Abstract
To a mesh function we associate the natural analogue of the Monge-Ampère measure. The latter is shown to be equivalent to the Monge-Ampère measure of the convex envelope. We prove that the uniform convergence to a bounded convex function of mesh functions implies the uniform convergence on compact subsets of their convex envelopes and hence the weak convergence of the associated Monge-Ampère measures. We also give conditions for mesh functions to have a subsequence which converges uniformly to a convex function. Our result can be used to give alternate proofs of the convergence of some discretizations for the second boundary value problem for the Monge-Ampère equation and was used for a recently proposed discretization of the latter. For mesh functions which are uniformly bounded and satisfy a convexity condition at the discrete level, we show that there is a subsequence which converges uniformly on compact subsets to a convex function. The convex envelopes of the mesh functions of the subsequence also converge uniformly on compact subsets. If in addition they agree with a continuous convex function on the boundary, the limit function is shown to satisfy the boundary condition strongly.
Similar content being viewed by others
References
Arutyunov, A.V., Obukhovskii, V.: Convex and Set-Valued Analysis. De Gruyter Graduate. De Gruyter, Berlin (2017). Selected topics
Awanou, G.: Discrete Aleksandrov solutions of the Monge-Ampère equation. arXiv:1408.1729
Awanou, G.: The second boundary value problem for a discrete Monge-Ampère equation with symmetrization. arXiv:1910.14376
Bakelman, I.J.: Convex Analysis and Nonlinear Geometric Elliptic Equations. Springer, Berlin (1994). With an obituary for the author by William Rundell, edited by Steven D. Taliaferro
Barles, G., Burdeau, J.: The Dirichlet problem for semilinear second-order degenerate elliptic equations and applications to stochastic exit time control problems. Commun. Partial Differ. Equ. 20(1–2), 129–178 (1995)
Barles, G., Souganidis, P.E.: Convergence of approximation schemes for fully nonlinear second order equations. Asymptot. Anal. 4(3), 271–283 (1991)
Benamou, J.-D., Duval, V.: Minimal convex extensions and finite difference discretisation of the quadratic Monge–Kantorovich problem. Eur. J. Appl. Math. 30(6), 1041–1078 (2019)
Benamou, J.-D., Froese, B.D.: Weak Monge-Ampère solutions of the semi-discrete optimal transportation problem. In: Topological Optimization and Optimal Transport. Radon Ser. Comput. Appl. Math., vol. 17, pp. 175–203. de Gruyter, Berlin (2017)
Caffarelli, L.A.: Boundary regularity of maps with convex potentials. Commun. Pure Appl. Math. 45(9), 1141–1151 (1992)
del Teso, F., Manfredi, J.J., Parviainen, M.: Convergence of dynamic programming principles for the p-Laplacian. Adv. Calc. Var. 1 (2020). (Ahead-of-print)
Evans, L.C., Gariepy, R.F.: Measure Theory and Fine Properties of Functions. Studies in Advanced Mathematics. CRC Press, Boca Raton (1992)
Froese, B., Oberman, A.: Convergent finite difference solvers for viscosity solutions of the elliptic Monge-Ampère equation in dimensions two and higher. SIAM J. Numer. Anal. 49(4), 1692–1714 (2011)
Gutiérrez, C.E.: The Monge-Ampère Equation. Progress in Nonlinear Differential Equations and Their Applications, vol. 44. Birkhäuser, Boston (2001)
Gutiérrez, C.E., van Nguyen, T.: On Monge-Ampère type equations arising in optimal transportation problems. Calc. Var. Partial Differ. Equ. 28(3), 275–316 (2007)
Hackbusch, W.: Elliptic Differential Equations. In: Theory and Numerical Treatment. Springer Series in Computational Mathematics, vol. 18. Springer, Berlin (2010). English edition, translated from the 1986 corrected German edition by Regine Fadiman and Patrick D. F. Ion
Hartenstine, D.: The Dirichlet problem for the Monge-Ampère equation in convex (but not strictly convex) domains. Electron. J. Differ. Equ. 2006 138, (2006). (Electronic)
