计算物理
查看 最近的 文章
显示 2025年10月03日， 星期五 新的列表
- [1] arXiv:2510.01205 [中文pdf, pdf, html, 其他]
-
标题： 化学中高精度方法的集成软硬件执行模型标题： Integrated Software/Hardware Execution Models for High-Accuracy Methods in ChemistryNicholas Bauman, Ajay Panyala, Libor Veis, Jiri Brabec, Paul Rigor, Randy Meyer, Skyler Windh, Craig Warner, Tony Brewer, Karol Kowalski主题： 计算物理 (physics.comp-ph) ; 化学物理 (physics.chem-ph) ; 量子物理 (quant-ph)
先进量子化学方法的有效部署和应用与新兴且高度多样化的计算资源的最佳使用密切相关。 本文研究了Micron内存技术和Azure Quantum Element云计算在密度矩阵重整化群(DMRG)模拟中的协同利用，该模拟基于双单位耦合簇(DUCC)假设的耦合簇(CC)下折叠/有效哈密顿量。 我们分析了DMRG-DUCC工作流的性能，强调了选择合适的硬件以反映与工作流特定组件相关的数值开销。 我们报告了一种混合方法，利用Micron CXL硬件进行内存容量密集的CC下折叠阶段，同时使用AQE云计算进行资源需求较低的DMRG模拟。 此外，我们分析了在Micron原型系统上进行的可扩展ExaChem电子模拟套件的性能。
The effective deployment and application of advanced methodologies for quantum chemistry is inherently linked to the optimal usage of emerging and highly diversified computational resources. This paper examines the synergistic utilization of Micron memory technologies and Azure Quantum Element cloud computing in Density Matrix Renormalization Group (DMRG) simulations leveraging coupled-cluster (CC) downfolded/effective Hamiltonians based on the double unitary coupled cluster (DUCC) Ansatz. We analyze the performance of the DMRG-DUCC workflow, emphasizing the proper choice of hardware that reflects the numerical overheads associated with specific components of the workflow. We report a hybrid approach that takes advantage of Micron CXL hardware for the memory capacity intensive CC downfolding phase while employing AQE cloud computing for the less resource-intensive DMRG simulations. Furthermore, we analyze the performance of the scalable ExaChem suite of electronic simulations conducted on Micron prototype systems.
新提交 (展示 1 之 1 条目 )
- [2] arXiv:2510.01370 (交叉列表自 cs.CV) [中文pdf, pdf, html, 其他]
-
标题： SPUS：一种轻量级且参数高效的偏微分方程基础模型标题： SPUS: A Lightweight and Parameter-Efficient Foundation Model for PDEs主题： 计算机视觉与模式识别 (cs.CV) ; 人工智能 (cs.AI) ; 机器学习 (cs.LG) ; 计算物理 (physics.comp-ph)
我们引入了小型PDE U-Net求解器（SPUS），这是一种紧凑且高效的基座模型（FM），被设计为一种统一的神经算子，用于解决各种偏微分方程（PDEs）。 与现有最先进的PDE FM—主要基于大型复杂Transformer架构，具有高计算和参数开销—SPUS利用了一种轻量级的残差U-Net架构，该架构在该领域作为基座模型架构尚未得到充分探索。 为了在这个极简框架中实现有效的学习，我们采用了一种简单但强大的自回归预训练策略，该策略紧密复制数值求解器的行为以学习底层物理。 SPUS在多样化的流体动力学PDE集上进行预训练，并在6个具有挑战性的未见过的下游PDE上进行评估，这些PDE跨越了各种物理系统。 实验结果表明，SPUS使用基于残差U-Net的架构在这些下游任务上实现了最先进的一般化性能，同时需要显著更少的参数和最少的微调数据，突显了其作为解决多样化PDE系统的高度参数高效FM的潜力。
We introduce Small PDE U-Net Solver (SPUS), a compact and efficient foundation model (FM) designed as a unified neural operator for solving a wide range of partial differential equations (PDEs). Unlike existing state-of-the-art PDE FMs-primarily based on large complex transformer architectures with high computational and parameter overhead-SPUS leverages a lightweight residual U-Net-based architecture that has been largely underexplored as a foundation model architecture in this domain. To enable effective learning in this minimalist framework, we utilize a simple yet powerful auto-regressive pretraining strategy which closely replicates the behavior of numerical solvers to learn the underlying physics. SPUS is pretrained on a diverse set of fluid dynamics PDEs and evaluated across 6 challenging unseen downstream PDEs spanning various physical systems. Experimental results demonstrate that SPUS using residual U-Net based architecture achieves state-of-the-art generalization on these downstream tasks while requiring significantly fewer parameters and minimal fine-tuning data, highlighting its potential as a highly parameter-efficient FM for solving diverse PDE systems.
