生成时间: 2026年07月17日 09:33:41
重点关注论文数量: 3
了解领域论文数量: 5
不相关论文数量: 20
重点关注论文(完整分析)
1. 基于区域分解与非侵入式神经模型降阶的多尺度混合维仿真
Multiscale Mixed-Dimensional Simulation via Domain Decomposition and Non-Intrusive Neural Model Order Reduction
作者: Nunzio Dimola, Piermario Vitullo, Paolo Zunino
类别: math.NA
发布日期: 2026-07-16
arXiv ID: 2607.15171v1
摘要翻译: 科学与工程中涌现的许多计算模型具有多尺度结构,使得全局问题的组装或直接求解在计算上不可行。区域分解(DD)方法通过将全局问题替换为一系列耦合的局部问题,并通过迭代求解重构全局响应,从而克服了这一局限。本文在区域分解降阶模型(DD-ROM)的框架下提出了一种方法,其基础是观察到DD不仅自然地局部化了求解算子,也局部化了其几何和参数依赖性。其核心思想是,DD将一个全局难以处理的解映射转化为一族局部可表示算子,这些算子可在与公共参考构型对应后,从可负担的局部数据中学习得到——我们通过局部可表示性的概念形式化了这一思路。然后,训练非侵入式神经代理模型来逼近细尺度的局部运算,并将其嵌入迭代求解器中。训练算法基于一种级联策略,旨在匹配部署后的代理迭代所遇到的分布。我们将得到的DD方法解释为扰动不动点迭代,并证明全局误差受代理逼近误差的控制。该框架针对混合维椭圆问题进行了实例化,这类问题将三维体区域与嵌入的一维夹杂耦合,采用双层非重叠Robin-Robin方法。数值实验表明,所得DD-ROM稳定,对未见微观几何形状能实现精确逼近,并具有关于子域数量的良好可扩展性,能缩放至大规模全局问题,同时在在线阶段避免了细尺度算子组装和局部高保真求解器。
Comment: 43 pages, 7 figures
链接: http://arxiv.org/abs/2607.15171v1
1. 研究对象和背景
许多科学与工程问题中的计算模型呈现多尺度结构,即宏观域中包含局部精细尺度特征,例如裂隙多孔介质 [1,2]、植物根系 [3,4]、血管化生物组织 [5,6,7,8] 以及复合材料 [9,10,11] 等。这些系统的数值模拟通常导致极大规模的系统,反复求解代价极其高昂。在很多情况下,无论是全局解析的精细模拟还是全局精细算子的组装,在计算上都不可承受。
本文聚焦于一类由参数化偏微分方程控制的多尺度问题,其系数与源项在细观尺度上呈现异质性,导致解中同时存在多个特征长度。设全局域(宏观域)为 $\Omega\subset\mathbb{R}^d$,特征尺度 $L$,其内部包含细观结构 $\Lambda$,尺度 $\ell\ll L$。问题的参数 $\mu\in\mathcal{P}$ 编码了 $\Lambda$ 的变异性。考虑稳态椭圆型问题:求 $u_\mu\in V(\Omega)$ 满足 $p_\mu(u_\mu)=0$,其解映射为 $s:\mathcal{P}\to V(\Omega)$,$s(\mu)=u_\mu$。直接离散该问题需要在整个 $\Omega$ 上解析细观结构,导致离散算子的组装和反复求解都极其昂贵。
已有多种方法应对这一挑战,包括均匀化 [12]、多尺度有限元 [13]、异质多尺度方法 [14]、局部正交分解 [15] 和数值升尺度 [11] 等,它们通过导出有效的粗尺度模型来保留细观结构的影响。基于投影的模型降阶 [16,17] 利用解流形的低维结构加速重复模拟。近年来,数据驱动和算子学习方法展示了直接从数据逼近解算子的能力 [18,19,20,21]。这些方法尽管差异显著,但通常依赖于特定问题的多尺度构造,或者依赖于离线阶段生成的高保真全局数据集。对于极大规模的多尺度系统,这种离线成本本身就可能成为主要瓶颈。
区域分解 (Domain Decomposition, DD) 方法提供了一种自然的机制,将降阶任务从全局系统转移到一系列局部问题上 [22,23]。这一思想在局部化和基于组件的模型降阶中已得到广泛认可。经典方法为子域或可复用原型组件构造降阶空间,并通过拉格朗日乘子、静态凝聚、降阶界面、间断格式、优化原理或 Schwarz 迭代等方式进行耦合。近期的综述 [24] 将这类通常称为 DD-ROM 的方法分为侵入式投影类和非侵入式数据驱动类,并强调了空间局部化、局部参数化、局部基构造和界面耦合的核心作用。例如文献 [25,26,27] 展示了区域分解与局部模型降阶相结合是一个活跃的研究方向。更一般地,科学机器学习的兴起正促进区域分解技术与数据驱动建模之间日益紧密的交互,产生了融合算法结构与习得局部表示的混合降阶框架 [28]。
在上述背景下,本文的贡献在于针对多尺度混合维问题,从算子层面实现了一种 DD-ROM。我们将二维非重叠 Robin–Robin 迭代中涉及的细尺度操作进行分解,并通过引入三个局部替代模型进行近似,即传输映射 $T_{\rm rom}$、局部解映射 $S_{\rm rom}$ 和粗残差贡献 $C_{\rm rom}$。它们的联合使用替代了在线区域分解迭代所需的所有细尺度算子计算。单纯对局部子问题进行替代仍需要在在线阶段进行界面传输量和粗残差贡献的高保真计算;通过学习这些额外的局部算法映射,本文方法消除了对细尺度局部算子的剩余依赖。同时,保留显式粗校正提供了一种全局通信机制,其重要性随子域数目的增加而愈发重要。由此得到的方法既非侵入式又可扩展。我们将这种双层算子‑替代构造称为多尺度 DD‑ROM 框架,简记为 MS‑DD‑ROM。
本文还将上述框架实例化到一个具体的混合维问题上:三维连续介质与嵌入的一维网络相耦合的椭圆问题。三维问题通过双层非重叠 Robin–Robin 方法求解,而一维结构进入局部参数化,并驱动基于微结构的替代模型。此设置虽然受血管化组织中的氧气传输与灌注问题启发 [6],但方法论在抽象层面表述,可广泛适用于具有可复用局部分解的参数化多尺度 PDE。
论文关注的核心空白在于:如何在不依赖全局精细模拟的前提下,构建可处理参数化多尺度混合维问题的、有理论保证的非侵入式 DD ‑ROM,并确保替代模型在 DD 迭代中的稳定性与可扩展性。
2. 主要定理或主要结果
本文的主要结果可概括为以下定理、假设与数值发现:
假设 1 (局部可表示性) 设存在参考域 $\Omega^\star=(0,1)^d$,局部化参数空间 $\widehat{\mathcal{P}}$,以及参考局部问题族 $\{p_{\widehat{\mu}}^\star\}$ 及其解算子 $s^\star:\widehat{\mathcal{P}}\to V(\Omega^\star)$,使得对每个子域 $i$、每次 DD 迭代 $k$ 以及每个全局参数 $\mu$,存在 $\widehat{\mu}_i^{(k)}\in\widehat{\mathcal{P}}$ 满足
$$ \mathcal{T}_i s_i\big(\widehat{\mu}_i^{(k)}\big) = s^\star\big(\widehat{\mu}_i^{(k)}\big) \quad\text{或等价地}\quad s_i\big(\widehat{\mu}_i^{(k)}\big) = \mathcal{T}_i^{-1} s^\star\big(\widehat{\mu}_i^{(k)}\big), $$其中 $\mathcal{T}_i$ 为通过双射 $\Phi_i:\Omega^\star\to\Omega_i$ 定义的拉回映射。该假设保证了所有局部问题可统一为单一参考参数化问题。
假设 2 (均匀逼近性) 参考替代模型 $\tilde{s}^\star$ 满足
$$ \sup_{\widehat{\mu}\in\widehat{\mathcal{P}}} \| e^\star(\widehat{\mu})\|_{V(\Omega^\star)} \le \varepsilon, $$其中 $e^\star := s^\star - \tilde{s}^\star$ 为局部替代误差。
假设 3 (DD 迭代的稳定性) 存在常数 $C_{\rm DD}>0$ 使得扰动 $\delta^{(k)} := \widetilde{DD}_\mu(\tilde{u}^{(k)}) - DD_\mu(\tilde{u}^{(k)})$ 满足
$$ \big\| \delta^{(k)} \big\|_\star \le C_{\rm DD} \max_{i\in\mathcal{I}} \big\| e^\star(\widehat{\mu}_i^{(k)}) \big\|_{V(\Omega^\star)} . $$该假设刻画了局部算子扰动通过界面交换与全局装配操作的传播方式,其成立依赖于 DD 算子的 Lipschitz 连续性。
在上述假设下,主要理论结论为:
命题 1 (替代扰动下的稳定性) 设 $DD_\mu$ 在范数 $\|\cdot\|_\star$ 下为压缩映射,压缩因子 $q<1$,且扰动满足假设 3。则替代‑扰动的 DD 迭代满足
$$ \limsup_{k\to\infty} \big\| u_\mu - \tilde{u}^{(k)} \big\|_\star \le \frac{C_{\rm DD}}{1-q}\,\varepsilon . $$该不等式将全局误差直接与局部替代逼近误差 $\varepsilon$ 联系起来,放大因子由压缩因子 $q$ 和稳定性常数 $C_{\rm DD}$ 决定。
级联训练策略 为确保训练数据分布与替代模型部署时 DD 迭代所遇到的状态分布一致,本文设计了一种级联训练算法:依次引入各替代模型,每个后续模型都在已被近似模型生成的迭代轨迹上进行训练。
实例化结果 对于混合维问题 (2),其中三维域采用双层非重叠 Robin–Robin DD,传输参数 $\rho>0$,局部子问题如 (7) 所示。数值实验表明:MS‑DD‑ROM 稳定,对于未见过的微观几何能够达到精确逼近,并且关于子域数目具有良好的可扩展性,能够处理大规模全局问题而在线阶段完全避免细尺度算子的组装和局部高保真求解。
3. 研究方法、关键技术和核心工具
本文的证明框架与技术工具围绕以下核心要素展开:
1. 局部可表示性与参考域构造
这是整个框架的基石。区域分解自然地将全局解算子的参数依赖性局部化:每个局部子问题不仅依赖于物理参数(几何、系数、源项),还依赖于 DD 迭代过程中产生的界面数据(算法状态)。因此,局部参数空间被扩充为 $\widehat{\mathcal{P}}$,其元素 $\widehat{\mu}_i^{(k)} = (\lambda_i, \nu_i^{(k)})$ 包含物理参数 $\lambda_i$(如细观结构)和算法参数 $\nu_i^{(k)}$(如 Robin 传输数据)。进一步,通过假设子域 $\Omega_i$ 均可通过双射 $\Phi_i$ 映到统一的参考域 $\Omega^\star$,可将所有局部解算子拉回到 $\Omega^\star$ 上,从而只需学习一个参考替代模型 $\tilde{s}^\star:\widehat{\mathcal{P}}\to V(\Omega^\star)$。这一“局部可表示性”假设极大地降低了学习任务的复杂度与数据需求。
2. 级联训练策略