Ioffe, A.D., Tihomirov, V.M.: Theory of Extremal Problems. Elsevier, Amsterdam (2009)
Mirebeau, J.-M.: Discretization of the 3D Monge-Ampère operator, between wide stencils and power diagrams. ESAIM: Math. Model. Numer. Anal. 49(5), 1511–1523 (2015)
Neilan, M., Zhang, W.: Rates of convergence in \(W^{2}_{p}\)-norm for the Monge-Ampère equation. SIAM J. Numer. Anal. 56(5), 3099–3120 (2018)
Nochetto, R.H., Zhang, W.: Discrete ABP estimate and convergence rates for linear elliptic equations in non-divergence form. Found. Comput. Math. 18(3), 537–593 (2018)
Nochetto, R.H., Zhang, W.: Pointwise rates of convergence for the Oliker-Prussner method for the Monge-Ampère equation. Numer. Math. 141(1), 253–288 (2019)
Roberts, A.W., Varberg, D.E.: Another proof that convex functions are locally Lipschitz. Am. Math. Mon. 81, 1014–1016 (1974)
Rockafellar, R.T.: Convex Analysis. Princeton Mathematical Series, vol. 28. Princeton University Press, Princeton (1970)
Tawarmalani, M., Sahinidis, N.V.: Convex extensions and envelopes of lower semi-continuous functions. Math. Program. 932, Ser. A, 247–263 (2002)
Acknowledgements
The author would like to thank the referees for their comments which lead to a better presentation with simpler proofs. In particular the statement of Lemmas 7 and 10 are due to one of the referees as well as Theorem 4 and its proof. The proof of Lemma 16 given was suggested by one of the referees. The author was partially supported by NSF grants DMS-1319640 and DMS-1720276. The author would like to thank the Isaac Newton Institute for Mathematical Sciences, Cambridge, for support and hospitality during the programme “Geometry, compatibility and structure preservation in computational differential equations” where part of this work was undertaken. Part of this work was supported by EPSRC grant no EP/K032208/1.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix
Appendix
In [10] convergence of numerical schemes for discrete approximations to viscosity solutions of the Dirichlet problem for dynamic programming principles for the \(p\)-Laplacian is given. The proof is based on a numerical analysis approach, the Barles-Souganidis framework which consists in checking stability, consistency and monotonicity of the numerical scheme. It is known that the Barles-Souganidis framework requires the so-called strong uniqueness property for the differential equation, which is a comparison principle for equations with the boundary condition in the viscosity sense. In [10] the strong uniqueness property for the Laplace equation is proved for smooth domains and convergence of the numerical scheme on a bounded Lipschitz domain is obtained through barriers on appropriate shrinking rings. We review below their approach for the standard discretization of the Laplace equation.
Recall that \(\varOmega \subset \mathbb{R}^{d}\) is a bounded Lipschitz domain and \(g \in C(\partial \varOmega )\). We consider the problem
We define for \(h>0\) the following sets:
Outer boundary strip: \(\varGamma _{h} = \{ \, x \in \mathbb{R}^{d}\setminus \varOmega , d(x, \partial \varOmega ) \geq h\, \}\) and put \(O=\varGamma _{1}\).
Inner boundary strips: \(I_{h} = \{ \, x \in \varOmega , d(x,\partial \varOmega ) \leq h \, \}\) and put \(I=I_{1}\).
Extended domain: \(\widetilde{\varOmega } = \varOmega \cup O\) and extended computational domain \(\widetilde{\varOmega }_{h} = \varOmega _{h} \cup O\).
Let \(G\) be a continuous extension of \(g\) to \(\widetilde{\varOmega }\). Note that \(\partial \varOmega _{h} \subset O\). Given a mesh function \(u_{h}\), we extend it to \(\varOmega _{h} \cup O\) by \(u_{h}(x)=G(x)\) for all \(x \in O \setminus \partial \varOmega _{h}\). Analogous to [10, (2.9)], we consider the discrete problem: find a mesh function \(u_{h}\) such that
A mesh function solves (27) if and only if it solves (29).
