arXiv subjects

Yanhui Yang

Publications and source records attributed to Yanhui Yang.

17 recordsLinked to original sources

arXiv2026

Mechanical response of quasi-two-dimensional colloidal clusters under uniaxial tension

Despite extensive studies of equilibrium conformations of colloidal clusters, little is known about their mechanical response. Here, we investigate the tensile behavior of a quasi-two-dimensional colloidal cluster subjected to uniaxial tension up to fracture. The sample is a ribbon-shaped assembly of 16 colloidal beads bound by short-range depletion attraction. Using multiple optical tweezers, we clamp the cluster at both ends and perform a tensile test along its long axis. Combining video microscopy with particle tracking, we measure the tensile stress, strain, and particle configurations during deformation. We observe diverse mechanical response behaviors, including elastic, plastic, and soft-mode deformation, with fracture occurring at a strain near 10\%. To explain these behaviors, we construct a spring-mass frame model with breakable elastic bonds. We perform canonical Monte Carlo simulations on the full model with 32 degrees of freedom and compute the statistical distributions of mechanical observables using a simplified model with only 7 degrees of freedom. Both the simulations and the theoretical calculations accurately reproduce the experimental stress--strain curves. Moreover, the configuration distributions predicted by the simplified model agree well with both experiment and simulation in the elastic and soft-mode regimes, with only minor discrepancies in the plastic regime. This work demonstrates that the simplified spring-mass model captures the essential physics governing the rich tensile response behavior of the colloidal cluster.

cond-mat.soft↗

arXiv2026

Matter Clustering in Astrid: Reduced Baryonic Suppression from Realistic Black Hole Dynamics

Baryonic feedback from active galactic nuclei (AGN) is often invoked as a major source of suppression in the matter power spectrum, with implications for precision cosmology and the $S_8$ tension. We present Astrid-DMO, the dark matter-only counterpart to the large-volume Astrid hydrodynamical simulation, and measure baryonic effects through $P_{\rm hydro}(k)/P_{\rm DMO}(k)$. We find no significant suppression at $z=0$ and mild suppression at $z=0.2$, weaker than in other state-of-the-art simulations. Using controlled small-volume runs, we identify a key driver of this discrepancy: the treatment of black hole (BH) dynamics. The widely used BH repositioning scheme artificially enhances BH mergers and boosts kinetic AGN feedback (e.g., by a factor of $2$ at $z=1.5$), leading to overly strong suppression. By contrast, a more physical dynamical friction model reduces feedback efficiency and weakens clustering suppression. Consequently, reconciling large-scale structure measurements with cosmic microwave background (CMB)-inferred $\Lambda$CDM cosmology, while matching observed halo gas fractions, becomes more challenging. Although strengthening AGN feedback can increase suppression, in our model this induces tensions with the observed galaxy stellar mass and AGN luminosity functions. These results sharpen the need for novel mechanisms that can efficiently eject gas from halos without compromising other galaxy properties.

astro-ph.GA↗

arXiv2026

A Comparative Evaluation of AI Agent Security Guardrails

This report presents a comparative evaluation of DKnownAI Guard in AI agent security scenarios, benchmarked against three competing products: AWS Bedrock Guardrails, Azure Content Safety, and Lakera Guard. Using human annotation as the ground truth, we assess each guardrail's ability to detect two categories of risks: threats to the agent itself (e.g., instruction override, indirect injection, tool abuse) and requests intended to elicit harmful content (e.g., hate speech, pornography, violence). Evaluation results demonstrate that DKnownAI Guard achieves the highest recall rate at 96.5\% and ranks first in true negative rate (TNR) at 90.4\%, delivering the best overall performance among all evaluated guardrails.

cs.CR↗

arXiv2025

Connecting current and future dual AGN searches to LISA and PTA gravitational wave detections