直接用理想条件下(使用精确局部算子)的 DD 迭代数据训练替代模型,会在部署时遇到分布偏移:替代模型自身的误差会改变后续迭代的输入分布,导致性能退化。本文的解决方案是级联训练:按照替代模型被引入 DD 迭代的先后顺序,依次训练。例如,先训练一个替代某个局部算子的模型,然后用包含该替代模型的迭代生成数据训练下一个替代模型。这样,每个模型的训练分布与它在最终 MS‑DD‑ROM 迭代中所面对的分布一致,从而在理论上支持全局逼近精度的要求(假设 2 需覆盖 DD 轨迹上的所有可能状态)。
3. 受扰不动点迭代与稳定性分析
将替代局部算子嵌入 DD 迭代后,全局迭代算子变为 $\widetilde{DD}_\mu$,它可看作精确迭代 $DD_\mu$ 加上一个依赖于替代误差的扰动项 $\delta^{(k)}$。在 $DD_\mu$ 为压缩映射(因子 $q<1$)的假设下,命题 1 通过标准的扰动收缩半群分析证明了全局误差的渐近界。该分析的关键在于分解 $\tilde{u}^{(k+1)} - u_\mu = DD_\mu(\tilde{u}^{(k)}) - DD_\mu(u_\mu) + \delta^{(k)}$,然后利用压缩性进行递推。扰动的大小由假设 3 引出的局部误差界 $\varepsilon$ 和稳定性常数 $C_{\rm DD}$ 控制。这一分析不仅提供了误差估计,还指导了训练要求:即需要确保替代模型在包含 DD 迭代状态的局部参数空间上有均匀精度。
4. 混合维问题的具体算子分解
在实例化到混合维椭圆问题 (2) 时,三维域上的双层 Robin–Robin DD 涉及三类核心局部操作:(a)Robin 传输算子 $T_{\rm rom}$,它将邻域界面状态转化为 Robin 右端项;(b)局部解算子 $S_{\rm rom}$,即求解带 Robin 条件的本地子问题;(c)粗残差贡献 $C_{\rm rom}$,它计算子域对粗网格校正方程的贡献。传统降阶若仅替代 $S_{\rm rom}$,仍需在线计算 $T_{\rm rom}$ 和 $C_{\rm rom}$,而它们同样涉及细尺度装配。本文的非侵入式 MS‑DD‑ROM 将三者全部用神经网络替代,从而在在线阶段完全摆脱了细尺度局部矩阵的装配与求解。神经网络训练数据仅来自本地精细模拟,无需全局模拟。
5. 有限元离散与代数结构
文中给出了基于有限元的弱形式 (10) 和迭代格式 (11)。由于采用一致网格且子域仿射等价,局部代数结构具有相似性,这为神经替代模型的输入输出设计提供了方便。传输数据 $f_{h,j}$ 表达为对偶表示,并通过 Robin 条件实现界面耦合。整个代数框架保持了 DD 方法的模块化优点,使神经替代模型可以即插即用。
4. 与之前工作的比较
本文与既有 DD‑ROM 及多尺度方法的研究对比可沿以下几个维度展开:
1. 与非侵入式 DD‑ROM 的对比
现有的非侵入式 DD‑ROM 方法通常需要全局精细解或全局精细模拟来提供训练数据(如文献 [24] 中所述的数据驱动类方法)。本文的关键不同在于:训练数据完全源自本地精细模拟,不需要任何全局精细解。其基础是局部可表示性假设,该假设将所有子域映射到参考域,使得只需学习一个参考替代模型,训练数据可由对参考域上参数化局部问题的精细求解生成。这大幅降低了离线成本,并使得即使全局问题规模大到无法进行精细模拟时,MS‑DD‑ROM 仍能构建。
2. 与侵入式投影 DD‑ROM 的对比
侵入式方法通常需要在本地构建降阶基(如 POD 基),并要求在线阶段能够访问精细离散算子以进行投影(见 [25,26,27])。本文方法完全非侵入,使用神经网络学习从局部参数到局部解(或所需量)的映射,无需任何内部代码修改。此外,通过同时替代传输映射和残差贡献,进一步消除了在线阶段对精细算子的依赖,而传统侵入式 DD‑ROM 往往仍需在线计算界面通量或残差。
3. 与混合维问题数值方法的对比
混合维 3D‑1D 耦合问题 (2) 已有较深入的研究,如降维建模 [30,31,32]。这些工作关注耦合格式本身(如交替求解)以及离散化方案,但并未将求解器与局部神经替代模型结合。本文在此类耦合问题之上引入 DD 与神经替代模型,是针对参数化情形求解效率的显著推进。特别是,对于拓扑变化或细观几何变化的参数化问题,以往需要重新计算全局耦合解,而本文只需对参考局部问题训练一次替代模型,即可在线响应该类变化。
4. 级联训练策略与分布偏移处理
机器学习辅助的迭代求解器常面临训练‑部署分布不一致的问题。在替代模型进入循环后,其自身误差会改变后续迭代的输入分布,导致性能下降。本文明确认识到这一点,并设计了级联训练策略,使每个替代模型的训练状态来自先前已被近似的迭代。这与普通离线训练(使用理想迭代数据)形成对比,普通离线训练可能在替代误差较大时失效。级联策略从原理上缓解了分布偏移,这是对 DD‑ROM 训练方法的一个重要补充。
5. 理论保证与适用范围
本文提供了基于压缩不动点迭代的误差传播分析,将全局误差显式地与局部替代误差 $\varepsilon$、收缩因子 $q$ 和稳定性常数 $C_{\rm DD}$ 联系起来。这与许多纯数据驱动方法的黑箱性质不同,为替代模型精度需求提供了指导。然而,这一分析依赖于几点假设:$DD_\mu$ 为压缩映射(取决于传输参数 $\rho$ 的适当选择)、局部可表示性成立(要求子域几何足够规则并能仿射映射到参考域)、以及局部替代模型的均匀逼近性能在 $\widehat{\mathcal{P}}$ 上实现。这些条件限制了当前框架直接应用于任意复杂几何或高度异质非仿射子域情形。文中实例采用张量积域和 Cartesian 分区,正是为了满足这些假设。与重叠型 Schwarz 方法相比,本文选择了非重叠 Robin 方法,其优点是统一的界面处理,适合复用替代模型,但代价是收敛性对 $\rho$ 敏感。
6. 与纯机器学习算子方法的对比
近年来,如 DeepONet、Fourier Neural Operator 等全局算子学习方法可以直接学习解算子,但它们通常需要大量全局解作为训练数据,且难以保证对未知参数的外推稳定性。本文的 MS‑DD‑ROM 通过保留 DD 的迭代结构和粗校正机制,将机器学习任务限制在具有物理意义的局部算子范围内,显著降低了对数据量的需求,并提升了长程相干性和可扩展性。局部算子的输入维数远低于全局算子,这也使得训练更易实现。
7. 局限与未来工作
作者明确指出,当前工作中仅对三维域进行了 DD 降阶,一维问题仍用全局求解;重叠型 Schwarz 变体未探索;$\rho$ 参数的优化未系统研究;理论分析依赖的 Lipschitz 稳定性假设 3 未给出严格证明,仅基于算子复合论证。此外,数值实验限于二维图示(图 1 为示意图),对于真正三维大规模并行计算的实际效益尚待全面评估。这些均为后续研究留出了方向。
2. 高反差非均质多孔介质中两相流模拟的自适应物理保持多尺度方法
An Adaptive and Physics-Preserving Multiscale Method for Two-Phase Flow Simulations in High-Contrast Heterogeneous Porous Media
作者: Junhao Huang, Eric Chung, Wing Tat Leung
类别: math.NA
发布日期: 2026-07-16
arXiv ID: 2607.15063v1
摘要翻译: 在本文中,我们针对高反差多孔介质中的不可压缩且不混溶两相流,提出了一种自适应物理保持多尺度方法。该方法将物理保持的隐式压力显式饱和度格式(P-IMPES)与混合约束能量极小化的广义多尺度有限元方法相结合。核心算法组件是针对饱和度依赖系数的自适应更新策略。由于有效渗透率 (κ_n=λ_t(S_w^n)K) 通过总流度依赖于变化的饱和度,我们引入了一个自适应更新算法,该算法监测流度加权系数的变化,并仅在超过预设容差时才重新生成多尺度空间。进一步采用局部后处理步骤来恢复细网格质量守恒。理论分析是本文的核心部分。我们证明了两个相位的局部守恒性、相位表达式的无偏性,以及在适当CFL条件下的有界保持性。对于对流主导的情形,我们建立了速度和饱和度的误差估计,这些估计清晰地识别出由自适应容差、粗网格尺寸、谱逼近和前沿层误差所带来的贡献。在不同高反差渗透率场上的数值实验证实了方法的物理特性,并表明较小的自适应容差能改善饱和度逼近,同时避免不必要的多尺度空间更新。
Comment: 33 pages, 9 figures, and 2 tables
链接: http://arxiv.org/abs/2607.15063v1
1. 研究对象和背景
本文研究不可压缩非混相两相流在多孔介质中的数值模拟,控制方程由质量守恒、达西定律、饱和度约束和毛管压力关系组成:
$$ \phi \frac{\partial S_\alpha}{\partial t} + \nabla\cdot\boldsymbol{u}_\alpha = F_\alpha, \quad \alpha = w,n, \qquad \boldsymbol{u}_\alpha = -\frac{k_{r\alpha}}{\mu_\alpha}\boldsymbol{K}\nabla p_\alpha, $$$$ S_n + S_w = 1, \qquad p_c(S_w) = p_n - p_w, $$其中 $\phi$ 为孔隙度,$S_\alpha$ 为相饱和度,$\boldsymbol{u}_\alpha$ 为相达西速度,$p_\alpha$ 为相压力,$\boldsymbol{K}$ 为绝对渗透率张量,$k_{r\alpha}$、$\mu_\alpha$、$F_\alpha$ 分别为相对渗透率、粘度和源汇项。总速度 $\boldsymbol{u}_t = \boldsymbol{u}_w + \boldsymbol{u}_n$,总流度 $\lambda_t = \lambda_w + \lambda_n$,分流函数 $f_\alpha = \lambda_\alpha / \lambda_t$。在实际地质介质中,$\boldsymbol{K}$ 常具有多尺度、高对比度特征(如高渗透通道与包裹体),导致压力和速度场强烈非均质,饱和度演化的前缘可能形成尖锐间断,直接精细网格模拟虽能分辨这些效应,但在长时间两相流模拟中反复求解压力-速度系统计算代价极高。
时间离散上,全隐式方法鲁棒但需每步求解大型非线性系统;隐式压力显式饱和度(IMPES)方法因分离压力求解与饱和度更新而效率更高,但标准 IMPES 方案可能面临稳定性条件苛刻、一个相质量不守恒、相位偏差及饱和度越界等问题,且毛管压力引起的介质界面处饱和度间断增加了难度。Chen 与 Sun 提出了物理保持的 IMPES(P-IMPES)方案,其核心思想是将相达西速度用总速度和与毛管势梯度相关的辅助速度表示,所得格式满足局部质量守恒、正确处理总速度的法向连续性、对润湿相和非润湿相无偏、且在适当 CFL 条件下保持饱和度在物理界限内。该方案的这些性质对饱和度输运方程的求解至关重要。
空间离散方面,经典粗化方法通过等效粗尺度系数减小精细尺度非均质性,但难以处理强非局部特征;多尺度有限元和多尺度有限体积方法通过局部精细尺度问题构造粗尺度基函数,提供了一种降阶策略。对于紊流模拟,局部质量守恒至关重要,这推动了混合多尺度有限元、多尺度 mortar 方法及保守通量后处理技术的发展。广义多尺度有限元方法(GMsFEM)通过局部谱分解构造多个基函数,增强了灵活性,但在高对比度介质中,高渗透通道等非局部特征难以被标准局部基函数捕捉。约束能量最小化广义多尺度有限元方法(CEM-GMsFEM)通过引入辅助空间和超样本区域的能量最小化基函数,能捕获通道特征并在粗网格尺寸下获得不依赖于对比度的收敛性,其混合形式利用辅助压力基函数构造局部速度基函数,首次获得细网格质量守恒的速度逼近和一阶收敛性。