- [3] arXiv:2510.01396 (交叉列表自 cs.LG) [中文pdf, pdf, html, 其他]
-
标题： 复杂化学系统的自由能计算的神经网络代理模型标题： Neural Network Surrogates for Free Energy Computation of Complex Chemical Systems评论： 6页，4图。本工作已获接受在2025年第29届国际计算机科学与工程会议（ICSEC）上发表，地点为泰国清迈，并将在IEEE Xplore上发表。主题： 机器学习 (cs.LG) ; 人工智能 (cs.AI) ; 化学物理 (physics.chem-ph) ; 计算物理 (physics.comp-ph)
自由能重构方法，如高斯过程回归（GPR），需要集体变量（CVs）的雅可比矩阵，这是一个限制复杂或机器学习CVs使用的问题。 我们引入了一个神经网络代理框架，该框架直接从笛卡尔坐标学习CVs，并使用自动微分提供雅可比矩阵，绕过了解析形式。 在MgCl2离子对系统中，我们的方法对于简单的距离CV和复杂的配位数CV都实现了高精度。 此外， 雅可比误差也遵循近似高斯分布，使其适用于GPR流程。 该框架使基于梯度的自由能方法能够包含复杂和机器学习的CVs，拓宽了生物化学和材料模拟的范围。
Free energy reconstruction methods such as Gaussian Process Regression (GPR) require Jacobians of the collective variables (CVs), a bottleneck that restricts the use of complex or machine-learned CVs. We introduce a neural network surrogate framework that learns CVs directly from Cartesian coordinates and uses automatic differentiation to provide Jacobians, bypassing analytical forms. On an MgCl2 ion-pairing system, our method achieved high accuracy for both a simple distance CV and a complex coordination-number CV. Moreover, Jacobian errors also followed a near-Gaussian distribution, making them suitable for GPR pipelines. This framework enables gradient-based free energy methods to incorporate complex and machine-learned CVs, broadening the scope of biochemistry and materials simulations.
- [4] arXiv:2510.01419 (交叉列表自 cond-mat.mtrl-sci) [中文pdf, pdf, 其他]
-
标题： 双层六方氮化硼中大扭曲铁电性和涡旋位错的多尺度分析标题： Multiscale analysis of large twist ferroelectricity and swirling dislocations in bilayer hexagonal boron nitride主题： 材料科学 (cond-mat.mtrl-sci) ; 计算物理 (physics.comp-ph)
以其原子薄结构和固有的铁电特性，异形双层六方氮化硼（hBN）在下一代非易失性存储应用中引起了广泛关注。 然而，迄今为止的研究几乎全部集中在小的异形变形上，关于铁电性在大异形变形下是否能够持续的问题尚未被探索。 在本工作中，我们利用史密斯标准型双晶学，确定了相对于高对称构型（如AA堆叠和21.786789$\circ$扭转构型）异形变形的双层hBN构型中铁电性的晶体学起源。 随后，我们展示了在靠近AA和$\Sigma 7$堆叠构型的双层hBN中存在面外铁电性。 原子模拟显示，靠近AA的系统在小扭转和小应变下都能支持铁电性，而在后者中，极化翻转由旋涡位错的变形所控制，而不是前者中观察到的直线界面位错。 对于靠近$\Sigma 7$的系统，由于缺乏可靠的原子间势能，我们开发了一个基于密度泛函理论的连续框架——双晶学启发的框架不变多尺度（BFIM）模型，该模型能够捕捉靠近$\Sigma 7$堆叠构型的异形变形中的面外铁电性。 这些大异形变形的双层构型中的界面位错表现出明显较小的伯格斯矢量，与小扭转和小应变双层hBN中的界面位错相比。 BFIM模型再现了原子模拟结果，并为预测大单元胞异质结构中的铁电性提供了一个强大且计算高效的框架，在这种结构中，原子模拟的成本过高。
With its atomically thin structure and intrinsic ferroelectric properties, heterodeformed bilayer hexagonal boron nitride (hBN) has gained prominence in next-generation non-volatile memory applications. However, studies to date have focused almost exclusively on small heterodeformations, leaving the question of whether ferroelectricity can persist under large heterodeformation entirely unexplored. In this work, we establish the crystallographic origin of ferroelectricity in bilayer hBN configurations heterodeformed relative to high-symmetry configurations such as the AA-stacking and the 21.786789 $\circ$ twisted configuration, using Smith normal form bicrystallography. We then demonstrate out-of-plane ferroelectricity in bilayer hBN across configurations vicinal to both the AA and $\Sigma 7$ stacking. Atomistic simulations reveal that AA-vicinal systems support ferroelectricity under both small twist and small strain, with polarization switching in the latter governed by the deformation of swirling dislocations rather than the straight interface dislocations seen in the former. For $\Sigma 7$-vicinal systems, where reliable interatomic potentials are lacking, we develop a density-functional-theory-informed continuum framework--the bicrystallography-informed frame-invariant multiscale (BFIM) model, which captures out-of-plane ferroelectricity in heterodeformed configurations vicinal to the $\Sigma 7$ stacking. Interface dislocations in these large heterodeformed bilayer configurations exhibit markedly smaller Burgers vectors compared to the interface dislocations in small-twist and small-strain bilayer hBN. The BFIM model reproduces atomistic simulation results and provides a powerful, computationally efficient framework for predicting ferroelectricity in large-unit-cell heterostructures where atomistic simulations are prohibitively expensive.
