Searcharxiv⌕ Search

arXiv subjects

Weihao Liu

Publications and source records attributed to Weihao Liu.

At least 19 recordsLinked to original sources

Comparative Analysis of Influences of Magnetic Field Inputs on Simulating the 2017 September 10 Solar Energetic Particle Event

Physics-based solar energetic particle (SEP) models simulate the ambient solar wind, coronal mass ejections (CMEs), and SEPs using photospheric magnetic field maps as the inner boundary condition. In this study, we select the 2017 September 10 SEP event, which produced the first ground level enhancement observed on the surfaces of both Earth and Mars, to examine how these inputs influence the modeled results. We use the SOlar wind with FIeld lines and Energetic particles (SOFIE) model suite, driven independently by the GONG and ADAPT-GONG magnetic field maps, to simulate the solar wind, CME, and SEPs from the low corona to beyond Mars' orbit. The two maps yield different open flux regions and hence different solar wind conditions, both globally comparable to in situ observations at Earth, STA, and Mars, with the GONG-driven results performing slightly better. We find that in the ADAPT-GONG-driven run, a more energetic CME flux rope propagates into a slower, wider stream, producing broader CME expansion and a more extended shock, which reaches both planets within ~4 hours of the observed arrival, while the GONG-driven run misses both arrivals. Both simulations reproduce the observed SEP fluxes at Earth within a factor of ~3, while the ADAPT-GONG-driven results compare better with observations at Mars, which connects magnetically to the shock flank. Therefore, for this event, the relative performance of the two simulations depends on the observer and observable, with SEP fluxes shaped by CME propagation through the ambient solar wind and by magnetic connectivity to the evolving shock.

physics.space-ph↗

ScienceIDE: Turning World's Scientific Codebase into Agent Learnable Environments

Scientific code repositories encode decades of human knowledge in executable models, methods, and tools. Yet fragmented toolchains, implicit domain conventions, and specialized correctness criteria make this knowledge difficult to convert into reliable learning experience-a challenge we call the scientific experience bottleneck. We introduce ScienceIDE, infrastructure for turning the world's scientific code into programmable environments for scientific agents. Guided by expert-defined scientific cases and acceptance criteria, agents transform repositories into executable environments that support task generation, execution, and scientific verification. These environments provide a shared foundation for supervised fine-tuning, reinforcement learning, and evaluation. Using verified interaction trajectories, we train PhAI-IDE-72B/9B/4B. The model family shows gains in held-out scientific-code repair and across selected general-purpose benchmarks in code, reasoning, and knowledge, providing evidence of positive transfer from scientific experience to broader capabilities. ScienceIDE lays the foundation for an integrated workspace for agent learning and scientific practice, making humanity's scientific software a shared substrate for developing scientific intelligence. Code: https://github.com/aitofound/ScienceIDE

cs.CL↗

An Implicit Conservative Finite Volume Scheme with Poisson Brackets for the Steady-State Transport of Galactic Cosmic Rays

Galactic cosmic rays (GCRs) pose a persistent radiation hazard to astronauts and spacecraft in the heliosphere. Modeling their transport is essential for understanding how the solar wind and heliospheric magnetic field shape their spatial distributions and energy spectra. Timely predictions require solving for the particle distribution function much faster than real time, but explicit numerical schemes are restricted by the stability limit on the time step. In this Letter, we present an implicit conservative finite-volume scheme for kinetic equations formulated with Poisson brackets. This scheme conserves particle number, maintains the total-variation-diminishing property, achieves second-order spatial accuracy, and remains stable for time steps far beyond the explicit limit, while reaching quasi-steady states in less than ~1/100 of the computational time required for the explicit scheme, as demonstrated by numerical tests. We apply this scheme to the Parker transport equation along magnetic field lines in the background solar wind during Carrington rotation 2211, where it reaches steady-state GCR distributions within 10 iterations, about 1000 times faster than the explicit scheme. The GCR proton spectrum averaged over 27 days agrees with AMS-02 measurements to within 10%. The simulation also yields a realistic radial intensity gradient and captures the Forbush decrease features associated with three stream interaction regions observed at Earth.

