Addressing prior dependence in hierarchical Bayesian modeling for PTA data analysis II: Noise and SGWB inference through parameter decorrelation Journal: Astronomy and Computing

Eleonora Villa, Luigi D’Amico, Aldo Barca, Fatima Modica Bittordo, Francesco Alì, Massimo Meneghetti, Luca Naso · 2025

Read Addressing prior dependence in hierarchical Bayesian modeling for PTA data analysis II: Noise and SGWB inference through parameter decorrelation Journal: Astronomy and Computing by Eleonora Villa, Luigi D’Amico, Aldo Barca, Fatima Modica Bittordo, Francesco Alì, Massimo Meneghetti, Luca Naso (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-01959.txt.

Introduction

Eleonora Villa and six colleagues continue a program to free pulsar timing array results from the analyst's prior choices. Array analyses must model each pulsar's noise before any gravitational wave background can be inferred, and fixing a uniform noise prior per pulsar can bias the combined inference. Their hierarchical Bayesian strategy instead parametrizes the noise priors by higher-level hyperparameters shared across the array.

The remaining danger is sensitivity to the hyperprior itself. The authors recast the model. They project the hyperparameters at right angles onto the physical parameter subspace. The projection runs through normalizing flows. The flows keep the transformation reversible. They also preserve shrinkage and information pooling between pulsars. Sampling uses the flow-guided i-nessai sampler. Applied to a minimal three-pulsar case, the method simultaneously inferred noise and background parameters, with the reparametrized treatment constraining the noise more stably.

Catalog record

CREATOREleonora Villa, Luigi D’Amico, Aldo Barca, Fatima Modica Bittordo, Francesco Alì, Massimo Meneghetti, Luca Naso
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE43 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-01959