- [5] arXiv:2510.01543 (交叉列表自 quant-ph) [中文pdf, pdf, html, 其他]
-
标题： 变分方法在具有长程竞争相互作用的开放量子系统中的应用标题： Variational approach to open quantum systems with long-range competing interactions主题： 量子物理 (quant-ph) ; 量子气体 (cond-mat.quant-gas) ; 计算物理 (physics.comp-ph)
短程和长程相互作用之间的竞争是自然界许多涌现现象的基础。 尽管在实验控制方面取得了快速进展，但能够准确模拟具有复杂长程相互作用的开放量子多体系统的计算方法仍然很少。 在此，我们通过引入一种高效且可扩展的方法来处理一维和二维的耗散量子格点，结合矩阵乘积算符和时间依赖变分蒙特卡罗方法，解决了这一限制。 我们通过模拟具有竞争性代数衰减相互作用的自旋-$\frac{1}{2}$格点在多达$N=200$个格点上的非平衡动力学和稳态，展示了该算法的多功能性、有效性和独特的方法论优势，揭示了远离平衡时空间调制磁序的出现。 这种方法为深入理解各种实验可实现的具有长程相互作用的量子系统的复杂非平衡特性提供了有前景的前景，包括里德伯原子、超冷偶极分子和被捕获的离子。
Competition between short- and long-range interactions underpins many emergent phenomena in nature. Despite rapid progress in their experimental control, computational methods capable of accurately simulating open quantum many-body systems with complex long-ranged interactions at scale remain scarce. Here, we address this limitation by introducing an efficient and scalable approach to dissipative quantum lattices in one and two dimensions, combining matrix product operators and time-dependent variational Monte Carlo. We showcase the versatility, effectiveness, and unique methodological advantages of our algorithm by simulating the non-equilibrium dynamics and steady states of spin-$\frac{1}{2}$ lattices with competing algebraically-decaying interactions for as many as $N=200$ sites, revealing the emergence of spatially-modulated magnetic order far from equilibrium. This approach offers promising prospects for advancing our understanding of the complex non-equilibrium properties of a diverse variety of experimentally-realizable quantum systems with long-ranged interactions, including Rydberg atoms, ultracold dipolar molecules, and trapped ions.
- [6] arXiv:2510.01875 (交叉列表自 cond-mat.mtrl-sci) [中文pdf, pdf, html, 其他]
-
标题： 增强时变密度泛函理论计算动态响应性质的效率标题： Enhancing the Efficiency of Time-Dependent Density Functional Theory Calculations of Dynamic Response PropertiesZhandos A. Moldabekov, Sebastian Schwalbe, Uwe Hernandez Acosta, Thomas Gawne, Jan Vorberger, Michele Pavanello, Tobias Dornheim主题： 材料科学 (cond-mat.mtrl-sci) ; 化学物理 (physics.chem-ph) ; 计算物理 (physics.comp-ph) ; 等离子体物理 (physics.plasm-ph)
X射线汤姆逊散射（XRTS）是诊断极端条件下材料性质的重要技术，例如高压和强激光加热。 时间依赖密度泛函理论（TDDFT）是建模XRTS谱以及一系列其他动态材料性质最准确的可用从头算方法之一。 然而，强烈的热激发，以及需要考虑温度和密度的变化以及探测器的有限尺寸，显著增加了与常规条件相比的TDDFT模拟的计算成本。 在本工作中，我们提出了一种广泛适用的方法，用于优化和提高TDDFT计算的效率。 我们的方法基于动态结构因子与虚时密度-密度关联函数之间的一一映射，这在费曼的量子多体理论路径积分公式中自然出现。 具体而言，我们将虚时域中的严格收敛测试与基于约束的噪声衰减技术相结合，以在不引入任何显著偏差的情况下提高TDDFT建模的效率。 因此，我们可以报告出高达一个数量级的速度提升，从而可能为建模极端条件下物质的单个XRTS测量节省数百万CPU小时。
X-ray Thomson scattering (XRTS) constitutes an essential technique for diagnosing material properties under extreme conditions, such as high pressures and intense laser heating. Time-dependent density functional theory (TDDFT) is one of the most accurate available ab initio methods for modeling XRTS spectra, as well as a host of other dynamic material properties. However, strong thermal excitations, along with the need to account for variations in temperature and density as well as the finite size of the detector significantly increase the computational cost of TDDFT simulations compared to ambient conditions. In this work, we present a broadly applicable method for optimizing and enhancing the efficiency of TDDFT calculations. Our approach is based on a one-to-one mapping between the dynamic structure factor and the imaginary time density--density correlation function, which naturally emerges in Feynman's path integral formulation of quantum many-body theory. Specifically, we combine rigorous convergence tests in the imaginary time domain with a constraints-based noise attenuation technique to improve the efficiency of TDDFT modeling without the introduction of any significant bias. As a result, we can report a speed-up by up to an order of magnitude, thus potentially saving millions of CPU hours for modeling a single XRTS measurement of matter under extreme conditions.
- [7] arXiv:2510.01918 (交叉列表自 quant-ph) [中文pdf, pdf, html, 其他]
-
标题： 基于混合量子-经典漫步的社区检测图表示学习标题： Hybrid Quantum-Classical Walks for Graph Representation Learning in Community Detection评论： 6页。被2025年IEEE国际量子人工智能会议接收主题： 量子物理 (quant-ph) ; 计算物理 (physics.comp-ph)
图表示学习（GRL）已成为分析跨不同领域复杂网络数据的核心技术，包括生物系统、社交网络和数据分析。 传统GRL方法在捕捉复杂图中的复杂关系时常常遇到困难，特别是那些表现出非平凡结构特性的图，如幂律分布或分层结构。 本文介绍了一种新颖的受量子启发的GRL算法，利用混合量子-经典行走来克服这些限制。 我们的方法结合了量子和经典动力学的优点，使行走者能够同时探索图中的高度局部和远距离连接。 针对网络社区检测的一个案例研究的初步结果表明，这种混合动力学使算法能够有效地适应复杂的图结构，为GRL任务提供了一个强大且多功能的解决方案。
Graph Representation Learning (GRL) has emerged as a cornerstone technique for analysing complex, networked data across diverse domains, including biological systems, social networks, and data analysis. Traditional GRL methods often struggle to capture intricate relationships within complex graphs, particularly those exhibiting non-trivial structural properties such as power-law distributions or hierarchical structures. This paper introduces a novel quantum-inspired algorithm for GRL, utilizing hybrid Quantum-Classical Walks to overcome these limitations. Our approach combines the benefits of both quantum and classical dynamics, allowing the walker to simultaneously explore both highly local and far-reaching connections within the graph. Preliminary results for a case study in network community detection shows that this hybrid dynamic enables the algorithm to adapt effectively to complex graph topologies, offering a robust and versatile solution for GRL tasks.
