Colloque national du WG Dark Energy 2026 - 10ème édition

2 nov. 2026, 04:30 → 4 nov. 2026, 22:50 Europe/Paris

Institut Henri Poincaré

Les problématiques autour de l’énergie noire apparaissent comme essentielles dans les axes de recherche de l’INSU, de l’IN2P3, de l’INP et de l'IRFU.

La communauté française en cosmologie, avec une forte implication dans plusieurs expériences ambitieuses dont le satellite Euclid et le télescope LSST, s’est très largement mobilisée sur cette problématique.

À l’issue des premiers colloques a émergé le besoin d’une animation approfondie selon trois axes : les aspects théoriques, les sondes et les outils.

Ce colloque sera l'occasion de présenter et discuter les derniers résultats et le fruit des réflexions dans ces trois axes, situés dans un panorama plus large du domaine.

Dates importantes

• 4 octobre 2026 : clôture de l'appel à contributions
• 25 octobre 2026 : clôture des inscriptions

Cette manifestation a reçu le soutien du CNRS, de l'INP, de l'IN2P3, de l'INSU, du CNES, de l'ATCG,  de l'ATGRAM et du GdR CoPhy.

---

The issues surrounding dark energy have become central to the research priorities of the INSU, IN2P3, INP, and IRFU.

The French cosmology community, deeply involved in several ambitious projects such as the DESI telescope, Euclid satellite and the LSST telescope, has widely mobilized around this topic.

Following the initial conferences, a need emerged for in-depth exploration along three axes: theoretical aspects, probes, and tools. This conference will provide an opportunity to present and discuss the latest results and insights from these three areas, set within the broader landscape of the field.

Important dates

• October 4th 2026 : deadline for abstract submission
• October 25th 2026 : deadline for registrations
  

This event has received support from the CNRS, INP, IN2P3, INSU, CNES, ATCG, ATGRAM, and the GdR CoPhy.

 

lun. 2 novembre

13:30 → 14:00 welcome

14:00 → 14:30 Pauline Vielzeuf

Orateur: Pauline VIELZEUF (cppm)

14:30 → 15:00 FOSSIL: Probing the cosmological model with spectral distortions

The Cosmic Microwave Background (CMB) is a pillar of modern cosmology. Since the results of the COBE/FIRAS satellite in the nineties the CMB spectrum is known to be a quasi-perfect blackbody. However, both standard and non-standard physics processes inevitably lead to deviations from blackbody, named spectral distortions.
FOSSIL, proposed to ESA as its 8th medium-class mission, aims at achieving an unprecedented leap forward in sensitivity, allowing us to revisit the measurement of the CMB spectrum and to detect these spectral distortions, and thereby providing us with a unique opportunity to reveal novel probes of the early Universe, particle physics, and structure formation.

Orateur: Nabila Aghanim (Institut d'Astrophysique Spatiale)

15:00 → 15:30 Euclid

Orateur: Marc Sauvage (CEA/DRF/Irfu/SAp)

15:30 → 16:00 coffee break

16:00 → 16:30 H0

Orateur: Louise Breuval (Observatoire de Paris, LESIA)

16:30 → 16:50 From synthetic source injection to simulated mocks: characterizing redshift calibration systematics for Euclid

Euclid is an ongoing space-based cosmological survey designed to map the large-scale structure of the Universe by observing billions of galaxies. Its primary goal is to investigate the nature of dark matter and dark energy and to place tight constraints on the cosmological model.
A key probe for Euclid is galaxy weak gravitational lensing, which measures the small distortions induced in the observed shapes of background galaxies by the gravitational fields of foreground matter. Combined with galaxy clustering, weak lensing provides a powerful means of probing the distribution and evolution of matter throughout the Universe.
A crucial ingredient for extracting robust cosmological constraints from these measurements is an accurate determination of the redshift distribution of the galaxies entering the analysis. Since Euclid primarily relies on photometric observations, galaxy redshifts are estimated from their observed colours and fluxes. Calibrating the relationship between galaxy colours and redshift is therefore an essential step in the analysis and represents one of the main sources of systematic uncertainty in the cosmological parameter constraints.
It is thus essential to properly characterize, control, and mitigate the various sources of uncertainty affecting the inferred redshift distributions.
In this talk, we use a synthetic source-injection pipeline to characterize the Euclid galaxy selection function and photometric noise properties in both the Wide and Deep surveys. The Deep Survey benefits from substantially deeper imaging, enabling more accurate redshift estimates and providing a reference sample for calibrating the redshift distributions of galaxies in the Wide Survey.
We then apply the Euclid selection function to the state-of-the-art N-body Flagship 2 (FS2) simulation to generate realistic Euclid-like galaxy mock catalogues. These mocks are used to perform the redshift calibration and quantify the different sources of systematic uncertainty. Finally, we investigate how these uncertainties can be consistently propagated into the cosmological analysis, providing informed priors for the likelihood and assessing their impact on the resulting cosmological parameter constraints.

