James Alvey

The thermal Sunyaev–Zel’dovich (tSZ) power spectrum is a sensitive probe of cosmology and cluster astrophysics, but its statistics are non-Gaussian because the signal receives a significant contribution from rare, massive, low-redshift galaxy clusters. As a result, a Gaussian likelihood fails to describe the statistics of its power spectrum on large scales. We use simulation-based inference (SBI) to test the accuracy of the standard Gaussian power-spectrum likelihood for a -like tSZ analysis. Using halo-based simulations of full-sky Compton- y maps, we train neural posterior and likelihood estimators and compare the resulting constraints with those from a Gaussian likelihood assumption. Using only multipoles ℓ < 1000 , we find that the Gaussian likelihood assumption gives unbiased cosmological constraints, while the SBI-based inference shows a mild broadening of the posterior distributions for the amplitudes of residual foregrounds. This suggests that the Gaussian likelihood assumption is sufficiently accurate for cosmological inference for a -like tSZ analysis, while SBI provides a useful validation tool to model non-Gaussian likelihoods beyond analytic approximations.

Extreme mass-ratio inspirals pose a difficult challenge in terms of both search and parameter estimation for upcoming space-based gravitational-wave detectors such as LISA. Their signals are long and of complex morphology, meaning they carry a large amount of information about their source, but their waveforms are expensive to compute and they occupy a vast and multimodal parameter space. We explore how sequential simulation-based inference methods, specifically truncated marginal neural ratio estimation, could offer solutions to some of the challenges surrounding extreme-mass-ratio inspiral data analysis. We show that this method can efficiently narrow down the volume of the complex 11-dimensional search parameter space by a factor of 10 6 – 10 7 and provide one-dimensional marginal proposal distributions for nonspinning extreme-mass-ratio inspirals. We discuss the current limitations of this approach and place it in the broader context of a global strategy for future space-based gravitational-wave data analysis.

Simulation-based inference (SBI) is emerging as a new statistical paradigm for addressing complex scientific inference problems. By leveraging the representational power of deep neural networks, SBI can extract the most informative simulation features for the parameters of interest. Sequential SBI methods extend this approach by iteratively steering the simulation process towards the most relevant regions of parameter space. This is typically implemented through an algorithmic structure, in which simulation and network training alternate over multiple rounds. This strategy is particularly well suited for high-precision inference in high-dimensional settings, which are commonplace in physics applications with growing data volumes and increasing model fidelity. Here, we introduce dynamic SBI, which implements the core ideas of sequential methods in a round-free, asynchronous, and highly parallelisable manner. At its core is an adaptive dataset that is iteratively transformed during inference to resemble the target observation. Simulation and training proceed in parallel: trained networks are used both to filter out simulations incompatible with the data and to propose new, more promising ones. Compared to round-based sequential methods, this asynchronous structure can significantly reduce simulation costs and training overhead. We demonstrate that dynamic SBI achieves significant improvements in simulation and training efficiency while maintaining inference performance. We further validate our framework on two challenging astrophysical inference tasks: characterising the stochastic gravitational wave background and analysing strong gravitational lensing systems. Overall, this work presents a flexible and efficient new paradigm for sequential SBI.

Simulation-based inference (SBI) offers a principled and flexible framework for conducting Bayesian inference in any situation where forward simulations are feasible. However, validating the accuracy and reliability of the inferred posteriors remains a persistent challenge. In this work, we point out a simple diagnostic approach rooted in ensemble learning methods to assess the internal consistency of SBI outputs that does not require access to the true posterior. By training multiple neural estimators under identical conditions and evaluating their pairwise Kullback–Leibler (KL) divergences, we define a consistency criterion that quantifies agreement across the ensemble. We highlight two core use cases for this framework: (a) for generating a robust estimate of the systematic uncertainty in parameter reconstruction associated with the training procedure, and (b) for detecting possible model misspecification when using trained estimators on real data. We also demonstrate the relationship between significant KL divergences and issues such as insufficient convergence due to, e.g. too low a simulation budget, or intrinsic variance in the training process. Overall, this ensemble-based diagnostic framework provides a lightweight, scalable, and model-agnostic tool for enhancing the trustworthiness of SBI in scientific applications.

