Skip to main content

The Robin Mean Value Equation I: A Random Walk Approach to the Third Boundary Value Problem


We study the family of integral equations, called the Robin mean value equations (RMV), that are local averaged approximations to the Robin-Laplace boundary value problem (RL). When posed on \(\mathcal {C}^{1,1}\)-regular domains, we prove existence, uniqueness, equiboundedness and the comparison principle for solutions to (RMV). For the continuous right hand side of (RL), we show that solutions to (RMV) converge uniformly, in the limit of the vanishing radius of averaging, to the unique W2,p solution, which coincides with the unique viscosity solution of (RL). We further prove the lower bound on solutions to (RMV), which is consistent with the optimal lower bound for solutions to (RL). Our proofs employ martingale techniques, where (RMV) is interpreted as the dynamic programming principle along a suitable discrete stochastic process, interpolating between the reflecting and the stopped-at-exit Brownian walks.

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

Code Availability

Not applicable


  1. Antunovic, T., Peres, Y., Sheffield, S., Somersille, S.: Tug-of-war and infinity laplace equation with vanishing Neumann boundary condition. Commun. Partial. Differ. Equ. 37(10), 1839–1869 (2012)

    Article  MathSciNet  Google Scholar 

  2. Bass, R., Burdzy, K., Chen, Z.-Q.: On the Robin problem in fractal domains. Proc. Lond. Math. Soc. 96(2), 273–311 (2008)

    Article  MathSciNet  Google Scholar 

  3. Bass, R., Hsu, P.: Some potential theory for reflecting Brownian motion in Hölder and Lipschitz domains. Ann. Probab. 19(2), 486–508 (1991)

    Article  MathSciNet  Google Scholar 

  4. Benchérif-Madani, A., Pardoux, É.: A probabilistic formula for a poisson equation with Neumann boundary condition. Stoch. Anal. Appl. 27(4), 739–746 (2009)

    Article  MathSciNet  Google Scholar 

  5. Brezis, H., Cabrè, X.: Some simple nonlinear PDEs without solutions. Bollettino dell’Unione Matematica Italiana 8(1-B,2), 223–262 (1998)

    MathSciNet  MATH  Google Scholar 

  6. Bucur, D., Giacomini, A., Trebeschi, P.: The Robin-Laplacian problem on varying domains. Calc. Var. 55(6), 55–133 (2016). Partial Differential Equations

    Article  MathSciNet  Google Scholar 

  7. Brosamler, G.A.: A probabilistic solution of the Neumann problem. Mathematica Scandinavica 38(1), 137–147 (1976)

    Article  MathSciNet  Google Scholar 

  8. Charro, F., Garcia, J., Rossi, J.: A mixed problem for the infinity Laplacian via tug-of-war games. Calc. Var. 34(3), 307–320 (2009)

    Article  MathSciNet  Google Scholar 

  9. Crandall, M., Ishii, H., Lions, P.-L.: User’s guide to viscosity solutions of second order partial differential equations. Bull. Amer. Math. Soc. 27, 1–67 (1992)

    Article  MathSciNet  Google Scholar 

  10. Doob, J.L.: Classical Potential Theory and its Probabilistic Counterpart. Springer, New York (1984)

    Book  Google Scholar 

  11. Gilbarg, D., Trudinger, N.: Elliptic Partial Differential Equations of Second Order. Classics in Mathematics, 2nd edn. Springer, Berlin Heidelberg (2001)

    Book  Google Scholar 

  12. Ishii, H.: On the equivalence of two notions of weak solutions, viscosity solutions and distributional solutions. Funkcialaj Ekvacioj 38, 101–120 (1995)

    MathSciNet  MATH  Google Scholar 

  13. Knothe, H.: Contributions to the theory of convex bodies. Mich. Math. J. 4, 39–52 (1957)

    MathSciNet  MATH  Google Scholar 

  14. Lewicka, M., Manfredi, J., Ricciotti, D.: Random walks and random tug of war in the Heisenberg group. Mathematische Annalen 377, 797–846 (2020)

  15. Lewicka, M., Peres, Y.: Which domains have two-sided supporting unit spheres at every boundary point?. Expositiones Mathematicae 38(4), 548–558 (2020)

    Article  MathSciNet  Google Scholar 

  16. Lewicka, M., Peres, Y.: The Robin mean value equation II: asymptotic Hölder regularity submitted (2019)

  17. Lieberman, G.: Oblique Derivative Problems for Elliptic Equations. World Scientific Publishing (2013)

  18. Lucas, K.: Submanifolds of dimension n − 1 in En with normals satisfying a lipschitz condition, studies in eigenvalue problems. Technical Report 18, Dept of Mathematics, University of Kansas (1957)