然而,将 P-IMPES 与混合 CEM-GMsFEM 直接结合用于两相流仍面临时变系数的挑战。在时间层 $t^n$,压力-速度系统的系数 $\kappa^n = \lambda_t(S_w^n)\boldsymbol{K}$ 依赖当前润湿相饱和度,因而随饱和度演化而改变。若多尺度空间仅由初始系数构造并保持不变,基函数可能不再适应当前系数场,从而恶化速度逼近和饱和度前缘模拟。针对此问题,本文提出一种自适应混合 CEM-GMsFEM 方法,其核心是监测流度加权系数的变化,当变化超过预设容差时重新生成多尺度空间,否则复用已有空间,从而在不显著增加计算量的前提下保持逼近精度。此外,通过仅对有非零细网格质量残差的粗单元进行局部后处理,恢复总速度的细网格质量守恒。本文试图填补的空白是:在保持 P-IMPES 物理性质的前提下,构建能有效应对流度变化的多尺度逼近,并给出包含自适应容差、粗网格尺寸、谱截断和前缘误差等贡献的严格误差分析。
2. 主要定理或主要结果
本文的主要理论结果分为两部分:物理性质分析和误差估计。
局部质量守恒与无偏性(第 4.1 节)
在任意时刻 $t^{n+1}$,对于选定的湿相饱和度更新公式 (4.1) 和 $S_n^{H,n+1}=1-S_w^{H,n+1}$,通过添加两相离散方程并利用总速度的离散守恒性,可以推导出非湿相也满足守恒方程 (4.3)。因此方案对两个相均保持局部质量守恒。此外,基于总速度和辅助毛管速度的无偏定义,可证得两相得到相同的离散解,即方案对湿相和非湿相无偏。
饱和度的界保持性质(第 4.2 节)
在适当的 CFL 条件(文中给出具体形式)下,离散饱和度 $S_w^{H,n}$ 和 $S_n^{H,n}$ 对所有时间层 $n$ 保持在物理区间 $[0,1]$ 内。该结论推广了 P-IMPES 在一致精细网格上的界保持性质,证明依赖于两个辅助引理(引理 4.1 和 4.2),局部守恒性和迎风型离散的极大值原理。
速度与饱和度误差估计(第 4.3 节)
针对对流主导情形,作者建立了速度误差和饱和度误差的先验估计。以总速度误差 $\|\boldsymbol{u}_t - \boldsymbol{u}_t^H\|_{L^2}$ 和湿相饱和度误差 $\|S_w - S_w^H\|_{L^2}$ 为例,最终估计可概括为
$$ \|\boldsymbol{u}_t - \boldsymbol{u}_t^H\|_{L^2} \leq C\big(\, \varepsilon_\text{update} + H^{\alpha} + \Lambda^{-1}_{\min} + \mathcal{E}_{\text{front}}\big), $$$$ \|S_w - S_w^H\|_{L^2} \leq C'\big(\, \varepsilon_\text{update} + H^{\alpha} + \Lambda^{-1}_{\min} + \mathcal{E}_{\text{front}}\big), $$其中 $\varepsilon_\text{update}$ 为自适应容差控制的部分(与 $\eta^n$ 相关),$H$ 为粗网格尺寸,$\alpha$ 为与离散阶数有关的正数,$\Lambda_{\min}$ 为谱截断对应的最小保留特征值(其倒数反映谱近似误差),$\mathcal{E}_{\text{front}}$ 表示饱和度前缘误差。这些估计清晰地分离了自适应更新策略、空间离散尺度、谱近似精度和物理界面演化带来的误差贡献,为方法的参数选择提供了理论依据。
3. 研究方法、关键技术和核心工具
本文的主要技术障碍在于:有效渗透率 $\kappa^n$ 随饱和度演化而变,导致初始构造的多尺度基函数在后续时间步精度下降,而频繁更新多尺度空间又会显著增加计算成本。作者的创新点在于设计了一个自适应更新准则和配套的局部后处理,并建立了相应的误差分析框架。
1. 基于系数变异的自适应更新策略
定义误差指示器
$$ \eta^n = \left\| \left(\kappa_i^H\right)^{-1/2} - \left(\kappa_n^H\right)^{-1/2} \right\|_{L^2(\Omega)}, $$其中 $i$ 为最近一次更新多尺度空间的时间层。当 $\eta^n$ 超过预设容差 $\varepsilon$ 时,利用当前系数 $\kappa_n^H$ 重新构造压力多尺度空间 $Q_{ms}$ 和速度多尺度空间 $V_{ms}$(通过求解局部谱问题和约束能量最小化问题);否则复用旧空间。该指示器直接度量了系数场的相对变化,其分析基础在于 $\kappa^{-1/2}$ 范数等价于能量范数,与混合变分形式的构造自然相容。这样,在饱和度变化剧烈时自动触发更新,而在平稳阶段避免不必要的重构,权衡了精度与效率。
2. 基于细网格残差标记的局部后处理
原始混合 CEM-GMsFEM 所得总速度仅对多尺度压力空间满足守恒,不一定满足细尺度质量守恒。本文提出仅在含有非零细网格质量残差 $R_{K_h}^{n+1} = (F_t, 1_{K_h}) - \sum_\alpha \beta_\alpha(\widehat{\boldsymbol{u}}_t^{H,n+1}, 1_{K_h}; S_w^{H,n})$ 的粗单元上进行局部后处理。在每个标记单元 $K$ 上,以原多尺度速度的法向通量为边界条件,求解局部 RT0-有限体积问题 (3.6)–(3.7),得到细网格守恒速度 $\boldsymbol{u}_{t,p}^{H,n+1}$。粗单元常量属于 $Q_{ms}$ 的事实保证了相容性条件 $\sum_{K_h\subset K} R_{K_h}^{n+1}=0$,从而局部问题适定。该方案避免了对所有粗单元进行后处理,大幅降低了附加计算量。
3. 物理性质证明
局部质量守恒与无偏性的证明采用直接代数推导:从湿相更新公式出发,利用饱和度约束和总速度的离散守恒性,得到非湿相的等价方程,并验证两相速度表达的一致性。有界性证明沿袭 P-IMPES 的框架,需要两个引理:引理 4.1 表明每个单元上饱和度更新可表示为前一层饱和度的凸组合;引理 4.2 给出确保该性质成立的具体 CFL 条件。
4. 误差估计的分解技术
误差分析的核心在于将总误差分解为若干可控部分:
- 自适应容差贡献:通过 $\eta^n$ 控制系数“过时”引起的能量误差。
- 多尺度逼近误差:利用 CEM-GMsFEM 的对比度无关收敛性质,将空间离散误差与 $H$ 和 $\Lambda_{\min}$ 相联系。
- 前缘层误差:源于饱和度间断附近数值扩散效应。作者通过广义混合方法的标准技巧(引入投影算子、局部对偶问题等)将速度误差与饱和度误差分别估计,并最终得出上述分离形式的界。估计过程中,系数随时间变化导致的额外项由自适应更新条件吸收。
4. 与之前工作的比较
本文与已有工作的比较体现在多个层面:
1. 与 P-IMPES 方案本身的比较
标准 P-IMPES(Chen & Sun)在一致精细网格上已具备局部守恒、无偏和界保持性质。本文在多尺度离散框架下完整保留了这些性质(第 4.1 和 4.2 节给出的证明与精细情形平行,但需处理多尺度速度场的额外误差),并通过自适应 CEM 基函数和局部后处理确保了在降低自由度的同时不失物理保真性。与文献[12]相比,本文的 P-IMPES 多尺度实现并未引入新的物理偏差。
2. 与现有 P-IMPES-多尺度方法[37]的区别
文献[37]将 P-IMPES 与 MsFEM 结合,采用残差驱动的基函数富集和基于源项的标记,且后处理针对整个粗网格层。本文的不同之处在于:
- 基空间构建:使用混合 CEM-GMsFEM,其超样本能量最小化基函数天然具有对比度鲁棒性,并且基函数个数可通过谱截断控制;[37]中 MsFEM 对高对比度适应性有限。
- 空间更新触发机制:本文基于系数变异指示器 $\eta^n$ 进行全局重建,而非局部富集;这更适合流度场全局变化的特性。
- 后处理标记:本文采用细单元残差 $R_{K_h}$ 是否为零进行标记,仅后处理有局部不守恒的粗单元,比[37]的源项标记更节约。
- 误差分析:本文首次给出了分离自适应容差、粗网格尺寸、谱误差和前缘误差的估计,为参数选择提供了定量指导;[37]中仅建立了速度与饱和度误差关系,未考虑系数更新误差。
3. 与标准混合 CEM-GMsFEM 的比较
标准混合 CEM-GMsFEM [36] 面向稳态或定常系数问题,不涉及系数随时间变化。本文通过引入自适应更新,将 CEM-GMsFEM 推广到两相流这类非定常系数情形,代价是在某些时间步需要重新求解局部谱问题和构造基函数,但自适应策略保证了总重构次数远小于总时间步数。此外,后处理仅标记有残差的单元,进一步控制了额外开销。
4. 适用范围与局限
本文方法适用于不可压缩非混相两相流,且假设毛管压力已知。与全隐式方案相比,P-IMPES 仍受 CFL 条件限制,但界保持性质确保了稳定性。误差估计针对对流主导情况,扩散主导情况下可能需另行分析。从提供内容中无法确认该方法对三维问题或非结构网格的推广性,有待后续工作验证。
3. 一种用于可压缩阻力霍尔磁流体动力学系统的保结构数值方法
A structure-preserving Numerical Method for the Compressible Resistive-Hall-MHD System
作者: Murtazo Nazarov, Rafael Rodriguez-Velasco, Ignacio Tomas
类别: math.NA, physics.comp-ph, physics.plasm-ph
发布日期: 2026-07-15
arXiv ID: 2607.14286v1
摘要翻译: 本文针对可压缩阻力霍尔磁流体动力学(MHD)模型提出了一种保结构方法。微分算子被分裂为两部分:由可压缩欧拉方程构成的流体动力学部分,以及由耦合洛伦兹力与感应方程的系统构成的磁场部分。该方法对欧拉部分采用连续拉格朗日元,对磁场部分采用curl协调有限元空间。流体动力学部分保持了密度和内能的正性、总能量守恒以及比熵的极小原理。由于有限元的选取,磁场部分保持了散度无源约束。流体部分使用显式强稳定性保持龙格库塔(SSP-RK)方法求解,磁场部分则采用克兰克-尼科尔森方法求解,后者需要使用牛顿法。给出了相应牛顿迭代的雅可比矩阵的强制性估计。我们引入了高阶人工电阻率来改善非线性残差的条件数及雅可比矩阵的可逆性。通过求解包括光滑惠斯勒波、用于比较阻力MHD与阻力霍尔MHD的奥萨格-唐涡旋,以及磁重联问题在内的多个具有挑战性的基准测试,验证了该方法的鲁棒性与精度。
Comment: 37 pages, 7 figures
链接: http://arxiv.org/abs/2607.14286v1
1. 研究对象和背景
本文研究可压缩电阻霍尔磁流体动力学(Resistive-Hall-MHD)系统的数值方法。该系统的控制方程为:
