DeepXDE is an open-source library for scientific machine learning and physics-informed learning, best known for popularising physics-informed neural networks (PINNs). Maintained by Lu Lu and contributors, it has earned more than 4,400 stars on GitHub and remains actively developed in 2026.
DeepXDE lets scientists and engineers solve forward and inverse ordinary and partial differential equations with neural networks that respect the governing physics. It implements a broad algorithm collection: PINNs for many problem types, fractional PINNs, DeepONets for learning operators, neural arbitrary polynomial chaos for stochastic PDEs, and hard-constraint variants for inverse design and topology optimisation. Unusually, it is backend-agnostic, running on TensorFlow, PyTorch, JAX and PaddlePaddle, so teams can keep their preferred framework. Extensive documentation, examples, published benchmark cases and an accompanying SIAM Review paper have made it a standard teaching and research tool in the scientific machine learning community.
The library is distributed through PyPI and conda-forge and has been applied across areas including fluid mechanics, materials science, geophysics and biomedical engineering. Its documentation includes beginner-friendly tutorials that start from a simple Poisson equation and progress to complex inverse problems, which lowers the entry barrier for scientists who are new to neural PDE solvers but already comfortable with classical numerical methods.
DeepXDE is free and open source under the LGPL-2.1 license. It installs from PyPI or conda-forge, and the full documentation is hosted on Read the Docs at no cost.
DeepXDE is for researchers, graduate students and engineers in computational science who solve differential equations: fluid dynamics, materials, geophysics and engineering design teams use it to prototype physics-informed models quickly instead of hand-writing training loops for every new PDE problem.