Graph Path Likelihood for Galaxy Formation on Layered Halo Graphs

Daneng Yang · 2026

Read Graph Path Likelihood for Galaxy Formation on Layered Halo Graphs by Daneng Yang (2026) 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-2603-15128.txt.

Introduction

Yang frames galaxy formation as a stochastic inference problem on layered halo merger graphs. Cutting a full simulation down to a few variables leaves unresolved parts that act like noise. His Graph Path Likelihood Model treats assembly histories as random paths on these graphs. Time edges carry causal transport, while same-age host edges carry environmental influence. A graph neural likelihood model trained on simulation data reproduced the main statistics of stellar and gas mass histories. It also captured linked fluctuations shaped by environment.

Catalog record

CREATORDaneng Yang
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE92054
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.15128

Browse the full library · About the Tower of Babel

Full text

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

Graph Path Likelihood for Galaxy Formation on Layered Halo Graphs Daneng Yang Abstract Likelihood-based forward modeling is standard in galaxy formation, but most implementations are formulated as forward maps rather than explicit trajectory-level likelihoods conditioned jointly on assembly history and environment. We introduce a Graph Path Likelihood Model (GPLM) on layered halo graphs, where temporal edges encode causal transport and coeval host edges encode environmental conditioning. On a fixed layered graph, the graph-conditioned path measure is written as P ⁡ ( 𝐱 ∣ G ) ∝ p attach ​ ( 𝐱 ∣ G ) ​ exp ⁡ [ − S ⁡ ( 𝐱 , G ) ] P(\mathbf{x}\mid G)\propto p_{\rm attach}(\mathbf{x}\mid G)\exp[-S(\mathbf{x};G)] , where S S is an effective action for dynamical increments, currently implemented as a Gaussian Onsager-Machlup term, and p attach p_{\rm attach} is a boundary measure for node entry. We also discuss a minimal preferential attachment-detachment prescription for the graph probability P ⁡ ( G ) P(G) , which facilitates placing the likelihood within a cosmological ensemble of layered graphs. Trained on layered graphs reconstructed from TNG50-1, GPLM improves stellar- and gas-mass predictions over transport-only baselines. As fixed-graph applications, we evaluate dark-matter-deficient-galaxy operator averages, compute gas-channel response under controlled deformations, and compare full and host-ablated path measures through likelihood-ratio diagnostics. In these examples, higher-order satellites show a higher incidence of dark-matter deficiency and broader graph-to-graph variation, while the gas-rich response indicates more diverse environmental processing histories. GPLM thus provides a bridge between astrophysical forward modeling, stochastic effective actions on structured histories, and path-integral diagnostics of environment-dependent galaxy evolution. I Introduction Galaxy formation models have successfully reproduced population statistics and galaxy catalogs by tuning key physical processes to match the observed numbers and properties of galaxies. These methods provide powerful tools for studying galaxy evolution and for reproducing observed distributions White and Frenk 1991 ; Cole et al. 2000 ; Benson 2012 ; Somerville and Davé 2015 ; Guo et al. 2016 ; Lacey et al. 2016 ; Nelson et al. 2019 ; Nelson et al. 2019 ; Robertson and Benson 2026 . However, they are typically formulated as forward maps rather than as explicit likelihoods for baryonic histories conditioned on assembly and environment. This makes quantities such as trajectory rarity, likelihood ratios between trajectories, and operator insertions difficult to define in a controlled manner. A useful way to make both assembly transport and local environment explicit within a single framework is to adopt a graph-based description Yang and Yu 2023 ; Makinen et al. 2022 ; Jespersen et al. 2022 . We therefore introduce a layered halo graph in which temporal edges encode progenitor-to-descendant transport and coeval host edges encode satellite-to-host conditioning. This structure supports a likelihood model that marginalizes unresolved baryonic and environmental variability into an effective graph-conditioned stochastic dynamics, shifting the central focus from a deterministic forward map to a path measure over layered merger histories. Related work in cosmology covers the reconstruction of dark-matter density, velocity, tidal fields, and even initial conditions from survey data and group