$$ \begin{aligned} &\partial_t \rho + \operatorname{div} \mathbf{m} = 0, \\ &\partial_t \mathbf{m} + \operatorname{div}(\rho^{-1}\mathbf{m}\mathbf{m}^\top + I p) = \mu \operatorname{curl} \mathbf{H} \times \mathbf{H}, \\ &\partial_t E + \operatorname{div}\left( \frac{\mathbf{m}}{\rho} (E + p) \right) = \mu (\operatorname{curl} \mathbf{H} \times \mathbf{H}) \cdot \frac{\mathbf{m}}{\rho} + r |\operatorname{curl} \mathbf{H}|^2, \\ &\partial_t \mathbf{H} - \operatorname{curl} \left( \frac{\mathbf{m}}{\rho} \times \mathbf{H} \right) = -\operatorname{curl} \left( \frac{r}{\mu} \operatorname{curl} \mathbf{H} + \frac{d_i}{\rho} \operatorname{curl} \mathbf{H} \times \mathbf{H} \right), \end{aligned} $$其中 $\rho$ 为密度,$\mathbf{m}=\rho\mathbf{v}$ 为动量,$E$ 为总机械能,$\mathbf{H}$ 为磁场,$p$ 为压强,$\mu$ 为磁导率,$r$ 为电阻率,$d_i$ 为离子惯性长度($d_i=m_i/q_i$)。右侧 $\operatorname{curl}(r \operatorname{curl} \mathbf{H})$ 为电阻项,$\operatorname{curl}(\frac{d_i}{\rho} \operatorname{curl} \mathbf{H} \times \mathbf{H})$ 为霍尔项。
MHD 系统在天体物理和磁约束核聚变等领域有广泛应用。霍尔项由 Lighthill 首次指出其重要性;进入本世纪以来,越来越多的文献表明电阻项和霍尔项对重现实际观测到的磁重联速率至关重要。数值求解 MHD 系统时,必须保持解的若干本质结构:正性(密度和内能非负)、总能量守恒、熵耗散以及磁场的散度无源约束($\operatorname{div}\mathbf{H}=0$)。对于理想可压缩 MHD 系统,保结构的数值方法已有较大发展;然而,关于可压缩电阻霍尔 MHD 系统的数学文献非常少,尤其是缺乏关注结构保持的数值格式。本文试图填补这一空白:在前序工作 [22] 的基础上,将算子分裂与保结构离散相结合,发展一个能在离散水平保持密度正性、内能正性、总能量守恒、比熵最小值原理以及散度无源约束的数值格式,并特别处理因霍尔项引入导致的非线性求解器鲁棒性问题。
2. 主要定理或主要结果
本文的主要结果包含两部分:算子分裂后各子系统的解析性质,以及全离散格式的保结构性质与数值验证。
算子性质:将系统 (1) 分裂为流体部分(算子1,即 Euler 方程)和磁部分(算子2,包含 Lorentz 力、电阻和霍尔项)。在适当边界条件或紧支初值及短时间内,算子1 保持质量、动量、总能量守恒,且 $\rho\ge 0$,比熵 $s$ 满足最小值原理,数学熵 $\eta=-\rho s$ 满足耗散不等式
$$ \partial_t \eta + \operatorname{div}(\eta \mathbf{v}) \le 0. $$算子2 在 $\rho>0$,内能 $e>0$ 且温度 $\theta>0$ 的初值下,具有以下点性质:
$$ \begin{aligned} &\frac{\partial}{\partial t} \int_\Omega \left( \frac12 \rho |\mathbf{v}|^2 + \frac{\mu}{2}|\mathbf{H}|^2 \right) dx + \int_{\partial\Omega} \left( r\operatorname{curl}\mathbf{H} - \mu(\mathbf{v}\times\mathbf{H}) + \mu\frac{d_i}{\rho}\operatorname{curl}\mathbf{H}\times\mathbf{H} \right) \cdot (\mathbf{H}\times\mathbf{n}) ds = -\int_\Omega r|\operatorname{curl}\mathbf{H}|^2 dx,\\ &\frac{\partial e}{\partial t} = \frac{r}{\rho}|\operatorname{curl}\mathbf{H}|^2 \ge 0,\quad \frac{\partial \theta}{\partial t} \ge 0,\quad \frac{\partial s}{\partial t} \ge 0,\quad \frac{\partial \eta}{\partial t} \le 0, \end{aligned} $$从而内能单调不减,温度与比熵单调非降,数学熵单调非增。
离散格式:流体变量用连续 $\mathbb{P}_1$ Lagrange 元,磁场用旋度协调的旋转 BDM$_1$ 元($\mathbf{H}_h \subset H(\operatorname{curl})$),使得离散散度无源自动保持($\operatorname{div}\mathbf{H}_h = 0$ 弱成立)。算子1 用显式强稳定 Runge–Kutta (SSP-RK) 推进,严格保正性、守恒和熵不等式(假设所选 Euler 求解器具有这些性质)。算子2 用 Crank–Nicolson 隐式格式离散,得到非线性残差并用 Newton 法求解。针对 Newton 迭代可能出现的 Jacobian 病态问题,作者给出了 Jacobian 强制性估计:当电子速度过大时,Jacobian 可能失去可逆性。为此引入一项与高阶导数相关的人工电阻率以改善 Newton 迭代的鲁棒性。全算法流程保证分裂顺序下组合格式的保结构性质。
数值验证:对光滑哨声波、Orszag–Tang 涡流和 GEM 磁重联问题等进行测试,最大分辨率达 $1024\times1024$ 单元。验证了格式的收敛性、能量行为以及网格(包括非结构网格)适应性。在 Orszag–Tang 涡流算例中直接对比了电阻 MHD 与电阻霍尔 MHD 的结果。
3. 研究方法、关键技术和核心工具
本文的核心技术路线基于算子分裂和相容有限元离散,并针对霍尔项的处理进行了关键创新。
算子分裂与半隐式时间推进:将原 PDE 系统分为可显式处理的 Euler 部分(算子1)和需隐式处理的磁部分(算子2)。分裂的设计使得各子算子能单独保持关键结构:Euler 部分可选用已有的保正性、保熵离散格式(本文采用 SSP-RK 加图粘性等方法,具体见 [22] 附录 B);磁部分因包含可能十分刚性的霍尔项,采用 Crank–Nicolson 格式进行时间离散。这种半隐式方法避免了整体隐式的高计算成本,同时保证了整体格式的稳定性。
磁部分的有限元离散与静磁约束:磁场 $\mathbf{H}$ 选用最低阶旋度协调空间(旋转 BDM$_1$),该空间包含所有线性向量值多项式,从而可预期 $L^2$ 二阶精度。更为重要的是,空间满足复合关系 $\nabla W_h \subset \mathbf{H}_h$,其中 $W_h$ 为连续的 $\mathbb{P}_2$ 空间。这一包含关系使得弱散度为零的条件等价于 $\mathbf{H}_h$ 与 $\nabla W_h$ 正交,从而在所给离散格式中自动保持 $\operatorname{div} \mathbf{H}_h = 0$ 的强制约束,无需额外散度清洗步骤。
Newton 迭代与人工电阻率:对于 Crank–Nicolson 离散得到的非线性方程,作者使用 Newton–Raphson 方法求解。在没有霍尔项时,Newton 迭代收敛极快(通常不超过 4 次每时间步);引入霍尔项后,在非结构网格上收敛行为变得不稳定,经常出现过多迭代或 Jacobian 奇异。作者对 Jacobian 进行了强制性分析(见附录 B),发现当 $\mathbf{J} = \operatorname{curl}\mathbf{H}$ 过大或电子速度 $\mathbf{u}_e$ 较大时,Jacobian 的强制性丧失。为克服此问题,提出在感应方程中添加人工电阻项 $\epsilon_h \operatorname{curl}(\dots)$,其中人工电阻系数 $\epsilon_h$ 与局部网格尺寸和电子速度的高阶导数相关,并设计为只在欠解析区域显著。该人工电阻改善了 Jacobian 的谱条件数,恢复了 Newton 迭代的二阶收敛性。
关键技术细节:质量矩阵使用集总格式(lumping)以保持正性并使步进局部化;磁场方程的弱形式利用分部积分将旋度转移至测试函数,自然耦合了速度场;Newton 迭代中 Jacobian 矩阵的推导需仔细处理旋度算子与有限元基的相互作用。整体算法在 3.6 节以伪代码形式给出。
4. 与之前工作的比较
本文在以下几个方面明确区别于已有研究:
与作者前序工作 [22] 的对比:[22] 仅针对理想可压缩 MHD(无电阻和霍尔项),而本文将其推广至电阻霍尔 MHD。更重要的技术区别在于非线性求解器行为:理想 MHD 中 Newton 迭代极佳;加入电阻项后变化不大,但霍尔项导致 Jacobian 在非结构网格上容易病态。本文由此提出人工电阻策略以稳定 Newton 迭代,这是对 [22] 方案的根本性改进。此外,本文在理论分析部分给出了更完整的算子2 保持温度、比熵、数学熵单调性的证明,以及 Jacobian 强制性估计,这些在 [22] 中均未涉及。
与其他电阻霍尔 MHD 数值格式的对比:现有文献(如 [63,20,54,24,59,4,14])中几乎没有以结构保持为首要目标的格式。多数方法建立在有限体积/差分框架下,常依赖散度清洗、约束传输等技术处理 $\operatorname{div}\mathbf{H}=0$,且鲜有对密度/内能正性的严格保持。本文的格式基于有限元,通过旋度协调空间天然满足散度约束,且流体部分继承保正性求解器,从离散层面保证了这些性质。
网格适应性与高分辨率:几乎所有已发表的电阻霍尔 MHD 计算均使用结构笛卡尔网格,且分辨率极低(如 GEM 问题通常仅 $128\times128$)。本文展示了格式在结构各向同性、各向异性、非结构拟一致网格、交叉网格等多种网格上的一致性结果,表明方案对网格印记不敏感。同时将 GEM 问题分辨率提升至 $1024\times1024$ 单元,为文献中最高水平,并计划将数据作为基准公开。
定量验证:由于可压缩电阻 MHD 缺乏精确行波解,以往工作仅限于定性图形比较。本文提出线性电阻区的光滑哨声波作为定量测试,用以验证格式的精度和收敛率,这在同类文献中尚属首次。
理论贡献:本文利用热力学稳定状态方程(凸内部能函数)将算子2 的性质从理想 MHD 的熵守恒/耗散推广到一般电阻霍尔情况,给出了内能增、温度升、比熵增的完整路径,为构造保结构格式提供了坚实的连续模型依据。这一分析在分析文献中也较为少见。
了解领域论文(摘要翻译)
