however, it is much slower, and I don’t recommend using it.
model name: tinygp_quasiperiodic_derivative
available since version 11.2.6
required common object:
activityimplemented using
tinygp(version 0.3.0, link to documentation)GPU acceleration supported (instruction incoming)
Read Caveats on the use of
tinyGPcarefully
model name: gp_quasiperiodic_derivative
required common object:
activitydirect implementation relying only on
numpyandscipyindependent covariance matrix for each dataset
Model parameters#
Name |
Parameter |
Common? |
Definition |
Notes |
|---|---|---|---|---|
|
Rotational period of the star |
common |
|
Replaced by |
|
Decay timescale of active regions |
common |
|
Replaced by |
|
Coherence scale |
common |
|
|
|
Amplitude of the quasi-periodic component |
dataset |
|
|
|
Amplitude of the derivative component |
dataset |
|
Keywords#
Model-wide keywords, with the default value in boldface.
hyperparameters_condition
accepted values:
True|Falseactivates the quasi-periodic hyperparameter condition described in the quasi-periodic kernel.
rotation_decay_condition
accepted values:
True|Falseif activated, requires
Pdec > 2 Prot.
halfrotation_decay_condition
accepted values:
True|Falseif activated, requires
Pdec > 0.5 Prot.
decay_rotation_factor or rotation_decay_factor
accepted values: float | not used
if provided, requires
Pdecto be larger than the specified factor timesProt.
use_stellar_rotation_period
accepted values:
True|Falsereplaces
Protwithrotation_periodfromstar_parameters.
use_stellar_activity_decay
accepted values:
True|Falsereplaces
Pdecwithactivity_decayfromstar_parameters.
Example#
1inputs:
2 RVdata:
3 file: datasets/star_RV_PyORBIT.dat
4 kind: RV
5 models:
6 - radial_velocities
7 - gp_qp_derivative
8common:
9 activity:
10 boundaries:
11 Prot: [10.0, 20.0]
12 Pdec: [20.0, 1000.0]
13 Oamp: [0.001, 1.0]
14models:
15 gp_qp_derivative:
16 model: gp_quasiperiodic_derivative
17 common: activity
18 hyperparameters_condition: True
19 rotation_decay_condition: True
20 boundaries:
21 Hamp: [0.0, 100.0]
22 Camp: [0.0, 100.0]