- [8] arXiv:2510.01977 (交叉列表自 physics.plasm-ph) [中文pdf, pdf, html, 其他]
-
标题： 用机器学习生成的初始条件加速动力学等离子体模拟标题： Accelerating kinetic plasma simulations with machine learning generated initial conditions主题： 等离子体物理 (physics.plasm-ph) ; 计算物理 (physics.comp-ph)
计算机辅助工程多时间尺度等离子体系统，这些系统表现出准稳态解，由于需要大量时间步才能达到收敛，因此具有挑战性。 结合机器学习技术与传统第一性原理模拟和高性能计算，为解决这一挑战提供了许多有趣的途径。 我们考虑通过机器学习生成的初始条件加速粒子等离子体模拟。 该方法通过建模与微电子工业相关的电容耦合等离子体放电进行了演示。 三个模型在设备驱动频率和工作压力的参数空间上的模拟数据上进行训练。 这些模型结合了多层感知器、主成分分析和卷积神经网络的元素，以预测最终的时间平均离子密度和速度分布函数的轮廓。 这些数据驱动的初始条件生成器（ICGs）在使用离线程序测量时，将收敛时间平均加快了17.1倍，或在使用在线程序时加快了4.4倍，其中卷积神经网络表现最佳。 本文还概述了一个连续数据驱动模型改进和模拟加速的工作流程，旨在生成足够的数据以实现完整的器件数字孪生。
Computer aided engineering of multi-time-scale plasma systems which exhibit a quasi-steady state solution are challenging due to the large number of time steps required to reach convergence. Machine learning techniques combined with traditional first-principles simulations and high-performance computing offer many interesting pathways towards resolving this challenge. We consider acceleration of kinetic plasma simulations via machine learning generated initial conditions. The approach is demonstrated through modeling of capacitively coupled plasma discharges relevant to the microelectronics industry. Three models are trained on simulations across a parameter space of device driving frequency and operating pressure. The models incorporate elements of a multi-layer perceptron, principal component analysis, and convolutional neural networks to predict the final time-averaged profiles of ion-density and velocity distribution functions. These data-driven initial condition generators (ICGs) provide a mean speedup of 17.1x in convergence time, when measured using an offline procedure, or a 4.4x speedup with an online procedure, with convolutional neural networks leading to the best performance. The paper also outlines a workflow for continuous data-driven model improvement and simulation speedup, with the aim of generating sufficient data for full device digital twins.
- [9] arXiv:2510.02051 (交叉列表自 quant-ph) [中文pdf, pdf, html, 其他]
-
标题： 改进神经网络在解决量子符号结构中的性能标题： Improving neural network performance for solving quantum sign structure评论： 8页，3图，将发表于《物理评论B》主题： 量子物理 (quant-ph) ; 强关联电子 (cond-mat.str-el) ; 计算物理 (physics.comp-ph)
神经量子态已成为研究非马尔可夫哈密顿量基态的广泛应用方法。 然而，现有方法通常依赖于对符号结构的先验知识，或需要单独预训练的相位网络。 我们引入了一种改进的随机再配置方法，该方法有效利用不同的虚时间步长来演化振幅和相位。 使用较大的时间步长进行相位优化，此方法实现了相位和振幅神经网络的同时高效训练。 我们的方法在海森堡 J_1-J_2 模型上得到了验证。
Neural quantum states have emerged as a widely used approach to the numerical study of the ground states of non-stoquastic Hamiltonians. However, existing approaches often rely on a priori knowledge of the sign structure or require a separately pre-trained phase network. We introduce a modified stochastic reconfiguration method that effectively uses differing imaginary time steps to evolve the amplitude and phase. Using a larger time step for phase optimization, this method enables a simultaneous and efficient training of phase and amplitude neural networks. The efficacy of our method is demonstrated on the Heisenberg J_1-J_2 model.