1. 基于结构保持神经网络和熵变量的Burgers方程亚网格尺度参数化
Subgrid-Scale Parameterization in Burgers’ Equation Using Structure-Preserving Neural Networks and Entropy Variables
作者: Aijaz Nazir, Ilya Timofeyev
类别: math.NA, cs.LG
发布日期: 2026-07-16
arXiv ID: 2607.14855v1
链接: http://arxiv.org/abs/2607.14855v1
摘要翻译
中文标题: 基于结构保持神经网络和熵变量的Burgers方程亚网格尺度参数化
摘要翻译: 我们提出一种机器学习方法,用于在偏微分方程的粗模拟中建立亚网格尺度(SGS)参数化。我们利用结构保持神经网络和熵变量,在Burgers方程的粗模拟中学习亚网格通量。特别地,我们采用解耦神经网络架构,显式地将亚网格修正分离为两个不同的组成部分:一个守恒通量势网络和一个涡粘性网络。我们证明,这种降阶框架保持了高物理保真度,准确地再现了全尺度系统的能量谱、时空相关函数和动力学特征。此外,我们表明该方法具有鲁棒性,并且适用于训练范围之外的参数。
2. 保持结构的拉格朗日乘子方法用于平均曲率流及其误差界
Structure-preserving Lagrange-multiplier methods for mean curvature flow and their error bounds
作者: Balázs Kovács, Buyang Li, Christian Lubich
类别: math.NA
发布日期: 2026-07-16
arXiv ID: 2607.14787v1
链接: http://arxiv.org/abs/2607.14787v1
摘要翻译
中文标题: 保持结构的拉格朗日乘子方法用于平均曲率流及其误差界
摘要翻译: 我们提出并分析了一类用于闭曲面平均曲率流的演化曲面有限元方法,该方法使用拉格朗日乘子以保持能量递减结构。算法基于解驱动形式,采用Huisken关于法向量和平均曲率的演化方程。时间离散对抛物几何变量使用线性隐式向后差分公式(BDF),对节点位置使用隐式Adams更新。该方法也可包含最小变形率类型的人工切向速度。所得的全离散算法在每一时间步上面积递减,其衰减率由计算得到的平均曲率预先确定。我们证明了在弱正则性假设下离散拉格朗日乘子的局部存在唯一性,以及用于其计算的简化牛顿迭代的收敛性。在较强的正则性假设下(与底层的演化曲面有限元法的收敛理论中所用的一致),对于多项式次数$k\ge 2$的有限元以及$q$步BDF和$q$步隐式Adams方法($2\le q\le 5$),无论是否带有最小变形率切向运动,我们都推导出了$H^1$范数下$h^k+\tau^q$阶的最优误差界。球面平均曲率流的数值实验验证了预期的收敛阶,并表明拉格朗日乘子修正仅带来很小的计算开销,且该开销本质上与网格尺寸和时间步长无关。
3. 针对内部与边界数据不等的椭圆型偏微分方程的算子分裂贝叶斯学习
Operator-Split Bayesian Learning for Elliptic PDEs with Unequal Interior and Boundary Data
作者: Emmanuel E. Oguadimma
类别: math.NA, math.ST
发布日期: 2026-07-16
arXiv ID: 2607.14680v1
链接: http://arxiv.org/abs/2607.14680v1
摘要翻译
中文标题: 针对内部与边界数据不等的椭圆型偏微分方程的算子分裂贝叶斯学习
摘要翻译: 我们针对内部和边界观测数量不等的二阶一致椭圆Dirichlet问题,提出了一种算子分裂贝叶斯学习框架。数据由域内源项的含噪测量和边界值的含噪测量组成。对这两个量赋予独立的贝叶斯神经网络(BNN)先验,并将得到的乘积后验通过椭圆解算子进行前推。我们证明了这种构造所诱导的后验在真实解附近收缩。收缩半径将受二阶椭圆算子控制的域贡献与受边界固有维数控制的边界贡献分开。与\cite{ZhaoLu2026}中的极小极大下界相结合,这给出了一个在对数因子范围内接近极小极大的上界。我们的数值实验展示了源项和边界不确定性的传播,以及不等的采样预算对后验重建的影响。
4. 基于层次化配点细化的物理信息神经网络新策略
A new strategy for physics-informed neural networks based on hierarchical collocation point refinement
作者: Minjae Choi, Dukhwan Shin, Youngsoo Yang, Eunjung Lee
类别: math.NA
发布日期: 2026-07-16
arXiv ID: 2607.14665v1
链接: http://arxiv.org/abs/2607.14665v1
摘要翻译
中文标题: 基于层次化配点细化的物理信息神经网络新策略
摘要翻译: 物理信息神经网络(PINNs)为求解偏微分方程提供了灵活的框架,但当需要大量配置点以精确施加控制方程时,训练会变得计算代价高昂。为缓解这一成本,我们引入了基于多重网格的参数更新PINN(MPU-PINNs),这是一种从粗到细的训练策略,在学习过程中逐步增加训练点的数量。该方法首先在粗配置点集上训练神经网络,然后将学习到的参数传递到逐步细化的层级。这种初始化策略使网络能够以较低的计算成本捕捉解的全局特征,然后再通过额外的训练点细化局部细节。为进一步提升高频问题的性能,我们加入了一种缩放技术,以缓解训练过程中的频谱偏差影响。我们在若干基准偏微分方程上评估了MPU-PINNs,包括二维和三维泊松方程、对流-扩散-反应方程以及亥姆霍兹方程。数值实验表明,MPU-PINNs显著减少了训练时间,同时达到了与传统PINNs以及其他代表性变体(如SA-PINNs和XPINNs)相当的精度。结果进一步表明,所提出的从粗到细的学习策略大幅减少了在细层级所需的优化工作量。总体而言,MPU-PINNs提供了一种高效的单网络训练框架,增强了PINNs在广泛偏微分方程问题上的计算效率和可扩展性。
5. 神经非常弱形式实现面向硬件的深度偏微分方程求解器
Neural Very Weak Formulations enabling Hardware-Oriented deep PDE solvers
作者: Gabriel Acosta, Francisco Bersetche
类别: math.NA
发布日期: 2026-07-16
arXiv ID: 2607.14498v1
链接: http://arxiv.org/abs/2607.14498v1
摘要翻译
中文标题: 神经非常弱形式实现面向硬件的深度偏微分方程求解器
摘要翻译: 作为概念验证,我们证明:只要从适当光滑空间中选取检验函数,椭圆问题的最小二乘非常弱形式即可由具有低正则性的神经网络有效离散化。除了避免自动微分所带来的直接计算收益外,该方法在多种神经网络空间中均表现良好,甚至在奇异解和高维设定等挑战性情境中亦如此。特别关注基于阶跃激活函数和一位量化线性函数的试探函数,因其适于高效的面向硬件实现。
不相关论文(基本信息)
1. 三维欧拉奇异性计算机辅助证明中奇异加权估计的解析有限秩修正
Analytic finite-rank corrections for singularly weighted estimates in a computer-assisted proof of 3D Euler singularity
作者: Jiajie Chen, Thomas Y. Hou
类别: math.AP, math.NA
发布日期: 2026-07-16
arXiv ID: 2607.15256v1
链接: http://arxiv.org/abs/2607.15256v1
摘要: Computer-assisted proofs of self-similar singularity formation for fluid equations often rely on numerically constructed approximate profiles. One effective approach to establishing stability of perturbations around a numerically constructed profile is to perform weighted energy estimates with singular weights near the singularity. However, the weighted norms require exact local vanishing conditions that are not automatically preserved by the equations nor the numerical construction. In this paper, we review an analytic low-rank correction method first developed in [ChenHou2023a,ChenHou2023b] to overcome this difficulty. The numerical step determines coefficients, rigorous bounds, and low-order defect modes in explicit global basis representations, while the required vanishing conditions are enforced analytically through low-rank corrections derived from Taylor expansions of the relevant quantities represented in a smooth basis. For completeness, we briefly review the singularly weighted estimates and a quantitative finite-rank perturbation method in the 2D Boussinesq / 3D Euler stability argument, where singular weights and the required vanishing order arise. Against this background, we formulate the local correction principle in a simplified setting, explain the correction of the residual error in numerical constructions of approximate space-time solutions and the stream function, and discuss its broader applicability to computer-assisted stability analysis for nonlocal PDEs.
