arXiv subjects

Xintong Shi

Publications and source records attributed to Xintong Shi.

12 recordsLinked to original sources

arXiv2026

Broad-angle photon-pair generation from flatband quasi-BIC resonant 3R-MoS$_{2}$ metasurfaces

Spontaneous parametric down-conversion (SPDC) in ultrathin optical resonant metasurfaces offers a promising platform for integrated quantum light sources. However, conventional resonances suffer from steep momentum dispersion, limiting photon-pair generation to narrow excitation and emission angles. Here, we demonstrate a flatband-resonant 3R-MoS$_{2}$ metasurface for high-efficiency, broad-angle SPDC photon-pair generation. By implementing a supercell symmetry-breaking strategy via Brillouin zone folding, we achieve a flatband quasi-bound states in the continuum (quasi-BICs) resonance in the near-infrared regime that maintains a robust quality factor across a wide range of incident angle. Exploiting the strong local electric field enhancement of resonance and the large second-order susceptibility of 3R-MoS$_{2}$, the metasurface boosts the degenerate SPDC pair-generation rate with orders of magnitude enhancement over an unpatterned film of identical thickness. Crucially, the near-zero band dispersion extends the effective emission angle to $\pm 5^{\circ}$ with negligible spectral drift $\sim 0.1$ nm and high directional tolerance. Under pulsed excitation with a finite spectral bandwidth, the photon pair yield remains high, demonstrating strong spectral locking to the quasi-BIC resonance against pump detunings. This work establishes a versatile strategy for momentum-independent quantum light sources, opening new avenues for scan-free quantum ghost imaging and spatial-frequency quantum processing.

physics.optics↗

arXiv2026

TRAJDEBUG: Tracing Error Lifecycle to Identify Critical Failures in Long-Horizon Agent Trajectories

LLM-based agentic systems have shown remarkable capabilities in complex domains, while suffering from cascading errors and difficulty in debugging. Critical error detection aims to locate the earliest error step in a failed trajectory that is responsible for the final failure. However, progress faces two main challenges. First, long trajectories make it difficult to identify individual errors, since the evidence for judging a step may be scattered across distant instructions, observations, and prior context. Second, failed trajectories often contain multiple local errors with different downstream effects, only some of which remain responsible for the final failure. In this work, we propose TrajDebug, an error-lifecycle tracing framework that addresses long-trajectory error discovery with multi-granularity history compression and evidence-based error identification, and supports critical attribution by tracing each error's resolution status and terminal impact. We further construct TrajErrBench, a benchmark of 486 manually annotated failed trajectories from Tau2Bench and SWE-Bench Pro, covering realistic tool-use and coding scenarios. Experiments across diverse agent benchmarks show that TrajDebug achieves the best overall performance over existing baselines, and application studies further demonstrate that its diagnoses provide actionable feedback for improving downstream agent success. We will release the codes and data to facilitate further research.

cs.AI↗

arXiv2026

Coexistence and manipulation of multiple singularities in a reconfigurable non-Hermitian metasurface

Non-Hermitian frameworks extend conventional Hermitian physics, offering a powerful paradigm for describing open systems. Central to this field are various singularities within the complex parameter space, such as exceptional points (EPs) and scattering zeros, which dictate exotic physical behaviors. As research shifts from isolated singularities toward multi-singularity interactions, conventional planar metasurfaces remain constrained by limited tuning dimensions. Here, we propose a mirror-coupled design that maps a metasurface into a quasi-high-dimensional parameter space. By employing a metallic plane to generate image resonators, this scheme multiplies the system degrees of freedom without increasing the number of physical resonators. Its implementation on a reconfigurable platform integrated with PIN diodes yields the coexistence and manipulation of an EP and multiple reflection zeros. Through simulations and microwave experiments, we characterize the dynamic evolution of these singularities and exploit their synergistic effects for two distinct applications. First, for tunable absorption, multiple reflection zeros are spectrally coordinated to achieve a near-perfect absorption band exceeding $99.9\%$ across the X-band, thereby dynamically suppressing target scattering. Second, for enhanced sensing, a reflection zero couples with the EP to form a hybrid singularity. This hybrid state inherits the power-law sensitivity of the EP while substantially boosting robustness against fluctuations, resolving the conventional trade-off between sensitivity and stability and simplifying detection to direct peak tracking rather than complex multimode eigenvalue fitting. Our work provides a general methodology to circumvent parameter competition among non-Hermitian singularities, opening new avenues for multifunctional metadevices across the electromagnetic spectrum.

physics.optics↗

arXiv2026

Can LLM-as-a-Judge Reliably Verify Rubrics in Agentic Scenarios?