Dual active galactic nuclei (DAGN) mark an observable stage of massive black hole (MBH) pairing in galaxy mergers and are precursors to the MBH binaries that generate low-frequency gravitational waves. Using the large-volume ASTRID cosmological simulation, we construct DAGN catalogs matched to current (COSMOS-Web, DESI) and forthcoming (AXIS, Roman) searches. With realistic selection functions applied, ASTRID reproduces observed dual fractions, separations, and host-galaxy properties across redshifts. We predict a substantial population of small-separation (<5 kpc) duals that current surveys fail to capture, indicating that the apparent paucity of sub-kpc systems in COSMOS-Web is driven primarily by selection effects rather than a physical deficit. By following each simulated dual forward in time, we show that dual AGN are robust tracers of MBH mergers: ~30-70% coalesce within $\lesssim 1$ Gyr, and 20-60% of these mergers produce gravitational-wave signals detectable by LISA. Duals accessible to AXIS and Roman are the progenitors of ~10% of low-redshift LISA events and ~30% of the PTA-band stochastic background. Massive green-valley galaxies with moderate-luminosity AGN, together with massive star-forming hosts containing bright quasars at $z>1$, emerge as the most likely environments for imminent MBH binaries. These results provide a unified cosmological framework linking dual AGN demographics, MBH binary formation, and gravitational-wave emission, and they identify concrete, high-priority targets for coordinated electromagnetic and GW searches in upcoming multi-messenger surveys.

astro-ph.GA↗

arXiv2025

DeepKnown-Guard: A Proprietary Model-Based Safety Response Framework for AI Agents

With the widespread application of Large Language Models (LLMs), their associated security issues have become increasingly prominent, severely constraining their trustworthy deployment in critical domains. This paper proposes a novel safety response framework designed to systematically safeguard LLMs at both the input and output levels. At the input level, the framework employs a supervised fine-tuning-based safety classification model. Through a fine-grained four-tier taxonomy (Safe, Unsafe, Conditionally Safe, Focused Attention), it performs precise risk identification and differentiated handling of user queries, significantly enhancing risk coverage and business scenario adaptability, and achieving a risk recall rate of 99.3%. At the output level, the framework integrates Retrieval-Augmented Generation (RAG) with a specifically fine-tuned interpretation model, ensuring all responses are grounded in a real-time, trustworthy knowledge base. This approach eliminates information fabrication and enables result traceability. Experimental results demonstrate that our proposed safety control model achieves a significantly higher safety score on public safety evaluation benchmarks compared to the baseline model, TinyR1-Safety-8B. Furthermore, on our proprietary high-risk test set, the framework's components attained a perfect 100% safety score, validating their exceptional protective capabilities in complex risk scenarios. This research provides an effective engineering pathway for building high-security, high-trust LLM applications.

cs.AI↗

arXiv2025

The ASTRID Simulation at z=0: From Massive Black Holes to Large-scale Structure

We present the $z=0$ results for the cosmological simulation ASTRID. Hosting $2\times 5500^3\approx$ 0.33 trillion particles in a box of $370\, {\rm Mpc}$ per side, ASTRID is one of the largest cosmological hydrodynamic simulations evolved to $z=0$. ASTRID features a large population of massive black holes (MBHs), covering a wide mass range $4\times10^{4}\sim 2\times 10^{11}\ M_{\odot}$. The adopted dynamical friction model provides a relatively accurate description of MBH dynamics, making ASTRID a powerful tool to study MBH growth and mergers in a cosmological context. ASTRID successfully captures the co-evolution of MBHs and their host galaxies, producing $M_{\rm BH}-M_{\star}$ and $M_{\rm BH}-\sigma$ relations in good agreement with observations. Notably, ASTRID generates scatter in these relations that is more consistent with observations than previous simulations, indicating a more realistic MBH diversity. The galaxy stellar mass function at $z=0$ is generally consistent with observational constraints. When dust attenuation is applied, the galaxy luminosity function also agrees well with observations, and the bimodality in galaxy colors is reproduced as well. ASTRID hosts a large population of massive galaxy groups and clusters: 7 halos have $M_{\rm 200c}>10^{15}\ M_{\odot}$, and 9709 halos have $M_{\rm 200c}>10^{13}\ M_{\odot}$. We quantify the stellar mass content in these halos, and find that the correlations between the stellar and halo mass match well with observational constraints. Finally, we present the $z=0$ power spectra of MBH and galaxies, as well as their bias with respect to the matter power spectrum. We find that MBHs with $M_{\rm BH}\geq 10^{8}\ M_{\odot}$ and galaxies with $M_{\star}\geq 10^{10.5}\ M_{\odot}$ serve as good tracers of large-scale structure.

astro-ph.GA↗

arXiv2025

Small-scale Lyman alpha forest cosmology with PRIYA: Constraints from XQ100 and KODIAQ-SQUAD one-dimensional flux power spectra

