2025-07-11 | | Total: 26
We present a multi-agent system for automation of scientific research tasks, cmbagent. The system is formed by about 30 Large Language Model (LLM) agents and implements a Planning & Control strategy to orchestrate the agentic workflow, with no human-in-the-loop at any point. Each agent specializes in a different task (performing retrieval on scientific papers and codebases, writing code, interpreting results, critiquing the output of other agents) and the system is able to execute code locally. We successfully apply cmbagent to carry out a PhD level cosmology task (the measurement of cosmological parameters using supernova data) and evaluate its performance on two benchmark sets, finding superior performance over state-of-the-art LLMs. The source code is available on GitHub, demonstration videos are also available, and the system is deployed on HuggingFace and will be available on the cloud.
Understanding variations in the dust extinction curve is imperative for using dust as a tracer of local structure in the interstellar medium, understanding dust chemistry, and observational color corrections where dust is a nuisance parameter. However, the extinction curve is complicated and exhibits features across a wide range of wavelength scales, from narrow atomic lines and diffuse interstellar bands ("DIBs"), to intermediate-scale and very broad structures ("ISS" and "VBS"), and the overall slope of the optical extinction curve, parameterized by R(V). Robust, population-level studies of variations in these features are only now possible with large, all-sky, spectroscopic surveys. However, these features are often studied independently because they require drastically different spectral resolution. In this work, we couple features with disparate wavelength scales by cross-matching precision catalogs of DIB measurements from APOGEE and Gaia RVS with low-resolution extinction-curve measurements from Gaia XP. Using this combination, we show that there are meaningful correlations between the strengths of extinction-curve features across all wavelength scales. We present a model that statistically explains part of the excess scatter in DIB strength versus extinction, and we show variation in line shapes of two DIBs as a function of R(V). We find that most DIBs increase in strength with increasing R(V) and/or increasing strength of the ISS, though we found one DIB that anomalously decreases in strength with increasing R(V). Using the behavior of the ensemble of DIBs in APOGEE, we present this as the first evidence of systematic chemical variation accompanying R(V) variation.
We present GokuNEmu, a ten-dimensional neural network emulator for the nonlinear matter power spectrum, designed to support next-generation cosmological analyses. Built on the Goku N-body simulation suite and the T2N-MusE emulation framework, GokuNEmu predicts the matter power spectrum with ∼0.5% average accuracy for redshifts 0≤z≤3 and scales 0.006≤k/(hMpc−1)≤10. The emulator models a 10D parameter space that extends beyond ΛCDM to include dynamical dark energy (characterized by w0 and wa), massive neutrinos (∑mν), the effective number of neutrinos (Neff), and running of the spectral index (αs). Its broad parameter coverage, particularly for the extensions, makes it the only matter power spectrum emulator capable of testing recent dynamical dark energy constraints from DESI. In addition, it requires only ∼2 milliseconds to predict a single cosmology on a laptop, orders of magnitude faster than existing emulators. These features make GokuNEmu a uniquely powerful tool for interpreting observational data from upcoming surveys such as LSST, Euclid, the Roman Space Telescope, and CSST.
Accurate and efficient simulation-based emulators are essential for interpreting cosmological survey data down to nonlinear scales. Multifidelity emulation techniques reduce simulation costs by combining high- and low-fidelity data, but traditional regression methods such as Gaussian processes struggle with scalability in sample size and dimensionality. In this work, we present T2N-MusE, a neural network framework characterized by (i) a novel 2-step multifidelity architecture, (ii) a 2-stage Bayesian hyperparameter optimization, (iii) a 2-phase k-fold training strategy, and (iv) a per-z principal component analysis strategy. We apply T2N-MusE to selected data from the Goku simulation suite, covering a 10-dimensional cosmological parameter space, and build emulators for the matter power spectrum over a range of redshifts with different configurations. We find the emulators outperform our earlier Gaussian process models significantly and demonstrate that each of these techniques is efficient in training neural networks or/and effective in improving generalization accuracy. We observe a reduction in validation error by more than a factor of five compared to previous work. This framework has been used to build the most powerful emulator for the matter power spectrum, GokuNEmu, and will also be used to construct emulators for other statistics in future.
