arXiv subjects

Siddharth Mishra-Sharma

Publications and source records attributed to Siddharth Mishra-Sharma.

At least 19 recordsLinked to original sources

arXiv2026

DiscoverPhysics: Benchmarking LLMs for Out-of-the-Box Scientific Thinking

Frontier LLMs now perform strongly across a wide range of physics evaluations, but it is hard to disentangle genuine reasoning from recall of established science. We introduce DiscoverPhysics, an interactive benchmark that asks a LLM agent to discover the laws of motion of a simulated world whose physics deliberately deviates from our own. We construct 22 worlds governed by, among others, screened and fractional-power gravity, multi-species couplings, hidden dark-matter-like particles, non-coordinate-free physics, and time-varying interactions. Each world is generated on demand by an N-body simulator, for which the agent proposes several rounds of experiments, observes raw trajectory data, and ultimately submits both a natural-language explanation of the world's physics and a Python implementation of the inferred law. Because solving a world requires the agent to design informative experiments and revise its hypotheses, the benchmark probes long-horizon reasoning over an experimental history. We evaluate submissions along two complementary axes: trajectory MSE on held-out particles and an LLM-judged explanation score following an expert-written rubric assessing conceptual understanding of each world. Across eleven frontier models, we find that the strongest agents pass only half of the worlds and consistently fail on those where latent structure must be uncovered. Open-source models lag substantially behind commercial models, both in their ability to design informative experiments and in extracting conclusions from the data. We further find that good predictive accuracy does not guarantee high explanation quality and that conceptual understanding depends on hypothesis refinement through well-chosen experiments.

stat.ML↗

arXiv2026

Deep Learning for Astrophysics: An Open Textbook from the NASA Cosmic Origins AI/ML Science and Technology Interest Group

Recent community assessments identify education as a principal barrier to adopting modern machine learning in astronomy. We present Deep Learning for Astrophysics, a freely available textbook at https://deeplearning4astro.com, curated from the NASA Cosmic Origins Artificial Intelligence and Machine Learning Science and Technology Interest Group (AI/ML STIG) lecture series. The book collects 23 chapters by 17 lecturers across six parts, moving from computational foundations and deep-learning architectures through generative modeling, simulation-based inference, reinforcement learning, and large-language-model agents to the practice of AI-laden science. Many include executable notebooks using astronomical data.

astro-ph.IM↗

arXiv2026

Program Synthesis for Simulation-Based Inference: Joint Model Selection and Parameter Estimation

Neural simulation-based inference enables parameter estimation for complex models, but typically requires the user to specify a simulator encoding a fixed model structure. We present a framework for joint model selection and parameter estimation that combines large language models for program synthesis with neural simulation-based inference. Given a natural language description of the system and data under investigation, an LLM proposes candidate simulator programs which are iteratively refined via feedback-driven mutation and evaluated using neural density estimation. The approach enables simulation-based inference over a pool of models, not just parameters within a fixed model. On benchmarks spanning deterministic dynamics, stochastic epidemic models, and dark matter substructure inference from gravitational-lensing images, the method identifies plausible model families from open-ended prompts, with accuracy that reflects the information content of the data and identifiability of candidate models.

cs.LG↗

arXiv2026

A Scientific Human-Agent Reproduction Pipeline

Reproducing scientific analyses is essential for preserving knowledge, building extensible codebases, and deepening researcher understanding - yet the effort often outweighs its academic recognition. We argue that the reproduction of scientific data analyses is fundamentally a translation task: converting human-readable knowledge (papers, documentation) into machine-readable analysis code. This makes it uniquely well-suited for AI agents. We present SHARP (Scientific Human-Agent Reproduction Pipeline), a structured framework for reproducing scientific analyses through human-agent collaboration. SHARP decomposes a reproduction task into discrete steps, which an AI agent executes autonomously using specialized subagents for code generation, testing, and quality assurance. At defined checkpoints, the researcher reviews progress, provides feedback, and steers the analysis - keeping the human firmly in control of scientific judgment while the agent handles implementation. We demonstrate SHARP by reproducing a jet classification task in particle physics from a published paper. We evaluate the reproduction along three axes: analysis performance against the original results, code quality and faithfulness, and the nature of the human-agent conversation. The latter is evaluated with a novel framework for characterizing human-agent interactions. Our work highlights a practical model for AI-assisted scientific reproduction where the researcher's role shifts from writing code to understanding, evaluating, and directing - elevating human understanding rather than replacing it.