交叉提交 (展示 8 之 8 条目 )
- [10] arXiv:2503.14020 (替换) [中文pdf, pdf, html, 其他]
-
标题： LAMMPS模拟软件的一套教程标题： A Set of Tutorials for the LAMMPS Simulation Package评论： 有关联文件，请参见 https://github.com/lammpstutorials/lammpstutorials-article期刊参考： 《计算分子科学生活期刊》（LiveCoMS），6(1):3037，2025主题： 计算物理 (physics.comp-ph) ; 统计力学 (cond-mat.stat-mech)
开放源代码分子模拟软件包的可用性使科学家和工程师能够专注于运行和分析模拟，而无需编写、并行化和验证自己的模拟软件。 尽管如此，分子模拟因此对更广泛的受众变得可访问，但这类软件包的“黑箱”性质以及众多选项和功能可能会使正确使用它们变得具有挑战性，尤其是对于MD模拟领域的初学者而言。 LAMMPS 是一种多功能的分子模拟代码，旨在建模跨材料科学和计算化学广泛应用的基于粒子的系统，包括原子尺度、粗粒度、介观尺度、无网格连续体和离散元模型。 LAMMPS 能够在从小型台式计算机到大规模超级计算环境的各种规模的模拟中高效运行。 其灵活性和可扩展性使其成为原子和分子系统以及更广泛领域复杂和大规模模拟的理想选择。 本文介绍了一套教程，旨在使新用户更容易学习 LAMMPS。 前四个教程涵盖了在 LAMMPS 中运行不同复杂度系统的分子模拟基础。 接下来的四个教程介绍了更高级的分子模拟技术，特别是反应力场的使用、巨正则蒙特卡罗方法、增强采样和 REACTER 协议的应用。 此外，我们介绍了 LAMMPS-GUI，一个增强的跨平台图形文本编辑器，专门用于与 LAMMPS 一起使用，并能够在编辑的输入上直接运行 LAMMPS。 LAMMPS-GUI 在教程中作为主要工具用于编辑输入、运行 LAMMPS、提取数据和可视化模拟系统。
The availability of open-source molecular simulation software packages allows scientists and engineers to focus on running and analyzing simulations without having to write, parallelize, and validate their own simulation software. While molecular simulations thus become accessible to a larger audience, the `black box' nature of such software packages and wide array of options and features can make it challenging to use them correctly, particularly for beginners in the topic of MD simulations. LAMMPS is one such versatile molecular simulation code, designed for modeling particle-based systems across a broad range of materials science and computational chemistry applications, including atomistic, coarse-grained, mesoscale, grid-free continuum, and discrete element models. LAMMPS is capable of efficiently running simulations of varying sizes from small desktop computers to large-scale supercomputing environments. Its flexibility and extensibility make it ideal for complex and extensive simulations of atomic and molecular systems, and beyond. This article introduces a suite of tutorials designed to make learning LAMMPS more accessible to new users. The first four tutorials cover the basics of running molecular simulations in LAMMPS with systems of varying complexities. The second four tutorials address more advanced molecular simulation techniques, specifically the use of a reactive force field, grand canonical Monte Carlo, enhanced sampling, and the REACTER protocol. In addition, we introduce LAMMPS-GUI, an enhanced cross-platform graphical text editor specifically designed for use with LAMMPS and able to run LAMMPS directly on the edited input. LAMMPS-GUI is used as the primary tool in the tutorials to edit inputs, run LAMMPS, extract data, and visualize the simulated systems.
- [11] arXiv:2506.17087 (替换) [中文pdf, pdf, html, 其他]
-
标题： 基于PCG的神经求解器用于周期性微观结构的高分辨率均质化标题： PCG-Informed Neural Solvers for High-Resolution Homogenization of Periodic Microstructures主题： 计算物理 (physics.comp-ph)
周期微结构的力学性能在各种工程应用中至关重要。 均质化理论是通过在代表性体积单元上平均复杂微结构的行为来预测这些性能的强大工具。 然而，传统的均质化问题数值求解器计算成本较高，尤其是在高分辨率和复杂拓扑及几何情况下。 现有的基于学习的方法虽然有前景，但在这种情况下往往在准确性和泛化能力上存在困难。 为了解决这些挑战，我们提出了CGINS，这是一种预条件共轭梯度求解器启发的神经网络，用于求解均质化问题。 CGINS利用稀疏和周期性的三维卷积，以确保结构周期性的同时实现高分辨率学习。 它具有多级网络架构，有助于在不同尺度上进行有效学习，并采用最小势能作为无标签的损失函数进行自监督学习。 集成的预条件共轭梯度迭代确保网络提供适合PCG的初始解，从而实现快速收敛和高精度。 此外，CGINS施加全局位移约束以确保物理一致性，解决了之前方法依赖Dirichlet锚点的关键局限性。 在包含多种拓扑和材料配置的大规模数据集上评估，CGINS实现了最先进的准确性（相对误差低于1%），并且优于基于学习的基线方法和GPU加速的数值求解器。 值得注意的是，它在保持物理可靠预测的同时，在分辨率达到$512^3$的情况下，比传统方法快2到10倍。
The mechanical properties of periodic microstructures are pivotal in various engineering applications. Homogenization theory is a powerful tool for predicting these properties by averaging the behavior of complex microstructures over a representative volume element. However, traditional numerical solvers for homogenization problems can be computationally expensive, especially for high-resolution and complicated topology and geometry. Existing learning-based methods, while promising, often struggle with accuracy and generalization in such scenarios. To address these challenges, we present CGINS, a preconditioned-conjugate-gradient-solver-informed neural network for solving homogenization problems. CGINS leverages sparse and periodic 3D convolution to enable high-resolution learning while ensuring structural periodicity. It features a multi-level network architecture that facilitates effective learning across different scales and employs minimum potential energy as label-free loss functions for self-supervised learning. The integrated preconditioned conjugate gradient iterations ensure that the network provides PCG-friendly initial solutions for fast convergence and high accuracy. Additionally, CGINS imposes a global displacement constraint to ensure physical consistency, addressing a key limitation in prior methods that rely on Dirichlet anchors. Evaluated on large-scale datasets with diverse topologies and material configurations, CGINS achieves state-of-the-art accuracy (relative error below 1%) and outperforms both learning-based baselines and GPU-accelerated numerical solvers. Notably, it delivers 2 times to 10 times speedups over traditional methods while maintaining physically reliable predictions at resolutions up to $512^3$.