In [10] a notion of viscosity solution of (28) is first given. The authors therein recall the existence and uniqueness of such a viscosity solution. They then introduce a generalized version of viscosity solution where the boundary condition is assumed in a weaker sense. That notion has become known as boundary condition in the viscosity sense.
An upper semi-continuous function on \(\overline{\varOmega }\) is a viscosity subsolution of (28) if whenever \(x_{0} \in \overline{\varOmega }\) and \(\phi \in C^{2}(\overline{\varOmega })\) satisfy
we have
If the condition (31) at the boundary is replaced by
the function \(u\) is said to be a generalized viscosity subsolution of (28).
A lower semi-continuous function on \(\overline{\varOmega }\) is a viscosity supersolution of (28) if whenever \(x_{0} \in \overline{\varOmega }\) and \(\phi \in C^{2}(\overline{\varOmega })\) satisfy
we have
If the condition (34) at the boundary is replaced by
the function \(u\) is said to be a generalized viscosity supersolution of (28).
A function \(u \in C(\overline{\varOmega })\) is a viscosity solution of (28) if it is both a viscosity subsolution and a viscosity supersolution of (28). Note that for this notion the boundary condition is taken in the usual sense.
Theorem 15
[10, Theorem 3.4] Let \(\varOmega \subset \mathbb{R}^{d}\) be a bounded Lipschitz domain and \(g\in C(\partial \varOmega )\). If \(u\) is a viscosity subsolution of (28) and \(v\) a supersolution of (28), then \(u\leq v\) on \(\overline{\varOmega }\).
The authors in [10] proved the strong uniqueness property for smooth domains.
Proposition 1
[5, 10] Let \(\varOmega \subset \mathbb{R}^{d}\) be a \(C^{2}\) domain and \(g \in C(\partial \varOmega )\). Let \(u\) and \(v\) be respectively generalized viscosity subsolution and supersolution of (28). Then \(u \leq v\) in \(\overline{\varOmega }\).
Using standard arguments we review below, the above proposition allows for \(\varOmega \) a \(C^{2}\) domain to claim the convergence of the solution \(u_{h}\) of (29) to the viscosity solution of (28). We define for a \(C^{2}\) function \(\phi \) on \(\varOmega _{E}\)
with the operator \(\Delta _{e}\) defined as for (9). We write (28) in the standard form \(S(h,x,u_{h}(x),u_{h})=0\), \(x \in \widetilde{\varOmega }_{h}\) with a slight abuse of notation. We have the following analogues of [10, (2.11)–(2.12)].
If \(u_{h}\) solves (28), we have
Let \(u_{h}^{1}\) and \(u_{h}^{2}\) solve
with \(G^{1} \leq G^{2}\). Then
Equation (36) says that the scheme is stable. It is a consequence of properties of the matrix of the discrete linear problem (28) [15, Theorem 4.77]. The proof is analogous to [15, Remark 4.37].
Equation (37) is the discrete comparison principle. Again, it is a consequence of [15, Theorem 4.77]. The proof is analogous to [15, Theorem 4.38 b].
Next, we recall the monotonicity of the scheme, i.e. if \(u_{h} \leq v_{h}\) on \(\widetilde{\varOmega }_{h}\) with \(u_{h}(x)=v_{h}(x)\) we have \(S(h,x,u_{h}(x),u_{h})\leq S(h,x,v_{h}(x),v_{h})\).
Next, we describe the form of consistency of the scheme needed to prove convergence when the boundary condition is taken in the viscosity sense. For \(x \in \overline{\varOmega }\) and \(\phi \in C^{2}(\varOmega _{E})\)
and
For \(x\in \varOmega \), we have the usual consistency property. For \(x \in \partial \varOmega \), one can approach \(x\) with points \(y\) in either \(\varOmega \) or \(O\).
Theorem 16
Assume that \(\varOmega \) is a \(C^{2}\) domain. The solution \(u_{h}\) of (29) converges uniformly on \(\overline{\varOmega }\) to the viscosity solution of (28).