  19. Manfredi, J., Parviainen, M., Rossi, J.: On the definition and properties of p-harmonious functions. Ann. Sc. Norm. Super. Pisa Cl. Sci. 11(2), 215–241 (2012)

    MathSciNet  MATH  Google Scholar 

  20. Morel, J.-M., Oswald, L.: A uniform formulation for hopf maximum principle. Unpublished preprint (1985)

  21. Nittka, R.: Quasilinear elliptic and parabolic robin problems on Lipschitz domains. Nonlinear Differ. Equ. Appl. 20(3), 1125–1155 (2013)

    Article  MathSciNet  Google Scholar 

  22. Papanicolaou, V.: The probabilistic solution of the third boundary value problem for second order elliptic equations. Probab. Theory Relat. Fields 87(1), 27–77 (1990)

    Article  MathSciNet  Google Scholar 

  23. Peres, Y., Schramm, O., Sheffield, S., Wilson, D.: Tug-of-war and the infinity Laplacian. J. Amer. Math. Soc. 22, 167–210 (2009)

    Article  MathSciNet  Google Scholar 

  24. Peres, Y., Sheffield, S.: Tug-of-war with noise: A game theoretic view of the p-Laplacian. Duke Math. J. 145(1), 91–120 (2008)

    Article  MathSciNet  Google Scholar 

  25. Rosenblatt, M.: Remarks on a multivariate transformation. Ann. Math. Stat. 23, 470–472 (1952)

    Article  MathSciNet  Google Scholar 

Download references


The authors are grateful to D. Bucur for bringing to their attention regularity questions related to the Robin problem, and to J. Manfredi for discussing viscosity solutions. M.L. acknowledges partial support from the NSF grant DMS-1613153. A large part of this work has been completed during M.L. visits to Microsoft Research in Redmond.


M.L. acknowledges partial support from the NSF grant DMS-1613153 and support through visits to MSR (Microsoft Research) in Redmond.

Author information

Authors and Affiliations


Corresponding author

Correspondence to Marta Lewicka.

Ethics declarations

Conflict of Interests


Additional information

Availability of data and material

Not applicable

Publisher’s Note

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

Appendix: : A Proof of Uniqueness of Viscosity Solutions to ??

Appendix: : A Proof of Uniqueness of Viscosity Solutions to ??

We first make an observation that relies on the assumed regularity of \(\partial {\mathcal {D}}\).

Lemma A.1

When \({\mathcal {D}}\subset {\mathbb {R}}^{N}\) satisfies the uniform outer supporting sphere condition:

$$ \text{for every} x\in\partial{\mathcal{D}} \text{exists} B_{r}(b)\subset{\mathbb{R}}^{n}\setminus \bar{\mathcal{D}} \quad \text{such that} \quad |x-b|=r, $$

with some radius r > 0, then the boundary requirements in Definition 4.1 (i) and (ii) can be reduced to: \(\langle p, \vec {n}(x)\rangle + \gamma u(x)\leq 0\) and \(\langle p, \vec {n}(x)\rangle + \gamma u(x)\geq 0\), respectively.


Let \(x\in \partial {\mathcal {D}}\) and \( (p,X)\in J_{\bar {\mathcal {D}}}^{2,+}u(x)\). For each large j, consider the jet:

$$ (p_{j}, X_{j})= \left( p-\frac{1}{j}\vec{n}(x), ~ X+r Id_{N} - (r+j)\vec{n}(x)^{\otimes 2}\right). $$

We claim that \((p_{j}, X_{j})\in J_{\bar {\mathcal {D}}}^{2,+}u(x)\). In this case, we have:

$$ - \text{trace} X_{j} = -\big(\text{trace} X + (N-1)r-j\big)> f(x), $$

so by (i) there must be:\(\displaystyle 0\geq \langle p_{j}, \vec {n}(x)\rangle +\gamma u(x) = \langle p, \vec {n}(x)\rangle +\gamma u(x)-\frac {1}{j}\). Passing to the limit with \(j\to \infty \) we get the claimed boundary condition. To show that (pj,Xj) is indeed a super-jet, let:

$$ \begin{array}{@{}rcl@{}} q_{j}(y-x) & = & \big\langle -\frac{1}{j}\vec{n}(x), y-x\big\rangle +\frac{1}{2}\big\langle r Id_{N} -(r+j)\vec{n}(x)^{\otimes 2} : (y-x)^{\otimes 2}\big\rangle \\ & = & -\frac{1}{j}\langle\vec{n}(x), y-x\rangle +\frac{r}{2} |(y-x)_{tan}|^{2} -\frac{j}{2}\langle \vec{n}(x), y-x\rangle^{2}. \end{array} $$