hep-ph↗

arXiv2026

High-dimensional inference for the $γ$-ray sky with differentiable programming

We motivate the use of differentiable probabilistic programming techniques in order to account for the large model-space inherent to astrophysical $γ$-ray analyses. Targeting the longstanding Galactic Center $γ$-ray Excess (GCE) puzzle, we construct differentiable forward model and likelihood that make liberal use of GPU acceleration and vectorization in order to simultaneously account for a continuum of possible spatial morphologies consistent with the GCE emission in a fully probabilistic manner. Our setup allows for efficient inference over the large model space using variational methods. Beyond application to $γ$-ray data, a goal of this work is to showcase how differentiable probabilistic programming can be used as a tool to enable flexible analyses of astrophysical datasets.

astro-ph.HE↗

arXiv2026

Collider-Bench: Benchmarking AI Agents with Particle Physics Analysis Reproduction

Autonomous language-model agents are increasingly evaluated on long-horizon tool-use tasks, but existing benchmarks rarely capture the complexity and nuance of real scientific work. To address this gap, we introduce Collider-Bench, a benchmark for evaluating whether LLM agents can reproduce experimental analyses from the Large Hadron Collider (LHC) using only public papers and open scientific software. Such analyses are often difficult to reproduce because the public toolchain only approximates the software used internally by the experimental collaborations, while the published papers inevitably omit implementation details needed for a faithful reconstruction. Agents must therefore rely on physical reasoning, domain knowledge, and trial-and-error to fill these gaps. Each task requires the agent to turn a published analysis into an executable simulation-and-selection pipeline and submit predicted collision event yields in specified signal regions. These predictions are evaluated with standard histogram metrics that provide continuous fidelity scores without a hand-written rubric. We also report the computational cost incurred by each agent per task. Finally, we evaluate the codebase and full session trace using an LLM judge to catch qualitative failure modes such as fabrications, hallucinations and duplications. We release an initial set of tasks drawn from LHC searches, together with a containerized sandbox and event simulation tools. We evaluate across a capability ladder of general purpose coding agents. Our results show that on average no agent reliably beats the physicist-in-the-loop solution.

cs.LG↗

arXiv2026

spectroxide: A code package for computing cosmic microwave background spectral distortions

We present spectroxide, a code package for computing cosmic microwave background spectral distortions in which all ${\sim}14{,}500$ lines of Rust code, Python interface, and ${\sim}400$ automated tests were written by an AI assistant (Claude Code) under human physicist supervision. The solver evolves the photon Boltzmann equation under Compton scattering, double Compton emission, and Bremsstrahlung from $z \sim 5 \times 10^6$ to the present, computing spectral distortions from arbitrary heat and photon injection within this redshift range. No fully open-source code of this kind is publicly available; we validate against analytic limits, published spectra, and publicly available precomputed Green's function tables. We document the development as a case study in AI-assisted scientific computing, highlighting how domain expertise caught physics bugs (incorrect dimensional prefactors, near-cancellation errors) that evaded the full automated test suite, and provide recommendations for best practices in human--AI collaborative development of scientific software. We make spectroxide publicly available on GitHub.

astro-ph.CO↗

arXiv2025

LINX: A Fast, Differentiable, and Extensible Big Bang Nucleosynthesis Package

We introduce LINX (Light Isotope Nucleosynthesis with JAX), a new differentiable public Big Bang Nucleosynthesis (BBN) code designed for fast parameter estimation. By leveraging JAX, LINX achieves both speed and differentiability, enabling the use of Bayesian inference, including gradient-based methods. We discuss the formalism used in LINX for rapid primordial elemental abundance predictions and give examples of how LINX can be used. When combined with differentiable Cosmic Microwave Background (CMB) power spectrum emulators, LINX can be used for joint CMB and BBN analyses without requiring extensive computational resources, including on personal hardware.