physics.space-ph↗

Hallucination Self-Play: Bootstrapping Reinforced Detector via Evolved Generator

Identifying faithfulness hallucinations in LLM-generated outputs remains challenging due to the scarcity of high-quality annotated data. Recent work relies on advanced LLMs to synthesize training data, including rationales, labels, and hallucinated claims. However, these methods treat the generator as a static component, limiting iterative improvement of the detector. To address this limitation, we introduce Hallucination Self-Play (HSP), a novel framework that enables the detector to bootstrap with an evolved generator. HSP involves two roles initialized from the same base model, a detector that assesses the faithfulness of model outputs, and a generator that produces increasingly hard-to-detect hallucinated responses. Specifically, the detector is first fine-tuned on human-labeled data and then employed as a reward model to train the generator via reinforcement learning from AI feedback (RLAIF). In turn, the evolved generator synthesizes hallucination data to further optimize the detector through rule-based reinforcement learning. Experiments on RAGTruth and LLM-AggreFact across three model families demonstrate that the proposed framework can progressively enhance a small LLM to match or even outperform advanced LLMs without external supervision. Our code is available at https://github.com/maybenotime/Hallucination_Self-Play.

cs.CL↗

Decoding Looped Transformers Better for (Almost) Free

Looped Transformers achieve parameter efficiency by repeatedly executing a shared block across recurrent loops. Each loop yields an intermediate representation decodable for the same next token, yet standard decoding discards earlier states. Because earlier loops embody less computation, recurrence inherently supplies aligned weak-and-strong prediction pairs without auxiliary models or external training. We introduce LoopCD, a training-free contrastive decoding framework that guides token selection by contrasting the final prediction with an earlier recurrent pass, operating either in logit space with one extra output pass (LoopCD-Logits) or in hidden-state space with zero output overhead (LoopCD-Hidden). Across four looped Transformer families, LoopCD delivers substantial, consistent gains at full recurrent depth: LoopCD-Logits raises Ouro-2.6B-Thinking's AIME 2024 pass@1 from 61.88% to 73.33%, while LoopCD-Hidden lifts Huginn's HumanEval pass@1 from 22.56% to 31.71%. Crucially, these performance gains enable halving the number of recurrent loops while still matching or exceeding full-depth unguided baselines, reducing forward FLOPs by 22.5% to 48.2%. By transforming intermediate recurrent states into effective guidance signals, LoopCD achieves superior decoding quality while substantially reducing inference compute.

cs.LG↗

Beyond Math and Code: Lightweight Corpus-Grounded Process Rewards for Factual Question Answering

Process supervision during reinforcement learning (RL) post-training matters in factual question answering (QA) because responses receiving positive response-level rewards can still contain sentence-level factual errors. Unlike math and code, factual QA lacks inexpensive programmatic checks, making reward computation a bottleneck. Existing methods rely on neural verifiers to score individual sentences, requiring extensive model inference as RL repeats these checks across many sampled responses at every update. We therefore introduce CorVer (Corpus Verify), a lightweight training-time approach to process supervision that derives sentence-level rewards from corpus co-occurrence statistics. A 0.5B extractor identifies subject-object pairs, indexed corpus queries supply their co-occurrence counts, and the resulting sentence rewards are assigned to the corresponding tokens for RL. On Qwen3-4B and Qwen3-8B, CorVer reduces mean complete training time by 5.5-10.4 times relative to the four factuality-RL baselines. Across the four models evaluated against these baselines, CorVer achieves the highest accuracy in 17 of 20 model-benchmark settings. CorVer outperforms the unmodified models in all 30 standard factual-QA settings (six models from three families across five benchmarks) and remains effective on two additional multi-hop QA datasets.

cs.CL↗

Counterintuitive Magnetic Connectivity and Energetic Particle Flux Differences among Nearby Spacecraft During the 2023 February 24 Solar Energetic Particle Event