catalogs. Methods include Wiener-filtering, halo-domain reconstruction, ELUCID, and recent machine-learning pipelines Zaroubi et al. 1995 ; Erdogdu et al. 2006 ; Wang et al. 2009 ; Wang et al. 2012 ; Wang et al. 2013 ; Wang et al. 2014 ; Wang et al. 2016 ; Krywonos et al. 2026 ; Qin et al. 2023 ; Dai et al. 2023 ; Parker et al. 2024 ; Shi et al. 2025 ; Cranmer et al. 2020 . CAMELS-style benchmark suites motivate robust cross-model evaluation by exposing subgrid and domain shift in a controlled setting Villaescusa-Navarro et al. 2021 ; Villaescusa-Navarro et al. 2023 . The CAMELS-SAM extension supports large SAM-based catalog ensemble comparisons within the same benchmarks Perez et al. 2023 . Probabilistic approaches to the halo–galaxy connection are also being developed at the level of conditional or joint distributions of galaxy properties given halo features Yang et al. 2008 ; Stiskalek et al. 2022 ; Rodrigues et al. 2023 ; Lovell et al. 2023 ; Zhang et al. 2023 ; Xu et al. 2024 ; Rodrigues et al. 2025 ; Jagvaral et al. 2025 . Graph-based studies have grown in recent years. This includes merger-tree and catalog graph neural network (GNN) emulators Jespersen et al. 2022 ; Huertas-Company and Lanusse 2023 ; Villanueva-Domingo and Villaescusa-Navarro 2022 ; Nguyen et al. 2025 ; Lee and Villaescusa-Navarro 2025 , graph IMNN compression for simulation-based cosmological inference Makinen et al. 2022 , graph models for halo clustering Yang and Yu 2023 , and recent cross-model or multimodal inference pipelines built around graph-structured cosmological data Jo et al. 2025 ; Nguyen et al. 2026 ; Rose et al. 2025 . Notably, Mangrove applies GNN to halo merger trees to predict galaxy properties from halo assembly histories Jespersen et al. 2022 , and FLORAH uses recurrent neural networks together with normalizing flows to learn the probability distribution of dark matter halo assembly histories Nguyen et al. 2025 . These methods connect to optimal-compression techniques for likelihood-free inference Heavens et al. 2000 ; Alsing et al. 2018 ; Charnock et al. 2018 . They mainly focus on Eulerian fields, catalog predictions, or compressed parameter summaries. In comparison, our goal is different. We seek an explicit trajectory-level likelihood on a fixed layered graph, allowing transport and environment channels to be represented separately. This enables targeted deformations in path space. To achieve this, we introduce a Graph Path Likelihood Model (GPLM) on layered halo graphs with conditional path weight P(\mathbf{x}\mid G)\propto p_{\rm attach}(\mathbf{x}\mid G)\exp\left[-S(\mathbf{x};G)\right], (1) where S is the within-graph effective action for dynamical increments, and p_{\rm attach} is a boundary measure for entry or infall interfaces. It defines a conditional ensemble of baryonic histories on a fixed layered graph with both assembly and environment information encoded. More generally, the fixed-graph construction can be embedded into a graph-trajectory measure P(\mathbf{x},G)=P(\mathbf{x}\mid G)P(G), (2) where P(G) is the graph-sector probability in a graph ensemble. As a minimal closure of the formalism, we introduce a preferential attachment-detachment (PAD) mechanism, extended from the attachment- and host-edges-only version in Ref. Yang and Yu 2023 . PAD provides an explicit generative rule, so the graph prior can be expressed as a product of conditional detachment probabilities. Nevertheless, the main applications in this work remain on fixed graphs. We train and test on layered graphs reconstructed from IllustrisTNG-50-1 Nelson et al. 2019 ; Pillepich et al. 2019 ; Nelson et al. 2019 . For this paper, we use stellar and gas masses as a minimal two-field demonstration, but the construction itself is not tied to those variables. The likelihood is realized by a discrete-time Gaussian Onsager-Machlup (OM) action Onsager and Machlup 1953 on top of deterministic merger transport. Building on the trained model, we then perform several example calculations on fixed graphs. We use a DMDG (dark-matter-deficient galaxy) operator to demonstrate quenched path-integral evaluation with and without p_{\rm attach} . We also introduce a controlled gas-channel deformation, implemented by host-edge ablation, to compute operator responses, susceptibilities, and projected log-probability profiles. We then add path-space likelihood-ratio diagnostics based on the ratios between full and host-ablated dynamics. Because the likelihood is defined on graph-conditioned trajectories rather than on a specific field choice, the same

Read the complete text