Proof
Since the scheme is stable, consistent and monotone, it follows from Proposition 1 and the framework in [6] that the solution \(u_{h}\) of (29) converges uniformly on \(\overline{\varOmega }\) to the viscosity solution of (28). □
To handle the case \(\varOmega \) Lipschitz, a delicate treatment at the boundary is done in [10] using barriers on appropriate shrinking rings. Denote by \(B_{r}(x)\) the open ball of center \(x\) and radius \(r\). The following regularity condition for Lipschitz domains is the one used in the proof.
There exists \(\overline{\delta } >0\) and \(\mu \in (0,1)\) such that for every \(\delta \in (0,\overline{\delta })\) and \(y \in \partial \varOmega \), there exists a ball \(B_{\mu \delta }(z)\) strictly contained in \(B_{\delta }(y) \setminus \varOmega \). The constant \(\mu \) is independent of \(y \in \partial \varOmega \).
We have the following analogue of [10, Corollary 4.5]
Lemma 17
Given \(\eta >0\), there exist \(\delta =\delta (\eta ,G,\overline{\delta })\), \(k_{0}=k_{0}(\eta ,\mu ,G)\), \(h_{0}=h_{0}(\eta ,\delta ,\mu ,k_{0})\) such that
for all \(y \in \partial \varOmega \), \(x \in B_{\delta /4^{k_{0}}}(y) \cap \varOmega _{h}\) and \(h\leq h_{0}\).
Proof
As in [10], we prove only the one sided inequality \(u_{h}(x) \leq G(y) + \eta /2\), the other being similar. Also, we consider only the case \(d\neq 2\). The case \(d=2\) is treated with similar arguments as indicated on [10, p. 16].
Part I: In this part we collect parts of the proof in [10] which do not deal with dynamic programming. Fix \(\delta \in (0,\overline{\delta })\). Let \(u_{h}\) solve (29). For \(y \in \partial \varOmega \) define
Define
and for \(k\geq 0\)
Define \(\delta _{k}=\delta /4^{k-1}\). By the regularity assumption on \(\varOmega \), one can find balls \(B_{\mu \delta _{k+1}}(z_{k}) \subset B_{\delta _{k+1}}(y) \setminus \varOmega \) for all \(k\). In particular \(||y-z_{k}|| < \delta _{k+1}\).
For notational convenience, denote \(m=m^{h}(y)\), \(M=M^{h}\) and \(M_{k}=M_{k}^{h}(y)\). We consider the problem
Define
and (see [10, Fig. 1])
It is proven in [10, page 17] that
and that for \(x \in B_{(2-\mu ) \delta _{k+1}}(z_{k})\)
where
Part II: We assume in this part that \(u_{h} \leq M_{k}\) on \(B_{\delta _{k}}(y) \cap \varOmega _{h}\) for all \(h< h_{k}\) for a fixed \(h_{k}\) and that \(M_{k}-m\geq \eta /4\). We prove that there exists \(h_{k+1}=h_{k+1}(\eta ,\mu ,\delta ,d,G) \in (0,h_{k})\) such that for all \(h< h_{k+1}\), we have
For \(h \leq \mu \delta _{k+1}/2\) the barrier \(U_{k}\) is extended to the ring
Let \(U_{k}^{h}\) be a mesh function which solves
where \(R_{k}=B_{\delta _{k}}(z_{k}) \setminus \overline{B}_{ \mu \delta _{k+1}}(z_{k})\). Note that \(R_{k,h}\setminus R_{k}\) is the outer \(h\)-neighborhood of \(R_{k}\). Since \(R_{k}\) is smooth, by Theorem 1, \(U_{k}^{h}\) converges uniformly to \(U_{k}\) in \(R_{k,h}\) as \(h \to 0\) (recall that \(U_{k}^{h}=U_{k}\) outside \(R_{k}\)). Therefore, given
there exists \(h_{k+1} = h_{k+1} (\gamma )\), \(0< h_{k+1} \leq \min \{ \, \mu \delta _{k+1}/2, h_{k} \, \}\) such that
for all \(h\leq h_{k+1}\) and \(x \in R_{k,h}\).