astro-ph.CO↗

arXiv2025

Cosmological Parameter Estimation with a Joint-Likelihood Analysis of the Cosmic Microwave Background and Big Bang Nucleosynthesis

We present the first joint-likelihood analysis of Big Bang Nucleosynthesis (BBN) and Cosmic Microwave Background (CMB) data. Bayesian inference is performed on the baryon abundance and the effective number of neutrino species, $N_{\rm eff}$, using a CMB Boltzmann solver in combination with LINX, a new flexible and efficient BBN code. We marginalize over Planck nuisance parameters and nuclear rates to find $N_{\rm{eff}} = 3.08_{-0.13}^{+0.13},\,2.94 _{-0.16}^{+0.16},$ or $2.98_{-0.13}^{+0.14}$, for three separate reaction networks. This framework enables robust testing of the Lambda Cold Dark Matter paradigm and its variants with CMB and BBN data.

astro-ph.CO↗

arXiv2025

FLORAH-Tree: Emulating Dark Matter Halo Merger Trees with Graph Generative Models

Merger trees track the hierarchical assembly of dark matter halos across cosmic time and serve as essential inputs for semi-analytic models of galaxy formation. However, conventional methods for constructing merger trees rely on ad-hoc assumptions and are unable to incorporate environmental information. Nguyen et al. (2024) introduced FLORAH, a generative model based on recurrent neural networks and normalizing flows, for modeling main progenitor branches of merger trees. In this work, we extend this model, now referred to as FLORAH-Tree, to generate complete merger trees by representing them as graph structures that capture the full branching hierarchy. We trained FLORAH-Tree on merger trees extracted from the Very Small MultiDark Planck cosmological N-body simulation. To validate our approach, we compared the generated merger trees with both the original simulation data and with semi-analytic trees produced using the Extended Press-Schechter (EPS) formalism. We show that FLORAH-Tree accurately reproduces key merger rate statistics across a wide range of mass and redshift, outperforming the conventional EPS-based approach. We demonstrate its utility by applying the Santa Cruz semi-analytic model (SAM) to generated trees and showing that the resulting galaxy-halo scaling relations, such as the stellar-to-halo-mass relation and supermassive black hole mass-halo mass relation, closely match those from applying the SAM to trees extracted directly from the simulation. FLORAH-Tree provides a computationally efficient method for generating merger trees that maintain the statistical fidelity of N-body simulations.

astro-ph.GA↗

arXiv2025

Trial by FIRE: Probing the dark matter density profile of dwarf galaxies with GraphNPE

The Dark Matter (DM) distribution in dwarf galaxies provides crucial insights into both structure formation and the particle nature of DM. GraphNPE (Graph Neural Posterior Estimator), first introduced in Nguyen et al. (2023), is a novel simulation-based inference framework that combines graph neural networks and normalizing flows to infer the DM density profile from line-of-sight stellar velocities. Here, we apply GraphNPE to satellite dwarf galaxies in the FIRE-2 Latte simulation suite of Milky Way-mass halos, testing it against both Cold and Self-Interacting DM scenarios. Our method demonstrates superior precision compared to conventional Jeans-based approaches, recovering DM density profiles to within the 95% confidence level even in systems with as few as 30 tracers. Moreover, we present the first evaluation of mass modeling methods in constraining two key parameters from realistic simulations: the peak circular velocity, $V_\mathrm{max}$, and the peak virial mass, $M_\mathrm{200m}^\mathrm{peak}$. Using only line-of-sight velocities, GraphNPE can reliably recover both $V_\mathrm{max}$ and $M_\mathrm{200m}^\mathrm{peak}$ within our quoted uncertainties, including those experiencing tidal effects ($\gtrsim$ 63% of systems are recovered with our 68% confidence intervals and $\gtrsim$ 92% within our 95% confidence intervals). The method achieves 10-20% accuracy in $V_\mathrm{max}$ recovery, while $M_\mathrm{200m}^\mathrm{peak}$ is recovered to 0.1-0.4 dex accuracy. This work establishes GraphNPE as a robust tool for inferring DM density profiles in dwarf galaxies, offering promising avenues for constraining DM models. The framework's potential extends beyond this study, as it can be adapted to non-spherical and disequilibrium models, showcasing the broader utility of simulation-based inference and graph-based learning in astrophysics.