Surface radio antenna-based measurements of cosmic-ray air showers present significant computational challenges in accurately reconstructing physics observables, in particular, the depth of shower maximum, Xmax. State-of-the-art template fitting methods rely on extensive simulation libraries, limiting scalability. This work introduces a technique utilizing graph neural networks to reconstruct key air-shower parameters, in particular, direction and shower-core, energy, and Xmax. For training and testing of the networks, we use a CoREAS simulation library made for a future enhancement of IceCube's surface array with radio antennas. The neural networks provide a scalable framework for large-scale data analysis for next-generation astroparticle observatories, such as IceCube-Gen2.
The IceCube Neutrino Observatory is a multi-messenger observatory at the South Pole. As preparation for an enhancement of its surface array, IceTop, a prototype station consisting of elevated scintillation panels and radio antennas has been installed and is operating since 2020. The radio antennas detect emissions from cosmic-ray-induced air showers, and their precise orientation is essential for an accurate reconstruction of the air-shower properties. This work presents a novel method to determine the orientation by analyzing periodic variations of the Galactic background noise recorded by the antennas. In particular, we examine noise level variations correlated with the Earth's rotation and the apparent position of the Galactic Center. The method can provide a potential alternative or augment GPS-based measurements of the alignment of radio antennas at the South Pole.
IceCube Upgrade, starting in the Antarctic summer season 2025/2026, will enhance the sensitivity of the current IceCube in the GeV range and improve understanding of the ice properties. Around 700 new modules will be deployed deep in the ice along seven strings, spaced 3m vertically within a horizontal region of 100m. IceCube Upgrade mainly consists of two types of Cherenkov photon detectors, ∼280 Dual optical sensors in an Ellipsoid Glass for Gen2 (D-Eggs) and ∼400 multi-PMT Digital Optical Modules (mDOMs). Here, we present the status of mDOM production and acceptance testing for mass-produced mDOMs. The majority of the modules have already been produced, and all 128 modules on the first 2 strings have been successfully shipped to the South Pole after multiple steps of testing. Each mDOM contains 24 photomultipliers nearly isotropically distributed on the surface to achieve uniform photon sensitivity, making the integration and testing complicated. To reduce the amount of testing time, we have developed parallelized and automated software.
We have investigated the lensing event KMT-2024-BLG-0404. The light curve of the event exhibited a complex structure with multiple distinct features, including two prominent caustic spikes, two cusp bumps, and a brief discontinuous feature between the caustic spikes. While a binary-lens model captured the general anomaly pattern, it could not account for a discontinuous anomaly feature between the two caustic spikes. To explore the origin of the unexplained feature, we conducted more advanced modeling beyond the standard binary-lens framework. This investigation demonstrated that the previously unexplained anomaly was resolved by introducing an additional lens component with planetary mass. The estimated masses of the lens components are Mp=17.3+25.5−8.8 ME for the planet, and Mh,A=0.090+0.133−0.046 M⊙ and Mh,B=0.026+0.038−0.013 M⊙ for the binary host stars. Based on these mass estimates, the lens system is identified as a planetary system where a Uranus-mass planet orbits a binary consisting of a late M dwarf and a brown dwarf. The distance to the planetary system is estimated to be DL=7.21+0.93−0.97~kpc, with an 82\% probability that it resides in the Galactic bulge. This discovery represents the ninth planetary system found through microlensing with a planet orbiting a binary host. Notably, it is the first case where the host consists of both a star and a brown dwarf.
These notes are very much work-in-progress and simply intended to showcase, in various degrees of details (and rigour), some of the cosmology calculations that class_sz can do. We describe the class_sz code in C, Python and Jax. Based on the Boltzmann code class, it can compute a wide range of observables relevant to current and forthcoming CMB and Large Scale Structure surveys. This includes galaxy shear and clustering, CMB lensing, thermal and kinetic Sunyaev and Zeldovich observables, Cosmic Infrared Background, cross-correlations and three-point statistics. Calculations can be done either within the halo model or the linear bias model. For standard ΛCDM cosmology and extensions, class_sz uses high-accuracy cosmopower emulators of the CMB and matter power spectrum to accelerate calculations. With this, along with efficient numerical integration routines, most class_sz output can be obtained in less than 500 ms (CMB Cℓ's or matter P(k) take O(1ms)), allowing for fast or ultra-fast parameter inference analyses. Parts of the calculations are "jaxified", so the software can be integrated into differentiable pipelines.