For \(y\in \bar {\mathcal {D}}\) satisfying \(\langle \vec {n}(x), y-x\rangle \leq 0\), we get: \(\displaystyle q_{j}(y-x) \geq |\langle \vec {n}(x), y-x\rangle |\Big (\frac {1}{j}-\frac {j}{2} |\langle \vec {n}(x), y-x\rangle |\Big )\geq 0,\) if |yx| is small enough. On the other other hand, for \(y\in \bar {\mathcal {D}}\) such that \(\langle \vec {n}(x), y-x\rangle \geq 0\), we get:

$$ \begin{array}{@{}rcl@{}} q_{j}(y-x) & \geq & -\frac{2}{j}\langle \vec{n}(x), y-x\rangle +\frac{r}{2} |(y-x)_{tan}|^{2}\\ & \geq & -\frac{2}{j}\big(r-\sqrt{r^{2}-|(y-x)_{tan}|^{2}}\big) + \frac{r}{2} |(y-x)_{tan}|^{2}\geq 0, \end{array} $$

as for small |yx| and j ≫ 1 there holds: \(\displaystyle r-\sqrt {r^{2}-|(y-x)_{tan}|^{2}}\leq \frac {1}{r}|(y-x)_{tan}|^{2}\leq \frac {jr}{4}|(y-x)_{tan}|^{2}.\) Thus, in both cases, the validity of Eq. ?? for (p,X), implies the same asymptotic bound for each (pj,Xj) with sufficiently large j. □

Lemma A.2

Assume the uniform outer supporting sphere condition (9.1) with radius r > 0. Then the Robin problem ?? with \(f\in \mathcal {C}(\bar {\mathcal {D}})\) has at most one viscosity solution.


1. Let u,v be two viscosity solutions to ??. We will prove that uv. In fact, the same analysis works when u is a viscosity sub-solution and v is a viscosity super-solution, in the sense of Definition 4.1 (i) and (ii), where u is assumed only to be upper-semicontinuous and v lower-semicontinuous, and where the jets sets \(J_{\bar {\mathcal {D}}}^{2,+}\) and \(J_{\bar {\mathcal {D}}}^{2,-}\) are replaced by their closures \(\bar {J}_{\bar {\mathcal {D}}}^{2,+}\) and \(\bar {J}_{\bar {\mathcal {D}}}^{2,-}\), respectively (see [9] for the details). Also, we recall that the requirements at boundary points in Definition 4.1 can be reduced as in Lemma 9.1, because of Eq. 9.1. The stated comparison principle is proved in three steps: by replacing u and v with strict sub- and super-solutions uδ and vδ, and by doubling the variables technique with an appropriate nonlinear corrector, separately in the cases when uδ and vδ achieves its maximum in \({\mathcal {D}}\) or on \(\partial {\mathcal {D}}\). We now sketch these arguments.

For a sufficiently large C ≫ 1 and each δ ≪ 1, define:

$$ u_{\delta}(x) = u(x) - \frac{\delta}{2}\big(|x|^{2}-C\big), \qquad v_{\delta}(x) = v(x) + \frac{\delta}{2}\big(|x|^{2}-C\big). $$

Assume that \((p,X)\in J_{\bar {\mathcal {D}}}^{2,+}u_{\delta }(x)\), which is equivalent to: \((p-\delta x, X-\delta Id_{N})\in J_{\bar {\mathcal {D}}}^{2,+}u(x)\). Then:

$$ \begin{array}{ll} -\text{trace} X\leq f(x)-N\delta \qquad\quad\text{when} x\in{\mathcal{D}}\\ \langle p, \vec{n}(x)\rangle +\gamma u_{\delta}(x)\leq -\delta \qquad\text{when} x\in\partial{\mathcal{D}}. \end{array} $$

Thus, each uδ is a strict sub-solution of ??. Similarly, each vδ is a strict super-solution, namely \((p,X)\in J_{\bar {\mathcal {D}}}^{2,-}v_{\delta }(x)\) implies:

$$ \begin{array}{ll} -\text{trace} X\geq f(x)+N\delta \qquad\text{when} x\in{\mathcal{D}}\\ \langle p, \vec{n}(x)\rangle +\gamma v_{\delta}(x)\geq \delta \quad\quad\text{when} x\in\partial{\mathcal{D}}. \end{array} $$