We present a new cosmological analysis of the small-scale Lyman alpha forest 1D flux power spectrum (P1D) using high-resolution quasar spectra from XQ100 and KODIAQ-SQUAD, interpreted through the PRIYA emulator. PRIYA is a suite of galaxy formation simulations spanning a range of cosmological and inhomogeneous HeII reionization parameters, enabling few-percent-level predictions of the P1D. These datasets, probing down to $k \sim 6\,h\,\mathrm{Mpc}^{-1}$ at $z = 2-5$, offer access to non-linear scales inaccessible to large-volume surveys like eBOSS. We find that the XQ100 P1D yields constraints on the primordial power spectrum parameters $(A_P, n_P)$ at pivot scale $k_0 = 0.78\,\mathrm{Mpc}^{-1}$ that are consistent with PRIYA results from eBOSS DR14 and Planck CMB, albeit with broader uncertainties. Notably, this is achieved without external IGM temperature data, showing that XQ100 alone provides stronger constraints on thermal history than eBOSS DR14. In contrast, the KODIAQ-SQUAD P1D favors a significantly higher $A_P$ value, driven by the selection bias toward high-column density absorbers (HCDs). We also find that the P1D at $k > 0.045\,\mathrm{s/km}$ is more sensitive to Lyman limit system contamination and thermal history. When imposing a prior on $(A_P, n_P)$, the reduced $\chi^2$ remains unchanged and the inferred mean IGM temperature is unaffected, suggesting that cosmological and thermal parameters are largely sensitive to different scales. The XQ100 P1D therefore provides complementary information on thermal nuisance parameters, which can be jointly fit with eBOSS or DESI P1D measurements to improve cosmological constraints.

astro-ph.CO↗

arXiv2025

Ten-dimensional neural network emulator for the nonlinear matter power spectrum

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 $\sim 0.5 \%$ average accuracy for redshifts $0 \leq z \leq 3$ and scales $0.006 \leq k/(h\,\mathrm{Mpc}^{-1}) \leq 10$. The emulator models a 10D parameter space that extends beyond $\Lambda$CDM to include dynamical dark energy (characterized by $w_0$ and $w_a$), massive neutrinos ($\sum m_\nu$), the effective number of neutrinos ($N_\text{eff}$), and running of the spectral index ($\alpha_\text{s}$). 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 $\sim $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.

astro-ph.CO↗

arXiv2025

Design and optimization of neural networks for multifidelity cosmological emulation

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 the mean error by more than a factor of five and in the worst-case error by approximately a factor of eight in leave-one-out cross-validation, relative 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.

astro-ph.CO↗

arXiv2025

DT4ECG: A Dual-Task Learning Framework for ECG-Based Human Identity Recognition and Human Activity Detection

This article introduces DT4ECG, an innovative dual-task learning framework for Electrocardiogram (ECG)-based human identity recognition and activity detection. The framework employs a robust one-dimensional convolutional neural network (1D-CNN) backbone integrated with residual blocks to extract discriminative ECG features. To enhance feature representation, we propose a novel Sequence Channel Attention (SCA) mechanism, which combines channel-wise and sequential context attention to prioritize informative features across both temporal and channel dimensions. Furthermore, to address gradient imbalance in multi-task learning, we integrate GradNorm, a technique that dynamically adjusts loss weights based on gradient magnitudes, ensuring balanced training across tasks. Experimental results demonstrate the superior performance of our model, achieving accuracy rates of 99.12% in ID classification and 90.11% in activity classification. These findings underscore the potential of the DT4ECG framework in enhancing security and user experience across various applications such as fitness monitoring and personalized healthcare, thereby presenting a transformative approach to integrating ECG-based biometrics in everyday technologies.

eess.SP↗

arXiv2025

Improved VMD Based Remote Heartbeat Estimation Utilizing 60GHz mmWave Radar

This study introduces an improved VMD based signal decomposition methodology for non-contact heartbeat estimation using millimeterwave (mmWave) radar. Specifically, we first analyze the signal model of the mmWave radar system. The Variational Mode Decomposition (VMD) integrated with the Newton-Raphson-based optimizer (NRBO) algorithm are sequentially utilized for cardiac mechanic signal (CMS) reconstruction. The estimation accuracy is enhanced by adaptively optimizing the VMD parameters including intrinsic mode functions (IMFs) and penalty factor. Eventually, the experimental results of 18 subjects validate the effectiveness of the proposed method by comparing with three commonly used baselines.

eess.SP↗

arXiv2025

The Gravitational Wave Background from Massive Black Holes in the ASTRID Simulation

