Search for: All records

Large language models (LLMs) have achieved remarkable success in natural language processing (NLP), demonstrating significant capabilities in processing and understanding text data. However, recent studies have identified limitations in LLMs’ ability to manipulate, program, and reason about structured data, especially graphs. We introduce GraphEval36K1 , the first comprehensive graph dataset, comprising 40 graph coding problems and 36,900 test cases to evaluate the ability of LLMs on graph problem solving. Our dataset is categorized into eight primary and four sub-categories to ensure a thorough evaluation across different types of graphs. We benchmark ten LLMs, finding that private models outperform open-source ones, though the gap is narrowing. We also analyze the performance of LLMs across directed vs undirected graphs, different kinds of graph concepts, and network models. Furthermore, to improve the usability of our evaluation framework, we propose Structured Symbolic Decomposition (SSD), an instruction-based method designed to enhance LLM performance on complex graph tasks. Results show that SSD improves the average passing rate of GPT-4, GPT4o, Gemini-Pro and Claude-3-Sonnet by 8.38%, 6.78%, 29.28% and 25.28%, respectively. 

Abstract Plasma irregularities in the ionosphere induce scintillation of radio signals. Radio occultation (RO) observations of the Global Navigation Satellite Systems (GNSS) signals from low Earth orbit (LEO) allow monitoring of the ionospheric scintillation. Under certain conditions, it is possible to localize (geolocate) plasma irregularities along the line‐of‐sight between the GNSS and LEO satellites. While several techniques have been considered for the localization, in this study we use the back propagation (BP) of complex RO signals (phase and amplitude) measured at a high rate (HR), 50–100 Hz. Our method is based on a numerical solution of the wave equation, originally proposed for geolocation in 2002, with some modifications. We consider theoretical aspects of the BP technique, including assumptions, approximations and limitations, and perform numerical modeling of radio wave propagation. We investigate geolocation by BP for two regions with aligned and mis‐aligned irregularities and explain multi‐valued geolocations. We focus on the equatorial F region, consistent with the COSMIC‐2 observation sampling and use the IGRF‐13 model of the Earth's magnetic field to define the orientation of plasma irregularities. We use our method for processing of COSMIC‐2 HR scintillation data collected from the precise orbit determination antennas for 2 years: 2021 and 2023 (years with low and high solar activity). The results, represented by gridded monthly maps of geolocations, show clear seasonal and interannual variations. Additionally, we present comparison of the geolocations obtained independently from L1 and L2 signals for a 2‐month period. 

Abstract The paper presents the effects of the storm‐time prompt penetration electric fields (PPEF) and traveling atmospheric disturbances (TADs) on the total electron content (TEC), foF2 and hmF2 in the American sector (north and south) during the geomagnetic storm on 23–24 April 2023. The data show a poleward shift of the Equatorial Ionization Anomaly (EIA) crests to 18°N and 20°S in the evening of 23 April (attributed to eastward PPEF) and the EIA crests remaining almost in the same latitudes after the PPEF reversed westward. The thermospheric neutral wind velocity, foF2, hmF2, and TEC variations show that TADs from the northern and southern high latitudes propagating equatorward and crossing the equator after midnight on 23 April. The meridional keograms of ΔTEC show the TAD structures in the north/south propagated with phase velocity 470/485 m/s, wave length 4,095/4,016 km and period 2.42/2.30 hr, respectively. The interactions of the TADs also appear to modify the wind velocities in low latitudes. The eastward PPEF and equatorward TADs also favored the development of a clear/not so clear F3 layer in northern/southern regions of the equator. 

