arXiv subjects

Dongchen Huang

Publications and source records attributed to Dongchen Huang.

11 recordsLinked to original sources

arXiv2026

Deep Research in Physical Sciences: A Multi-Agent Framework and Comprehensive Benchmark

Deep research agents are Large Language Model (LLM)-based systems designed for autonomous, multi-step scientific reasoning, and they hold immense potential for accelerating research in the physical sciences. However, comprehensive and in-depth evaluations of their capabilities within this domain remain lacking. To address this gap, we introduce PhySciBench, a benchmark highly relevant to physical science research, comprising 200 expert-curated questions, balanced between physics and chemistry, across six task categories that reflect real-world scientific workflows. Evaluations of state-of-the-art models and agent systems on PhySciBench reveal limited performance; even the strongest baseline, Gemini Deep Research, achieves an accuracy of only 33.5%. Analysis of failure cases identifies three recurrent deficiencies: fragility in extended reasoning chains, limited knowledge transfer across steps, and a lack of physics-grounded self-verification. Motivated by these findings, we develop DelveAgent, a modular multi-agent framework equipped with an adaptive planning loop, dual-granularity memory, and a hierarchical physics-grounded reflection mechanism. Across four scientific benchmarks, DelveAgent improves accuracy by up to 7.5 percentage points while reducing inference costs to approximately one-third of the strongest baseline. These results establish the significance of PhySciBench as a critical benchmark for evaluating AI systems in the physical sciences and demonstrate that architectural specialization can effectively enhance the reliability of autonomous scientific research. Our data and code are publicly available at https://github.com/yigengjiang/physci-deepresearch.

physics.comp-ph↗

arXiv2026

PRISM: Deriving a White-Box Transformer as a Signal-Noise Decomposition Operator via Maximum Coding Rate Reduction

Deep learning models, particularly Transformers, are often criticized as "black boxes" and lack interpretability. We propose Prism, a white-box attention-based architecture derived from the principles of Maximizing Coding Rate Reduction ($\text{MCR}^2$). By modeling the attention mechanism as a gradient ascent process on a distinct signal-noise manifold, we introduce a specific irrational frequency separation ($\pi$-RoPE) to enforce incoherence between signal (semantic) and noise (syntactic) subspaces. We show empirical evidence that these geometric inductive biases can induce unsupervised functional disentanglement alone. Prism spontaneously specializes its attention heads into spectrally distinct regimes: low-frequency heads capturing long-range causal dependencies (signal) and high-frequency heads handling local syntactic constraints and structural artifacts. To provide a theoretical grounding for these spectral phenomena, we draw an analogy between attention mechanism and a Hamiltonian dynamical system and identify that the standard geometric progression of Rotary Positional Embeddings (RoPE) induces dense resonance networks (Arnold Tongues), leading to feature rank collapse. Empirical validation on 124M-parameter models trained on OpenWebText demonstrates that Prism spontaneously isolates the Attention Sink pathology and maintains isentropic information flow across layers. Further, we suggest a physics-informed plug-and-play intervention KAM-RoPE for large language models (LLMs). Our results suggest that interpretability and performance can be unified through principled geometric construction, offering a theoretically grounded alternative to heuristic architectural modifications

cs.LG↗

arXiv2025

CMPhysBench: A Benchmark for Evaluating Large Language Models in Condensed Matter Physics

We introduce CMPhysBench, designed to assess the proficiency of Large Language Models (LLMs) in Condensed Matter Physics, as a novel Benchmark. CMPhysBench is composed of more than 520 graduate-level meticulously curated questions covering both representative subfields and foundational theoretical frameworks of condensed matter physics, such as magnetism, superconductivity, strongly correlated systems, etc. To ensure a deep understanding of the problem-solving process,we focus exclusively on calculation problems, requiring LLMs to independently generate comprehensive solutions. Meanwhile, leveraging tree-based representations of expressions, we introduce the Scalable Expression Edit Distance (SEED) score, which provides fine-grained (non-binary) partial credit and yields a more accurate assessment of similarity between prediction and ground-truth. Our results show that even the best models, Grok-4, reach only 36 average SEED score and 28% accuracy on CMPhysBench, underscoring a significant capability gap, especially for this practical and frontier domain relative to traditional physics. The code anddataset are publicly available at https://github.com/CMPhysBench/CMPhysBench.

cs.LG↗

arXiv2024

Generalized Spectral Decomposition for Quantum Impurity Problems

Solving quantum impurity problems may advance our understanding of strongly correlated electron physics, but its development in multi-impurity systems has been greatly hindered due to the presence of shared bath. Here, we propose a general operation strategy to disentangle the shared bath into multiple auxiliary baths and relate the problem to a spectral decomposition problem of function matrix for applying the numerical renormalization group (NRG). We prove exactly that such decomposition is possible for models satisfying (block) circulant symmetry, and show how to construct the auxiliary baths for arbitrary impurity configuration by mapping its graph structure to the subgraph of a regular impurity configuration. We further propose an approximate decomposition algorithm to reduce the number of auxiliary baths and save the computational workload. Our work reveals a deep connection between quantum impurity problems and the graph theory, and provides a general scheme to extend the NRG applications for realistic multi-impurity systems.