2. We will show that uδvδ in \(\bar {\mathcal {D}}\), for all δ ≪ 1. By contradiction, fix δ > 0 and assume that:

$$ \max_{\bar{\mathcal{D}}}\big(u_{\delta}-v_{\delta}\big)>0. $$

We first treat the case of \(\max \limits _{\bar {\mathcal {D}}}\big (u_{\delta }-v_{\delta }\big ) > \max \limits _{\partial \bar {\mathcal {D}}}\big (u_{\delta }-v_{\delta }\big )\). We apply [9, Proposition 3.7] to: Φ(x,y) = uδ(x) − vδ(y) and \({\Psi }(x,y) = \frac {1}{2}|x-y|^{2}\) and obtain a sequence \(\{(x_{\alpha }, y_{\alpha })\}_{\alpha \to \infty }\) of maximizers to Φ − αΨ on \(\bar {\mathcal {D}}\times \bar {\mathcal {D}}\) that converges to some diagonal element (z0,z0) such that z0 is a maximizer of uδvδ and thus \(z_{0}\in {\mathcal {D}}\). Applying [9, Theorem 3.2] to: w(x,y) = uδ(x) − vδ(y) and \(\phi (x,y) = \frac {\alpha }{2}|x-y|^{2}\), we further obtain sequences of matrices \(X_{\alpha }, Y_{\alpha }\in {\mathbb {R}}^{N\times N}_{\text {sym}}\) satisfying:

$$ \begin{array}{ll} \big(\alpha (x_{\alpha} - y_{\alpha}), X_{\alpha}\big) \in J_{{\mathcal{D}}}^{2,+}u_{\delta}(x_{\alpha}),\qquad \big(\alpha (x_{\alpha} - y_{\alpha}), Y_{\alpha}\big) \in J_{\mathcal{D}}^{2,-}v_{\delta}(y_{\alpha})\\ \left[\begin{array}{cc} X_{\alpha}&0\\ 0& -Y_{\alpha} \end{array}\right] \leq 3\alpha \left[\begin{array}{cc} Id_{N}& - Id_{N}\\ - Id_{N}& Id_{N} \end{array}\right]. \end{array} $$

The first two assertions above, together with (9.2), (9.3) yield:

$$ -\text{trace} (X_{\alpha} - Y_{\alpha})\leq f(x_{\alpha}) - f(y_{\alpha})-2N\delta\to -2\delta \quad \text{as} \alpha\to\infty, $$

which contradicts the last assertion in Eq. 9.4 that implies: trace(XαYα) ≤ 0.

3. We now treat the remaining case, namely that of:

$$ \max_{\bar{\mathcal{D}}}\big(u_{\delta}-v_{\delta}\big) = \big(u_{\delta}-v_{\delta}\big)(z_{0})>0 \quad \text{for some} z_{0}\in\partial{\mathcal{D}}. $$

Applying [9, Proposition 3.7] to:

$$ {\Phi}(x,y) = u_{\delta}(x) - v_{\delta}(y) - \gamma u_{\delta}(z_{0})\langle y-x, \vec{n}(z_{0})\rangle - |x-z_{0}|^{2},\qquad {\Psi}(x,y) = \frac{1}{2}|x-y|^{2}, $$

we obtain a sequence \(\{(x_{\alpha }, y_{\alpha })\}_{\alpha \to \infty }\) of maximizers to Φ − αΨ that converges to some (z,z), where z is a maximizer of Φ(z,z) = uδ(z) − vδ(z) −|zz0|4. Hence there must be z = z0. Also:

$$ \alpha|x_{\alpha}-y_{\alpha}|^{2}\to 0\quad \text{as}\alpha\to\infty. $$

We now apply [9, Theorem 3.2] to:

$$ w(x,y) = u_{\delta}(x) - v_{\delta}(y), \qquad \phi(x,y) = \frac{\alpha}{2}|x-y|^{2} +\gamma u_{\delta}(z_{0})\langle y-x, \vec{n}(z_{0})\rangle, $$

which yields existence of sequences of matrices \(X_{\alpha }, Y_{\alpha }\in {\mathbb {R}}^{N\times N}_{\text {sym}}\) satisfying:

$$ \begin{array}{ll} \big(\alpha (x_{\alpha} - y_{\alpha}) - \gamma u_{\delta}(z_{0})\vec{n}(z_{0})+4|x_{\alpha}-z_{0}|^{2}(x_{\alpha}-z_{0}), X_{\alpha}\big) \in J_{\bar{\mathcal{D}}}^{2,+}u_{\delta}(x_{\alpha}),\\ \big(\alpha (x_{\alpha}- y_{\alpha}) - \gamma u_{\delta}(z_{0})\vec{n}(z_{0}), Y_{\alpha}\big) \in J_{\bar{\mathcal{D}}}^{2,-}v_{\delta}(y_{\alpha}), \\ \left[\begin{array}{cc} X_{\alpha}&0 \\ 0& -Y_{\alpha} \end{array}\right] \leq \nabla^{2}\phi(x_{\alpha}, y_{\alpha}) +\frac{1}{\alpha}\nabla^{2}\phi(x_{\alpha}, y_{\alpha})^{2}. \end{array} $$