ABSTRACT We revisit the Quasar Main Sequence (QMS) by investigating the impact of the stellar component from the host galaxy (HG) on the emission line spectra of the active galactic nuclei (AGNs). We first detect spectra with broad emission lines using a line ratio method for a sample of $$\sim$$3000 high SNR ($>$20) black hole mapper objects (part of the fifth phase of the Sloan Digital Sky Survey). We then built the Index diagram, a novel diagnostic tool using the z-corrected spectra, model-free, designed to easily identify spectra with significant stellar HG contributions and to classify the AGN spectra into three categories based on AGN–HG dominance: HG-dominated (HGD), Intermediate (INT), and AGN-dominated (AGND) sources. A colour-z diagram was used to refine the AGN–HG classification. We subtract the stellar contributions from the HGD and INT spectra before modelling the AGN spectrum to extract the QMS parameters. Our QMS reveals that HGD galaxies predominantly occupy the Population B region with no R$$_{\rm {Fe\,{\small II}}}$$, with outliers exhibiting R$$_{\rm {Fe\,{\small II}}}$$  $>$ 1, likely due to HG subtraction residuals and a faint contribution of h$$\beta _{\rm BC}$$. INT and AGND spectra show similar distributions in the Population A region, while in Population B, a tail of AGND sources becomes apparent. Cross-matching with radio, infrared, and X-ray catalogues, we find that the strongest radio emitters are associated with HGD and INT groups. Strong X-ray emitters are found in INT and AGND sources, also occupying the AGN region in the Wide-field Infrared Survey Explorer colour diagram. 

Abstract The spectral line profile of the atomic oxygen O¹D₂—³P₂transition near 6300 Å in the airglow has been used for more than 50 years to extract neutral wind and temperature information from the F‐region ionosphere. A new spectral model and recent samples of this airglow emission in the presence of the nearby lambda‐doubled OH Meinel (9‐3) P₂(2.5) emission lines underscores earlier cautions that OH can significantly distort the OI line center position and line width observed using a single‐etalon Fabry‐Perot interferometer (FPI). The consequence of these profile distortions in terms of the emission profile line width and Doppler position is a strong function of the selected etalon plate spacing. Single‐etalon Fabry‐Perot interferometers placed in the field for thermospheric measurements have widely varying etalon spacings, so that systematic wind biases caused by the OH line positions differ between instruments, complicating comparisons between sites. Based on the best current determinations of the OH and O¹D line positions, the ideal gap for a single‐etalon FPI wind measurements places the OH emissions in the wings of the O¹D spectral line profile. Optical systems that can accommodate prefilters with square passbands less than ∼3 Å in the optical beam can effectively block the OH contamination. When that is not possible, a method to fit for OH contamination and remove it in the spectral background of an active Fabry‐Perot system is evaluated. 

Abstract Midlatitude thermospheric wind observations from the Michelson Interferometer for Global High‐resolution Thermospheric Imaging on board the Ionospheric Connections Explorer (ICON/MIGHTI) and from the ground‐based Boulder, Urbana, Millstone Hill and Morocco Fabry‐Perot interferometers (FPIs) are used to study a distinct solar local time (SLT) evolution in the nighttime wind field around the December solstice period. Our results show, to the best of our knowledge for the first time, strong non‐migrating tides in midlatitude thermospheric winds using coincident from different observing platforms. These observations exhibited a structure of strong (∼50–150 m/s) eastward and southward winds in the pre‐midnight sector (20:00–23:00 SLT) and in the post‐midnight sector (02:00–03:00 SLT), with a strong suppression around midnight. Tidal analysis of ICON/MIGHTI data revealed that the signature before midnight was driven by diurnal (D0, DE1, DE2, DW2) and semidiurnal (SE2, SE3, SW1, SW4) tides, and that strong terdiurnal (TE2, TW1, TW2, TW5) and quatradiurnal (QW2, QW3, QW6) tides were important contributors in the mid‐ and post‐midnight sectors. ICON/MIGHTI tidal reconstructions successfully reproduced the salient structures observed by the FPI and showed a longitudinal dual‐peak variation with peak magnitudes around 200°–120°W and 30°W–60°E. The signature of the structure extended along the south‐to‐north direction from lower latitudes, migrated to earlier local times with increasing latitude, and strengthened above 30°N. Tidal analysis using historical FPI data revealed that these structures were often seen during previous December solstices, and that they are much stronger for lower solar flux conditions, consistent with an upward‐propagating tidal origin.