2. 面向投资组合优化的SciPhy强化学习
SciPhy Reinforcement Learning for Portfolio Optimization
作者: Igor Halperin, Andrey Itkin
类别: q-fin.PM, math.AP, math.NA, q-fin.CP, q-fin.MF
发布日期: 2026-07-16
arXiv ID: 2607.15195v1
链接: http://arxiv.org/abs/2607.15195v1
摘要: This paper introduces a dynamic portfolio optimization framework for large institutional investors using Scientific Physics-Informed Reinforcement Learning (SciPhyRL). Formulated in continuous time over an extended state space that includes explicit cumulative costs, the approach leverages offline historical data to learn optimal, distribution-aware strategies. A core innovation reduces the optimization challenge to solving an HJB equation by projecting it onto observed trajectories as a pathwise Hamilton-Jacobi equation. This is solved directly from data using PINN in a single offline sweep, eliminating the need for traditional value or policy iteration. To make the method effective at practical short horizons, the control variable is recast from a continuous trading rate to a discrete target holding. This ensures signal-implied positions are reached immediately, while execution costs are evaluated against a microstructure-grounded quadratic price impact model. Evaluated on a $14$-asset ETF universe using an engineered oracle signal, the learned Gibbs policy yields substantial out-of-sample Sharpe ratio improvements over static and myopic baselines. The results demonstrate that the proposed framework successfully translates known signal quality into a robust, multi-period, and cost-aware allocation mechanism with strictly controlled volatility and turnover.
3. Friedrichs系统的高效高阶局部时间积分
Efficient higher-order local time integration for Friedrichs’ systems
作者: Marlis Hochbruck, Jonas Köhler, Malik Scheifinger
类别: math.NA
发布日期: 2026-07-16
arXiv ID: 2607.15192v1
链接: http://arxiv.org/abs/2607.15192v1
摘要: In this paper, we construct an efficient higher-order local time integration scheme for spatially discretized linear Friedrichs’ systems. In particular, our interest is in problems where only a few of the mesh elements are small while the majority of the elements is much larger. The special combination of two methods like the leapfrog method on the coarse part of the mesh and the Crank-Nicolson method on the fine part as was done in Hochbruck, Sturm 2016 and Hochbruck, Köhler 2022 is not suitable for higher-order time integration. Therefore, we suggest to approximate the solution of the linear systems arising in each time step by a preconditioned Krylov subspace method, e.g., the quasi-minimal residual method by Freund and Nachtigal 1991. The techniques developed here for linear problems also carry over to nonlinear problems, where linear systems of the same type arise within a Newton-type iteration. Motivated by the analysis of locally implicit methods by Hochbruck and Sturm 2016, we show how to construct a preconditioner in such a way that the number of iterations required by the Krylov subspace method to achieve a certain accuracy is bounded independently of the diameter of the small mesh elements. We prove this behavior by using Faber polynomials and complex approximation theory. The cost to apply the preconditioner consists of the solution of a small linear system, whose dimension corresponds to the degrees of freedom within the fine part of the mesh (and its next coarse neighbors). If this dimension is small compared to the size of the full mesh, the preconditioner is very efficient. We conclude by verifying our theoretical results with numerical experiments for the fourth-order Gauss-Legendre Runge–Kutta method.
4. 由Artin-Schreier多项式生成的完全均匀分布的准蒙特卡罗序列
Perfectly equidistributed Quasi-Monte Carlo sequences from Artin-Schreier polynomials
作者: Nicolas Bonneel, David Coeurjolly, Victor Ostromoukhov
类别: cs.DM, math.NA
发布日期: 2026-07-16
arXiv ID: 2607.15141v1
链接: http://arxiv.org/abs/2607.15141v1
摘要: To numerically integrate a function, one may resort to Quasi-Monte Carlo estimators, that average integrand values at pseudo-random well-distributed uniform sampling locations. Better uniformity improves the worst-case integration-error bound. A standard measure of uniformity is given by an integer $t$ value, where $t=0$ yields the best uniformity. Producing sequences of samples with bounded $t$ values can be achieved with Sobol’ recursive construction, that uses coefficients of irreducible polynomials. While $b$-dimensional sequences with $t=0$ can be obtained by taking $b$ polynomials of degree $1$ over the Galois Field $\mathrm{GF}(b)$, we show conditions that guarantee $t=0$ for specific higher degree polynomials. In particular, we relate the Sobol’ construction to tensorized powers of Pascal matrices when the chosen polynomials only differ by a constant and exhibit simple conditions to guarantee $t=0$ in this case. We then focus on Artin-Schreier irreducible polynomials, in the form $p_i(x) = x^b - x + c_i$, where $i \in \{1, \dots, b-1\}$ and $b$ is prime, and we make explicit conditions that always guarantees $t=0$ in $b-1$ dimensions. Combining $b$-dimensional Sobol’ of degree $1$ and our $(b-1)$-dimensional Artin-Schreier sequence of degree $b$, we provide a fast greedy procedure that optimizes the $(2b-1)$-dimensional combined $t$ value, while guaranteeing $t=0$ projection in subspaces.
5. 非线性逼近流形上基于残差的时间离散化:分析与高斯应用
Residual-Based Time Discretization on Nonlinear Approximation Manifolds: Analysis and Gaussian Applications
作者: Eddy de Leon, Caroline Lasser
类别: math.NA, quant-ph
发布日期: 2026-07-16
arXiv ID: 2607.15086v1
链接: http://arxiv.org/abs/2607.15086v1
摘要: We study time-discrete parametric approximations of evolution equations in Hilbert spaces based on residual minimization. The solution is represented by a parametrized ansatz belonging to a low-dimensional nonlinear manifold, and time stepping is performed by minimizing suitably defined residuals at each step. Two natural residual formulations are considered: discretization followed by parametrization of the evolution equation, and discretization of the Dirac–Frenkel variational principle governing the parameter dynamics. A unified error analysis is developed for both approaches within the family of $ζ$-methods. The resulting bounds separate the effects of time discretization from those of residual minimization and yield first- and second-order convergence under Lipschitz, one-sided Lipschitz, and dissipativity assumptions. For the variational formulation, additional stability conditions involving the conditioning of the parametrization map arise naturally. The framework is applied to Gaussian approximation manifolds, for which residual norms and gradients admit explicit closed-form expressions when polynomial operators are involved. This enables efficient implementation without spatial discretization. Numerical experiments for time-dependent Schrödinger equations illustrate the theoretical convergence rates and the influence of residual accuracy on conservation properties.
6. Monge-Ampère方程的全离散最小二乘分裂格式:有限元分析与收敛性
Fully discrete least-squares splitting scheme for the Monge-Ampère equation: finite element analysis and convergence
作者: Anna Peruso
类别: math.NA
发布日期: 2026-07-16
arXiv ID: 2607.15024v1
链接: http://arxiv.org/abs/2607.15024v1
摘要: The least-squares splitting algorithm for the Monge-Ampère equation has been used successfully in computations for several years, but a convergence theory for fully discrete splitting schemes of this type has remained unavailable. In this work, we introduce and analyze a finite element framework for smooth solutions of the Dirichlet Monge-Ampère equation in two dimensions. The proposed schemes combine a discrete Hessian reconstruction with a local projection onto the determinant constraint. Under a discrete Miranda-Talenti estimate and standard approximation properties of the Hessian reconstruction, we prove local convergence of the iterative scheme and optimal-order convergence of its limit to the exact solution in an $H^2$-type norm. We verify the estimates for conforming $C^1$ schemes, including the Argyris element, and for $C^0$-interior penalty and DG schemes of degree at least three; quadratic $C^0$-interior penalty and DG schemes are also covered when $|u|_{H^3(Ω)}$ is sufficiently small. To the best of our knowledge, these discretizations have not previously been proposed or analyzed for least-squares splitting methods. Numerical experiments confirm the theoretical convergence rates.
