Abstract Physics informed neural networks (PINNs) are a deep learning approach designed to solve partial differential equations (PDEs). Accurately learning the initial conditions is crucial when employing PINNs to solve PDEs. However, simply adjusting weights and imposing hard constraints may not always lead to better learning of the initial conditions; sometimes it even makes it difficult for the neural networks to converge. To enhance the accuracy of PINNs in learning the initial conditions, this paper proposes a novel strategy named causally enhanced initial conditions (CEICs). This strategy works by embedding a new loss in the loss function: the loss is constructed by the derivative of the initial condition and the derivative of the neural network at the initial condition. Furthermore, to respect the causality in learning the derivative, a novel causality coefficient is introduced for the training when selecting multiple derivatives. Additionally, because CEICs can provide more accurate pseudo-labels in the first subdomain, they are compatible with the temporal-marching strategy. Experimental results demonstrate that CEICs outperform hard constraints and improve the overall accuracy of pre-training PINNs. For the 1D-Korteweg-de Vries, reaction and convection equations, the CEIC method proposed in this paper reduces the relative error by at least 60% compared to the previous methods.
Wenshu Zha(查文舒), Dongsheng Chen(陈东升), Daolun Li(李道伦), Luhang Shen(沈路航), and Enyuan Chen(陈恩源) Causally enhanced initial conditions: A novel soft constraints strategy for physics informed neural networks 2025 Chin. Phys. B 34 040701
[1] Lagaris I E, Likas A and Fotiadis D I 1998 IEEE Trans. Neural Netw. 9 987 [2] Long Z, Lu Y and Dong B 2019 J. Comput. Phys. 399 108925 [3] Liu Z, Yang Y and Cai Q 2019 Appl. Math. Mech.-Engl. Ed. 40 237 [4] Irignano J and Spiliopoulos K 2018 J. Comput. Phys. 375 1339 [5] Lu L, Jin P, Pang G, Zhang Z and Karniadakis G E 2021 Nat. Mach. Intell. 3 218 [6] Li Z, Kovachki N, Azizzadenesheli K, Liu B, Bhattacharya K, Stuart A and Anandkumar A 2021 International Conference on Learning Representations [7] Raissi M, Perdikaris P and Karniadakis G E 2019 J. Comput. Phys. 378 686 [8] Pang G, Lu L and Karniadakis G E 2019 SIAM J. Sci. Comput. 41 A2603 [9] Shen L H, Li D L, Zha W S, Li X and Liu X L 2022 J. Pet. Sci. Eng. 213 110460 [10] Mao Z and Meng X 2023 Appl. Math. Mech.-Engl. Ed. 44 1069 [11] Wang S F, Sankaran S and Perdikaris P 2024 Comput. Methods Appl. Mech. Eng. 421 116813 [12] Gao Z, Tang T, Yan L and Zhou T 2024 Commun. Appl. Math. Comput. 6 1720 [13] Lu L, Meng X, Mao Z and Karniadakis G E 2021 SIAM Review 63 208 [14] Wu C, Zhu M, Tan Q, Kartha Y and Lu L 2023 Comput. Methods Appl. Mech. Eng. 403 115671 [15] Yu J, Lu L, Meng X and Karniadakis G E 2022 Comput. Methods Appl. Mech. Eng. 393 114823 [16] Wang S, Teng Y and Perdikaris P 2021 SIAM J. Sci. Comput. 43 A3055 [17] Jacot A, Gabriel F and Hongler C 2018 Advances in Neural Information Processing Systems 31 8571 [18] Mattey R and Ghosh S 2022 Comput. Methods Appl. Mech. Eng. 390 114474 [19] Wight C L and Zhao J 2020 arXiv:2007.04542v1math.NA] [20] Krishnapriyan A, Gholami A, Zhe S, Kirby R and Mahoney M W 2021 Advances in neural information processing systems 34 26548 [21] Guo J, Yao Y, Wang H and Gu T 2023 J. Comput. Phys 489 112258 [22] Jagtap A D and Karniadakis G E 2020 Commun. Comput. Phys. 28 2002 [23] Wang S, Yu X and Perdikaris P 2022 J. Comput. Phys. 449 110768 [24] Penwarden M, Jagtap A D, Zhe S, Karniadakis G E and Kirby R M 2023 J. Comput. Phys. 493 112464 [25] Guo J, Wang H F, Gu S L and Hou C P 2024 Chin. Phys. B 33 050701 [26] Wang Q, Li D L and Zha W S 2025 Phys. Fluids 37 013607 [27] Zha W, Li X, Xing Y, He L and Li D L 2020 Advances in Geo-Energy Research 4 107 [28] Zha W, Li D and Lu Z 2018 Advances in Geo-Energy Research 2 218 [29] Liu J N, Hou Q Z,Wei J G and Sun Z W 2023 Chin. Phys. B 32 070702 [30] Tian S F, Li B and Zhao Z 2024 Chin. Phys. Lett. 41 030201 [31] Peng K Y, Jing Y, Zhang W and Li J 2023 Chin. Phys. B 32 040208
Altmetric calculates a score based on the online attention an article receives. Each coloured thread in the circle represents a different type of online attention. The number in the centre is the Altmetric score. Social media and mainstream news media are the main sources that calculate the score. Reference managers such as Mendeley are also tracked but do not contribute to the score. Older articles often score higher because they have had more time to get noticed. To account for this, Altmetric has included the context data for other articles of a similar age.
