How to embed any likelihood into SBI: Application to Planck + Stage IV galaxy surveys and Dynamical Dark Energy

Guillermo Franco Abellán, Noemi Anau Montel, Oleg Savchenko, Christoph Weniger · 2025

Read How to embed any likelihood into SBI: Application to Planck + Stage IV galaxy surveys and Dynamical Dark Energy by Guillermo Franco Abellán, Noemi Anau Montel, Oleg Savchenko, Christoph Weniger (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-2507-22990.txt.

Introduction

Guillermo Franco Abellán and co-authors show how to fold any ready-made likelihood into simulation-based inference, in this 2025 paper. Simulation-based methods are fast but demand that nature be written as a simulator, which some prized likelihoods, like the Planck satellite's, resist. Their trick builds a stand-in simulator from the samples of an earlier chain run, which can then join any true simulator.

They wed Planck's data to a mock Stage IV galaxy survey and test evolving dark energy. The joint scan matched the classic method while skipping its heavy solver calls entirely.

Catalog record

CREATORGuillermo Franco Abellán, Noemi Anau Montel, Oleg Savchenko, Christoph Weniger
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE55 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2507-22990

Browse the full library · About the Tower of Babel

Full text

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

How to embed any likelihood into SBI: Application to Planck + Stage IV galaxy surveys and Dynamical Dark Energy Preprint: APS/123-QED Guillermo Franco Abellán, Noemi Anau Montel, Oleg Savchenko, Christoph Weniger Abstract Simulation-based inference (SBI) allows fast Bayesian inference for simulators encoding implicit likelihoods. However, some explicit likelihoods cannot be easily reformulated as simulators, hindering their integration into combined analyses within SBI frameworks. One key example in cosmology is given by the Planck CMB likelihoods. We present a simple method to construct an effective simulator for any explicit likelihood using samples from a previously converged Markov Chain Monte Carlo (MCMC) run. This effective simulator can subsequently be combined with any forward simulator. To illustrate this method, we combine the full Planck CMB likelihoods with a 3 × \times 2pt simulator (cosmic shear, galaxy clustering and their cross-correlation) for a Stage IV survey like Euclid , and test evolving dark energy parameterized by the w 0 ​ w a w_{0}w_{a} equation-of-state. Assuming the w 0 ​ w a w_{0}w_{a} CDM cosmology hinted by DESI BAO DR2 + Planck 2018 + PantheonPlus SNIa datasets, we find that future 3 × \times 2pt data alone could detect evolving dark energy at 5 ​ σ 5\sigma , while its combination with current CMB, BAO and SNIa datasets could raise the detection to almost 7 ​ σ 7\sigma . Moreover, thanks to simulation reuse enabled by SBI, we show that our joint analysis is in excellent agreement with MCMC while requiring zero Boltzmann solver calls. This result opens up the possibility of performing massive global scans combining explicit and implicit likelihoods in a highly efficient way. I Introduction The most stringent and robust cosmological results of the next decade will be obtained from the combination of multiple astronomical surveys. This includes data from past and current experiments like Planck Aghanim et al. 2020a , ACT Louis et al. 2025 , SPT Camphuis et al. 2025 , KiDS Wright et al. 2025 , DES Abbott et al. 2022 , BOSS Alam et al. 2021 , and DESI Abdul Karim et al. 2025 , as well as forthcoming surveys such as Euclid Mellier et al. 2024 , LSST Ivezić et al. 2019 , Simons Observatory Ade et al. 2019 , and CMB-S4 Abazajian et al. 2016 . Performing Bayesian inference from these surveys typically involves evaluating a large number of theoretical models – each requiring costly calls to Boltzmann solvers like CAMB Lewis et al. 2000 or CLASS Lesgourgues 2011 -- and marginalizing over numerous nuisance parameters. In the context of Markov Chain Monte Carlo (MCMC) methods, this time consuming exercise needs to be repeated for each new combination of experiments, 1 1 1 To name just a few concrete examples, the recent DESI BAO DR2 analysis Abdul Karim et al. 2025 explored 14 configurations of experiments to test the w_{0}w_{a} CDM model; the ‘ H_{0} Olympics’ paper Schöneberg et al. 2022 analyzed 12 data combinations across 8 different ‘finalist’ models; and 30 combinations of experiments were considered in Brinckmann et al. 2019 to derive forecast constraints on neutrino masses. which