skip to main content
DOE PAGES title logo U.S. Department of Energy
Office of Scientific and Technical Information

Title: Solving high-dimensional partial differential equations using deep learning

Abstract

Developing algorithms for solving high-dimensional partial differential equations (PDEs) has been an exceedingly difficult task for a long time, due to the notoriously difficult problem known as the “curse of dimensionality.” This paper introduces a deep learning-based approach that can handle general high-dimensional parabolic PDEs. To this end, the PDEs are reformulated using backward stochastic differential equations and the gradient of the unknown solution is approximated by neural networks, very much in the spirit of deep reinforcement learning with the gradient acting as the policy function. Numerical results on examples including the nonlinear Black–Scholes equation, the Hamilton–Jacobi–Bellman equation, and the Allen–Cahn equation suggest that the proposed algorithm is quite effective in high dimensions, in terms of both accuracy and cost. Furthermore, this opens up possibilities in economics, finance, operational research, and physics, by considering all participating agents, assets, resources, or particles together at the same time, instead of making ad hoc assumptions on their interrelationships.

Authors:
ORCiD logo [1];  [2];  [3]
  1. Princeton Univ., Princeton, NJ (United States)
  2. ETH Zurich, Zurich (Switzerland)
  3. Princeton Univ., Princeton, NJ (United States); Beijing Inst. of Big Data Research, Beijing (China)
Publication Date:
Research Org.:
Princeton Univ., NJ (United States)
Sponsoring Org.:
USDOE Office of Science (SC)
OSTI Identifier:
1540276
Grant/Contract Number:  
SC0009248
Resource Type:
Accepted Manuscript
Journal Name:
Proceedings of the National Academy of Sciences of the United States of America
Additional Journal Information:
Journal Volume: 115; Journal Issue: 34; Journal ID: ISSN 0027-8424
Publisher:
National Academy of Sciences
Country of Publication:
United States
Language:
English
Subject:
97 MATHEMATICS AND COMPUTING; Science & Technology; Other Topics; partial differential equations; backward stochastic differential equations; high dimension; deep learning; Feynman–Kac

Citation Formats

Han, Jiequn, Jentzen, Arnulf, and E, Weinan. Solving high-dimensional partial differential equations using deep learning. United States: N. p., 2018. Web. doi:10.1073/pnas.1718942115.
Han, Jiequn, Jentzen, Arnulf, & E, Weinan. Solving high-dimensional partial differential equations using deep learning. United States. doi:https://doi.org/10.1073/pnas.1718942115
Han, Jiequn, Jentzen, Arnulf, and E, Weinan. Mon . "Solving high-dimensional partial differential equations using deep learning". United States. doi:https://doi.org/10.1073/pnas.1718942115. https://www.osti.gov/servlets/purl/1540276.
@article{osti_1540276,
title = {Solving high-dimensional partial differential equations using deep learning},
author = {Han, Jiequn and Jentzen, Arnulf and E, Weinan},
abstractNote = {Developing algorithms for solving high-dimensional partial differential equations (PDEs) has been an exceedingly difficult task for a long time, due to the notoriously difficult problem known as the “curse of dimensionality.” This paper introduces a deep learning-based approach that can handle general high-dimensional parabolic PDEs. To this end, the PDEs are reformulated using backward stochastic differential equations and the gradient of the unknown solution is approximated by neural networks, very much in the spirit of deep reinforcement learning with the gradient acting as the policy function. Numerical results on examples including the nonlinear Black–Scholes equation, the Hamilton–Jacobi–Bellman equation, and the Allen–Cahn equation suggest that the proposed algorithm is quite effective in high dimensions, in terms of both accuracy and cost. Furthermore, this opens up possibilities in economics, finance, operational research, and physics, by considering all participating agents, assets, resources, or particles together at the same time, instead of making ad hoc assumptions on their interrelationships.},
doi = {10.1073/pnas.1718942115},
journal = {Proceedings of the National Academy of Sciences of the United States of America},
number = 34,
volume = 115,
place = {United States},
year = {2018},
month = {8}
}

Journal Article:
Free Publicly Available Full Text
Publisher's Version of Record

Citation Metrics:
Cited by: 49 works
Citation information provided by
Web of Science

Save / Share:

Works referenced in this record:

Option Pricing and Replication with Transactions Costs
journal, December 1985


A numerical algorithm for a class of BSDEs via the branching process
journal, February 2014

  • Henry-Labordère, Pierre; Tan, Xiaolu; Touzi, Nizar
  • Stochastic Processes and their Applications, Vol. 124, Issue 2
  • DOI: 10.1016/j.spa.2013.10.005

Forward-backward stochastic differential equations and quasilinear parabolic PDEs
journal, May 1999

  • Pardoux, Etienne; Tang, Shanjian
  • Probability Theory and Related Fields, Vol. 114, Issue 2
  • DOI: 10.1007/s004409970001