Orateur: Tyann Dumerchat (IFAE / Universitat Autonoma de Barcelona)

16:50 → 17:10 Field-level Inference for the Joint Analysis of Galaxy Clustering and CMB Lensing

We present our work towards a field-level inference (FLI) pipeline for the joint analysis of DESI galaxy clustering and CMB lensing from Planck and ACT. We demonstrate an end-to-end Bayesian framework that jointly samples the posterior of the initial density field, galaxy bias, and local primordial non-Gaussianity ($f_{\mathrm{NL}}$). Our differentiable forward model propagates initial conditions through Lagrangian Perturbation Theory (2LPT), applies a Lagrangian bias expansion, and computes the CMB lensing convergence along the line of sight via the Born approximation. The joint likelihood combines galaxy number counts with the convergence map observations, incorporating ACT DR6 reconstruction noise and analytically accounting for the unmodeled high-z lensing contribution beyond the simulation volume. The high-dimensional posterior is explored efficiently with the MCLMC sampler, the joint problem converging as well as galaxies alone. The pipeline is validated against AbacusLensing simulations. Preliminary results show that CMB lensing improves the reconstruction of the initial field outside the galaxy survey volume, where clustering alone carries no information. This work is a step toward applying FLI to real data.

Orateur: Jonathan Hawla

17:10 → 17:30 Kinessence, phantom crossing and sign-switching dark energy

We introduce Kinessence models where combining K-mouflage and quintessence leads to phantom-
crossing of the dark energy equation of state and sign-switching of the dark energy density. This
is realised by joining a dark scalar sector comprising K-essence kinetic terms and a quintessence
potential together with a Yukawa coupling to matter. We investigate the dynamics of these models
in the Jordan frame both at the background and perturbative levels. We concentrate on the effective
equation of state and the matter fraction for the background. The Cosmic Microwave Background
(CMB) spectra, the linear power spectrum and the growth of structure via fσ8 are also analysed
at the perturbative level. The quintessence potential drives the acceleration of the Universe at low
redshift with an equation of state greater than minus one. The K-essence kinetic term leads to
a phantom crossing behaviour depending on the strength of the Yukawa coupling. The potential
fifth force mediated by the scalar field is screened by the K-mouflage mechanism and passes solar
system tests, i.e. the Cassini bound and the periastron of the Moon. This allows for relatively large
Yukawa couplings to matter β∼0.1, large enough to lead to the crossing of the phantom divide at
small redshift z ∼0.5. Despite the presence of a significant scalar interaction on large scales, we
find that the power spectrum and the growth of structure are reduced compared to ΛCDM. This
is also the case for the matter fraction which deviates from the standard model by a few percent.
Similarly the Hubble rate is larger at low redshift than at recombination whilst the CMB spectra
show small deviations from ΛCDM. In these models, the dark energy fraction becomes negative at
redshifts greater than two. All these features should alleviate the various tensions such as the larger
value of H0 at small redshift, a possible reduction of growth as measured by S8 and the puzzling
bound on the sum of the neutrino masses. Deviations from Λ-CDM of the CMB lensing spectrum
may provide a detectable window for these models.

Orateur: philippe brax (IPHT Saclay)

17:30 → 17:50 A late universe model designed to fit the cosmological data: the Szekeres exact GR solution

In the current era of precision cosmology, it is well-known that the present standard $\Lambda$CDM model, despite its successes, is an insufficient average of our actual Universe. Given the recorded tensions, it appears that a more precise representation of the local universe is mandatory. The Szekeres exact solution of General Relativity possesses all the nice properties allowing it to represent an inhomogeneous small scale universe becoming naturally homogeneous at some large transition scale. However, a raw use of the Szekeres model is challenging, since, besides a few constant parameters, it is described by a set of five metric functions, hence a difficult inference problem when we want to fit it to cosmological observations. However, a number of data analyses displayed in the recent literature have lead to the conclusion that the matter distribution and expansion are anisotropic roughly in the direction of the CMB dipole. This can be represented by the Szekeres axially symmetric version where two defining functions vanish. However, the inference issue remains even if reduced to three functions. To simplify such a fitting issue, we propose to replace this difficult inference problem by a spectral approximation problem and to use truncated Chebyshev polynomial expansions of the metric functions. As a first step, a truncation of the expansion at order two implies that a given Szekeres model can be determined by only nine constant parameters which greatly simplifies the fitting to observations. Moreover, beyond the transition to homogeneity, the $\Lambda$CDM model applies and the early universe picture is unchanged.