Simulation-based inference (SBI) is emerging as a new statistical paradigm for addressing complex scientific inference problems. By leveraging the representational power of deep neural networks, SBI can extract the most informative simulation features for the parameters of interest. Sequential SBI methods extend this approach by iteratively steering the simulation process towards the most relevant regions of parameter space. This is typically implemented through an algorithmic structure, in which simulation and network training alternate over multiple rounds. This strategy is particularly well suited for high-precision inference in high-dimensional settings, which are commonplace in physics applications with growing data volumes and increasing model fidelity. Here, we introduce dynamic SBI, which implements the core ideas of sequential methods in a round-free, asynchronous, and highly parallelisable manner. At its core is an adaptive dataset that is iteratively transformed during inference to resemble the target observation. Simulation and training proceed in parallel: trained networks are used both to filter out simulations incompatible with the data and to propose new, more promising ones. Compared to round-based sequential methods, this asynchronous structure can significantly reduce simulation costs and training overhead. We demonstrate that dynamic SBI achieves significant improvements in simulation and training efficiency while maintaining inference performance. We further validate our framework on two challenging astrophysical inference tasks: characterising the stochastic gravitational wave background and analysing strong gravitational lensing systems. Overall, this work presents a flexible and efficient new paradigm for sequential SBI.

It is widely established that a lower bound on the dark matter particle mass, m , can be obtained by demanding that the de Broglie wavelength in a given galaxy must be smaller than the virial radius of the galaxy, leading to m ≳ 10 − 22 eV when applied to typical dwarf galaxies. This lower limit has never been derived precisely or rigorously. We use stellar kinematical data for the Milky Way satellite galaxy Leo II to self-consistently reconstruct a statistical ensemble of dark matter wave functions and corresponding density profiles. By comparison to a data-driven, model-independent reconstruction, and using a variant of the maximum mean discrepancy as a statistical measure, we determine that a self-consistent description of dark matter in the local Universe requires m > 2.2 × 10 − 21 e V ( C L > 95 % ). This lower limit is free of any assumptions pertaining to cosmology, microphysics (including spin), or dynamics of dark matter, and only assumes that it is predominantly composed of a single bosonic particle species.

Model misspecification analysis strategies, such as anomaly detection, model validation, and model comparison are a key component of scientific model development. Over the last few years, there has been a rapid rise in the use of simulation-based inference (SBI) techniques for Bayesian parameter estimation, applied to increasingly complex forward models. To move toward fully simulation-based analysis pipelines, however, there is an urgent need for a comprehensive simulation-based framework for model misspecification analysis. In this work, we provide a solid and flexible foundation for a wide range of model discrepancy analysis tasks, using . From a theoretical perspective, we introduce the statistical framework built around performing many hypothesis tests for distortions of the simulation model. We also make explicit analytic connections to classical techniques: anomaly detection, model validation, and goodness-of-fit residual analysis. Furthermore, we introduce an efficient self-calibrating training algorithm that is useful for practitioners. We demonstrate the performance of the framework in multiple scenarios, making the connection to classical results where they are valid. Finally, we show how to conduct such a distortion-driven model misspecification test for real gravitational wave data, specifically on the event GW150914. Published by the American Physical Society 2025

In recent years, there has been a remarkable development of simulation-based inference (SBI) algorithms, and they have now been applied across a wide range of astrophysical and cosmological analyses. There are a number of key advantages to these methods, centred around the ability to perform scalable statistical inference without an explicit likelihood. In this work, we propose two technical building blocks to a specific sequential SBI algorithm, truncated marginal neural ratio estimation (TMNRE). In particular, first, we develop autoregressive ratio estimation with the aim to robustly estimate correlated high-dimensional posteriors. Secondly, we propose a slice-based nested sampling algorithm to efficiently draw both posterior samples and constrained prior samples from ratio estimators, the latter being instrumental for sequential inference. To validate our implementation, we carry out inference tasks on three concrete examples: a toy model of a multidimensional Gaussian, the analysis of a stellar stream mock observation, and finally, a proof-of-concept application to substructure searches in strong gravitational lensing. In addition, we publicly release the code for both the autoregressive ratio estimator and the slice sampler.

The next generation of space- and ground-based facilities promise to reveal an entirely new picture of the gravitational wave sky: thousands of galactic and extragalactic binary signals, as well as stochastic gravitational wave backgrounds (SGWBs) of unresolved astrophysical and possibly cosmological signals. These will need to be disentangled to achieve the scientific goals of experiments such as LISA, Einstein Telescope, or Cosmic Explorer. We focus on one particular aspect of this challenge: reconstructing an SGWB from (mock) LISA data. We demonstrate that simulation-based inference (SBI)—specifically truncated marginal neural ratio estimation (TMNRE)—is a promising avenue to overcome some of the technical difficulties and compromises necessary when applying more traditional methods such as Monte Carlo Markov Chains (MCMC). To highlight this, we show that we can reproduce results from traditional methods both for a template-based and agnostic search for an SGWB. Moreover, as a demonstration of the rich potential of SBI, we consider the injection of a population of low signal-to-noise ratio supermassive black hole transient signals into the data. TMNRE can implicitly marginalize over this complicated parameter space, enabling us to directly and accurately reconstruct the stochastic (and instrumental noise) contributions. We publicly release our TMNRE implementation in the form of the code saqqara.