Algorithms for the resolution of stochastic control problems in high dimension by using probabilistic and max-plus methods

Abstract : Stochastic optimal control problems with finite horizon are a class of optimal control problems where intervene stochastic processes in a bounded time. As many optimal control problems, they are often solved using a dynamic programming approach which results in a second order Partial Differential Equation (PDE) called the Hamilton-Jacobi-Bellman equation. Grid-based methods, probabilistic methods or more recently max-plus methods can be used then to solve this PDE. However, the first type of methods default in a space of high dimension because of the curse of dimensionality while the second type of methods allowed till now to solve only problems where the nonlinearity of the PDE with respect to the second order derivatives is not very high. As for the third type of method, it results in an explosion of the complexity of the value function. We introduce two new probabilistic schemes in order to enlarge the class of problems that can be solved with probabilistic methods. One is adapted to PDE with bounded coefficients while the other can be applied to PDE with bounded or unbounded coefficients. We prove the convergence of the two probabilistic scheme and obtain error estimates in the case of a PDE with bounded coefficients. We also give some results about the behavior of the second probabilistic scheme in the case of a PDE with unbounded coefficients. After that, we introduce a completely new type of method to solve stochastic optimal control problems with finite horizon that we call the max-plus probabilistic method. It allows to add the non linearity feature of max-plus methods to a probabilistic method while controlling the complexity of the value function. An application to the computation of the optimal super replication price of an option in an uncertain correlation model is given in a 5 dimensional space.
Complete list of metadatas

Cited literature [68 references]  Display  Hide  Download

https://pastel.archives-ouvertes.fr/tel-01891835
Contributor : Abes Star <>
Submitted on : Wednesday, October 10, 2018 - 9:19:05 AM
Last modification on : Wednesday, April 17, 2019 - 1:32:49 AM
Long-term archiving on : Friday, January 11, 2019 - 12:38:20 PM

File

58994_FODJO_2018_archivage.pdf
Version validated by the jury (STAR)

Identifiers

  • HAL Id : tel-01891835, version 1

Citation

Eric Fodjo. Algorithms for the resolution of stochastic control problems in high dimension by using probabilistic and max-plus methods. Analysis of PDEs [math.AP]. Université Paris-Saclay, 2018. English. ⟨NNT : 2018SACLX034⟩. ⟨tel-01891835⟩

Share

Metrics

Record views

214

Files downloads

66