Orateur: Dr Marie-Noëlle Célérier (Observatoire de Paris)

17:50 → 18:10 Dark Gravity cosmology: an Upgrade

Dark Gravity is a natural extension of General Relativity in the presence of a flat, non-dynamical background. We present an upgraded version of the theory in which the transition from the matter-dominated expansion law, a(t)∝t2/3, to the late-time accelerated regime, a(t)∝t2, occurs progressively rather than abruptly.

Compared with the version presented at the MG17 conference, the new model provides a significantly improved description of the main cosmological observations, including Type Ia supernovae, baryon acoustic oscillations (BAO), the cosmic microwave background (CMB), age estimates of the oldest stars, and direct measurements of the Hubble constant. The upgraded model introduces three free parameters—H0 and two transition redshifts—instead of the single transition redshift of the previous version.

After briefly reviewing the theoretical foundations of Dark Gravity and the original mechanism responsible for cosmic acceleration, we introduce the new transition scenario and discuss its observational performance.

Orateur: Frederic HENRY COUANNIER (cppm)

18:10 → 18:30 Extracting cosmological parameters from multi-wavelength cross-correlations

Galaxy clustering and weak lensing on small scales hold much of the cosmological information in the coming surveys of large scale structure. They also carry the imprint of baryons: gas expelled from haloes by feedback suppresses the matter power spectrum at the level of the statistical errors, and this effect is degenerate with . The same gas is observed directly, in X-rays, through the thermal Sunyaev–Zel'dovich effect and in 21-cm emission, and galaxies and active nuclei trace the haloes that hold it. We use a differentiable halo model that describes galaxies, AGN, hot and neutral gas and matter within one population of haloes and one baryon budget. With it we predict every auto- and cross-spectrum among these tracers, together with their joint covariance. Because each prediction is differentiable with respect to cosmological and astrophysical parameters, we compute Fisher forecasts by automatic differentiation. The gain comes from constraining feedback with the data instead of marginalising over it.

Orateur: johan comparat

mar. 3 novembre

09:00 → 09:30 Impact of a wCDM cosmology on the tSZ power spectrum

The distribution of matter in the Universe is a powerful probe of cosmology. Measuring the efficiency with which gravity produces clusters against expanding Universe is the key to understanding, e.g. the equation of state of dark energy. The thermal Sunyaev-Zel’dovich (tSZ) effect, being sensitive to the baryonic content of the Universe, also shows sensitivity to dark energy. In this work, I investigate the impact of dark energy on the tSZ effect. To this end, I conducted new simulations varying the equation of state of dark energy within RAMSES, and compared different quantities such as the matter and tSZ power spectrum. In this talk, I will present and discuss my preliminary results on these comparisons across different cosmologies.

Orateur: Emma Ayçoberry (IAP)

09:30 → 09:50 Impacts of alternative Dark Energy models on the formation of galaxies

Recent cosmological measurements have suggested that the dark energy density might not be constant but evolving. In this study, we explored the interplay between the properties of dark energy and the physical processes responsible for the formation of galaxies which has remained poorly explored. We carried out the analysis on merger trees generated from a semi-analytical approach based on the Excursion Set, using four different dark energy models. We find that the equation of state parameter for dark energy has a clear impact on the formation and growth of galaxies. Typically, quintessence models of dark energy are the most structured, i.e, they form evolved galaxies earlier. Quite the opposite, phantom models of dark energy are the least structured, i.e, form structures later. Dynamical models of dark energy on the other hand remain close to $\Lambda$CDM, even though they are slightly more structured. These predictions motivate further investigations using cosmological simulations, that will enable a more in-depth analysis of the effect of dark energy on the formation and evolution of galaxies.

Orateur: Adrian SZPILFIDEL (LUX)

09:50 → 10:10 Deep Learning Detection of Galaxy Clusters with YOLOv10: From SDSS to Rubin/LSST

Galaxy clusters are powerful and key cosmological probes that trace the evolution of large-scale structures of the universe and inform models of cosmic evolution, but their identification in the increasingly large imaging datasets produced by current and upcoming surveys requires detection methods that are both accurate and scalable. The Vera C. Rubin Observatory Legacy Survey of Space and Time (LSST) will provide deep, wide-field optical imaging over an unprecedented survey area, enabling the detection of thousands of galaxy clusters up to high redshifts and motivating the development of new approaches for identifying galaxy clusters directly from astronomical images.

Building upon the YOLO-CL framework, originally based on YOLOv3, I have developed an updated galaxy-cluster detection pipeline using the more recent YOLOv10 object-detection architecture. The model detects and localizes galaxy clusters directly in multiband optical images, without requiring a preliminary galaxy catalog. Applied to Sloan Digital Sky Survey (SDSS) imaging using redMaPPer clusters as reference detections, the method achieves high completeness and purity, demonstrating the effectiveness of modern object-detection architectures for this task.