Since \(\nabla ^{2}\phi (x_{\alpha }, y_{\alpha }) = \alpha \left [\begin {array}{cc} Id_{N}& - Id_{N}\\ - Id_{N}& Id_{N} \end {array}\right ] + \mathcal {O}\big (|x_{\alpha }-z_{0}|^{2}\big )\), the last assertion above implies:

$$ \text{trace} (X_{\alpha}-Y_{\alpha})\leq \mathcal{O}\big(|x_{\alpha}-z_{0}|^{2} + \frac{1}{\alpha}|x_{\alpha}-z_{0}|^{4}\big) \quad \to 0\quad \text{as }\alpha\to\infty. $$

Note that for large α ≫ 1 there must be \(x_{\alpha }, y_{\alpha }\in {\mathcal {D}}\). Indeed, if \(x_{\alpha }\in \partial {\mathcal {D}}\) then Eq. 9.2 and the first assertion in Eq. 9.7 yield the contradiction in:

$$ \begin{array}{@{}rcl@{}} -\delta & \geq & \alpha \langle x_{\alpha} - y_{\alpha}, \vec{n}(x_{\alpha})\rangle - \gamma u_{\delta}(z_{0})\langle \vec{n}(z_{0}), \vec{n}(x_{\alpha})\rangle\\ &&+ 4|x_{\alpha} - z_{0}|^{2}\langle x_{\alpha}-z_{0}, \vec{n}(x_{\alpha})\rangle + \gamma u_{\delta}(x_{\alpha}) \geq -\frac{\alpha}{2r}|x_{\alpha} - y_{\alpha}|^{2} \\&&+ \gamma u_{\delta}(z_{0})\langle \vec{n}(z_{0}), \vec{n}(x_{\alpha})\rangle + \mathcal{O}\big(|x_{\alpha} - z_{0}|^{3}\big) + \gamma u_{\delta}(x_{\alpha}) \\ & \to& 0\quad \text{as }\alpha\to\infty, \end{array} $$

where we used (9.1) for the bound \( \langle y_{\alpha } - x_{\alpha }, \vec {n}(x_{\alpha })\rangle \leq \frac {1}{2r}|x_{\alpha } - y_{\alpha }|^{2}\), followed by Eq. 9.8. Similarly, if \(y_{\alpha }\in \partial {\mathcal {D}}\) then Eq. 9.3 and the second assertion in Eq. 9.7 brings the contradiction with Eq. 9.5, as:

$$ \begin{array}{@{}rcl@{}} \delta & \leq & \alpha \langle x_{\alpha} - y_{\alpha}, \vec{n}(y_{\alpha})\rangle - \gamma u_{\delta}(z_{0})\langle \vec{n}(z_{0}), \vec{n}(y_{\alpha})\rangle + \gamma v_{\delta}(y_{\alpha})\\ & \leq & \frac{\alpha}{2r}|x_{\alpha} - y_{\alpha}|^{2} - \gamma u_{\delta}(z_{0})\langle \vec{n}(z_{0}), \vec{n}(y_{\alpha})\rangle +\gamma v_{\delta}(y_{\alpha})\\ &\to& \gamma\big(v_{\delta}(z_{0}) - u_{\delta}(z_{0})\big)\quad \text{as} \alpha\to\infty. \end{array} $$

The fact of \(x_{\alpha }, y_{\alpha }\in {\mathcal {D}}\) established, we use Eq. 9.7 together with Eqs. 9.29.3, to obtain:

$$ -\text{trace} \big(X_{\alpha} - Y_{\alpha}\big)\leq f(x_{\alpha})-f(y_{\alpha}) - 2N\delta \to -2N\delta\quad \text{as} \alpha\to\infty, $$

contradicting (9.8). This ends the proof of the Lemma. □

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Lewicka, M., Peres, Y. The Robin Mean Value Equation I: A Random Walk Approach to the Third Boundary Value Problem. Potential Anal (2022).

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI:


  • Robin problem
  • Third boundary value problem
  • Oblique boundary conditions
  • Dynamic programming principle
  • Random walk
  • Finite difference approximations
  • Viscosity solutions

Mathematics Subject Classification (2010)

  • 35J05
  • 35J25
  • 60G50