astro-ph.GA↗

arXiv2025

Detecting Model Misspecification in Cosmology with Scale-Dependent Normalizing Flows

Current and upcoming cosmological surveys will produce unprecedented amounts of high-dimensional data, which require complex high-fidelity forward simulations to accurately model both physical processes and systematic effects which describe the data generation process. However, validating whether our theoretical models accurately describe the observed datasets remains a fundamental challenge. An additional complexity to this task comes from choosing appropriate representations of the data which retain all the relevant cosmological information, while reducing the dimensionality of the original dataset. In this work we present a novel framework combining scale-dependent neural summary statistics with normalizing flows to detect model misspecification in cosmological simulations through Bayesian evidence estimation. By conditioning our neural network models for data compression and evidence estimation on the smoothing scale, we systematically identify where theoretical models break down in a data-driven manner. We demonstrate a first application to our approach using matter and gas density fields from three CAMELS simulation suites with different subgrid physics implementations.

astro-ph.CO↗

arXiv2025

Auditing language models for hidden objectives

We study the feasibility of conducting alignment audits: investigations into whether models have undesired objectives. As a testbed, we train a language model with a hidden objective. Our training pipeline first teaches the model about exploitable errors in RLHF reward models (RMs), then trains the model to exploit some of these errors. We verify via out-of-distribution evaluations that the model generalizes to exhibit whatever behaviors it believes RMs rate highly, including ones not reinforced during training. We leverage this model to study alignment audits in two ways. First, we conduct a blind auditing game where four teams, unaware of the model's hidden objective or training, investigate it for concerning behaviors and their causes. Three teams successfully uncovered the model's hidden objective using techniques including interpretability with sparse autoencoders (SAEs), behavioral attacks, and training data analysis. Second, we conduct an unblinded follow-up study of eight techniques for auditing the model, analyzing their strengths and limitations. Overall, our work provides a concrete example of using alignment audits to discover a model's hidden objective and proposes a methodology for practicing and validating progress in alignment auditing.

cs.AI↗

arXiv2024

A point cloud approach to generative modeling for galaxy surveys at the field level

We introduce a diffusion-based generative model to describe the distribution of galaxies in our Universe directly as a collection of points in 3-D space (coordinates) optionally with associated attributes (e.g., velocities and masses), without resorting to binning or voxelization. The custom diffusion model can be used both for emulation, reproducing essential summary statistics of the galaxy distribution, as well as inference, by computing the conditional likelihood of a galaxy field. We demonstrate a first application to massive dark matter haloes in the Quijote simulation suite. This approach can be extended to enable a comprehensive analysis of cosmological data, circumventing limitations inherent to summary statistic -- as well as neural simulation-based inference methods.

astro-ph.CO↗

arXiv2024

Inferring the Morphology of the Galactic Center Excess with Gaussian Processes

Descriptions of the Galactic Center using Fermi gamma-ray data have so far modeled the Galactic Center Excess (GCE) as a template with fixed spatial morphology or as a linear combination of such templates. Although these templates are informed by various physical expectations, the morphology of the excess is a priori unknown. For the first time, we describe the GCE using a flexible, non-parametric machine learning model -- the Gaussian process (GP). We assess our model's performance on synthetic data, demonstrating that the model can recover the templates used to generate the data. We then fit the \Fermi data with our model in a single energy bin from 2-20 GeV (leaving a spectral GP analysis of the GCE for future work) using a variety of template models of diffuse gamma-ray emission to quantify our fits' systematic uncertainties associated with diffuse emission modeling. We interpret our best-fit GP in terms of GCE templates consisting of an NFW squared template and a bulge component to determine which bulge models can best describe the fitted GP and to what extent the best-fit GP is described better by an NFW squared template versus a bulge template. The best-fit GP contains morphological features that are typically not associated with traditional GCE studies. These include a localized bright source at around $(\ell,b) = (20^{\circ}, 0^{\circ})$ and a diagonal arm extending Northwest from the Galactic Center. In spite of these novel features, the fitted GP is explained best by a template-based model consisting of the bulge presented in Coleman et al. (2020) and a squared NFW component. Our results suggest that the physical interpretation of the GCE in terms of stellar bulge and NFW-like components is highly sensitive to the assumed morphologies, background models, and the region of the sky used for inference.