On \(\varGamma _{1}^{h}\), we have \(u_{h}=G \leq m\) and so, using (40) and \(\varGamma _{1}^{h} \subset R_{k,h}\) (which follows from \(||y-z_{k}|| < \delta _{k+1}\)), we get
Since \(h<\delta _{k+1}/2\) and \(||y-z_{k}|| < \delta _{k+1}\) we have \(B_{\delta _{k}/2+h}(z_{k}) \subset B_{\delta _{k}}(y)\). If \(u_{h} \leq M_{k}\) on \(B_{\delta _{k}}(y) \cap \varOmega _{h}\), using (38) and (40) we get on \(\varGamma _{2}^{h}\)
In summary, we have
Since the \(h\)-boundary of \(B_{\delta _{k}/2}(y) \cap \varOmega _{h}\) is contained in \(\varGamma _{1}^{h} \cup \varGamma _{2}^{h}\) (see [10, Fig. 1]), by the discrete comparison principle (37), we obtain
Using again (40) we have under the assumption \(u_{h} \leq M_{k}\) on \(B_{\delta _{k}}(y) \cap \varOmega _{h}\)
Next, since \(B_{\delta _{k+1}}(y) \subset B_{\delta _{k}/2}(z_{k})\) we get by (41)
As \(B_{\delta _{k+1}}(y) \subset B_{(2-\mu ) \delta _{k+1}}(z_{k})\), we have by (39) \(a u_{h} + b m \leq b' m + a'M_{k}+ 2 \gamma \) on \(B_{\delta _{k+1}}(y) \cap \varOmega _{h}\). Since by assumption \(M_{k}-m\geq \eta /4\), we have
Using in addition \(b'-b+a'=a\), we obtain for \(h< h_{k+1}\) in \(B_{\delta _{k+1}}(y) \cap \varOmega _{h}\)
Part III: As \(G\) is uniformly continuous on the compact set \(\varGamma _{1}\), there exists \(0<\delta < \overline{\delta }\) such that for all \(y \in \partial \varOmega \),
By the stability result (36)
for all \(h\). Thus we can take \(h_{0}=1\).
If \(M^{h}_{0}(y)-m^{h}(y)=M^{h}-m^{h}(y) < \eta /4\), then by (44), we obtain \(u_{h} \leq M^{h} < m^{h}(y)+\eta /4 < G(y)+\eta /2\). Otherwise, there exists \(0< h_{1}< h_{0}\) such that for all \(h< h_{1}\), \(u_{h}\leq M_{1}^{h}(y)\) in \(B_{\delta _{1}}(y) \cap \varOmega \). If \(M^{h}_{1}-m^{h}(y) < \eta /4\) we proceed as before to obtain the desired inequality.
As \(0<\theta <1\), for some integer \(s\), we have \(\theta ^{s}( M^{h}-m^{h}(y) ) <\eta /4\) which with (44) gives the desired inequality if \(u_{h} \leq M^{h}_{s}\) on \(B_{\delta _{s}}(y) \cap \varOmega \). If for some \(k< s\) we have \(M^{h}_{k}-m^{h}(y) < \eta /4\), we take \(k_{0}=k\). Otherwise \(k_{0}=s\). □
We can now state
Theorem 17
Assume that \(\varOmega \) is a Lipschitz domain. The solution \(u_{h}\) of (29) converges uniformly on compact subsets of \(\varOmega \) to the viscosity solution of (28).
Proof
The proof is as on [10, p. 18]. It follows from Lemma 17 that the half relaxed limits satisfy the boundary condition in the classical sense. One can then use Theorem 15 and the framework in [6]. □
Rights and permissions
About this article
Cite this article
Awanou, G. On the Weak Convergence of Monge-Ampère Measures for Discrete Convex Mesh Functions. Acta Appl Math 172, 6 (2021). https://doi.org/10.1007/s10440-021-00400-x
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s10440-021-00400-x