For solar energetic particles (SEPs), it is generally expected that observers magnetically closer to the eruption source region exhibit higher particle intensities than those poorly connected to the eruption site. However, the 2023 February 24 SEP event departs from this simple picture: Earth and STA, near 1 au, are nominally better connected to the source region, whereas Solar Orbiter (SolO), at 0.77 au but less favorably connected, observed SEP fluxes more than an order of magnitude higher. This difference cannot be simply explained by nominal magnetic connectivity or radial scaling of SEP fluxes alone. To investigate this behavior, we perform a global magnetohydrodynamic simulation of the associated coronal mass ejection (CME) using the Alfvén Wave Solar-atmosphere Model-Realtime (AWSoM-R). The simulation reveals that the CME flux rope originates close to a coronal streamer and as it propagates and expands, the CME-driven shock is effectively distorted, developing into two distinct flanks with different strengths. Although the three spacecraft are separated by only $\lesssim$30$^{\circ}$ in heliolongitude, their magnetic footpoints differ by $\gtrsim$50$^{\circ}$ in longitude because of a nearby stream interaction region. Specifically, Earth and STA connect to a weaker shock region, while SolO connects to the shock nose with a higher compression ratio and more efficient particle acceleration. We further simulate SEPs using the Multiple-Field-Line Advection Model for Particle Acceleration (M-FLAMPA) coupled with AWSoM-R, obtaining results that reproduce the observed flux differences among the three spacecraft, demonstrating that this counterintuitive behavior results from their connections to different regions of the inhomogeneous CME-driven shock.

physics.space-ph↗

Simulating the Solar Corona with Multiple Solar Photospheric Magnetic Maps during the 8 April 2024 Total Solar Eclipse

The 8 April 2024 total solar eclipse (TSE) provides a unique opportunity to study the solar corona. This work presents our simulations of the solar corona at the time of the eclipse based on magnetohydrodynamic (MHD) modeling performed with the Alfvén Wave Solar atmosphere Model (AWSoM) in the Space Weather Modeling Framework, developed at the University of Michigan. We performed multiple simulations based on photospheric magnetic maps from four sources, i.e., ADAPT-GONG, Lockheed Martin ESFAM-HMI, HipFT-HMI, and NSO-NRT-HMI maps. Our study focuses on how differences in the magnetic field maps affect the coronal magnetic field structure and coronal heating properties in the simulation. The synthesized observables show remarkable differences due to the distinct magnetic coronal topologies, which stem from the different local magnetic flux distributions. We analyze the properties of the open magnetic flux regions of the models. We also study the coronal heating rate in the models. The total volume integrated heating rate yields a difference of $20\%$ across the models. The results also show that the differential emission measure in the high-temperature regions is sensitive to the magnetic field maps. Our findings underscore the importance of comprehensive photospheric magnetic field data in improving future solar coronal models.

astro-ph.SR↗

Latent Thoughts Tuning: Bridging Context and Reasoning with Fused Information in Latent Tokens

While explicit Chain-of-Thought (CoT) equips Large Language Models (LLMs) with strong reasoning capabilities, it constrains the model's thoughts to a discrete vocabulary space. Recently, reasoning in continuous latent space has emerged as a promising alternative, but current paradigms suffer from feature collapse and instability due to distribution mismatch when recurrently reusing hidden states, or alignment issues when relying on assistant models. To address this, we propose Latent Thoughts Tuning (LT-Tuning), a post-training framework that redefines how latent thoughts are constructed and deployed. Instead of relying solely on raw hidden states, our method introduces a Context-Prediction-Fusion mechanism that jointly leverages contextual hidden states and predictive semantic guidance from the vocabulary embedding space. Combined with a progressive three-stage curriculum learning pipeline, LT-Tuning also enables dynamic switching between latent and explicit thinking modes. Experiments demonstrate that our method outperforms existing latent reasoning baselines, effectively mitigating feature collapse and achieving robust reasoning accuracy.