- [12] arXiv:2209.12462 (替换) [中文pdf, pdf, html, 其他]
-
标题： 高熵合金中固溶强化的精确从头算建模标题： Accurate ab initio modeling of solid solution strengthening in high entropy alloys评论： 20页；15图；发表于《物理评论材料》期刊参考： 物理评论材料 6，103602 (2022)主题： 材料科学 (cond-mat.mtrl-sci) ; 计算物理 (physics.comp-ph)
高熵合金（HEA）代表一类具有前景特性的材料，如高强度和延展性、抗辐射损伤等。 同时，成分的组合多样性以及复杂的结构使得它们很难用传统方法进行研究。 在本工作中，我们提出了一种基于协同势近似内的从头算计算的计算效率高的方法。 为了使该方法具有预测性，我们对状态方程应用了交换-关联修正，并考虑了磁态和平衡体积的热效应。 该方法与现有固溶体整体性质的实验数据有良好的一致性。 作为具体案例，该工作流程应用于一系列铁族高熵合金，以在无需参数的模型基础上，研究其固溶强化。 结果揭示了合金组分之间的复杂相互作用，我们通过局部键合的简单模型对其进行分析。 由于计算效率高，该方法可以作为高熵合金最优设计的自适应学习工作流的基础。
High entropy alloys (HEA) represent a class of materials with promising properties, such as high strength and ductility, radiation damage tolerance, etc. At the same time, a combinatorially large variety of compositions and a complex structure render them quite hard to study using conventional methods. In this work, we present a computationally efficient methodology based on ab initio calculations within the coherent potential approximation. To make the methodology predictive, we apply an exchange-correlation correction to the equation of state and take into account thermal effects on the magnetic state and the equilibrium volume. The approach shows good agreement with available experimental data on bulk properties of solid solutions. As a particular case, the workflow is applied to a series of iron-group HEA to investigate their solid solution strengthening within a parameter-free model based on the effective medium representation of an alloy. The results reveal intricate interactions between alloy components, which we analyze by means of a simple model of local bonding. Thanks to its computational efficiency, the methodology can be used as a basis for an adaptive learning workflow for optimal design of HEA.
- [13] arXiv:2503.01446 (替换) [中文pdf, pdf, html, 其他]
-
标题： 双折叠随机分支环状聚合物的涌现动力学标题： The emergent dynamics of double-folded randomly branching ring polymers主题： 软凝聚态物理 (cond-mat.soft) ; 计算物理 (physics.comp-ph)
随机分支双折叠环状聚合物的统计特性与RNA的二级结构、螺旋DNA的大规模分支（因此是细菌染色体）、单环状聚合物通过障碍物阵列时的构象，以及真核染色体和皱缩的非连接环状聚合物熔体的构象统计有关。 双折叠环状聚合物根据双折叠所依赖的随机树状图是淬火的还是退火的，以及这些树在空间嵌入中是否可以自由进行布朗运动，而属于不同的动态普适类。 从局部来看，可以区分（i）围绕固定树的类似repton的质量传输，（ii）侧枝的自发产生和删除，以及（iii）树节点的位移，在这种情况下互补的环段在空间中一起扩散。 在此，我们采用适合的弹性晶格聚合物模型的动态蒙特卡洛模拟，研究不同组合的上述局部模式在三种不同系统中的介观动力学：理想不相互作用的环、自避环以及熔融状态下的环。 我们观察到环状链的预期标度区域，即在障碍物阵列中双折叠环的动力学，以及Rouse-like树动力学作为极限情况。 特别值得注意的是，对于具有$\nu=1/3$的皱缩环，$g_1\sim t^{0.4}$的单体均方位移类似于细菌染色体中观察到的亚扩散区域。 在我们的分析中，我们关注不同局部动力学模式对出现的动力学的贡献在多大程度上是简单的叠加。 值得注意的是，当所有三种类型的局部运动都存在时，我们揭示了相互作用环的动力学出现了非平凡的加速。
The statistics of randomly branching double-folded ring polymers are relevant to the secondary structure of RNA, the large-scale branching of plectonemic DNA (and thus bacterial chromosomes), the conformations of single-ring polymers migrating through an array of obstacles, as well as to the conformational statistics of eukaryotic chromosomes and melts of crumpled, non-concatenated ring polymers. Double-folded rings fall into different dynamical universality classes depending on whether the random tree-like graphs underlying the double-folding are quenched or annealed, and whether the trees can undergo unhindered Brownian motion in their spatial embedding. Locally, one can distinguish (i) repton-like mass transport around a fixed tree, (ii) the spontaneous creation and deletion of side branches, and (iii) displacements of tree node, where complementary ring segments diffuse together in space. Here we employ dynamic Monte Carlo simulations of a suitable elastic lattice polymer model of double-folded, randomly branching ring polymers to explore the mesoscopic dynamics that emerge from different combinations of the above local modes in three different systems: ideal non-interacting rings, self-avoiding rings, and rings in the melt state. We observe the expected scaling regimes for ring reptation, the dynamics of double-folded rings in an array of obstacles, and Rouse-like tree dynamics as limiting cases. Of particular interest, the monomer mean-square displacements of $g_1\sim t^{0.4}$ observed for crumpled rings with $\nu=1/3$ are similar to the subdiffusive regime observed in bacterial chromosomes. In our analysis, we focus on the question to which extent contributions of different local dynamical modes to the emergent dynamics are simply additive. Notably, we reveal a non-trivial acceleration of the dynamics of interacting rings, when all three types of local motion are present.