In parallel, I am investigating the incorporation of astrophysical prior information through class-activation-map-based auxiliary supervision, encouraging the network to focus on physically relevant galaxy populations and providing greater interpretability of its detections.

This effort aims to contribute a scalable and interpretable, high-fidelity cluster detection pipeline for next-generation cosmological analyses and to assess its robustness as we move from controlled cluster samples toward realistic next-generation survey data.

Orateur: Vinh Phat Tran (APC/UPCité)

10:10 → 10:30 Evolving only the observed spacetime region: fast Euclid-scale lightcones with sCOLA

Precision constraints on dark energy from large-scale-structure surveys require large ensembles of cosmological simulations for covariance estimation, forward modelling, and simulation-based inference. Producing such ensembles at Euclid volumes remains computationally challenging with conventional $N$-body methods. We present a fast and scalable framework based on the spatially discretised COmoving Lagrangian Acceleration (sCOLA) method that reduces this cost by evolving only the spacetime region contributing to the observed lightcone.

sCOLA decomposes large simulation volumes into independent tiles. Only tiles intersecting the survey volume are evolved, and each tile is stopped once it exits the observer's past lightcone. The resulting algorithm is perfectly parallel and avoids unnecessary spatial and temporal evolution. We validate sCOLA lightcones against $\texttt{Gadget-4}$, finding percent-level agreement in the dark-matter power spectrum up to $k=0.5\,h\,\mathrm{Mpc}^{-1}$ and agreement at the $\sim10\%$ level in halo statistics. For our target lightcone configurations, this approach yields speed-ups of up to $\sim10^3\!-\!10^4$ relative to Tree-PM simulations.

Using this framework, we produced Silmarilli, a suite of $\sim100$ independent lightcone simulations covering a volume equivalent to the Galactic North Cap of the Euclid Wide Survey over $0.84<z<1.88$, with a particle mass resolution of $10^{10}\,M_\odot$, for a total cost of 6 million CPU hours.

Silmarilli provides survey-scale dark-matter and halo realizations for empirical covariance estimation, emulator training, and simulation-based inference, demonstrating how sCOLA can make large ensembles of Euclid-scale lightcones computationally tractable for precision cosmology.

Orateur: Mayeul Aubin (PhD student at Institut d'Astrophysique de Paris (IAP))

10:30 → 11:00 coffee break

11:00 → 11:20 Full-Shape Analysis of DESI DR2 and Cross-Correlation with CMB Lensing

After a brief overview of the DESI survey and its timeline, I will present the full-shape analysis of DESI DR2 galaxy clustering data. This analysis constrains the growth rate of structure from the galaxy power spectrum and, for the first time in a DESI collaboration analysis, the bispectrum. I will highlight the methodological developments underpinning this analysis and the resulting (blinded) cosmological constraints. I will then present the cross-correlation analysis with the CMB lensing map from Planck and ACT.

Orateur: Arnaud de Mattia

11:20 → 11:40 Cosmology at z>2 with the DESI Lyman-alpha forest

The Lyman-alpha (Lya) forest provides a unique probe of large-scale structure, extending cosmological measurements to redshifts and scales currently inaccessible to other tracers. The new DESI Lya data offer an unprecedented opportunity for precision cosmology at high redshift.

I will present new DESI Lya results: a new z>2 DR2 BAO measurement in three redshift bins, providing a direct observational test of the Einstein–de Sitter expansion history (2607.19619); the DR2 full-shape analysis, yielding a %-level Alcock–Paczynski measurement, and its implications for the BAO-CMB tension (2607.27410); and constraints from DR1 P1D measurements on departures of the primordial power spectrum from a power law at high k (2601.21432).

Orateur: Eric Armengaud (IRFU - CEA)

11:40 → 12:10 Beyond standard galaxy clustering methods with DESI

Orateur: Pauline Zarrouk (LPNHE)

14:00 → 14:30 TBD

Orateur: Alice PISANI (Aix Marseille University, CNRS/IN2P3, CPPM)

14:30 → 15:00 UNIONS

Orateurs: Cail Daley (CosmoStat, CEA Paris-Saclay), Calum Murray (CEA)

15:00 → 15:30 Dark energy with supernovae

Orateur: Nicolas Regnault (LPNHE)

15:30 → 16:00 coffee break

16:00 → 16:30 Euclid meets CMB lensing: first cosmological constraints from the Euclid Quick Data Release