The IceCube Neutrino Observatory searches for the origins of astrophysical neutrinos using various techniques to overcome the significant backgrounds produced by cosmic-ray air showers. One such technique involves combining the neutrino data with other cosmic messengers to identify spatial and temporal correlations. IceCube contributes to multi-messenger astrophysics (MMA) by providing alerts for interesting events observed in the detector. The Gamma-ray Follow-Up (GFU) cluster alert system is one stream that identifies potential neutrino flares in realtime, producing around 20 alerts per year. GFU-cluster alerts have been privately shared with Imaging Air Cherenkov Telescopes (IACTs) through memoranda of understanding since IceCube's predecessor, AMANDA. To preserve blindness to the full behavior of our data, the current system mutes updates from sources following the initial GFU-cluster alert sent, preventing further updates until the activity drops below the alert threshold. With growing knowledge of the potential environments that produce astrophysical neutrinos and to foster open collaboration, the GFU-cluster alerts will shift to be publicly shared. Additionally, the new alert platform will provide all above-threshold information such that the source behavior after the initial alert is not obscured. The above threshold data will be distributed through an interactive website that will update the community on the status of active GFU-cluster alerts. This presentation will introduce the new GFU-cluster platform and the accompanying website, soon to be accessible to the MMA community.
We analyse ESPRESSO data for the stars HD10700, HD20794, HD102365, and HD304636 acquired via its Guaranteed Time Observations (GTO) programme. We characterise the stars' radial velocity (RV) signals down to a precision of 10 cm/s on timescales ranging from minutes to planetary periods falling within the host's habitable zone (HZ). We study the RV signature of pulsation, granulation, and stellar activity, inferring the potential presence of planets around these stars. Thus, we outline the population of planets that while undetectable remain compatible with the available data. A simple model of stellar pulsations successfully reproduced the intra-night RV scatter of HD10700 down to a few cm/s. For HD102365 and HD20794, an additional source of scatter at the level of several 10 cm/s remains necessary to explain the data. A kima analysis was used to evaluate the number of planets supported by the nightly averaged time series of each of HD10700, HD102365, and HD304636, under the assumption that a quasi-periodic Gaussian process (GP) regression is able to model the activity signal. While a frequency analysis of HD10700 RVs is able to identify a periodic signal at 20d, when it is modelled along with the activity signal the signal is formally non-significant. ESPRESSO data on their own do not provide conclusive evidence for the existence of planets around these three stars. ESPRESSO is shown to reach an on-sky RV precision of better than 10 cm/s on short timescales (<1h) and of 40 cm/s over 3.5 yr. A subdivision of the datasets showcases a precision reaching 20-30 cm/s over one year. These results impose stringent constraints on the impact of granulation mechanisms on RV. In spite of no detections, our analysis of HD10700 RVs demonstrates a sensitivity to planets with a mass of 1.7M⊕ for periods of up to 100 d, and a mass of 2-5M⊕ for the star's HZ. (abridged)
FRB 20180916B is a repeating Fast Radio Burst (FRB) which produces bursts in a 5.1 day active window which repeats with a 16.34 day period. Models have been proposed to explain the periodicity using dynamical phenomena such as rotation, precession or orbital motion. Polarization Position Angle (PA) of the bursts can be used to distinguish and constraint the origin of the long term periodicity of the FRB. We aim to study the PA variability on short (within an observation) and long timescales (from observation to observation). We aim to compare the observed PA variability with the predictions of various dynamical progenitor models for the FRB. We use the calibrated burst dataset detected by uGMRT in Band 4 (650 MHz) which have been published in arXiv:2409.12584 . We transform the PA measured at 650 MHz to infinite frequency such that PAs measured in different observations are consistent, and finally measure the changes within and across active windows. We find that PA of the bursts vary according to the periodicity of the source. We constrain the PA variability to be within seven degrees on timescales less than four hours for all MJDs. In addition, we also tentatively note the PA measured at the same phase in the active window varies from one cycle to another. Using the findings, we constrain rotational, precession and binary progenitor models. Rotational model partially agrees with observed PA variability but requires further study to fully constrain. We robustly rule out all flavors of precessional models where either precession explains the periodicity of the FRB or the variability from one cycle to another. Lastly, we draw similarities between FRB 20180916B and a X-ray binary system, Her X 1, and explicitly note that both the sources exhibit a similar form of PA variability.