- [14] arXiv:2504.14187 (替换) [中文pdf, pdf, html, 其他]
-
标题： 从实验激子能量中检索单层过渡金属二硫属化物的基本材料参数：一种解析方法标题： Retrieval of fundamental material parameters of monolayer transition metal dichalcogenides from experimental exciton energies: An analytical approach评论： 11页，5图，4表。将发表于《物理评论B》主题： 中尺度与纳米尺度物理 (cond-mat.mes-hall) ; 材料科学 (cond-mat.mtrl-sci) ; 强关联电子 (cond-mat.str-el) ; 计算物理 (physics.comp-ph)
我们提出了一种直接且高度准确的方法，从单层过渡金属二硫属化物（TMDCs）的实验磁激子能量中提取材料参数，如屏蔽长度、带隙能量、激子约化质量和周围介质的介电常数。 我们的方法依赖于解析公式，使我们能够直接从实验的$s$状态激子能量$E_{1s}$、$E_{2s}$和$E_{3s}$计算出屏蔽长度$r_0$和带隙能量$E_g$。 我们还建立了周围介电常数$\kappa$与激子约化质量$\mu$之间的关系。 这种关系简化了TMDC单层中磁激子的薛定谔方程，将其转化为仅依赖于单一材料参数$\mu$的一参数方程。 此外，我们开发了一个高精度的解析公式，用于计算作为激子约化质量函数的磁激子能量：$E(B,\mu)$。 然后，该公式的逆可以用来从磁激子能量的实验数据中计算激子约化质量。 通过应用此方法，我们从单层TMDC的磁激子能量中提取关键材料参数，$E_g$，$r_0$，$\mu$，和$\kappa$，包括由六方氮化硼（hBN）薄层封装的WSe$_2$，WS$_2$，MoSe$_2$，和MoS$_2$，在各种当前实验中。 我们检索到的材料特性补充并修正了现有的实验和理论数据。 此外，我们开发了一种分析方法，用于以高精度计算抗磁系数和激子半径，与数值计算相比。 基于此方法，我们提供了使用提取的材料参数计算得到的抗磁系数和激子半径。
We propose a straightforward and highly accurate method for extracting material parameters such as screening length, bandgap energy, exciton reduced mass, and the dielectric constant of the surrounding medium from experimental magnetoexciton energies available for monolayer transition metal dichalcogenides (TMDCs). Our approach relies on analytical formulations that allow us to calculate the screening length $r_0$ and bandgap energy $E_g$ directly from the experimental $s$-state exciton energies $E_{1s}$, $E_{2s}$, and $E_{3s}$. We also establish a relationship between the surrounding dielectric constant $\kappa$ and the exciton reduced mass $\mu$. This relationship simplifies the Schr{\"o}dinger equation for a magnetoexciton in a TMDC monolayer, transforming it into a one-parameter equation that depends solely on the single material parameter $\mu$. Furthermore, we develop an analytical formula with high accuracy for magnetoexciton energies as a function of the exciton reduced mass: $E(B,\mu)$. Then, the inverse of this formula allows us to calculate the exciton reduced mass from experimental data on magnetoexciton energies. By applying this method, we extract key material parameters, $E_g$, $r_0$, $\mu$, and $\kappa$, from the magnetoexciton energies of monolayer TMDCs, including WSe$_2$, WS$_2$, MoSe$_2$, and MoS$_2$, encapsulated by hexagonal boron nitride (hBN) slabs in various current experiments. The material properties we retrieve complement and correct existing experimental and theoretical data. Additionally, we develop an analytical method for calculating diamagnetic coefficients and exciton radii with high accuracy compared to numerical calculations. Based on this method, we provide diamagnetic coefficients and exciton radii computed using the extracted material parameters.
- [15] arXiv:2506.05292 (替换) [中文pdf, pdf, html, 其他]
-
标题： 超越经验的学习：利用储备计算推广到未见过的状态空间标题： Learning Beyond Experience: Generalizing to Unseen State Space with Reservoir Computing评论： 18页，12图。更新以包含RC在未见过的分离和不对称吸引域以及未见过的混沌吸引子上的泛化结果主题： 机器学习 (cs.LG) ; 动力系统 (math.DS) ; 混沌动力学 (nlin.CD) ; 计算物理 (physics.comp-ph)
机器学习技术为仅从观测数据建模动态系统提供了一种有效的方法。 然而，如果没有显式的结构先验——关于底层动态的内置假设——这些技术通常难以推广到训练数据中表示不佳的动态方面。 在这里，我们证明了储备计算——一种简单、高效且多功能的机器学习框架，常用于动态系统的数据驱动建模——可以在没有显式结构先验的情况下推广到状态空间的未探索区域。 首先，我们描述了一种多轨迹训练方案用于储备计算机，该方案支持在一组不相交的时间序列上进行训练，从而有效利用可用的训练数据。 然后，将这种训练方案应用于多稳态动态系统，我们展示了在单一吸引子轨迹上训练的RC可以通过捕捉完全未观察到的吸引子中的系统行为来实现域外泛化。
Machine learning techniques offer an effective approach to modeling dynamical systems solely from observed data. However, without explicit structural priors -- built-in assumptions about the underlying dynamics -- these techniques typically struggle to generalize to aspects of the dynamics that are poorly represented in the training data. Here, we demonstrate that reservoir computing -- a simple, efficient, and versatile machine learning framework often used for data-driven modeling of dynamical systems -- can generalize to unexplored regions of state space without explicit structural priors. First, we describe a multiple-trajectory training scheme for reservoir computers that supports training across a collection of disjoint time series, enabling effective use of available training data. Then, applying this training scheme to multistable dynamical systems, we show that RCs trained on trajectories from a single basin of attraction can achieve out-of-domain generalization by capturing system behavior in entirely unobserved basins.
