SPLIT-PINN: Separable probability learning technique via physics-informed neural networks for high-dimensional probabilistic modeling
Source abstract
Abstract
We present a probabilistic modeling framework for incorporating small-scale spatial heterogeneity into macroscopic descriptions of material behavior for polycrystalline metallic materials. Spatially heterogeneous material state fields are represented using probability density functions (PDFs), providing a principled statistical description of microstructural variability and state evolution across different computational polycrystalline realizations. The framework is built on the inverse identification of a probabilistic transport model, formulated as a Liouville equation with an unknown drift term. To enable accurate, stable, and interpretable inference of this drift field in high-dimensional, transport-dominated settings, we develop a Separable Probability Learning Technique via Physics-Informed Neural Networks (SPLIT-PINN). This method incorporates a marginal-correction drift decomposition, orthogonality constraints, and residual-based adaptive training to enhance well-posedness, numerical stability, and physical consistency without imposing restrictive parametric assumptions. Using SPLIT-PINN, the drift field governing the temporal evolution of joint state PDFs is inferred directly from data. After benchmark validation, the framework is applied to physical computational datasets describing the evolution of polycrystalline microstructural states, including von Mises stress, dislocation density, and equivalent plastic strain rate. The learned Liouville model, trained on a single dataset, is subsequently used in forward predictions of the temporal evolution of joint and marginal PDFs for multiple unseen polycrystal realizations. Quantitative comparisons with reference PDFs demonstrate that the proposed framework yields accurate and robust probabilistic predictions and generalizes effectively across datasets.
Source: openaire · Quality: candidate unverified · Source record ↗
Research context
OpenAlex topics: Machine Learning in Materials Science · Model Reduction and Neural Networks · Gaussian Processes and Bayesian Inference
SciVX discovery tags are terminology-based. OpenAlex topics are source-provided classifications.
References (72)
Show source-provided references
1. Enabling strain hardening simulations with dislocation dynamics10.1088/0965-0393/15/6/001 ↗
2. Gpu-accelerated dislocation dynamics using subcycling time-integration10.1088/1361-651X/ab3a03 ↗
3. Learning dislocation dynamics mobility laws from large-scale MD simulations10.1038/s41524-024-01378-4 ↗
4. Enhanced mobility of dislocation network nodes and its effect on dislocation multiplication and strain hardening10.1016/j.actamat.2024.119884 ↗
5. Multi-objective loss balancing for physics-informed deep learning10.1016/j.cma.2025.117914 ↗
6. Chaotic motion and stochastic excitation10.1007/BF00044984 ↗
7. Local micro-mechanical stress conditions leading to pore nucleation during dynamic loading10.1016/j.ijplas.2020.102903 ↗
8. Runge-kutta methods10.4249/scholarpedia.3147 ↗
9. Generalized thermodynamics and fokker-planck equations: applications to stellar dynamics and two-dimensional turbulence10.1103/PhysRevE.68.036108 ↗
10. Stochastic response analysis of a spar-type FOWT subjected to extreme waves by a novel filter wave model and the DR-PDEE
11. Globally-evolving-based generalized density evolution equation for nonlinear systems involving randomness from both system parameters and excitations10.1098/rspa.2022.0356 ↗
12. Solving inverse stochastic problems from discrete particle observations using the fokker–planck equation and physics-informed neural networks10.1137/20M1360153 ↗
13. Anomalous plasticity of body-centered-cubic crystals with non-Schmid effect10.1016/j.ijsolstr.2018.01.029 ↗
14. Numerical methods for high-dimensional probability density function equations10.1016/j.jcp.2015.10.030 ↗
15. The loss surfaces of multilayer networks
16. Physics-informed neural networks with adaptive localized artificial viscosity10.1016/j.jcp.2023.112265 ↗
17. Scientific machine learning through physics–informed neural networks: where we are and what’s next10.1007/s10915-022-01939-z ↗
18. Limiting velocities and transonic dislocations in mg10.1016/j.commatsci.2022.111786 ↗
19. Physical analyses of crystal plasticity by DD simulations10.1016/j.scriptamat.2005.10.066 ↗
20. Numerical methods for kinetic equations10.1017/S0962492914000063 ↗
21. Attribution of heterogeneous stress distributions in low-grain polycrystals under conditions leading to damage10.1016/j.ijplas.2025.104258 ↗
22. The liouville equation and its potential usefulness for the prediction of forecast skill. part i: theory10.1175/1520-0493(1994)122<0703:TLEAIP>2.0.CO;2 ↗
23. Fesser, L., D’Amico-Wong, L., Qiu, R., 2023. Understanding and mitigating extrapolation failures in physics-informed neural networks. arxiv preprint arXiv: 2306.09478.