becomes increasingly challenging given the diversity and complexity of datasets. In recent years, simulation-based inference (SBI) Cranmer et al. 2020 has emerged as a powerful cosmological tool for Bayesian analysis in complex settings (see e.g. Lemos et al. 2024 ; Jeffrey et al. 2024 ; von Wietersheim-Kramsta et al. 2025 ; Novaes et al. 2025 ; Zeghal et al. 2024 ; Nguyen et al. 2024 ; Zubeldia et al. 2025 ; Franco Abellán et al. 2024 ; Savchenko et al. 2025 ). By using forward simulations that encode implicit likelihoods, SBI offers several key benefits. First, it supports amortized inference for rapid re-analysis and cross-validation. Second, it can flexibly incorporate complex effects through forward modeling that would be expensive to compute or intractable to express analytically in a likelihood. Third, it can efficiently yield marginal parameter estimates by directly integrating out nuisance variables. Finally, because simulations can be reused across inference tasks, SBI eliminates the need for repeated model evaluations, greatly enhancing computational efficiency. However, some legacy cosmological likelihoods, even if explicit in form, cannot be readily reformulated as forward simulators. A notable example are the Planck cosmic microwave background (CMB) likelihoods Aghanim et al. 2020b . Specifically, the low- \ell likelihoods ( Commander , SimAll ) are computed at the pixel map level due to the non-Gaussian nature of the power spectrum at large scales; and the high- \ell likelihoods ( Plik ) rely on “pseudo- C_{\ell} ’s” from various frequency channels, which introduce 47 nuisance parameters to describe instrument noise and foregrounds. Such dataset is too complex to be simulated in large numbers, preventing its integration into SBI workflows and thus limiting the ability to leverage the benefits previously discussed. In this work, we present a simple yet general trick to transform any explicit likelihood into an effective simulator . Our method introduces an auxiliary-observable , derived from samples of a pre-converged MCMC run, to construct an effective simulator that faithfully reproduces the original explicit likelihood. This effective simulator can be directly used for SBI, inheriting the full suite of benefits offered by such framework. Specifically, from a single MCMC, one can rapidly generate many effective simulations, which can be reused across different SBI tasks. Moreover, SBI allows seamless combination of likelihoods, whether explicit ones (turned into effective simulators) or implicit. This directly addresses the computational challenge of performing massive global scans, as it enables inference without the need for additional model evaluations each time new data is added. Alternatively, our approach can be used to speed up traditional sampling-based inference, by leveraging our effective simulator to train a fast likelihood emulator. This connects naturally to prior work on emulation in cosmology, where surrogate models are trained to approximate computationally expensive theory codes Spurio Mancini et al. 2022 ; Nygaard et al. 2023 ; Günther et al. 2025 or likelihoods McClintock and Rozo 2019 ; Bevins et al. 2023 and then used as fast drop-in replacements in MCMC samplers (see also Mootoovaloo et al. 2024 ; Taylor et al. 2024 for different methods to accelerate joint cosmological analyses). We validate our method through a series of cosmological applications. First, we build an effective Planck \Lambda CDM simulator and apply it within SBI, recovering posteriors that are in excellent agreement with the original MCMC. Next, we combine this effective simulator with a forecast simulator of 3 \times 2pt probes (weak lensing, galaxy clustering, and their cross-correlation) for a Stage IV photometric galaxy survey, and perform a joint \Lambda CDM analysis. Thanks to simulation reuse enabled by SBI, we find accurate posteriors for this data combination without requiring any new model evaluations. Finally, we demonstrate our approach on a directly relevant physics application, by extending the SBI analysis to probe dynamical dark energy parameterized by the Chevallier-Polarski-Linder (CPL) equation-of-state Chevallier and Polarski 2001 ; Linder 2003 . This is motivated by the recent exciting results by the DESI collaboration, which hint at departures from a cosmological constant ( Adame et al. 2025 ) . Specifically, the combination of data from CMB + DESI baryonic acoustic oscillation (BAO) + Type Ia supernovae (SNIa) shows a preference for

Read the complete text