cond-mat.str-el↗

arXiv2024

Training-set-free two-stage deep learning for spectroscopic data de-noising

De-noising is a prominent step in the spectra post-processing procedure. Previous machine learning-based methods are fast but mostly based on supervised learning and require a training set that may be typically expensive in real experimental measurements. Unsupervised learning-based algorithms are slow and require many iterations to achieve convergence. Here, we bridge this gap by proposing a training-set-free two-stage deep learning method. We show that the fuzzy fixed input in previous methods can be improved by introducing an adaptive prior. Combined with more advanced optimization techniques, our approach can achieve five times acceleration compared to previous work. Theoretically, we study the landscape of a corresponding non-convex linear problem, and our results indicates that this problem has benign geometry for first-order algorithms to converge.

cond-mat.mtrl-sci↗

arXiv2023

Superconducting fluctuations and charge-4$e$ plaquette state at strong coupling

We apply the static auxiliary field Monte Carlo approach to study phase correlations of the pairing fields in a microscopic model with spin-singlet pairing interaction. We find that the short- and long-range phase correlations are well captured by the phase mutual information, which allows us to construct a theoretical phase diagram containing the uniform $d$-wave superconducting region, the phase fluctuating region, the local pairing region, and the disordered region. We show that the gradual development of phase coherence has a number of consequences on spectroscopic measurements, such as the development of the Fermi arc and the anisotropy in the angle-resolved spectra, scattering rate, entropy, specific heat, and quasiparticle dispersion, in good agreement with experimental observations. For strong coupling, our Monte Carlo simulation reveals an unexpected charge-4$e$ plaquette state with $d$-wave bonds, which competes with the uniform $d$-wave superconductivity and exhibits a U-shaped density of states.

cond-mat.supr-con↗

arXiv2023

Spectroscopic data de-noising via training-set-free deep learning method

De-noising plays a crucial role in the post-processing of spectra. Machine learning-based methods show good performance in extracting intrinsic information from noisy data, but often require a high-quality training set that is typically inaccessible in real experimental measurements. Here, using spectra in angle-resolved photoemission spectroscopy (ARPES) as an example, we develop a de-noising method for extracting intrinsic spectral information without the need for a training set. This is possible as our method leverages the self-correlation information of the spectra themselves. It preserves the intrinsic energy band features and thus facilitates further analysis and processing. Moreover, since our method is not limited by specific properties of the training set compared to previous ones, it may well be extended to other fields and application scenarios where obtaining high-quality multidimensional training data is challenging.

cond-mat.mtrl-sci↗

arXiv2023

Removing grid structure in angle-resolved photoemission spectra via deep learning method

Spectroscopic data may often contain unwanted extrinsic signals. For example, in ARPES experiment, a wire mesh is typically placed in front of the CCD to block stray photo-electrons, but could cause a grid-like structure in the spectra during quick measurement mode. In the past, this structure was often removed using the mathematical Fourier filtering method by erasing the periodic structure. However, this method may lead to information loss and vacancies in the spectra because the grid structure is not strictly linearly superimposed. Here, we propose a deep learning method to effectively overcome this problem. Our method takes advantage of the self-correlation information within the spectra themselves and can greatly optimize the quality of the spectra while removing the grid structure and noise simultaneously. It has the potential to be extended to all spectroscopic measurements to eliminate other extrinsic signals and enhance the spectral quality based on the self-correlation of the spectra solely.

cond-mat.mtrl-sci↗

arXiv2022

Learning disentangled representation for classical models

Finding disentangled representation plays a predominant role in the success of modern deep learning applications, but the results lack a straightforward explanation. Here we apply the information bottleneck method and its $β$-VAE implementation to find the disentangled low-dimensional representation of classical models. For the Ising model, our results reveal a deep connection between the disentangled features and the physical order parameters, and the widely-used Bernoulli decoder is found to be learning a mean-field Hamiltonian at fixed temperature. This analogy motivates us to extend the application of $β$-VAE to more complex classical models with non-binary variables using different decoder neural network and propose a modified architecture $β^2$-VAE to enforce thermal fluctuations in generated samples. Our work provides a way to design novel physics-informed algorithm that can yield learned features in potential correspondence with real physical properties.

cond-mat.str-el↗

arXiv2022

Learned Optimizers for Analytic Continuation

Traditional maximum entropy and sparsity-based algorithms for analytic continuation often suffer from the ill-posed kernel matrix or demand tremendous computation time for parameter tuning. Here we propose a neural network method by convex optimization and replace the ill-posed inverse problem by a sequence of well-conditioned surrogate problems. After training, the learned optimizers are able to give a solution of high quality with low time cost and achieve higher parameter efficiency than heuristic fully-connected networks. The output can also be used as a neural default model to improve the maximum entropy for better performance. Our methods may be easily extended to other high-dimensional inverse problems via large-scale pretraining.

cs.LG↗

arXiv2021

Mutual information, quantum phase transition, and phase coherence in Kondo systems

We propose a static auxiliary field approximation to study the hybridization physics of Kondo systems without the sign problem and use the mutual information to measure the intersite hybridization correlations. Our method takes full account of the spatial fluctuations of the hybridization fields at all orders and overcomes the artificial (first-order) phase transition predicted in the mean-field approximation. When applied to the two-impurity Kondo model, it reveals a logarithmically divergent amplitude mutual information near the so-called "Varma-Jones" fixed point and a large phase mutual information manifesting the development of intersite phase coherence in the Kondo regime, with observable influences on physical properties. These highlight the importance of hybridization fluctuations and confirm the mutual information as a useful tool to explore the hybridization physics in Kondo systems.

cond-mat.str-el↗