24. Description of a turbulent cascade by a fokker-planck equation10.1103/PhysRevLett.78.863 ↗
25. Gardiner · 2009 · Vol. 4
26. Gardiner · 2004 · Vol. 3
27. A digital representation environment for the analysis of microstructure in 3d10.1186/2193-9772-3-5 ↗
28. Multiscale modeling of plastic deformation of molybdenum and tungsten: I. atomistic studies of the core structure and glide of 1/2<111> screw dislocations at 0k10.1016/j.actamat.2008.07.018 ↗
29. Multiscale modeling of plastic deformation of molybdenum and tungsten: II. yield criterion for single crystals based on atomistic studies of glide of 1/2<111> screw dislocations10.1016/j.actamat.2008.07.037 ↗
30. Hairer · 1993
31. Solving high-dimensional partial differential equations using deep learning10.1073/pnas.1718942115 ↗
32. Multi-output multi-physics-informed neural network for learning dimension-reduced probability density evolution equation with unknown spatio-temporal-dependent coefficients10.1016/j.ymssp.2024.111683 ↗
33. Dr-pde-net: a time-varying inverse multi-physics-informed neural network paradigm for solving dimension-reduced probability density evolution equation in noisy data regimes10.1016/j.cma.2025.118059 ↗
34. Pc-srgan: physically consistent super-resolution generative adversarial network for general transient simulations10.1109/TPAMI.2025.3596647 ↗
35. A comparative study of shear band tracking strategies in three-dimensional finite elements with embedded weak discontinuities10.1016/j.finel.2018.11.001 ↗
36. Three-dimensional explicit finite element formulation for shear localization with global tracking of embedded weak discontinuities10.1016/j.cma.2019.05.011 ↗
37. Inverse and ill-posed problems: theory and applications
38. Kalidindi · 2015
39. Kalnay · 2003
40. Physics-informed machine learning10.1038/s42254-021-00314-5 ↗
41. Khalil · 2002 · Vol. 3
42. Lasota · 2013 · Vol. 97
43. Deformation, dislocation evolution and the non-schmid effect in body-centered-cubic single- and polycrystal tantalum10.1016/j.ijplas.2023.103529 ↗
44. LeVeque · 2002 · Vol. 31
45. Probability density evolution method: background, significance and recent developments10.1016/j.probengmech.2015.09.013 ↗
46. The principle of preservation of probability and the generalized density evolution equation10.1016/j.strusafe.2006.08.001 ↗
47. Strain localization and dynamic recrystallization in polycrystalline metals: thermodynamic theory and simulation framework10.1016/j.ijplas.2019.03.005 ↗
48. Probabilistic structural dynamics
49. Discontinuity computing using physics-informed neural networks10.1007/s10915-023-02412-1 ↗
50. The role of collinear interaction in dislocation-induced hardening10.1126/science.1085477 ↗
51. Dislocation strengthening in FCC metals and in BCC metals at high temperatures10.1016/j.actamat.2016.12.040 ↗
52. Estimates on the generalization error of physics-informed neural networks for approximating a class of inverse problems for PDEs10.1093/imanum/drab032 ↗
53. Estimates on the generalization error of physics-informed neural networks for approximating PDEs10.1093/imanum/drab093 ↗
54. Modeling and simulation framework for dynamic strain localization in elasto-viscoplastic metallic materials subject to large deformations10.1016/j.ijplas.2016.09.009 ↗
55. Pavliotis · 2008
56. Physics-informed neural network reconciles australian displacements and tectonic stresses10.1038/s41598-023-50759-0 ↗
57. Slip tendency analysis from sparse stress and satellite data using physics-guided deep neural networks10.1029/2024GL109524 ↗
58. On the spectral bias of neural networks
59. Physics-informed neural networks: a deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations10.1016/j.jcp.2018.10.045 ↗
60. Fokker-planck equation
61. Statistical evaluation of microscale stress conditions leading to void nucleation in the weak shock regime10.1016/j.ijplas.2025.104318 ↗
62. Scalable algorithms for physics-informed neural and graph networks10.1017/dce.2022.24 ↗
63. On the construction and comparison of difference schemes10.1137/0705041 ↗
64. Variational assimilation of meteorological observations with the adjoint vorticity equation. i: theory10.1002/qj.49711347812 ↗
65. Tarantola · 2005
66. Microstructure and mechanical properties in quenched and tempered low-carbon steel
67. Algorithms for solving high dimensional PDEs: from nonlinear monte carlo to machine learning
68. A comprehensive study of non-adaptive and residual-based adaptive sampling for physics-informed neural networks10.1016/j.cma.2022.115671 ↗
69. Aggregated residual transformations for deep neural networks
70. Scale invariance in plastic flow of crystalline solids10.1080/00018730600583514 ↗
71. Data-driven statistical reduced-order modeling and quantification of polycrystal mechanics leading to porosity-based ductile damage10.1016/j.jmps.2023.105386 ↗
72. A comprehensive review of advances in physics-informed neural networks and their applications in complex fluid dynamics10.1063/5.0226562 ↗