Skip to main content

Solving Elliptic Equations with Brownian Motion: Bias Reduction and Temporal Difference Learning


The Feynman-Kac formula provides a way to understand solutions to elliptic partial differential equations in terms of expectations of continuous time Markov processes. This connection allows for the creation of numerical schemes for solutions based on samples of these Markov processes which have advantages over traditional numerical methods in some cases. However, naïve numerical implementations suffer from issues related to statistical bias and sampling efficiency. We present methods to discretize the stochastic process appearing in the Feynman-Kac formula that reduce the bias of the numerical scheme. We also propose using temporal difference learning to assemble information from random samples in a way that is more efficient than the traditional Monte Carlo method.

This is a preview of subscription content, access via your institution.

Fig. 1
Fig. 2
Fig. 3
Fig. 4
Fig. 5
Fig. 6

Materials Availability

The datasets generated during and/or analysed during the current study, as well as Python 3 implementations of Algorithms 1, 2, and 3, are available from the corresponding author on reasonable request.


  1. Note that the first authors of these two papers are two different people: Jiequn Han in Han et al. (2020a) and Jihun Han in Han et al. (2020b).


Download references


We gratefully acknowledge that this research was supported by the Fields Institute for Research in Mathematical Sciences. Its contents are solely the responsibility of the authors and do not necessarily represent the official views of the Institute. We acknowledge the support of the Natural Sciences and Engineering Research Council of Canada (NSERC): RGPIN-2019-06946 for ARS and PDF-502287-2017 for MN. Thank you to the anonymous reviewer for the helpful comments.

Author information

Authors and Affiliations


Corresponding author

Correspondence to Adam R Stinchcombe.

Additional information

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Appendix: Expected Exit Time of a Brownian Bridge: Proof of Eq. 7

Appendix: Expected Exit Time of a Brownian Bridge: Proof of Eq. 7

We establish the following inequality about the exit time of a Brownian bridge: Let 0 < a < x. Consider a Brownian bridge with initial position B(0) = 0 and final position B(Δτ) = x. Let \(T_{a} = \inf \{t>0: B(t) > a\}\) be the first time the Brownian bridge crosses a barrier at a. Then, we have that \(\mathbb {E}[T_{a}]\) obeys the inequality:

$$ \frac{1}{1 +x^{-2}{\varDelta} \tau } \leq \frac{\mathbb{E}[T_{a}]}{\frac{a}{x}{\varDelta} \tau} \leq 1. $$

In our setting, Eq. 7 follows immediately from this fact by taking the barrier \(a = |\rho _{\partial {\varOmega }}(\vec B_{\text {old}})|\) and the final position \(x = {\varDelta }\rho = \rho _{\partial {\varOmega }}(\vec B_{\text {new}}) - \rho _{\partial {\varOmega }}(\vec B_{\text {old}})\).

To prove Eq. 13, we use the probability density of Ta from Eq. 5, to find that \(\mathbb {E}{\left [T_{a}\right ]}\) is given by

$$ \begin{array}{@{}rcl@{}} \mathbb{E}\left[T_{a}\right] &=& a{\int}_{0}^{{\varDelta}\tau}\frac{\sqrt{{\varDelta} \tau}}{\sqrt{2\pi}t^{1/2}({\varDelta}\tau - t)^{1/2}}\exp\left( \frac{x^{2}}{2{\varDelta} \tau}-\frac{(a-x)^{2}}{2({\varDelta}\tau - t)} - \frac{a^{2}}{2t}\right)~\ d t \\ & =& a {\int}_{0}^{{\varDelta}\tau}\rho\left( t,a \right)~\ d t = a\mathbb{E}\left[L_{a}\right], \end{array} $$

where \(\rho \left (t,a \right )\) denotes the probability density of the Brownian bridge to be at B(t) = a at time t, and La is the local time at a of this Brownian bridge. The probability density for this local time has an explicit formula from Equation (3) in Pitman (1999), namely,

$$\mathbb{P}(L_{a} > y) = \exp\left( -\frac{1}{2{\varDelta}\tau}\left( (|a| + |x-a| + y)^{2} - x^{2}\right)\right).$$

For 0 < a < x, we have |a| + |xa| = x, which yields

$$ \mathbb{E}\left[T_{a}\right] = a{\int}_{0}^{\infty} \exp\left( -\frac{1}{2{\varDelta}\tau}\left( 2xy + y^{2}\right)\right)~\ d y. $$

Finally, we can compute by a change of variable that

$$ \begin{array}{@{}rcl@{}} \frac{\mathbb{E}\left[T_{a}\right]}{\frac{a}{x}{\varDelta}\tau} &=& \frac{x}{{\varDelta}\tau}{\int}_{0}^{\infty} \exp\left( -\frac{1}{2{\varDelta}\tau}\left( 2xy + y^{2}\right)\right)~\ d y \\ & =& \sqrt{2\pi}x\exp\left( \frac{x^{2}}{2{\varDelta}\tau}\right){\int}_{0}^{\infty} \frac{1}{\sqrt{2\pi}{\varDelta}\tau}\exp\left( -\frac{1}{2}\left( \frac{y + x}{{\varDelta}\tau}\right)^{2}\right)~\ d y \\ & =& \sqrt{2\pi}\frac{x}{\sqrt{{\varDelta} \tau}}\exp\left( \frac{x^{2}}{2{\varDelta}\tau}\right)\mathbb{P}\left( X > \frac{x}{\sqrt{{\varDelta}\tau}}\right), \textrm{ where } X\sim \mathcal{N}(0,1). \end{array} $$

The Mill’s ratio inequality from Lemma 12.9 in Mörters and Peres (2012), which holds for all c > 0, gives

$$ \frac{1}{\sqrt{2\pi}}\frac{1}{c+c^{-1}} e^{-c^{2} / 2} \leq \mathbb{P}(X > c) \leq \frac{1}{\sqrt{2\pi}} \frac{1}{c} e^{-c^{2} / 2} . $$

This gives the desired result of Eq. 13 by setting \(c = x/{\sqrt {{\varDelta } \tau }}\).

Rights and permissions

Reprints and Permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Martin, C., Zhang, H., Costacurta, J. et al. Solving Elliptic Equations with Brownian Motion: Bias Reduction and Temporal Difference Learning. Methodol Comput Appl Probab 24, 1603–1626 (2022).

Download citation

  • Received:

  • Revised:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI:


Mathematics Subject Classification (2010)