The Euclid Quick Data Release (Q1) offers the first opportunity to test the cosmological potential of Euclid on real data. In this talk I will present the first cosmological analysis performed on Euclid data, based on the tomographic clustering of photometric galaxies in the three Euclid Deep Fields and their cross-correlation with CMB lensing from Planck and ACT. Despite covering only about 60 square degrees, the data already deliver a combined detection significance above 30σ. I will show how cross-correlating with CMB lensing provides a powerful handle on the systematics that affect early data, from imaging artefacts to photometric redshift errors, and how it helps break the degeneracy between galaxy bias and the growth of structure. The resulting constraints on the matter density, the amplitude of fluctuations and the growth factor are consistent with current state-of-the-art results, and reach comparable precision with a sky area hundreds of times smaller. These results provide a first glimpse of what Euclid, combined with CMB lensing, will achieve with its first major data release.

Orateur: Dr Margherita Lembo (Institut d'Astrophysique de Paris (IAP))

16:30 → 16:50 Growth, geometry, and early-universe split of the matter density parameter Ωm

While the $\Lambda$CDM model can successfully reproduce the measurements of many cosmological probes, some discrepancies have recently emerged. It is therefore necessary to test the standard cosmological model for consistency.

An important stress test is to separate the effect of different cosmological regimes on the parameter inference. We treat three regimes separately here: geometry, growth, and the early Universe. The geometric regime concerns the expansion and curvature history, the growth regime governs structure formation, and the early-universe regime affects physics prior to recombination. Previous analyses have performed the split between geometry and growth, whereas we also consider the influence of the early Universe separately.

We perform this split for the present-day matter density parameter $\Omega_{\rm m}$ using multiple cosmological observables. The data we use are galaxy clustering and weak lensing statistics (3x2pt) from the Dark Energy Survey, cosmic microwave background data from Planck, baryon acoustic oscillations from the Dark Energy Spectroscopic Instrument, type Ia supernovae samples from Pantheon+, and redshift-space distortions from a collection of galaxy surveys. For each of these probes, we introduce a phenomenological split into these three regimes.

The correlation between the geometric and the early regime for the matter density is strong, but that between the growth regime and the others is not. All regimes are compatible in the posterior distribution, but the difference between the geometry and the early regimes, $\Delta \Omega_{\rm m}^{\rm geo,early}$, has a 2$\sigma$ discrepancy with 0.

Orateur: Felicitas Keil (IRAP Toulouse)

16:50 → 17:10 UNIONS: constraining the galaxy-halo shape connection in 3D for intrinsic alignments studies

The 3D morphology of galaxies provides direct insights into the processes governing galaxy formation and evolution within the large-scale structures of the Universe. More precisely, the connection between the morphology of galaxies and halos is a key ingredient in cosmology and astrophysics. It is in particular interesting to understand the intrinsic alignments (IA) of galaxies, the correlation between the intrinsic 3D shapes of galaxies and the underlying tidal field of dark matter, which is known to be an astrophysical systematic effect for cosmic shear analyses. However, information about the 3D morphology of galaxies is lost along the line-of-sight due to the projection of the 3D shapes in 2D. In this talk, I present a novel method that I developed to infer the distribution of 3D galaxy shapes from their observed 2D images using simulation-based inference. I use halo catalogues of the N-body simulation AbacusSummit to extract the 3D shapes and 3D orientations of dark matter halos. Then, I use a Halo Occupation Distribution (HOD) model and galaxy clustering to populate these halos with galaxies. Finally, I present a forward model for the 3D shape of central galaxies that live inside these halos. I applied my method to the galaxy shape catalogue of the Ultraviolet Near-Infrared Optical Northern Survey (UNIONS), a multi-band optical survey which is going to cover 6250 deg^2 in the Northern sky. I obtain constraints on the distribution of 3D galaxy shapes and on the 3D galaxy-halo shape connection from 5800 deg^2 of observed sky. In future work, my idea is to predict an IA signal using this model of 3D morphology, in order to reveal, additionally to the observed galaxy alignments, extra information about the connection between the morphology of galaxies and halos in 3D.

Orateur: Antonin Corinaldi (CEA Paris-Saclay / AIM / CosmoStat)

17:10 → 17:30 Cosmological dependence of the magnification bias from N-body simulations and relativistic ray-tracing