Data collected so far by the Pierre Auger Observatory have enabled major advances in ultra-high energy cosmic ray physics and demonstrated that improved determination of masses of primary cosmic-ray particles, preferably on an event-by-event basis, is necessary for understanding their origin and nature. Improvement in primary mass measurements was the main motivation for the upgrade of the Pierre Auger Observatory, called AugerPrime. As part of this upgrade, scintillator detectors are added to the existing water-Cherenkov surface detector stations. By making use of the differences in detector response to the electromagnetic particles and muons between scintillator and water-Cherenkov detectors, the electromagnetic and muonic components of cosmic-ray air showers can be disentangled. Since the muonic component is sensitive to the primary mass, such combination of detectors provides a powerful way to improve primary mass composition measurements over the original Auger surface detector design. In this paper, the so-called Scintillator Surface Detectors are discussed, including their design characteristics, production process, testing procedure and deployment in the field.
The next generation neutrino telescope, IceCube-Gen2, will be sensitive to the astrophysical and cosmogenic flux of neutrinos across a broad energy range, from the TeV to the EeV scale. The planned design includes 8 cubic kilometers of ice instrumented with approximately 10,000 optical sensors, a surface array, and a radio array of antennas embedded in the ice laid out sparsely over 500 km^2. The radio array provides sensitivity to ultra-high energy neutrinos using independent radio stations that can trigger on Askaryan emission from neutrino interactions in the ice. In this contribution, we present the design for the radio array along with its planned implementation, which is expected to increase sensitivity to neutrinos with energies beyond 100PeV by at least an order of magnitude over existing arrays. Furthermore, we will quantify the expected science output by presenting measurement forecasts for the main science cases of diffuse flux and point source discovery, as well as cross-section and flavor measurements.
We propose a method for estimating the Fisher score--the gradient of the log-likelihood with respect to model parameters--using score matching. By introducing a latent parameter model, we show that the Fisher score can be learned by training a neural network to predict latent scores via a mean squared error loss. We validate our approach on a toy linear Gaussian model and a cosmological example using a differentiable simulator. In both cases, the learned scores closely match ground truth for plausible data-parameter pairs. This method extends the ability to perform Fisher forecasts, and gradient-based Bayesian inference to simulation models, even when they are not differentiable; it therefore has broad potential for advancing cosmological analyses.
The identification of potential sources of ultra-high-energy cosmic rays (UHECRs) remains challenging due to magnetic deflections and propagation losses, which are particularly strong for nuclei. In previous iterations of this work, we proposed an approach for UHECR astronomy based on Bayesian inference through explicit modelling of propagation and magnetic deflection effects. The event-by-event mass information is expected to provide tighter constraints on these parameters and to help identify unknown sources. However, the measurements of the average mass through observations from the surface detectors at the Pierre Auger Observatory already indicate that the UHECR masses are well represented through its statistical average. In this contribution, we present our framework which uses energy and mass moments of lnA to infer the source parameters of UHECRs, including the mass composition at the source. We demonstrate the performance of our model using simulated datasets based on the Pierre Auger Observatory and Telescope Array Project. Our model can be readily applied to currently available data, and we discuss the implications of our results for UHECR source identification.
