高对比度椭圆界面的随机特征配点方法
Random Feature Collocation for High-Contrast Elliptic Interface Problems
摘要: 针对具有分片材料系数和给定界面跳跃条件的高对比度 H( div ) 型椭圆界面问题,研究共享输出系数的随机特征配点方法及其高对比度离散误差。引入取值为−1和1的分类变量,使同一组随机特征能够表示界面两侧不同的向量场,并统一离散区域方程、法向边界条件及两类界面跳跃。进一步引入辅助变量 p=α ∇⋅u ,构造一阶混合配点形式,并依据能量尺度对辅助关系残差施加 α 权重。二维数值结果表明,在 ρ=1 0 6 时,基本混合形式的相对 L 2 误差中位数为7.30 × 10−6,约为增广单网络二阶形式的1/61;但双网络二阶形式仍达到1.89 × 10−6。通量增强可显著降低原始界面通量残差,却会增大场误差。结果说明,方程改写的效果依赖于分片表示、残差权重和随机特征尺度,不能推广为所有高对比度问题的统一优势。
Abstract: This paper studies random feature collocation with shared output coefficients for high-contrast H( div ) -type elliptic interface problems with piecewise material coefficients and prescribed interface jumps, and analyzes its discretization error at high contrast. A categorical variable taking the values −1 and 1 is introduced so that one set of random features can represent different vector fields on the two sides of the interface; the subdomain equations, the normal boundary condition and the two interface jumps are then discretized in a unified way. An auxiliary variable p=α ∇⋅u is introduced to construct a first-order mixed collocation formulation, and the residual of the auxiliary relation is weighted by α according to the energy scale. Two-dimensional results show that, at ρ=1 0 6 , the median relative L 2 error of the basic mixed formulation is 7.30 × 10−6, about 1/61 of that of the augmented single-network second-order formulation, whereas the two-network second-order formulation still reaches 1.89 × 10−6. Flux enhancement substantially reduces the original interface flux residual but increases the field error. These results indicate that the effect of the equation reformulation depends on the piecewise representation, the residual weight and the random feature scale, and cannot be generalized into a uniform advantage for all high-contrast problems.
文章引用:王嘉伟. 高对比度椭圆界面的随机特征配点方法[J]. 应用数学进展, 2026, 15(11): 213-222. https://doi.org/10.12677/aam.2026.1511444

参考文献

[1] Hiptmair, R., Li, J. and Zou, J. (2010) Convergence Analysis of Finite Element Methods for H(div;Ω)-Elliptic Interface Problems. Journal of Numerical Mathematics, 18, 187-218.
https://doi.org/10.1515/jnum.2010.010
[2] Boffi, D., Brezzi, F. and Fortin, M. (2013) Mixed Finite Element Methods and Applications. Springer.
[3] Liu, H., Zhang, L., Zhang, X., et al. (2020) Interface-Penalty Finite Element Methods for Interface Problems in H1, H(curl), and H(div). Computer Methods in Applied Mechanics and Engineering, 367, Article 113137.
https://doi.org/10.1016/j.cma.2020.113137
[4] Li, R., Liu, Q. and Yang, F. (2023) A Reconstructed Discontinuous Approximation on Unfitted Meshes to H(curl) and H(div) Interface Problems. Computer Methods in Applied Mechanics and Engineering, 403, Article 115723.
https://doi.org/10.1016/j.cma.2022.115723
[5] Huang, G.B., Zhu, Q.Y. and Siew, C.K. (2006) Extreme Learning Machine: Theory and Applications. Neurocomputing, 70, 489-501.
https://doi.org/10.1016/j.neucom.2005.12.126
[6] Chen, J., Chi, X., E, W. and Yang, Z. (2022) Bridging Traditional and Machine Learning-Based Algorithms for Solving PDEs: The Random Feature Method. Journal of Machine Learning, 1, 268-298.
https://doi.org/10.4208/jml.220726
[7] Dong, S. and Li, Z. (2021) Local Extreme Learning Machines and Domain Decomposition for Solving Linear and Nonlinear Partial Differential Equations. Computer Methods in Applied Mechanics and Engineering, 387, Article 114129.
https://doi.org/10.1016/j.cma.2021.114129
[8] Dong, S. and Yang, J. (2022) On Computing the Hyperparameter of Extreme Learning Machines: Algorithm and Application to Computational PDEs, and Comparison with Classical and High-Order Finite Elements. Journal of Computational Physics, 463, Article 111290.
https://doi.org/10.1016/j.jcp.2022.111290
[9] Wang, Z. and Zhang, Z. (2020) A Mesh-Free Method for Interface Problems Using the Deep Learning Approach. Journal of Computational Physics, 400, Article 108963.
https://doi.org/10.1016/j.jcp.2019.108963
[10] He, C., Hu, X. and Mu, L. (2022) A Mesh-Free Method Using Piecewise Deep Neural Network for Elliptic Interface Problems. Journal of Computational and Applied Mathematics, 412, Article 114358.
https://doi.org/10.1016/j.cam.2022.114358
[11] Hu, W.F., Lin, T.S. and Lai, M.C. (2022) A Discontinuity Capturing Shallow Neural Network for Elliptic Interface Problems. Journal of Computational Physics, 469, Article 111576.
https://doi.org/10.1016/j.jcp.2022.111576
[12] Song, W., Chi, X., Yang, Z., Cheng, W. and Chen, J. (2026) Discontinuity-Capturing Random Feature Method for Interface Problems. Computer Methods in Applied Mechanics and Engineering, 453, Article 118841.
https://doi.org/10.1016/j.cma.2026.118841