7. 非自治随机抛物方程的有限维反馈镇定
Finite-Dimensional Feedback Stabilization of Nonautonomous Stochastic Parabolic Equations
作者: Behzad Azmi, Jonas von der Heydt, Sergio Rodrigues
类别: math.OC, math.AP, math.NA, math.PR
发布日期: 2026-07-16
arXiv ID: 2607.14911v1
链接: http://arxiv.org/abs/2607.14911v1
摘要: We investigate finite-dimensional feedback stabilization for nonlinear nonautonomous stochastic parabolic equations driven by $Q$-Wiener, covering both additive and multiplicative perturbations. The control is given by a finite linear combination of localized indicator-type actuators whose supports are selected as part of the construction and may have arbitrarily small total measure. The feedback law is constructed by means of oblique projections onto suitable finite-dimensional subspaces. Within the variational Gelfand triple framework, we prove well-posedness of the closed-loop system under standard coercivity, growth, and global Lipschitz assumptions. By appropriately choosing the actuator configuration and feedback strength, we establish exponential mean-square stabilization of the stochastic dynamics and, for pure multiplicative noise, almost-sure stabilization. A fully discrete three-layer implementation complements the theoretical results. Numerical experiments illustrate the influence of number of actuators, noise intensity, and nonlinear effects on the closed-loop stabilization behavior.
8. 驯化随机梯度哈密顿蒙特卡洛
Tamed Stochastic Gradient Hamiltonian Monte Carlo
作者: Zhuoran Wang, Ying Zhang
类别: math.OC, math.NA, stat.ML
发布日期: 2026-07-16
arXiv ID: 2607.14862v1
链接: http://arxiv.org/abs/2607.14862v1
摘要: In this paper, we propose a novel tamed stochastic gradient Hamiltonian Monte Carlo (tSGHMC) algorithm for sampling and stochastic optimization problems with superlinearly growing stochastic gradients. Under a certain continuity in average condition and a strong convexity condition, we establish a non-asymptotic error bound in Wasserstein-2 distance for tSGHMC with the rate of convergence equal to $1/4$. Then, we derive an upper estimate for the associated expected excess risk, which provides a theoretical guarantee for the performance of tSGHMC. To illustrate the effectiveness of the proposed algorithm, we apply tSGHMC to practical examples, including a newsvendor problem and a Conditional Value-at-Risk minimization problem, using synthetic and real-world datasets. Numerical results support our theoretical findings. Furthermore, we compare tSGHMC with its first-order counterpart, namely, the tamed unadjusted stochastic Langevin algorithm. Simulation results demonstrate that tSGHMC achieves lower root mean square error and expected excess risk across a range of tasks.
9. 基于非残差估计子的二阶线性椭圆型偏微分方程自适应有限元方法的最优复杂度, 第一部分: 对称偏微分方程
Optimal complexity of adaptive FEM for second-order linear elliptic PDEs driven by non-residual estimators, Part I: Symmetric PDEs
作者: Philipp Bringmann, Aleksandar Dadic, Dario Ferloni, Gregor Gantner, Dirk Praetorius, Julian Streitberger
类别: math.NA
发布日期: 2026-07-16
arXiv ID: 2607.14820v1
链接: http://arxiv.org/abs/2607.14820v1
摘要: We consider adaptive finite element methods for symmetric second-order linear elliptic PDEs, where the adaptive algorithm steers the local mesh refinement as well as an iterative algebraic solver. Under abstract assumptions on the underlying a-posteriori error estimator and the solver, we prove that the usual adaptive algorithm leads to unconditional full R-linear convergence, independently of the user-chosen adaptivity parameters. For sufficiently small parameters, this guarantees optimal complexity in the sense that the decay rate of an appropriate quasi-error is optimal with respect to the overall computation cost (and hence time) measured in terms of the usual nonlinear approximation classes. Unlike available results in the literature, the main focus is on the analytical understanding of non-residual estimators like averaging-based estimators as proposed by Zienkiewicz and Zhu or estimators based on equilibrated fluxes.
10. 带小加性噪声的随机Korteweg-de Vries方程时间逼近的强误差分析
Strong error analysis of a temporal approximation for stochastic Korteweg-de Vries equation with small additive noise
作者: Jianbo Cui, Raffaele D’Ambrosio, Stefano Di Giovacchino, Liying Sun
类别: math.NA
发布日期: 2026-07-16
arXiv ID: 2607.14813v1
链接: http://arxiv.org/abs/2607.14813v1
摘要: We study strong temporal approximation of periodic stochastic Korteweg–de Vries equation driven by small additive (Q)-Wiener noise of amplitude (\mathcal O(\varepsilon)), (0<\varepsilon\ll1). Strong error analysis for temporal approximations of stochastic KdV is a challenging problem, due to the additional derivative term in the nonlinearity and thanks to the lack of suitable exponential moment bounds for the exact solutions. Exploiting the small-noise regime, we first decompose the solution into a deterministic KdV flow and a stochastic component; then we linearize the obtained stochastic equation and approximate the resulting equation by means of Fourier analytic techniques. Combining the small-noise linearization error, the discretization error of the linearized equation, and the deterministic temporal approximation error, we prove strong convergence rates of order (\mathcal O(\max(\varepsilon^2,τ,\varepsilonτ^{1/2}))) under (H^1)-regularity and (\mathcal O(\max(\varepsilon^2,τ))) under (H^2)-regularity, for the obtained approximation of the original stochastic KdV. To the best of our knowledge, these are the first explicit strong convergence rates shown for numerical time approximations of the stochastic KdV.
11. 演化隐式几何的静态-动态缩减非拟合边界代数方程方法
An unfitted boundary algebraic equation method with static-dynamic reduction for evolving implicit geometries
作者: Yizhen Huang, Qing Xia
类别: math.NA
发布日期: 2026-07-16
arXiv ID: 2607.14808v1
链接: http://arxiv.org/abs/2607.14808v1
摘要: Repeated elliptic solves on domains with evolving boundaries arise in moving-interface simulation, design, and reactive navigation. Even when a fixed Cartesian grid avoids remeshing, rebuilding all boundary interactions for every configuration can limit the efficiency of repeated solves. We develop a static–dynamic boundary reduction for an unfitted lattice Green’s function method on prescribed moving planar domains. Like boundary integral and boundary element methods, the formulation reduces the problem to boundary-supported unknowns through a Green representation. Its construction, however, reverses the usual order: the Cartesian operator is discretized before the Green representation is formed, rather than representing the continuous problem first and then discretizing the boundary. This discretize-then-represent viewpoint avoids boundary meshes and singular quadrature. The method also separates interactions associated with stationary geometry from those affected by motion, reuses the invariant part throughout a simulation, and updates only couplings involving the changing boundary. Boundary conditions are imposed at true interface intersections, lattice-kernel data are reused, and the interior field is reconstructed by a fast sine-transform solver. The principal contribution is an implemented and validated update strategy for translating, deforming, appearing, and topology-changing obstacles.
12. 关于浅水线性化矩方程的双曲随机Galerkin投影
On Hyperbolic Stochastic Galerkin Projections of Shallow Water Linearised Moment Equations
作者: Safiere Kuijpers, Julian Koellermeier
类别: math.NA
发布日期: 2026-07-16
arXiv ID: 2607.14750v1
链接: http://arxiv.org/abs/2607.14750v1
摘要: In this work, we present an intrusive stochastic Galerkin formulation of the one-dimensional shallow water linearised moment equations expressed in conservative variables, using the pseudospectral product for generalised polynomial chaos expansions. The shallow water linearised moment equations constitute a hyperbolic system of partial differential equations with an arbitrary number of equations that enhance the accuracy of the standard shallow water equations. Without loss of generality, we assume for both the theoretical analysis and the simulations that the uncertain parameter is the friction coefficient. For the new stochastic Galerkin shallow water linearised moment equations, we derive an energy equation, analyse the hyperbolicity - since this property is not preserved by the stochastic Galerkin projection - and introduce a regularisation to ensure hyperbolicity for the linear case of the shallow water linearised moment equations. Through numerical tests, we demonstrate the accuracy of the new stochastic Galerkin formulation in comparison with a non-intrusive Monte Carlo method, showing that the stochastic Galerkin approach achieves comparable accuracy with significantly faster run times.
13. 基于牛顿法的大规模稀疏连续时间代数Riccati方程的混合精度迭代精化
Newton-Based Mixed Precision Iterative Refinement for Large-Scale Sparse Continuous-Time Algebraic Riccati Equations
作者: Juan Zhang, Yang Zhou
类别: math.NA
发布日期: 2026-07-16
arXiv ID: 2607.14742v1
链接: http://arxiv.org/abs/2607.14742v1
摘要: We propose a Newton-based mixed precision iterative refinement framework for solving large-scale sparse continuous-time algebraic Riccati equations (CAREs). The framework computes the initial approximation and the inner Lyapunov correction equations in lower precision, while evaluating residuals and updating the solution in higher precision. To handle indefinite residuals and Newton correction terms in low-rank form, we introduce factor decomposition procedures with truncation strategies that preserve positive semidefiniteness and control rank growth. A first-order rounding error analysis derives a residual recurrence for the refinement process and relates stable mixed precision refinement to a Lyapunov operator conditioning threshold governed by the unit roundoff of the lower precision inner solves. We then present a concrete ADI-based realization, using NLR-ADI for the initial CARE approximation and LR-ADI for the inner Lyapunov correction equations. Compared with dense Lyapunov correction implementations, this realization reduces the main computations to shifted linear solves and low-rank factor operations, and we provide a solver-dependent complexity analysis. Numerical experiments on dense CARE over a range of condition numbers illustrate the conditioning effect described by the error analysis, and experiments on large-scale sparse CAREs show that the mixed precision framework is faster than the full double precision implementation while maintaining the same level of accuracy.