cs.CL↗

BC-DIR: Bandit-Controlled Deadline-Aware Incremental Redundancy for QUIC in V2X Networks

Vehicle-to-Everything (V2X) communications require timely and reliable message delivery under highly dynamic wireless conditions. Existing approaches that integrate forward error correction (FEC) into QUIC rely mainly on proactive redundancy and fall back to retransmission once losses exceed the correction capability of the configured code, leading to inefficient recovery under burst loss and unnecessary overhead when network conditions are favorable. This paper presents a Bandit-Controlled Deadline-Aware Incremental Redundancy (BC-DIR) framework for QUIC-based V2X transport. BC-DIR combines rateless coding with a soft decoding deadline and a redundancy margin, enabling repair to be triggered within the available delivery budget while injecting additional repair symbols beyond the immediate deficit to improve recovery under burst loss. A contextual bandit controller further adapts the redundancy configuration online according to end-to-end feedback. We also develop a deadline-constrained reliability analysis under burst loss, showing the advantage of the proposed repair mechanism over conventional retransmission and the existence of an optimal redundancy margin. Monte Carlo simulations validate the analytical results. BC-DIR is implemented in a QUIC-based transport stack and evaluated in Veins/OMNeT++ under both congested urban V2X scenarios and stable network conditions. Experimental results show that, across different traffic congestion levels, BC-DIR improves completion ratio by 10\%--40\% over benchmark schemes in congested V2X scenarios, while under favorable network conditions it can even reduce overhead by about 1\% compared with native QUIC.

cs.NI↗

Radio Spectral Imaging and MHD Modeling of a CME-Driven Shock: Connecting Solar Type II Radio Bursts with Shock-Surface Magnetic Geometry

Solar type II radio bursts are widely regarded as signatures of shock waves propagating in the solar corona and are of particular importance for understanding shock-driven particle acceleration processes. Type II radio bursts often exhibit complex multi-lane and split-band features. The detailed spectral, temporal, and spatial structures carry key information about the shock properties and evolution. However, the physical origin of the multi-lane and split-band features remains unclear, largely due to a lack of spatially resolved data and understanding of the concurrent shock morphology and its magnetic-field context. In this work, we combine radio imaging spectroscopy of a multi-lane, split-band type II burst event with a three-dimensional global magnetohydrodynamic simulation of the associated coronal mass ejection-driven shock using the Alfvén Wave Solar atmosphere Model-Realtime. In this event, the burst intensity evolves from fundamental-emission dominated to harmonic-emission dominated. Meanwhile, the preferential emission source region moves from the Earth-facing side to the limb or far side, coinciding with quasi-perpendicular shock regions with enhanced Mach numbers. The observed spatial offset between the fundamental and harmonic sources is generally aligned with the projected shock-surface magnetic field from the simulation, consistent with anisotropic scattering in a magnetized turbulent plasma. These results establish a physical connection between type II radio sources and coronal shock magnetic geometry, providing new insight into the origin of the multi-lane features and their diagnostics of coronal shocks.

astro-ph.SR↗

Defects and Inconsistencies in Solar Flare Data Sources: Implications for Machine Learning Forecasting

Machine learning models for forecasting solar flares have been trained and evaluated using a variety of data sources, including Space Weather Prediction Center (SWPC) operational and science-quality data. Typically, data from these sources is minimally processed before being used to train and validate a forecasting model. However, predictive performance can be affected if defects and inconsistencies between these data sources are ignored. For a set of commonly used data sources, along with the software that queries and outputs processed data, we identify their defects and inconsistencies, quantify their extent, and show how they can affect predictions from data-driven machine-learning forecasting models. We also outline procedures for fixing these issues or at least mitigating their impacts. Finally, based on thorough comparisons of the effects of data sources on the trained forecasting model's predictive skill scores, we offer recommendations for using different data products in operational forecasting.

astro-ph.SR↗

Simulated Operational Testing of the Prototype Implementation of the SOFIE Model: The 2025 Space Weather Prediction Testbed Exercise