The new generation of galaxy surveys will reach an unprecedented precision, allowing to test cosmology and gravity theories from the relativistic effects on the clustering of matter in the universe. In particular, the magnification bias, the change of observed galaxy density caused by lensing magnification, can be used as a cosmological probe. In this work, we investigate the impact of the magnification bias within the w0wa-nu CDM cosmological framework using a suite of high-resolution N-body simulations. We explore the parameter space using a Latin Hypercube Sampling (LHS), and simulate 80 distinct cosmological models where the different cosmological parameters are varied. Building on the work of Breton et al., 2019 and Rasera et al., 2022, we perform relativistic ray-tracing for each cosmology, to generate lightcone catalogs that include the relativistic effects on the matter distribution, and measure the set of relativistic 3x2pt correlation functions (galaxy clustering, galaxy–galaxy lensing and cosmic shear). We can then compare the observables computed in the relativistic catalog to the same observables computed assuming the Born approximation, to isolate the effect of the magnification bias on these observables.
We then construct an emulator based on Gaussian Processes, trained on the measured correlation functions, to extrapolate the cosmological dependence of the amplitude of the magnification bias. Our emulator provides fast and accurate predictions of the relativistic corrections on the 3x2 pts correlation functions, and can be directly used to perform MCMC from correlation functions measured by observations. We show that the magnification bias induces measurable deviations in 3x2pt statistics, and that the relativistic correction to the galaxy-lensing correlation function is particularly sensitive to the dark energy equation of state. In future work, we plan to extend this analysis to models of modified gravity, to use these relativistic effects as a test of general relativity.

Orateur: Fabien Castillo (LUX, Observatoire de Paris)

17:30 → 17:50 DES: What have we learned about Weak-Lensing analysis within ΛCDM?

The DES collaboration pioneered the combination of cosmic shear, galaxy clustering, and galaxy–galaxy lensing in a 3×2pt cosmological analysis with DES Year 1, and has since performed two more major analyses. In this talk, I will first review the results of the DES Year 6 3×2-point analysis within ΛCDM, together with its combination with other DES probes, which constitutes the final 'standard' cosmological analysis of DES data.
I will then discuss the main challenges faced by 3×2point analyses and highlight some methodological advances made over the years, with a particular focus on shear calibration and redshift calibration.
Finally, I will draw some lessons from the DES experience that are relevant for the analysis of upcoming Stage IV weak-lensing surveys.

Orateur: William D'Assignies D.

17:50 → 18:10 Cheaper neural summary statistics: an ensemble of CNNs trained on low-resolution simulations

Machine learning techniques can be used to build near-optimal summary statistics for extracting cosmological information from the large-scale structure of the Universe. Their training, however, requires large numbers of simulated samples, making them expensive to construct. We aim to build a neural summary statistic trained on fast but approximate low-resolution simulations that remains informative when applied to higher-resolution ones, reducing the training-set cost. We train an ensemble of convolutional neural networks with different architectures to compress the matter overdensity field of low-resolution particle-mesh simulations into a few coefficients. The outputs of a subset of networks are concatenated to build a summary statistic, and we compare three selection strategies for the smallest informative sub-ensemble. We apply this neural summary to derive cosmological constraints on Ωm and σ8 from high-resolution particle-mesh simulations, using a Gaussian process emulator of the neural summary within a Gaussian likelihood, and compare it with a reference power spectrum analysis performed with the same procedure. We also consider dark matter halo catalogues from independent high-resolution N-body simulations. The neural summary extracts information beyond the binned power spectrum content when applied to higher-resolution data. Using multiple networks is essential, with a single network providing worse constraints than the power spectrum in several cases. Among the tested methods, a greedy search on the posterior volume provides the tightest constraints with the smallest subset of networks. The neural summary outperforms the power spectrum when applied to dark matter haloes, for different number densities and redshifts, despite being trained on low-resolution matter fields at z = 0. Using an ensemble of networks rather than a single one is therefore key, and offers a route to cheaper neural summaries.

Orateur: Iñigo Sáez-Casares (Università degli Studi di Milano)

18:10 → 18:30 The ADELINE suite: simulations of massive galaxy clusters in wCDM dark energy models

Understanding how modifications to the dark energy model affect the formation and evolution of galaxy clusters is of crucial importance to avoid potential biases when interpreting current and upcoming cosmological observations. However, this question remains relatively unexplored, particularly when realistic AGN feedback models are taken into account. In this presentation, I introduce the ADELINE suite of simulations, the first set of zoom-in simulations of galaxy clusters performed in alternative dark energy models and incorporating realistic baryonic physics, including AGN feedback.

These simulations provide a unique framework to investigate the sensitivity of the internal properties of galaxy clusters to the underlying dark energy model within the wCDM framework. I will discuss how modifications to the dark energy model impact the thermodynamic and structural properties of galaxy clusters throughout their redshift evolution, and investigate whether these signatures can be degenerate with the effects of AGN feedback.

Finally, I will examine the observational implications of these signatures, focusing on galaxy-cluster observables associated with the Sunyaev–Zeldovich and X-ray signals. These results provide a first step toward assessing the potential of the internal matter distribution of galaxy clusters as a probe of dark energy beyond LambdaCDM, while highlighting the importance of accurately modelling baryonic processes.