14. 具有收敛性分析的非马尔可夫随机薛定谔方程的低秩层次框架
A low-rank hierarchical framework for the non-Markovian stochastic Schrödinger equation with convergence analysis
作者: Zhuohan Zhang, Zhenning Cai
类别: math.NA
发布日期: 2026-07-16
arXiv ID: 2607.14689v1
链接: http://arxiv.org/abs/2607.14689v1
摘要: We propose and analyze a novel numerical framework for the non-Markovian stochastic Schrödinger equation (NMSSE) based on a low-rank approximation of the bath correlation functions. By decomposing the memory kernel into a finite-dimensional representation, we derive a truncated system of hierarchical equations that effectively balances computational tractability with physical fidelity. A rigorous convergence analysis is established for the hierarchical framework under mild assumptions. We demonstrate that our formulation serves as a mathematical generalization of the Hierarchy of Pure States (HOPS), encompassing it as a special case while offering a more flexible representation of non-Markovian effects. Numerical experiments across several benchmark models are presented to illustrate the validity and efficacy of the proposed method.
15. 用于拓扑优化的轨迹感知流匹配
Trajectory-Aware Flow Matching for Topology Optimisation
作者: Shusheng Xiao, Jinshuai Bai, Hyogu Jeong, Yunfei Xi, Yilin Gui, YuanTong Gu
类别: cs.LG, math.NA
发布日期: 2026-07-16
arXiv ID: 2607.14652v1
链接: http://arxiv.org/abs/2607.14652v1
摘要: Topology optimisation (TO) often requires repeated finite element analysis and sensitivity-based material updates, which can be costly when multiple candidate designs are needed under varying physical and design conditions. Generative TO offers a route to rapid design exploration, but existing models may rely on adversarial training, long reverse-diffusion sampling, or external guidance to maintain structural feasibility and physical consistency. This study develops a flow matching-based topology optimisation (FMTO) framework for conditional topology generation. Linear FMTO is first formulated as an endpoint-based baseline by interpolating between a Gaussian source field and the BESO reference topology. To introduce mechanically meaningful intermediate states, a trajectory-aware FMTO formulation is proposed, where volume-fraction-indexed BESO states are used to construct the probability path and target velocity field. This incorporates physics-guided optimisation history into generative flow learning without adding inference-time optimisation. A path–velocity mismatch analysis explains why moderate trajectory weighting can improve generation stability, whereas excessive guidance may over-constrain the learned transport. Numerical examples show that FMTO generates diverse topology candidates with improved compliance-related performance, volume-fraction satisfaction, topology fidelity, and substantially fewer sampling steps than a diffusion-based baseline. Under limited training data, trajectory-aware FMTO achieves the best overall performance with a moderate trajectory weight. Studies on trajectory-anchor density and three-dimensional topology generation further demonstrate the influence of path design and the applicability of the proposed framework beyond two-dimensional problems.
16. 梯度增强的样条维度分解用于有限训练样本下的不确定性量化
Gradient-enhanced spline dimensional decomposition for uncertainty quantification with limited training samples
作者: Eunho Heo, Dongjin Lee
类别: math.NA
发布日期: 2026-07-16
arXiv ID: 2607.14615v1
链接: http://arxiv.org/abs/2607.14615v1
摘要: A spline dimensional decomposition (SDD) surrogate effectively represents high-dimensional engineering responses with localized features and complex nonlinearities in uncertainty quantification (UQ). However, limited training data can make coefficient estimation from function values severely ill-conditioned. We propose gradient-enhanced SDD (GE-SDD), which trains the surrogate using function values and partial derivatives. A diagonal row-weight matrix balances the function and derivative blocks by their Frobenius norms. We solve the balanced system through ridge regression in probability-weighted Sobolev coordinates and select the regularization parameter using grouped K-fold cross-validation to prevent information leakage. Mapping the solution back to the L2-orthonormal SDD basis preserves closed-form mean and variance estimates. We evaluate the proposed GE-SDD on a two-dimensional continuous exponential function, a linear dynamical system with three uncertain parameters, and a 30-dimensional 25-bar truss. GE-SDD is more accurate than standard SDD and uses gradients more robustly than gradient-enhanced Kriging. GE-SDD achieves a median NRMSE of 1.022% on the nonsmooth benchmark, compared with 8.731% for Kriging. For the truss, GE-SDD yields lower NRMSE and more accurate standard-deviation estimates than Kriging at moderate training sizes and above. Overall, the benefits of gradient augmentation depend on input dimension, basis resolution, training size, and the target UQ quantity.
17. 通过控制理论引导动态网络中心性
Steering dynamic network centrality via control theory
作者: Fabio Durastante, Beatrice Meini, Luca Saluzzi
类别: math.NA
发布日期: 2026-07-16
arXiv ID: 2607.14610v1
链接: http://arxiv.org/abs/2607.14610v1
摘要: Time-evolving networks, or temporal networks, play a crucial role in modeling dynamic interactions across various domains, including biology, social sciences, and information technology. Unlike static networks, these systems undergo continuous changes in topology and edge weights, influencing processes such as information flow, transportation efficiency, and neural activity. Understanding and controlling these networks are essential for predicting future behavior and optimizing dynamic processes. This work focuses on the problem of dynamic centrality, a measure of node importance in time-dependent networks. Specifically, we address how to steer network centrality to a desired state by making minimal modifications to the network structure. This problem is formulated as an optimal control problem for an ordinary differential equation, either matrix- or vector-based, where the control acts on network edges. The proposed framework generalizes centrality control problems studied in static networks and leverages the Pontryagin Maximum Principle for efficient solutions. For large-scale problems, the required matrix-function actions are approximated by Krylov-type techniques, avoiding the explicit formation of dense matrix functions. Numerical experiments on synthetic and real temporal networks show that the proposed framework can effectively steer receive centrality under prescribed control constraints.
18. 迎风嵌入边界SBP算子:面向笛卡尔网格下任意形状区域的新型高阶数值格式
Upwind embedded boundary SBP operators: New high order numerical schemes for arbitrarily shaped domains with Cartesian grids
作者: Conner Dailey
类别: gr-qc, math-ph, math.NA
发布日期: 2026-07-15
arXiv ID: 2607.14403v1
链接: http://arxiv.org/abs/2607.14403v1
摘要: Embedded boundary summation by parts (SBP) methods define finite differencing based derivative operators with the added feature that the boundary need not coincide with a grid cell, allowing a boundary to be embedded on a regular Cartesian grid. This is achieved by the introduction of interpolation/extrapolation operators that match the accuracy of the boundary closure. These methods have been used to perform black hole excision simulations on a domain with a spherical boundary embedded in a regular Cartesian grid, demonstrating their usefulness for nonlinear problems. In this work, new operators are derived using this embedded boundary framework to increase the order of accuracy of the interior and boundary closure while minimizing the boundary error. Additionally, these novel operators improve the spectral properties on the grid by generalizing to an upwind scheme that has better dispersion relation preserving properties compared to traditional SBP schemes for wave equations. These operators are tested with the curvilinear scalar wave equation on a 3D multiblock grid with an excision sphere embedded in the center block to demonstrate the robustness and accuracy of these novel embedded operators.
19. 使用循环神经网络在铁磁叠片铁心有限元仿真中计入磁滞与涡流
Accounting for Hysteresis and Eddy Currents in Finite Element Simulations of Ferromagnetic Laminated Cores using a Recurrent Neural Network
作者: Florent Purnode, Louis Denis, François Henrotte, Gilles Louppe, Christophe Geuzaine
类别: cs.CE, cs.AI, math.NA
发布日期: 2026-07-15
arXiv ID: 2607.14321v1
链接: http://arxiv.org/abs/2607.14321v1
摘要: Incorporating hysteresis and eddy currents into finite element simulations of laminated-core electrical machines is computationally challenging. Resolving the fields inside the laminations at each integration point and at every nonlinear iteration leads to computational costs several orders of magnitude higher than anhysteretic simulations, making such approaches impractical for design applications. Conversely, simplified models accounting only for magnetic saturation are becoming increasingly inadequate as electrical machine topologies and operating conditions grow in complexity. In this context, machine learning surrogate modeling has emerged as a promising alternative, offering efficient and accurate approximations of complex electromagnetic behaviors. In this paper, a recurrent neural network is trained as a surrogate of a laminated-core material model for an isotropic laminated core, and is integrated into realistic two-dimensional magnetodynamic finite element simulations based on a magnetic vector potential formulation. The proposed approach achieves excellent agreement with the reference laminated-core model while limiting the computational cost to about twice that of an anhysteretic simulation. By training the recurrent neural network on a sufficiently diverse set of artificially generated magnetic field sequences designed to mimic those encountered in electrical machine simulations, the proposed approach can be readily applied across a wide range of finite element simulations. Furthermore, the trained surrogate model is provided as a standalone component that can be easily incorporated into existing computational frameworks. It is publicly available at https://gitlab.onelab.info/getdp/lamnet.
20. NGMRES 收敛性分析及对收缩和非收缩迭代的加速证明
NGMRES convergence analysis and proof of acceleration for contractive and noncontractive iterations
作者: Y. He, L. Rebholz, M. Xiao
类别: math.NA
发布日期: 2026-07-15
arXiv ID: 2607.14268v1
链接: http://arxiv.org/abs/2607.14268v1
摘要: This paper gives the first convergence analysis and proof of acceleration for nonlinear GMRES (NGMRES) applied to contractive and noncontractive fixed point iterations (FPIs) for solving general nonlinear systems. Our main results are that in both the contractive and noncontractive cases, the ratio gain of the optimization problem is the mechanism responsible for accelerating (or enabling) convergence. Our analysis also reveals a second important quantity related to the optimization problem, which directly predicts the linear convergence rate at each iteration and proves it is at most 1; hence only higher order terms are responsible for NGMRES non-convergence. Numerical results for several challenging nonlinear test problems are given that illustrate the theory, show how the acceleration improves convergence, show that the quantity predicting the linear convergence rate is remarkably accurate and moreover can be useful for adaptively choosing NGMRES depth, show how restarts can improve convergence in noncontractive iterations, show how NGMRES is naturally suited for finding distinct solutions of a multi-solution PDE, and that NGMRES can perform better than Anderson acceleration when applied to superlinear FPIs.