The LiteBIRD satellite mission aims at detecting Cosmic Microwave Background B modes with unprecedented precision, targeting a total error on the tensor-to-scalar ratio r of δr∼0.001. Operating from the L2 Lagrangian point of the Sun-Earth system, LiteBIRD will survey the full sky across 15 frequency bands (34 to 448 GHz) for 3 years.The current LiteBIRD baseline configuration employs 4508 detectors sampling at 19.1 Hz to achieve an effective polarization sensitivity of 2μK−arcmin and an angular resolution of 31 arcmin (at 140 GHz).We describe the first release of the official LiteBIRD simulations, realized with a new simulation pipeline developed using the LiteBIRD Simulation Framework.This pipeline generates 500 full-sky simulated maps at a Healpix resolution of nside=512. The simulations include also one year of Time Ordered Data for approximately one-third of LiteBIRD's total detectors.
The Decadal Survey on Astronomy and Astrophysics 2020 highlights the importance of advancing research focused on discovering and characterizing habitable worlds. In line with this priority, our goal is to investigate how planetary systems evolve through atmospheric escape and to develop methods for identifying potentially Earth-like planets. By leveraging the ultraviolet (UV) capabilities of the Habitable Worlds Observatory (HWO), we can use transit spectroscopy to observe atmospheric escape in exoplanets and explore the processes that shape their evolution, assess the ability of small planets to retain their atmospheres, and search for signs of Earth-like atmospheres. To achieve this, we support the development of a UV spectrograph with moderate- to high-resolution capabilities for point-source observations, coverage of key spectral features in the 100-300 nm range, and detectors that can register high count rates reliably. This article is an adaptation of a science case document developed for the Characterizing Exoplanets Steering Committee within HWO's Solar Systems in Context Working Group.
The instrument payload of the future Habitable Worlds Observatory (HWO) will span a wide range of wavelengths, including the ultraviolet (UV) region that cannot be easily accessed from the ground (< 350 nm). Along with its primary mission to characterize the habitability of candidate exo-Earths, HWO will be well suited for observations of potentially habitable icy ocean worlds in our Solar System, in particular with an integral field spectrograph (IFS). Here, we discuss future HWO observations of ocean worlds including Ceres, Europa, Enceladus, Ariel, and Triton. We explore the observational requirements for capturing ongoing and sporadic geyser activity and for measuring the spectral signatures of astrobiologically-relevant compounds, including water, salts, organics, and other bioessential components. We consider the key observing requirements for an IFS, including wavelength coverage, resolving power (R), angular resolution, and field-of-view (FOV). We also outline some of the potential measurements that would define incremental, substantial, and breakthrough progression for characterizing habitability at ocean worlds, primarily focusing on UV and visible (VIS) wavelengths (90 - 700 nm). Our investigation concludes that a UV/VIS IFS on HWO could make some groundbreaking discoveries, in particular for detection and long-term monitoring of geyser activity and interior-surface exchange of components critical for understanding habitability at ocean worlds.
We evaluate 9 Retrieval Augmented Generation (RAG) agent configurations on 105 Cosmology Question-Answer (QA) pairs that we built specifically for this purpose.The RAG configurations are manually evaluated by a human expert, that is, a total of 945 generated answers were assessed. We find that currently the best RAG agent configuration is with OpenAI embedding and generative model, yielding 91.4\% accuracy. Using our human evaluation results we calibrate LLM-as-a-Judge (LLMaaJ) system which can be used as a robust proxy for human evaluation. These results allow us to systematically select the best RAG agent configuration for multi-agent system for autonomous scientific discovery in astrophysics (e.g., cmbagent presented in a companion paper) and provide us with an LLMaaJ system that can be scaled to thousands of cosmology QA pairs. We make our QA dataset, human evaluation results, RAG pipelines, and LLMaaJ system publicly available for further use by the astrophysics community.
GRAND (the Giant Radio Array for Neutrino Detection) is a proposed next-generation observatory targetting primarily the detection of ultra-high-energy neutrinos, with energies exceeding about 100 PeV. GRAND is envisioned as a collection of large-scale ground arrays of self-triggered radio antennas that target the radio emission from extensive air showers initiated by UHE particles. Three prototype arrays are presently in operation: GRANDProto300 in China, with 65 units running since end of 2024, GRAND@Auger in Argentina with 10 units deployed on the site of the Pierre Auger Observatory, and GRAND@Nançay in France, a 4-unit setup installed at the Nançay radio-observatory and used for test purposes. The main objective of the GRAND prototype phase is to validate the detection principle and technology of GRAND, in preparation for its next phase, GRAND10k. GRAND10k will consist of two arrays of 10'000 antennas each, covering both the Northern and Southern hemispheres, to be deployed from 2030 on. Here we give an overview of the GRAND concept, its science goals, the status of the prototypes, their performances and first detection of cosmic rays, and the technical perspective they open for the future.