The CLEAR Space Weather Center of Excellence's solar energetic particle (SEP) model, SOlar wind with FIeld lines and Energetic particles (SOFIE), was run and evaluated on-site during the Space Weather Prediction Testbed (SWPT) exercise at the National Oceanic and Atmospheric Administration's Space Weather Prediction Center (NOAA/SWPC) in May 2025. As a physics-based SEP model, SOFIE simulates the acceleration and transport of energetic particles by the coronal mass ejection (CME)-driven shock in the solar corona and inner heliosphere, and has been validated against historical events. However, questions remain regarding whether a physics-based model, traditionally considered computationally expensive, could meet operational needs. The SWPT exercise offered a valuable opportunity to evaluate SOFIE under simulated operational conditions. On-site interactive feedback from SWPC forecasters, Space Radiation Analysis Group (SRAG) console operators, Community Coordinated Modeling Center (CCMC) personnel, and Moon-to-Mars Space Weather Analysis Office (M2M SWAO) analysts led to significant strategic improvements in the model configuration. The simulation grid was optimized by combining a coarser background grid with higher-resolution regions along the CME path and toward Earth, reducing computational cost without compromising accuracy. In this work, we present the simulated operational performance of SOFIE and its capability to predict SEP fluxes significantly faster than real time. During the SWPT exercise, SOFIE completed a 4-day SEP simulation within 5 hours using 1,000 central processing unit cores, although the earliest SEP forecast was obtained a few hours after CME onset. This marks a milestone in demonstrating SOFIE's operational usefulness and robustness to support future human space exploration.

astro-ph.SR↗

Evidence of Time-Dependent Diffusive Shock Acceleration in the 2022 September 5 Solar Energetic Particle Event

On 2022 September 5, a large solar energetic particle (SEP) event was detected by Parker Solar Probe (PSP) and Solar Orbiter (SolO), at heliocentric distances of 0.07 and 0.71 au, respectively. PSP observed an unusual velocity-dispersion signature: particles below $\sim$1 MeV exhibited a normal velocity dispersion, while higher-energy particles displayed an inverse velocity arrival feature, with the most energetic particles arriving later than those at lower energies. The maximum energy increased from about 20-30 MeV upstream to over 60 MeV downstream of the shock. The arrival of SEPs at PSP was significantly delayed relative to the expected onset of the eruption. In contrast, SolO detected a typical large SEP event characterized by a regular velocity dispersion at all energies up to 100 MeV. To understand these features, we simulate particle acceleration and transport from the shock to the observers with our newly developed SEP model - Particle ARizona and MIchigan Solver on Advected Nodes (PARMISAN). Our results reveal that the inverse velocity arrival and delayed particle onset detected by PSP originate from the time-dependent diffusive shock acceleration processes. After shock passage, PSP's magnetic connectivity gradually shifted due to its high velocity near perihelion, detecting high-energy SEPs streaming sunward. Conversely, SolO maintained a stable magnetic connection to the strong shock region where efficient acceleration was achieved. These results underscore the importance of spatial and temporal dependence in SEP acceleration at interplanetary shocks, and provide new insights to understand SEP variations in the inner heliosphere.

astro-ph.SR↗

Time-Dependent Radiation Quality Factor of Galactic Cosmic Rays in Deep Space and Shielding Environments: Modeling and Measurements