Recent pulsar timing array (PTA) observations have detected nanohertz gravitational waves, likely originating from massive black hole binaries (MBHBs). The detected amplitude is unexpectedly higher than inferred from the electromagnetic measurements. We present new gravitational wave background (GWB) results from the ASTRID simulation. Its large volume and on-the-fly dynamical friction for MBHs provide new insights into the MBHB population, offering a more accurate assessment of its contribution to the observed GWB. ASTRID predicts a GWB from MBHBs of $h_c=2.8\times10^{-15}$, or $\sim45\%$ of the observed amplitude at $\sim 4\,{\rm nHz}$ and $h_c=2.5\times10^{-16}$ ($5\%$) with $h_c\propto f^{-1.6}$ at $\sim 30\,{\rm nHz}$. These predictions remain below current PTA constraints but align with previous empirical models based on the observed MBH mass functions. By comparison, TNG300 with post-processed MBH dynamics yields a range between $70-90\%$ ($20\% - 30\%$) of the observed levels at low (high) frequencies. At low frequencies, ASTRID predicts that the bulk of the GWB originates from MBHB with masses $M_{\rm tot}=1-3\times 10^9\,M_\odot$ peaking at $z\approx 0.3$, consistent with TNG300. Notably, both simulations predict significant GWB contribution from minor mergers ($q<0.2$) by up to $\sim 40\%$. By tracing the full merger trees of local MBHs in ASTRID, we show that they generate GWs at $\sim 10\%-80\%$ of the maximum signal assuming no accretion and recent equal-mass mergers. Finally, we demonstrate the importance of on-the-fly MBH dynamics, the lack of which leads to $3- 5$ times excessive mass growth by merger, and a similar boost to the GWB prediction.

astro-ph.GA↗

arXiv2025

Central Cluster Galaxies: A Hotspot for Detectable Gravitational Waves from Black Hole Mergers

After Pulsar Timing Arrays (PTAs) have announced the evidence for a low-frequency gravitational wave background (GWB), the continuous waves (CWs) are the next anticipated gravitational wave (GW) signals. In this work, we model CW sources detectable by PTAs based on the massive black hole (MBH) merger population in the ASTRID cosmological simulation. We evolve MBH binaries, simulate their GW emissions, and calculate their detection probability (DP) for PTAs. The most detectable CW sources are produced by MBH mergers with masses M_BH > 10^10 solarmass in the lowest frequency bins with f<10 nHz. Remarkably, these mergers occur within massive galaxies with the stellar mass larger than 10^12 solarmass located at the center of galaxy clusters. Particularly striking in ASTRID is a triple merger event, wherein two consecutive mergers occur within 500 Myr interval in the same cluster core, generating high-DP CW signals at ~ 2nHz and ~ 10nHz. We also investigate the electromagnetic (EM) signatures associated with these events: either single or dual active galactic nuclei (AGN) in the massive host galaxies that are undergoing star formation. This research provides new insights into the low-frequency GW sky and informs future multi-messenger searches for PTA CW sources.

astro-ph.GA↗

arXiv2025

Ten-parameter simulation suite for cosmological emulation beyond $\Lambda$CDM

We present Goku, a suite of cosmological $N$-body simulations, and the corresponding 10-dimensional emulator, GokuEmu, for the nonlinear matter power spectrum. The simulations span the base parameters of $\Lambda$ Cold Dark Matter ($\Lambda$CDM) cosmology and its extensions, including dynamical dark energy ($w_0$, $w_a$), the sum of the neutrino masses ($\sum m_\nu$), the effective number of neutrinos ($N_\text{eff}$), and the running of the scalar spectral index ($\alpha_\text{s}$), enabling tests of new physics with data from upcoming surveys like the Roman Space Telescope, Euclid, and LSST. Designed within the MF-Box framework, which integrates multi-scale and multi-fidelity simulations, the suite includes high-fidelity simulations evolving $3000^3$ particles in $1\,(\text{Gpc}/h)^3$ volumes and low-fidelity simulations with $750^3$ particles across varying box sizes. This approach achieves percent-level accuracy in high-likelihood regions and 5% accuracy across broader parameter ranges, while reducing computational costs by 94% compared to single-fidelity methods. The simulations adopt an accurate treatment of massive neutrinos, enhancing predictions of the matter power spectrum on nonlinear scales. Key innovations include an adaptive sampling strategy and the use of beam search to optimize generalization accuracy. The emulator is valid for redshifts $z \leq 3$ and scales $0.01 \lesssim k / (h \, \text{Mpc}^{-1}) \lesssim 10$. Beyond the matter power spectrum, the simulations also support analyses of other statistical measures, such as the halo mass function. The emulator and its training data are publicly available at https://github.com/astro-YYH/GokuEmu, providing a valuable resource for cosmological parameter inference and model testing.