Rubric-based scoring has become a widely used paradigm in model evaluation, typically with LLM-as-a-Judge (LaaJ) for rubric scoring. However, the reliability of LaaJ for rubric scoring remains underexplored. This concern is especially pronounced in agentic scenarios, where long, complex outputs further challenge reliable scoring. To address this, we conduct a systematic meta-evaluation of LaaJ reliability for rubric verification. We introduce RuVerBench, the first benchmark for assessing LaaJ reliability in rubric verification for agentic scenarios. RuVerBench covers two prevalent agentic domains, deep research and agentic coding, with 2,458 instances, each containing a model-generated output, a rubric, and a human-annotated label indicating whether the output satisfies the rubric. Using RuVerBench, we evaluate numerous frontier LLMs and find that even the most advanced models achieve strong performance but still exhibit substantial noise. We further analyze the impact of key LaaJ strategies, including prompt design, batching, and majority voting, on rubric verification. We find that weaker models are more sensitive to prompt variations, batched verification presents a trade-off between accuracy and efficiency, and majority voting yields effective but diminishing returns. We have released our dataset and code to facilitate future research: https://github.com/THU-KEG/RuVerBench.

cs.CL↗

arXiv2026

Flat optics for analog computing: from fundamental mechanisms to advanced meta-processors

As the explosive growth of visual data increasingly strains the latency and energy limits of conventional electronic computing, optical analog computing has re-emerged as a disruptive paradigm for zero-power, speed-of-light information processing. Propelled by the unprecedented wave-manipulation capabilities of optical metasurfaces, this field is undergoing a rapid transition from macroscopic physical optics to ultra-compact, on-chip meta-processors. This Review examines the fundamental mechanisms of metasurface-empowered optical computing spanning Fourier-domain, nonlocal spatial-domain, and interferometric architectures that perform mathematical operations, with a particular focus on spatial differentiation and edge detection as representative computing tasks. By emphasizing recent breakthroughs, we highlight the evolution of meta-processors from static, linear regimes to dynamically reconfigurable, nonlinear, and quantum-assisted multidimensional platforms. We also envision how the synergy of AI-driven inverse design and the integration of analog meta-front-ends with optical neural networks will synergistically revolutionize next-generation intelligent machine vision.

physics.optics↗

arXiv2026

Doubly resonant nonlinear metasurfaces enabling NIR-to-UV upconversion for reconfigurable Fourier optical processing

Fourier optical processing underpins optical information manipulation, yet extending such operations to short wavelengths within compact platforms remains challenging. Here, we address this challenge by embedding reconfigurable Fourier-domain processing within IR-to-UV upconversion in a doubly resonant nonlinear metasurface. When coherently illuminated at the Fourier plane with an image-bearing signal and a spatially structured pump, the metasurface generates UV images via degenerate four-wave mixing. Crucially, the spatial-frequency content of these upconverted images is selectively shaped by the tailored spectrum of the pump. To boost the efficiency of this nonlinear process, the metasurface is designed to simultaneously support a toroidal dipole bound state in the continuum and a magnetic dipole resonance, providing spectrally aligned and independently enhanced field localization for signal and pump beams, respectively. Building on this architecture, we experimentally demonstrate directional and continuously tunable filtering at the upconverted UV wavelengths. These results establish nonlinear metasurfaces as a versatile platform for Fourier optics and reconfigurable all-optical image processing.

physics.optics↗

arXiv2026

On the Paradoxical Interference between Instruction-Following and Task Solving

Instruction following aims to align Large Language Models (LLMs) with human intent by specifying explicit constraints on how tasks should be performed. However, we reveal a counterintuitive phenomenon: instruction following can paradoxically interfere with LLMs' task-solving capability. We propose a metric, SUSTAINSCORE, to quantify the interference of instruction following with task solving. It measures task performance drop after inserting into the instruction a self-evident constraint, which is naturally met by the original successful model output and extracted from it. Experiments on current LLMs in mathematics, multi-hop QA, and code generation show that adding the self-evident constraints leads to substantial performance drops, even for advanced models such as Claude-Sonnet-4.5. We validate the generality of the interference across constraint types and scales. Furthermore, we identify common failure patterns, and by investigating the mechanisms of interference, we observe that failed cases allocate significantly more attention to constraints compared to successful ones. Finally, we use SUSTAINSCORE to conduct an initial investigation into how distinct post-training paradigms affect the interference, presenting empirical observations on current alignment strategies. We will release our code and data to facilitate further research

cs.CL↗

arXiv2025

Super-robust telecommunications enabled by topological half-supermodes

Topological photonics offers transformative potential for robust integrated waveguide devices due to their backscattering-immune properties. However, their integration faces two fundamental challenges: mode symmetry mismatch with conventional waveguides and prohibitive dimensions. We successfully overcome these two critical challenges by introducing a novel valley-ridge gap waveguide based on topological half-supermode engineering. By strategically hybridizing ridge waveguide modes and valley kink states, we create an exotic odd-symmetric supermode enabling robust propagation and ultra-compact operation. The further implementation of a perfect electric conductor boundary halves lateral dimensions while eliminating radiation loss. Crucially, our proposed valley-ridge interface achieves direct transverse electric mode matching with standard waveguides without transition structures, enabling seamless integration. Experimental results demonstrate reflection losses lower than -15 dB in realistic telecommunication scenarios with super-robust signal propagation through sharp bends. This work innovatively conceptualizes topological half-supermodes and pioneers their practical applications for integrated waveguide devices, establishing a completely new waveguide class that uniquely combines robust backscattering immunity with deep subwavelength compactness.