Understanding the long-term variation of the galactic cosmic ray (GCR) radiation environment is critical for assessing radiation risks in space exploration missions. In this study, we systematically model the linear energy transfer (LET) spectra of GCRs and the corresponding radiation quality factor, , in deep space and shielding environments. The Badhwar-O'Neill 2020 (BON20) model is used to represent GCR fluxes under different solar modulation potentials (phi), which characterize the level of solar activity. GCR interactions with spherical shielding of different thicknesses are simulated to obtain the LET spectra, absorbed dose, dose equivalent, and . We present a comprehensive dataset of these quantities for a range of phi values and shielding thicknesses. The results show that depends strongly on the shielding thickness but only weakly on solar activity. Furthermore, model predictions are validated against long-term measurements from the Cosmic Ray Telescope for the Effects of Radiation (CRaTER) orbiting the Moon, and the Liulin-MO detector on board the ExoMars Trace Gas Orbiter (TGO) orbiting Mars. In this comparison, we consider factors for anomalous cosmic ray (ACR) contributions and radial gradients of both GCRs and ACRs, applying scaling factors of 6.3% at 1 AU and 11.0% at 1.5 AU to the calculated absorbed dose rate. With these corrections, the modeled absorbed dose rate and exhibit consistent temporal variations with the observations under both thin and thick shielding conditions. Moreover, we investigate the distinct temporal evolution of for light and heavy GCR nuclei, revealing how solar modulation influences the elemental radiation quality factor across GCR species. These results offer new insights into the temporal and environmental dependence of the space radiation quality factor, with implications for radiation dose estimate and crewed mission design.

physics.space-ph↗

What Shapes a Creative Machine Mind? Comprehensively Benchmarking Creativity in Foundation Models

The meteoric rise of foundation models (FMs) has expanded their capabilities far beyond conventional tasks. Creativity, long regarded as a hallmark of human intelligence and a driver of innovation, is now increasingly recognized as a critical dimension of machine intelligence in the era of generative FMs, complementing traditional measures of accuracy. However, existing evaluation frameworks for creativity remain fragmented, relying on ad hoc metrics not firmly grounded in established theories. To address this gap, we introduce C^2-Eval, a holistic benchmark for unified assessment of creativity in FMs. C^2-Eval distinguishes between two complementary forms of creativity: convergent creativity, where tasks admit constrained solutions (e.g., code generation), and divergent creativity, where tasks are open-ended (e.g., storytelling). It evaluates both dimensions using fine-grained criteria derived from social-science theory, focusing on Usefulness, Originality, and Surprise (U-O-S). Through extensive experiments on leading proprietary and open-source models, we analyze trade-offs in their creative capabilities. Our results highlight both the strengths and challenges of current FMs in pursuing a creative machine mind, showing that C^2-Eval is an effective lens for examining the evolving landscape of creative AI.

cs.AI↗

SPA: Towards A Computational Friendly Cloud-Base and On-Devices Collaboration Seq2seq Personalized Generation with Casual Inference

Large language models(LLMs) have shown its outperforming ability on various tasks and question answering. However, LLMs require substantial memory storage on low-resource devices. More critically, the computational speed on these devices is also severely limited. In this paper, we propose SPA(Side Plugin Adaption), a lightweight architecture for fast on-devices inference on the constraints of strict on-devices computation and memory constraints. Compared with other on-devices seq2seq generation, SPA could make a fast and stable inference on low-resource constraints, allowing it to obtain cost effiency. Our method establish an interaction between a pretrained LLMs on-cloud and additive parameters on-devices, which could provide the knowledge on both pretrained LLMs and featured personal feature. Further more, SPA provides a framework to keep feature-base parameters on low computational devices while leave the parameters containing general information on the high computational devices.

cs.CL↗

Selected Languages are All You Need for Cross-lingual Truthfulness Transfer

Truthfulness stands out as an essential challenge for Large Language Models (LLMs). Although many works have developed various ways for truthfulness enhancement, they seldom focus on truthfulness in multilingual scenarios. Meanwhile, contemporary multilingual aligning technologies struggle to balance numerous languages and often exhibit serious truthfulness gaps across different languages, especially those that differ greatly from English. In our work, we extend truthfulness evaluation to multilingual contexts and propose a practical method for cross-lingual truthfulness transfer called Fact-aware Multilingual Selective Synergy (FaMSS). FaMSS is able to select an optimal subset of all tested languages by language bias and transfer contributions, and then employ translation instruction tuning for cross-lingual truthfulness transfer. Experimental results demonstrate that our approach can effectively reduce the multilingual representation disparity and boost cross-lingual truthfulness transfer of LLMs.

cs.CL↗