Deep learning
journal, May 2015

  • LeCun, Yann; Bengio, Yoshua; Hinton, Geoffrey
  • Nature, Vol. 521, Issue 7553
  • DOI: 10.1038/nature14539

Algorithms for overcoming the curse of dimensionality for certain Hamilton–Jacobi equations arising in control theory and elsewhere
journal, September 2016


A Primal-Dual Algorithm for Bsdes: a Primal-Dual Algorithm for Bsdes
journal, June 2015

  • Bender, Christian; Schweizer, Nikolaus; Zhuo, Jia
  • Mathematical Finance, Vol. 27, Issue 3
  • DOI: 10.1111/mafi.12100

The Pricing of Options and Corporate Liabilities
journal, May 1973

  • Black, Fischer; Scholes, Myron
  • Journal of Political Economy, Vol. 81, Issue 3
  • DOI: 10.1086/260062

Recursive valuation of defaultable securities and the timing of resolution of uncertainty
journal, November 1996

  • Duffie, Darrell; Schroder, Mark; Skiadas, Costis
  • The Annals of Applied Probability, Vol. 6, Issue 4
  • DOI: 10.1214/aoap/1035463324

Mastering the game of Go with deep neural networks and tree search
journal, January 2016

  • Silver, David; Huang, Aja; Maddison, Chris J.
  • Nature, Vol. 529, Issue 7587
  • DOI: 10.1038/nature16961

Adaptive importance sampling in least-squares Monte Carlo algorithms for backward stochastic differential equations
journal, April 2017


Approximation theory of the MLP model in neural networks
journal, January 1999


Pricing and hedging derivative securities in markets with uncertain volatilities
journal, June 1995


The randomized information complexity of elliptic PDE
journal, April 2006


Option Pricing with Differential Interest Rates
journal, April 1995


Deep Neural Networks for Acoustic Modeling in Speech Recognition: The Shared Views of Four Research Groups
journal, November 2012


Counterparty risk and Funding: the four Wings of the tva
journal, March 2013

  • CrÉPey, StÉPhane; Gerboud, RÉMi; Grbac, Zorana
  • International Journal of Theoretical and Applied Finance, Vol. 16, Issue 02
  • DOI: 10.1142/S0219024913500064

Backward Stochastic Differential Equations in Finance
journal, January 1997


Counterparty Risk Valuation: A Marked Branching Diffusion Approach
journal, January 2012


Implicit solution of uncertain volatility/transaction cost option pricing models with discretely observed barriers
journal, March 2001


    Works referencing / citing this record:

    Multiscale Modeling Meets Machine Learning: What Can We Learn?
    journal, February 2020

    • Peng, Grace C. Y.; Alber, Mark; Buganza Tepole, Adrian
    • Archives of Computational Methods in Engineering
    • DOI: 10.1007/s11831-020-09405-5

    A Deeper Look into ‘Deep Learning of Aftershock Patterns Following Large Earthquakes’: Illustrating First Principles in Neural Network Physical Interpretability
    book, May 2019

    • Mignan, Arnaud; Broccardo, Marco; Rojas, Ignacio
    • Advances in Computational Intelligence: 15th International Work-Conference on Artificial Neural Networks, IWANN 2019, Gran Canaria, Spain, June 12-14, 2019, Proceedings, Part I, p. 3-14
    • DOI: 10.1007/978-3-030-20521-8_1

    Neural networks catching up with finite differences in solving partial differential equations in higher dimensions
    journal, January 2020


    Solving partial differential equations by a supervised learning technique, applied for the reaction–diffusion equation
    journal, November 2019


    Data-driven acceleration of photonic simulations
    journal, December 2019


    Integrating machine learning and multiscale modeling—perspectives, challenges, and opportunities in the biological, biomedical, and behavioral sciences
    journal, November 2019

    • Alber, Mark; Buganza Tepole, Adrian; Cannon, William R.
    • npj Digital Medicine, Vol. 2, Issue 1
    • DOI: 10.1038/s41746-019-0193-y

    Solving Fokker-Planck equation using deep learning
    journal, January 2020

    • Xu, Yong; Zhang, Hao; Li, Yongge
    • Chaos: An Interdisciplinary Journal of Nonlinear Science, Vol. 30, Issue 1
    • DOI: 10.1063/1.5132840

    Using machine learning to predict extreme events in complex systems
    journal, December 2019

    • Qi, Di; Majda, Andrew J.
    • Proceedings of the National Academy of Sciences, Vol. 117, Issue 1
    • DOI: 10.1073/pnas.1917285117

    Machine-learning solver for modified diffusion equations
    journal, November 2018


    Numerical Solution of a Class of Nonlinear Partial Differential Equations by Using Barycentric Interpolation Collocation Method
    journal, December 2018

    • Wu, Hongchun; Wang, Yulan; Zhang, Wei
    • Mathematical Problems in Engineering, Vol. 2018
    • DOI: 10.1155/2018/7260346

    Variational Monte Carlo -- Bridging concepts of machine learning and high dimensional partial differential equations
    text, January 2018