physics.optics↗

arXiv2024

Polarization-controlled non-Hermitian metasurfaces for ultra-sensitive terahertz sensing

Non-Hermitian systems offer significant advantages in sensor design, especially at the exceptional points. However, the extreme sensitivity near these points poses great challenges due to fabrication errors and system noises, which degrade sensing performance. To address this, we introduce a novel approach leveraging the polarization degrees of freedom in non-Hermitian systems. In this work, we establish a direct relation between the incident polarization and the transmission phase of a coupled metasurface system and achieve the polarization-controlled phase singularity even post-fabrication. The incident polarization angle can be utilized as a sensing index, which enables indirect and accurate measurement. The theoretical approach is experimentally validated using a general design of THz non-Hermitian metasurface sensors. Our method enhances robustness and sensitivity, opening new avenues for practical applications in ultra-sensitive sensing.

physics.optics↗

arXiv2024

Extremely asymmetric absorption and reflection near the exceptional point of three-dimensional metamaterial

In recent years, particular physical phenomena enabled by non-Hermitian metamaterial systems have attracted significant research interests. In this paper, a non-Hermitian three-dimensional metamaterial near the exceptional point (EP) is proposed to demonstrate extremely asymmetric absorption and reflection. Unlike its conventional counterparts, this proposed metamaterial is constructed with a loss-assisted design. Localized losses are introduced into the structure by combining our technique of graphene-based resistive inks with conventional printed circuit board (PCB) process. Extremely asymmetric absorption and reflection near the EP are experimentally observed by tuning the loss between split ring resonators (SRRs) in the meta-atoms. Simultaneously, by linking the equivalent circuit model (ECM) with the Hamiltonian quantum physical model, the equivalent non-Hermitian Hamiltonian is obtained and a non-Hermitian transmission matrix is constructed. We show that tuning the structure and circuit parameters of the ECM produces a metamaterial system with EP response. Our system can be used in the design of asymmetric metamaterial absorbers. Our work lays down the way for the manipulation of EP to develop perfect absorption, sensing and other applications in the 3D metamaterial platform.

physics.app-ph↗

arXiv2023

A topological gap waveguide based on unidirectional locking of pseudo-spins

Photonic topological insulators (PTIs) have been widely studied due to the robustness of energy transport via supported edge modes immune to structural disorder. In this work, a topological gap waveguide is constructed by introducing line defect into a topological photonic crystal structure and combining it with a gap waveguide structure, which design therefore combines the advantages of both topological and gap waveguides. Not only does it give high transmission efficiency, but also enables high robustness for energy transmission under structural defects and sharp bends. Our proposed topological waveguide design can be implemented with conventional semiconductor technology and integrated into optical circuits for communication systems.

physics.app-ph↗

arXiv2023

Deep Neural Networks for Rank-Consistent Ordinal Regression Based On Conditional Probabilities

In recent times, deep neural networks achieved outstanding predictive performance on various classification and pattern recognition tasks. However, many real-world prediction problems have ordinal response variables, and this ordering information is ignored by conventional classification losses such as the multi-category cross-entropy. Ordinal regression methods for deep neural networks address this. One such method is the CORAL method, which is based on an earlier binary label extension framework and achieves rank consistency among its output layer tasks by imposing a weight-sharing constraint. However, while earlier experiments showed that CORAL's rank consistency is beneficial for performance, it is limited by a weight-sharing constraint in a neural network's fully connected output layer, which may restrict the expressiveness and capacity of a network trained using CORAL. We propose a new method for rank-consistent ordinal regression without this limitation. Our rank-consistent ordinal regression framework (CORN) achieves rank consistency by a novel training scheme. This training scheme uses conditional training sets to obtain the unconditional rank probabilities through applying the chain rule for conditional probability distributions. Experiments on various datasets demonstrate the efficacy of the proposed method to utilize the ordinal target information, and the absence of the weight-sharing restriction improves the performance substantially compared to the CORAL reference approach. Additionally, the suggested CORN method is not tied to any specific architecture and can be utilized with any deep neural network classifier to train it for ordinal regression tasks.

cs.LG↗