The NASA Great Observatories Maturation Program is a development plan to efficiently and effectively develop large, strategic astrophysics missions. Suborbital rocket and balloon programs have long been a key development tool for enabling large missions in NASA astrophysics. We review the significance of these suborbital missions in the preceding decades to demonstrate their contributions to the Great Observatories Maturation Program for the Habitable Worlds Observatory and beyond. We show that suborbital instruments have obtained new science observations of astrophysical sources across the electromagnetic spectrum, matured high-priority component technologies, and served as a training ground for principal investigators of Explorer-class astrophysics satellites. A brief discussion of emerging CubeSat and SmallSat missions and their place in the NASA astrophysics portfolio is also provided.
As of early 2025, the GRAND Collaboration operates three prototype arrays: GRANDProto300 in China, GRAND@Nançay in France, and GRAND@Auger in Argentina. The GRAND@Auger prototype was established through an agreement between the GRAND and Pierre Auger Collaborations, repurposing ten Auger Engineering Radio Array (AERA) stations as GRAND detection units. This setup provides a unique opportunity for coincident air-shower detection, enabling direct event-by-event comparison between GRAND@Auger and the well-established Pierre Auger detectors. Such comparisons allow for a detailed assessment of the detection principle and reconstruction capabilities of GRAND. In this contribution, we present an overview of the commissioning and preliminary results of GRAND@Auger, including the measurement of the Galactic background noise and the detection of the first self-triggered candidate event in coincidence with the Pierre Auger Observatory. Both results underscore the role of GRAND@Auger in advancing the GRAND project and refining the techniques required for large-scale radio detection of ultra-high-energy particles.
The Giant Radio Array for Neutrino Detection (GRAND) aims to detect and study ultra-high-energy (UHE) neutrinos by observing the radio emissions produced in extensive air showers. The GRANDProto300 prototype primarily focuses on UHE cosmic rays to demonstrate the autonomous detection and reconstruction techniques that will later be applied to neutrino detection. In this work, we propose a method for reconstructing the arrival direction and energy with high precision using state-of-the-art machine learning techniques from noisy simulated voltage traces. For each event, we represent the triggered antennas as a graph structure, which is used as input for a graph neural network (GNN). To significantly enhance precision and reduce the required training set size, we incorporate physical knowledge into both the GNN architecture and the input data. This approach achieves an angular resolution of 0.14{\deg} and a primary energy reconstruction resolution of about 15%. Additionally, we employ uncertainty estimation methods to improve the reliability of our predictions. These methods allow us to quantify the confidence of the GNN predictions and provide confidence intervals for the direction and energy reconstruction. Finally, we explore strategies to evaluate the consistency and robustness of the model when applied to real data. Our goal is to identify situations where predictions remain trustworthy despite domain shifts between simulation and reality.
The intensity and energy spectrum of galactic cosmic rays in the heliosphere are significantly influenced by the 11-year solar cycle, a phenomenon known as solar modulation. Understanding this effect and its underlying physical mechanisms is essential for assessing radiation exposure and associated risks during space missions. Starting from a previously developed effective predictive model of solar modulation, validated using cosmic ray flux measurements from space-based detectors such as PAMELA and AMS-02, we build a generalizable forecasting strategy for the long-term evolution of cosmic ray fluxes. This strategy is based on identifying delayed cross-correlation relationships between solar proxies and the model's parameters. It integrates recent findings on time lags between cosmic ray fluxes and solar activity, and incorporates advanced time-series signal processing techniques. The framework not only performs well in reproducing observed data, but also shows strong potential for applications in space radiation monitoring and forecasting. By efficiently capturing the long-term variability of galactic cosmic rays, our approach contributes valuable insights for evaluating radiation risks, ultimately supporting safer and more effective space exploration.