- [16] arXiv:2507.11276 (替换) [中文pdf, pdf, html, 其他]
-
标题： 通过时间尺度纠缠诊断相变标题： Diagnosing phase transitions through time scale entanglementStefan Rohshap, Hirone Ishida, Frederic Bippus, Anna Kauch, Karsten Held, Hiroshi Shinaoka, Markus Wallerberger主题： 强关联电子 (cond-mat.str-el) ; 统计力学 (cond-mat.stat-mech) ; 计算物理 (physics.comp-ph)
波函数的空间纠缠已经发展成为一个令人着迷且非常活跃的研究领域。 在这里，我们发现了一种完全不同的纠缠形式，即不同时间尺度之间的纠缠。 这可以通过量子张量列车诊断（QTTD）实现，在此过程中，对于$n$-粒子关联函数的键维数允许诊断时间纠缠。 作为例子，我们研究了Hubbard二聚体、带有和不带有最近邻跃迁的四站点Hubbard环以及单杂质Anderson模型的时间尺度纠缠。 除了介绍QTTD方法外，我们的主要发现是，在相变和交叉处，时间尺度纠缠通常是最大的。 这与所研究的关联函数无关。 因此，QTTD是一种用于检测量子相变、有限系统中的基态交叉和热交叉的通用工具。
Spatial entanglement of wave functions has matured into an enthralling and very active research area. Here, we unearth a completely different kind of entanglement, the entanglement between different time scales. This is feasible through quantics tensor train diagnostics (QTTD), wherein the bond dimension for an $n$-particle correlation function allows diagnosing the temporal entanglement. As examples, we study time-scale entanglement of the Hubbard dimer, the four-site Hubbard ring with and without next-nearest neighbor hopping and the single-impurity Anderson model. Besides introducing the QTTD method, our major finding is that the time-scale entanglement is generically maximal at phase transitions and crossovers. This is independent of the correlation function studied. Thus, QTTD is a universal tool for detecting quantum phase transitions, ground state crossings in finite systems, and thermal crossovers.
- [17] arXiv:2510.00116 (替换) [中文pdf, pdf, html, 其他]
-
标题： 追逐轨道，而非时间：一种用于长期偏心引力波代理模型的可扩展范式标题： Chase Orbits, not Time: A Scalable Paradigm for Long-Duration Eccentric Gravitational-Wave SurrogatesAkash Maurya, Prayush Kumar, Scott E. Field, Chandra Kant Mishra, Peter James Nee, Kaushik Paul, Harald P. Pfeiffer, Adhrit Ravichandran, Vijay Varma评论： 12页，7图，1表；更新的参考文献主题： 广义相对论与量子宇宙学 (gr-qc) ; 高能天体物理现象 (astro-ph.HE) ; 计算物理 (physics.comp-ph)
非圆形双黑洞波形的代理建模仍然具有挑战性。 由于轨道偏心率时间尺度变化导致这些波形的形态复杂，使得构建准确且高效的代理模型变得困难，尤其是对于足够长的波形，以覆盖当前地面引力波探测器的敏感频段。 我们提出了一种新颖且可扩展的代理构建技术，使长期持续的非圆形双黑洞波形的代理建模既可行又高效。 该技术旨在通过将中间偏心代理数据片段建模为一个称为平均近点角的角轨道元素，而不是时间，从而简化其谐波内容。 我们表明，这种新参数化方法相比使用时间的当代参数化方法，能减少一个数量级的代理基函数。 我们表明，当用瞬时波形偏心率和平均近点角表示时，代理数据片段在参数空间中的变化变得更加规律，大大简化了它们的参数空间拟合。 本文提出的这些方法使得为当前以及未来第三代引力波探测器构建长期持续的偏心代理模型成为可能。
Surrogate modeling of eccentric binary black hole waveforms has remained challenging. The complicated morphology of these waveforms due to the eccentric orbital timescale variations makes it difficult to construct accurate and efficient surrogate models, especially for waveforms long enough to cover the sensitivity band of the current ground-based gravitational wave detectors. We present a novel and scalable surrogate building technique which makes surrogate modeling of long-duration eccentric binary black hole waveforms both feasible and highly efficient. The technique aims to simplify the harmonic content of the intermediate eccentric surrogate data pieces by modeling them in terms of an angular orbital element called the mean anomaly, instead of time. We show that this novel parameterization yields an order of magnitude fewer surrogate basis functions than using the contemporary parameterization in terms of time. We show that variations in surrogate data-pieces across parameter space become much more regular when expressed in terms of the instantaneous waveform eccentricity and mean anomaly, greatly easing their parameter-space fitting. The methods presented in this work make it feasible to build long-duration eccentric surrogates for the current as well as future third-generation gravitational wave detectors.