1

Physics-Informed Machine Learning: A Survey

on Problems, Methods and Applications
Zhongkai Hao, Songming Liu, Yichi Zhang, Chengyang Ying, Yao Feng,
Hang Su, Jun Zhu

Abstract—Recent advances of data-driven machine learning have revolutionized fields like computer vision, reinforcement learning,
and many scientific and engineering domains. In many real-world and scientific problems, systems that generate data are governed by
physical laws. Recent work shows that it provides potential benefits for machine learning models by incorporating the physical prior and
collected data, which makes the intersection of machine learning and physics become a prevailing paradigm. By integrating the data
and mathematical physics models seamlessly, it can guide the machine learning model towards solutions that are physically plausible,
arXiv:2211.08064v2 [cs.LG] 7 Mar 2023

improving accuracy and efficiency even in uncertain and high-dimensional contexts. In this survey, we present this learning paradigm
called Physics-Informed Machine Learning (PIML) which is to build a model that leverages empirical data and available physical prior
knowledge to improve performance on a set of tasks that involve a physical mechanism. We systematically review the recent
development of physics-informed machine learning from three perspectives of machine learning tasks, representation of physical prior,
and methods for incorporating physical prior. We also propose several important open research problems based on the current trends
in the field. We argue that encoding different forms of physical prior into model architectures, optimizers, inference algorithms, and
significant domain-specific applications like inverse engineering design and robotic control is far from being fully explored in the field of
physics-informed machine learning. We believe that the interdisciplinary research of physics-informed machine learning will
significantly propel research progress, foster the creation of more effective machine learning models, and also offer invaluable
assistance in addressing long-standing problems in related disciplines.

Index Terms—Physics-Informed Machine Learning, AI for Science, PDE/ODE, Symmetry, Intuitive Physics

1 I NTRODUCTION of data. The powerful abstraction ability of deep neural

networks also motivates researchers to apply them on sci-
The paradigm of scientific research in recent decades has
entific problems in modeling physical systems. For example,
undergone a revolutionary change with the development of
AlphaFold 2 [7] has revolutionized the paradigm of protein
computer technology. Traditionally, researchers used theo-
structure prediction. Similarly, FourCastNet [8] has built an
retical derivation combined with experimental verification
ultra-large learning-based weather forecasting system that
to study natural phenomena. With the development of
surpasses traditional numerical forecasting systems. Deep
computational methods, a large number of methods based
Potential [9] proposed neural models for learning large-scale
on computer numerical simulation have been developed
molecular potential satisfying symmetry. The integration
to understand complex real systems. Nowadays, with the
of prior knowledge of physics, which represents a high-
automation and batching of scientific experiments, scientists
level abstraction of natural phenomena or human behaviors,
have accumulated a large amount of observational data. The
with data-driven machine learning models is becoming a
paradigm of (data-driven) machine learning is to understand and
new paradigm since it has the potential to facilitate novel
build models that leverage empirical data to improve performance
discoveries and solutions to challenges across a diverse
on some set of tasks [1]. It is an important research area to pro-
range of domains.
mote the development of modern science and engineering
Moreover, despite the impressive advancements of ma-
technology with the aid of learning from observational data
chine learning based models, there remain significant lim-
since we could extract a lot of information from data.
itations when deploying purely data-driven models in
As part of the remarkable progress of machine learning
real-world applications. In particular, data-driven machine
in recent years, deep neural networks [2] have achieved
learning models can suffer from several limitations such
milestone breakthroughs in the fields of computer vision
as a lack of robustness, interpretability, and adherence to
[3], natural language processing [4], speech processing [5],
physical constraints or commonsense reasoning. In com-
and reinforcement learning [6]. Their flexibility and scala-
puter vision, recognizing and understanding the geometry,
bility allow neural networks to be easily applied to many
shape, texture, and dynamics from images or videos can
different domains, as long as there is a sufficient amount
pose a significant challenge for deep neural networks, which
can lead to limitations in their ability to extrapolate beyond
• Zhongkai Hao, Songming Liu, Yichi Zhang, Chengyang Ying, Yao Feng, their training data. Additionally, such models have demon-
Hang Su, Jun Zhu are with Dept. of Comp. Sci. & Techn., Institute for AI,
BNRist Center, Tsinghua-Bosch Joint ML Center, Tsinghua University, strated suboptimal performance outside of their training
Email: {hzj21, liusm18, zyc22, ycy21, y-feng20}@mails.tsinghua.edu.cn, distribution [10] and are susceptible to adversarial attacks
{suhangss, dcszj}@mail.tsinghua.edu.cn, via human-imperceptible noise [11]. In deep reinforcement
• Jun Zhu is also with RealAI.
learning, an agent may learn to take actions that result in