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Addressing prior dependence in hierarchical Bayesian modeling for PTA data analysis II: Noise and SGWB inference through parameter decorrelation Journal: Astronomy and Computing Eleonora Villa, Luigi D’Amico, Aldo Barca, Fatima Modica Bittordo, Francesco Alì, Massimo Meneghetti, Luca Naso Abstract Pulsar Timing Arrays provide a powerful framework to measure low-frequency gravitational waves, but accuracy and robustness of the results are challenged by complex noise processes that must be accurately modeled. Standard PTA analyses assign fixed uniform noise priors to each pulsar, an approach that can introduce systematic biases when combining the array. To overcome this limitation, we adopt a hierarchical Bayesian modeling strategy in which noise priors are parametrized by higher-level hyperparameters. We further address the challenge posed by the correlations between hyperparameters and physical noise parameters, focusing on those describing red noise and dispersion measure variations. To decorrelate these quantities, we introduce an orthogonal reparametrization of the hierarchical model implemented with Normalizing Flows. We also employ i-nessai , a flow-guided nested sampler, to efficiently explore the resulting higher-dimensional parameter space. We apply our method to a minimal 3-pulsar case study, performing a simultaneous inference of noise and SGWB parameters. Despite the limited dataset, the results consistently show that the hierarchical treatment constrains the noise parameters more tightly and partially alleviates the red-noise–SGWB degeneracy, while the orthogonal reparametrization further enhances parameter independence without affecting the correlations intrinsic to the power-law modeling of the physical processes involved. † † graphicalabstract: † † highlights: Research highlight 1 : Implementation of Hierarchical Bayesian modeling and integration of flow-guided nested sampling into Pulsar Timing Array Bayesian inference pipelines Research highlight 2 : Decorrelation in the parameter space implemented via Normalizing Flows to address prior dependence 1 Introduction Understanding physical phenomena through parameter inference from observational data is a core task in modern astronomy. To account for poorly constrained effects, unmodeled physics, noises, and/or some systematics, it is standard practice to introduce so-called nuisance parameters that capture unknown or subdominant contributions. Although not of direct scientific interest, nuisance parameters are often correlated with the target physical parameters, and this gives rise to degeneracies in the posteriors. As a result, the inferred values of the parameters of interest may be biased or misrepresented, depending on the modeling chosen for the nuisance contributions. This issue is particularly critical when the data lack the constraining power to disentangle nuisance effects from the parameters of interest. This problem plays a particularly important role in the context of Pulsar Timing Arrays (PTA) analyses. PTA experiments measure the times of arrival (TOAs) of radio pulses emitted by highly stable millisecond pulsars. These measurements are modeled using a deterministic timing model that accounts for effects such as the pulsar spin evolution, astrometric parameters, and proper motion. Any deviation between the observed TOAs and the timing model predictions — referred to as timing residuals — is attributed to a combination of noise processes and potential gravitational wave signals. Recently, PTA experiments across the world have reported evidence for a stochastic gravitational wave background (SGWB), marking a major milestone in low-frequency gravitational wave astronomy, see Agazie et al. 2023 , EPTA Collaboration et al. 2023 , Xu et al. 2023 , and Reardon et al. 2023 . However, extracting meaningful information from the SGWB signal critically depends on accurately modeling the various noise contributions in the data, making this an especially delicate task. PTA analyses rely on modeling multiple sources of noise, such as intrinsic red noise and dispersion measure (DM) variations, which are both modeled as power-law signals and described by two parameters each, the \log_{10} of the amplitude and the spectral index \gamma , for each pulsar. The same kind of parameters describe the SGWB signal too, as it is modeled as a single power law common process across all the pulsars in the array. In a single-pulsar analysis, each pulsar represents a single realization of data with respect to the noise stochastic processes. The crucial point here is that it could happen that a certain type of noise is not significantly detectable in the data of a single pulsar whereas it becomes detectable in the data of the full array: this means that the entire array of pulsars represents a different realization of data with respect to the same noise parameters. By contrast, since the SGWB is modeled as a single common signal shared across all pulsars, the full array provides a single realization with respect to the SGWB parameters, unlike the case for noise parameters. The above considerations imply that using the results of single-pulsar analyses — such as assuming the absence of a given noise process — as priors in a full-array analysis can be problematic. It is not only an example of the so-called circular use of data but it also applies a prior derived from the wrong statistical realization. The discrepancy between the statistical description of the noise processes in the single pulsars and their statistical description in the entire array may accumulate across pulsars and lead to biased inference, e.g. it may introduce systematic errors in the inferred values of \log_{10}A and \gamma for the power law model of the SGWB. This is especially crucial for the pulsar red noise which is highly correlated with the SGWB, both being low-frequency stochastic processes modeled in the same way. The solution of this problem is then clear: the distribution of the noise parameters should be parametrized for each pulsar and inferred simultaneously with that of the two SGWB parameters in the analysis of the full array. This means that we have to choose a prior for all the parameters to infer, including noise priors. To account for the uncertainty in the noise distributions we parametrize noises prior by introducing an additional hierachical prior encoded in additional parameters, called hyperparameters. They describe the distribution of noise properties across the array and act as nuisance parameters in the inference problem. We remark that the hierarchical Bayesian modeling becomes necessary in PTA analyses, as it allows for the correct inference of both the astrophysical signal of interest, such as the SGWB, and the population-level noise properties encoded in the hyperpriors. All the above considerations were recently pointed out, although with slightly different approaches, in Goncharov and Sardana 2025 , van Haasteren 2024a , and van Haasteren 2024b . For a discussion about previous literature implicitly showing this problem, referred to as prior misspecification , also see Goncharov and Sardana 2025 . At this point, a well-known aspect of Bayesian modeling becomes relevant: posterior distributions can be significantly influenced by the choice of priors. This problem, known as prior dependence or prior-driven projection effects, affects every level of hierarchical modeling. Various strategies have been explored to address this issue in many different contexts. In general, one approach is to impose physically motivated priors, for instance informed by simulations or empirical models. Another possibility is to employ priors considered "non-informative", such as uniform or Jeffreys priors. Yet these choices come with their own limitations: uniform priors may implicitly encode arbitrary assumptions e.g. on prior bounds, and Jeffreys priors, although invariant under reparametrization, are often impractical in complex analyses. While uniform priors are of standard use in pulsar astronomy, in Laal et al. 2025 is

Read the complete text