Orateur: Félix Tornatore (Observatoire de Paris)

mer. 4 novembre

09:00 → 09:30 Martin Kunz

Orateur: Martin Kunz

09:30 → 10:00 Clotilde Laigle

10:00 → 10:30 The status of dynamical dark energy after the final Dark Energy Survey results

The Dark Energy Survey had a key role in the potential hints for a dynamical dark energy over the past two years. The collaboration recently released its final analysis on the CPL parametrisation of the equation of state, achieving its goal set 20 years ago to test dark energy with multi-probes. After summarizing the recent progress on constraining dynamical dark energy, I will present DES Y6 results on the dark energy equation of state from https://arxiv.org/abs/2605.27221v1, and specifically explain our approach to cosmological inference I co-led in https://arxiv.org/abs/2601.14859 and the challenges we will need to solve for LSST, Euclid and Roman analyses.

Orateur: Agnès Ferté (LPNHE)

10:30 → 11:00 coffee break

11:00 → 11:20 From Precision Cosmology to Accurate Spacetimes

While a new generation of surveys, including LSST and Euclid, is mapping the Universe with unprecedented precision, increasingly sophisticated analysis pipelines are being developed to extract every bit of information from these data. Yet some of the foundational assumptions used to interpret them have, in many respects, remained largely unchanged for decades. We still describe the Universe primarily as perturbations around an FLRW spacetime, model gravitational lensing through layers of approximations and idealized matter distributions, and run most cosmological simulations using Newtonian or weak-field gravity rather than solving the full Einstein equations. In this talk, I will discuss whether our ever-growing statistical and computational sophistication has been matched by a comparable evolution of the physical description of the Universe itself. I will review possible paths beyond the standard modeling framework, from both theoretical and numerical perspectives, and highlight the rich landscape of well-defined problems they open up across physics, mathematics, and computational cosmology. I will argue that, as tensions and anomalies persist and datasets enter an entirely new regime, the time may have come to rethink cosmological modeling from the ground up, taking general relativity in its full generality as the natural foundation. In practice, I will revisit several critical steps in cosmological modeling from a general-relativistic perspective, from the construction of initial conditions and the choice of cosmic topology to non-perturbative spacetime evolution and light propagation, and examine how the assumptions built into our standard framework may constrain the questions we can ask and, ultimately, the conclusions we can draw about the evolution of the Universe.

Orateur: Vincent Reverdy (Laboratoire d'Annecy de Physique des Particules)

11:20 → 11:40 Are we correctly estimating uncertainties in cluster analyses ?

The Legacy Survey of Space and Time (LSST) has now started its ten-year survey. Its promise: detecting of order 100,000 galaxy clusters up to z~1, while providing weak-lensing information for a substantial fraction of them. The weak-lensing measurements will allow us to calibrate the relation between cluster observables, such as richness and halo mass. Combined with complementary Stage IV surveys, these data will offer an unprecedented opportunities for cosmological analyses.

The abundance of galaxy clusters as a function of mass and redshift is a powerful probe of the growth of cosmic structure and expansion history of the Universe, being sensitive to parameters such as OmegaM, sigma8.

As statistical power increases, the accuracy of the cosmological inference increasingly depends on our control of selection effects, calibration uncertainties, astrophysical and theoretical modeling, and statistical methodology.

The current challenge is therefore shifting from an era where the need was to obtain more objects to an era where the need is to establish that the mapping from observables to cosmological parameters is unbiased. In this context, could an apparently significant deviation from ΛCDM or at the contrary an apparently good agreement between ΛCDM and our data, be an artifact of a biased mapping ?

Using the example of galaxy clusters, this talk will review the analysis steps that go from imaging to cosmological inference, focusing on the various assumptions, approximations and potential circularities and hidden dependencies that enter the analysis and may bias the final result. Ideas on how to make the measurement more robust will then be discussed.

Orateur: Narei Lorenzo Martinez (LAPP)

11:40 → 12:00 Gravitational Lensing Beyond Standard Approximations

Are the standard approximations used in gravitational lensing still adequate in the era of precision cosmology? If not, how can we move beyond them?