astro-ph.HE↗

arXiv2024

A Cosmic-Scale Benchmark for Symmetry-Preserving Data Processing

Efficiently processing structured point cloud data while preserving multiscale information is a key challenge across domains, from graphics to atomistic modeling. Using a curated dataset of simulated galaxy positions and properties, represented as point clouds, we benchmark the ability of graph neural networks to simultaneously capture local clustering environments and long-range correlations. Given the homogeneous and isotropic nature of the Universe, the data exhibits a high degree of symmetry. We therefore focus on evaluating the performance of Euclidean symmetry-preserving ($E(3)$-equivariant) graph neural networks, showing that they can outperform non-equivariant counterparts and domain-specific information extraction techniques in downstream performance as well as simulation-efficiency. However, we find that current architectures fail to capture information from long-range correlations as effectively as domain-specific baselines, motivating future work on architectures better suited for extracting long-range information.

cs.LG↗

arXiv2024

Full-shape analysis with simulation-based priors: constraints on single field inflation from BOSS

Perturbative, or effective field theory (EFT)-based, full-shape analyses of galaxy clustering data involve ``nuisance parameters'' to capture various observational effects such as the galaxy-dark matter connection (galaxy bias). We present an efficient approach to set informative physically motivated priors on these parameters. We extract these priors from simulated galaxy catalogs based on halo occupation distribution (HOD) models. First, we build a joint distribution between EFT galaxy bias and HOD parameters from a set of 10,500 HOD mock catalogs. We use the field level EFT technique that allows for cosmic variance cancellation, enabling a precision calibration of EFT parameters from computationally inexpensive small-volume simulations. Second, we use neural density estimators -- normalizing flows -- to model the marginal probability density of the EFT parameters, which can be used as a prior distribution in full shape analyses. As a first application, we use our HOD-based priors in a new analysis of galaxy power spectra and bispectra from the BOSS survey in the context of single field primordial non-Gaussianity. We find that our priors lead to a reduction of the posterior volume of bias parameters by an order of magnitude. We also find $f_{\rm NL}^{\rm equil} = 320\pm 300$ and $f_{\rm NL}^{\rm ortho} = 100\pm 130$ (at 68\% CL) in a combined two-template analysis, representing a $\approx 40\%$ improvement in constraints on single field primordial non-Gaussianity, equivalent to doubling the survey volume.

astro-ph.CO↗

arXiv2024

How DREAMS are made: Emulating Satellite Galaxy and Subhalo Populations with Diffusion Models and Point Clouds

The connection between galaxies and their host dark matter (DM) halos is critical to our understanding of cosmology, galaxy formation, and DM physics. To maximize the return of upcoming cosmological surveys, we need an accurate way to model this complex relationship. Many techniques have been developed to model this connection, from Halo Occupation Distribution (HOD) to empirical and semi-analytic models to hydrodynamic. Hydrodynamic simulations can incorporate more detailed astrophysical processes but are computationally expensive; HODs, on the other hand, are computationally cheap but have limited accuracy. In this work, we present NeHOD, a generative framework based on variational diffusion model and Transformer, for painting galaxies/subhalos on top of DM with an accuracy of hydrodynamic simulations but at a computational cost similar to HOD. By modeling galaxies/subhalos as point clouds, instead of binning or voxelization, we can resolve small spatial scales down to the resolution of the simulations. For each halo, NeHOD predicts the positions, velocities, masses, and concentrations of its central and satellite galaxies. We train NeHOD on the TNG-Warm DM suite of the DREAMS project, which consists of 1024 high-resolution zoom-in hydrodynamic simulations of Milky Way-mass halos with varying warm DM mass and astrophysical parameters. We show that our model captures the complex relationships between subhalo properties as a function of the simulation parameters, including the mass functions, stellar-halo mass relations, concentration-mass relations, and spatial clustering. Our method can be used for a large variety of downstream applications, from galaxy clustering to strong lensing studies.

astro-ph.GA↗