astro-ph.CO↗

arXiv2024

The Astrid Simulation: Evolution of black holes and galaxies to z=0.5 and different evolution pathways for galaxy quenching

We present new results from the ASTRID simulation from $z=3$ to $z=0.5$, covering the epoch of cosmic noon. The galaxy stellar mass function, as well as the black hole mass and luminosity functions in ASTRID, exhibit good agreement with recent observational constraints. We study the $M_{\rm BH}$-$M_*$ scaling relation and its connections to AGN luminosity, galaxy color, and star formation rate, demonstrating that AGN feedback plays a crucial role in the quenching of massive galaxies ($M_*>10^{10.5} M_{\odot}$). Although AGN feedback suppresses star formation through quenching, AGN-host galaxies still exhibit statistically higher levels of star formation compared to inactive ones, due to the positive correlation between AGN activity and star formation, both fueled by a shared gas reservoir. The fraction of quiescent galaxies in ASTRID increases with both galaxy mass and redshift evolution, aligning well with observational trends. We find that different quenching mechanisms can leave distinct morphological imprints on quenched galaxies. Massive, compact quiescent galaxies typically experience shorter quenching timescales, have younger central regions, and host overmassive black holes. This is usually due to a compaction-like quenching mechanism that funnels gas into the galaxy center, leading to starbursts and triggering AGN kinetic feedback. In contrast, quiescent galaxies with more diffuse morphologies generally experience `inside-out' quenching, which is characterized by older central regions compared to the outskirts. These galaxies typically experience longer quenching timescales due to quenching processes operating on a larger halo scale, which gradually deplete the galactic star-forming gas. Data of the \astrid simulation down to $z=0.5$ is available at \url{https://astrid.psc.edu}.

astro-ph.GA↗

arXiv2022

Radiative turbulent mixing layers at high Mach numbers

Radiative turbulent mixing layers (TMLs) are ubiquitous in astrophysical environments, e.g., the circumgalactic medium (CGM), and are triggered by the shear velocity at interfaces between different gas phases. To understand the shear velocity dependence of TMLs, we perform a set of 3D hydrodynamic simulations with an emphasis on the TML properties at high Mach numbers $\mathcal{M}$. Since the shear velocity in mixing regions is limited by the local sound speed of mixed gas, high-Mach number TMLs develop into a two-zone structure: a Mach number-independent mixing zone traced by significant cooling and mixing, plus a turbulent zone with large velocity dispersions which expands with greater $\mathcal{M}$. Low-Mach number TMLs do not have distinguishable mixing and turbulent zones. The radiative cooling of TMLs at low and high Mach numbers is predominantly balanced by enthalpy consumption and turbulent dissipation respectively. Both the TML surface brightness and column densities of intermediate-temperature ions (e.g., O VI) scale as $\propto\mathcal{M}^{0.5}$ at $\mathcal{M} \lesssim 1$, but reach saturation ($\propto \mathcal{M}^0$) at $\mathcal{M} \gtrsim 1$. Inflow velocities and hot gas entrainment into TMLs are substantially suppressed at high Mach numbers, and strong turbulent dissipation drives the evaporation of cold gas. This is in contrast to low-Mach number TMLs where the inflow velocities and hot gas entrainment are enhanced with greater $\mathcal{M}$, and cold gas mass increases due to the condensation of entrained hot gas.

astro-ph.GA↗

arXiv2022

QuoteR: A Benchmark of Quote Recommendation for Writing

It is very common to use quotations (quotes) to make our writings more elegant or convincing. To help people find appropriate quotes efficiently, the task of quote recommendation is presented, aiming to recommend quotes that fit the current context of writing. There have been various quote recommendation approaches, but they are evaluated on different unpublished datasets. To facilitate the research on this task, we build a large and fully open quote recommendation dataset called QuoteR, which comprises three parts including English, standard Chinese and classical Chinese. Any part of it is larger than previous unpublished counterparts. We conduct an extensive evaluation of existing quote recommendation methods on QuoteR. Furthermore, we propose a new quote recommendation model that significantly outperforms previous methods on all three parts of QuoteR. All the code and data of this paper are available at https://github.com/thunlp/QuoteR.

cs.CL↗