As cosmological tensions sharpen, a new generation of surveys such as Euclid and LSST is set to deliver an unprecedented volume of data, with billions of galaxies analyzed through increasingly sophisticated pipelines, including machine-learning techniques. Despite the technical progress, the theoretical framework used to interpret weak lensing observations still relies largely on approximations that have remained essentially unchanged for several decades. In this talk, we revisit the relativistic foundations of gravitational lensing and explore a framework that moves beyond these standard assumptions and towards the precision needs of the next generation of cosmological surveys. In particular, we introduce EXCALIBUR, a numerical approach being developed and designed to model light propagation in ΛCDM cosmological settings through the direct integration of null geodesics in first-order perturbed FLRW spacetimes. This method enables the extraction of fully relativistic lightcones for a given cosmological background and provides a unified description of weak and strong lensing within a single pipeline. It relies on evaluating the optical tidal matrix along photon trajectories using local computations of the Riemann tensor, from which the full Jacobi mapping is reconstructed without relying on perturbative expansions. This framework gives direct access to lensing observables such as convergence and shear, while naturally extending to higher-order distortions beyond the standard weak-lensing formalism. We will also present new results from ray-tracing simulations studying the effect of halo triaxiality on gravitational lensing observables. We will discuss the numerical implementation, the physical setup and interpretation of the method, as well as its implications for current and upcoming cosmological surveys.

Orateur: M. Laurent MAGRI-STELLA (LAPP)

12:00 → 12:20 From 3×2pt to 4×3pt: bispectra as calibrators of non-linear bias and intrinsic alignment

Upcoming photometric surveys such as Euclid and LSST will combine galaxy clustering and weak lensing to constrain dark energy, but exploiting mildly non-linear scales requires additional nuisance parameters for galaxy bias and intrinsic alignments, which degrade cosmological constraints. We present a framework combining 1-loop power spectra (3×2pt) with tree-level auto- and cross-bispectra of galaxy clustering and weak lensing (4×3pt), including intrinsic alignments in the TATT model. Second-order nuisance parameters enter power spectra only through loop corrections, but contribute to bispectra already at tree level, so that bispectra can calibrate them efficiently. Performing a Fisher forecast for a Stage IV survey, we show that adding the 4×3pt to the 3×2pt improves dark energy constraints, mostly through this calibration. Since the full 4×3pt becomes computationally prohibitive for a large number of tomographic bins, we also show that a small set of skew-spectra recovers most of its information.

Orateur: Raphaël Kou (IRAP)

12:20 → 12:40 Local PNG with DESI DR2: pipeline, validation, and forward modeling of survey selection effects

Primordial non-Gaussianity (PNG) is one of the few accessible windows into inflation: a significant detection of local PNG, $\left\vert f_\mathrm{NL}^\mathrm{loc} \right\vert \gtrsim 1$, would rule out standard single-field slow roll inflation. The tightest current constraints come from the CMB bispectrum, and are close to the cosmic variance limit. Three-dimensional galaxy surveys such as DESI offer tremendously more modes to exploit, and $f_\mathrm{NL}^\mathrm{loc}$ leaves a distinct signature in the large scales of the galaxy power spectrum through the scale-dependent bias. DESI is forecast to reach CMB-competitive constraints with five years of data.
Measuring PNG via the scale-dependent bias requires rigorous modeling of the large scales of the power spectrum (survey geometry, integral constraints) and good control of observational systematics (imaging systematics). Since the first DESI $f_\mathrm{NL}^\mathrm{loc}$ constraints with DR1 in 2024, the collaboration has substantially revised the modeling of these effects. I will present an overview of the DESI DR2 PNG pipeline, its validation, and the expected $f_\mathrm{NL}^\mathrm{loc}$ constraints. In particular, I will introduce a new forward model method for the power spectrum which accounts simultaneously for survey geometry, integral constraints, and power loss from imaging systematics mitigation, without approximation.

Orateur: Domitille Chebat (CEA Saclay - IRFU - DPhP)

12:40 → 13:10 N-Body Simulations of Luminal Horndeski Theories in the EFTofDE Formalism

The Effective Field Theory of Dark Energy (EFTofDE) is an efficient framework to describe a wide range of modified gravity theories with a limited number of parameters. To robustly constrain them by comparison with clustering and weak lensing data from upcoming large-scale structure surveys, high-resolution cosmological N-body simulations are required to obtain accurate predictions for the matter distribution on non-linear scales. We introduce two new N-body simulation codes for EFTofDE cosmologies: PySCo-EFT, a Python-based particle mesh code, and ECOSMOG-EFT, a RAMSES-based code with adaptive mesh refinement. We consider Horndeski models with a luminal gravitational wave speed. We use iterative solvers and multigrid schemes to solve for the additional scalar field equation in both codes, incorporating the non-linear Vainshtein screening mechanism. We present validation and convergence tests of the codes. We obtain a sub-0.5 percent agreement with linear theory on large scales and a similar agreement between the two codes on non-linear scales. We investigate the impact of the EFTofDE parameters on the matter-power-spectrum ratios between EFTofDE and ΛCDM cases. Depending on the EFTofDE parameters, the screening plays a negligible or dominant role compared to the linearised field equations. We present a preliminary power spectrum emulator for these EFTofDE scenarios.

Orateur: Himanish Ganjoo