Rubin-LSST

Abstract:

Modern astronomical surveys deliver immense volumes of transient detections, yet distinguishing real astrophysical signals (for example, explosive events) from bogus imaging artefacts remains a challenge. Convolutional neural networks are effectively used for real versus bogus classification; however, their reliance on opaque latent representations hinders interpretability. Here we show that large language models (LLMs) can approach the performance level of a convolutional neural network on three optical transient survey datasets (Pan-STARRS, MeerLICHT and ATLAS) while simultaneously producing direct, human-readable descriptions for every candidate. Using only 15 examples and concise instructions, Google’s LLM, Gemini, achieves a 93% average accuracy across datasets that span a range of resolution and pixel scales. We also show that a second LLM can assess the coherence of the output of the first model, enabling iterative refinement by identifying problematic cases. This framework allows users to define the desired classification behaviour through natural language and examples, bypassing traditional training pipelines. Furthermore, by generating textual descriptions of observed features, LLMs enable users to query classifications as if navigating an annotated catalogue, rather than deciphering abstract latent spaces. As next-generation telescopes and surveys further increase the amount of data available, LLM-based classification could help bridge the gap between automated detection and transparent, human-level understanding.

Abstract:

The production cross-section of high-mass τ-lepton pairs is measured as a function of the dilepton visible invariant mass, using 140 fb−1 of s=13 TeV proton-proton collision data recorded with the ATLAS detector at the Large Hadron Collider. The measurement agrees with the predictions of the Standard Model. A fit to the invariant mass distribution is performed as a function of b-jet multiplicity, to constrain the non-resonant production of new particles described by an effective field theory or in models containing leptoquarks or Z′ bosons that couple preferentially to third-generation fermions. The constraints on new particles improve on previous results, and the constraints on effective operators include those affecting the anomalous magnetic moment of the τ-lepton.

Abstract:

Standard cosmological weak lensing analyses using cosmic shear are inevitably sensitive to small-scale, non-linear clustering from low-redshift structures. The need to adequately model the clustering of matter on this non-linear regime, accounting for both gravitational and baryonic effects, adds significant uncertainty to weak lensing studies, particularly in the context of near-future Stage-IV datasets. In this paper, inspired by previous work on so-called “nulling” techniques, we present a general method that selects the linear combinations of a given tomographic cosmic shear dataset that are least sensitive to small-scale non-linearities, by essentially suppressing the contribution from low-redshift structures. We apply this method to the latest public cosmic shear data from the Dark Energy Survey, DES-Y3, that corresponds to 3 years of observation, and show: a) that a large fraction of the signal is dominated by the single mode that is most affected by non-linear scales, and b) that removing this mode leads to a ∼ 1σ upwards shift in the preferred value of S 8 ≡ σ 8√(ΩM/0.3), alleviating the tension with current CMB data. However, the removal of the most contaminated mode also results in a significant increase in the statistical uncertainties. Taking this into account, we find this shift to be compatible with a random fluctuation caused by removing this most-contaminated mode at the ∼ 1.4σ level. We also show that this technique may be used by future Stage-IV surveys to mitigate the sensitivity of the final constraints to baryonic effects, trading precision for robustness.

Abstract:

The Simons Observatory (SO) Small Aperture Telescopes (SATs) will observe the Cosmic Microwave Background (CMB) temperature and polarization at six frequency bands. Within these bands, the angular response of the telescope (beam) is convolved with the instrument's spectral response (commonly called bandpass) and the signal from the sky, which leads to the band-averaged telescope beam response, which is sampled and digitized. The spectral properties of the band-averaged beam depend on the natural variation of the beam within the band, referred to as beam chromaticity. In this paper, we quantify the impact of the interplay of beam chromaticity and intrinsic frequency scaling from the various components that dominate the polarized sky emission on the tensor-to-scalar ratio, r, and foreground parameters. We do so by employing a parametric power-spectrum-based foreground component separation algorithm, namely BBPower, to which we provide beam-convolved time domain simulations performed with the beamconv software while assuming an idealized version of the SO SAT optics. We find a small, 0.02σ, bias on r, due to beam chromaticity, which seems to mostly impact the dust spatial parameters, causing a maximum 0.77σ bias on the dust B-mode spectra amplitude, Ad , when employing Gaussian foreground simulations. However, we find all parameter biases to be smaller than 1σ at all times, independently of the foreground model. This includes the case where we introduce additional uncertainty on the bandpass shape, which accounts for approximately half of the total allowed gain uncertainty, as estimated in previous work for the SO SATs.

Abstract:

Abstract We investigate the clustering of Lyman-break galaxies at redshifts of 3 ≲ z ≲ 5 within the COSMOS field by measuring the angular two-point correlation function. Our robust sample of ~60,000 bright (mUV ≲ 27) Lyman-break galaxies was selected based on spectral energy distribution fitting across 14 photometric bands spanning optical and near-infrared wavelengths. We constrained both the 1- and 2-halo terms at separations up to 300 arcsec, finding an excess in the correlation function at scales corresponding to <20 kpc, consistent with enhancement due to clumps in the same galaxy or interactions on this scale. We then performed Bayesian model fits on the correlation functions to infer the Halo Occupation Distribution parameters, star formation duty cycle, and galaxy bias in three redshift bins. We examined several cases where different combinations of parameters were varied, showing that our data can constrain the slope of the satellite occupation function, which previous studies have fixed. For an MUV-limited sub-sample, we found galaxy bias values of $b_g=3.18^{+0.14}_{-0.14}$ at z ≃ 3, $b_g=3.58^{+0.27}_{-0.29}$ at z ≃ 4, $b_g=4.27^{+0.25}_{-0.26}$ at z ≃ 5. The duty cycle values are $0.62^{+0.25}_{-0.26}$, $0.40^{+0.34}_{-0.22}$, and $0.39^{+0.31}_{-0.20}$,respectively. These results suggest that, as the redshift increases, there is a slight decrease in the host halo masses and a shorter timescale for star formation in bright galaxies, at a fixed rest-frame UV luminosity threshold.