Multidimensional martingale optimal transport.

Abstract : In this thesis, we study various aspects of martingale optimal transport in dimension greater than one, from duality to local structure, and finally we propose numerical approximation methods.We first prove the existence of irreducible intrinsic components to martingal transport between two given measurements, as well as the canonicity of these components. We have then proved a duality result for optimal martingale transport in any dimension, point by-point duality is no longer true but a form of quasi safe duality is demonstrated. This duality makes it possible to demonstrate the possibility of decomposing the quasi-safe optimal transport into a series of optimal transport subproblems point by point on each irreducible component. Finally, this duality is used to demonstrate a principle of martingale monotony, analogous to the famous monotonic principle of classical optimal transport. We then study the local structure of optimal transport, deduced from differential considerations. We thus obtain a characterization of this structure using tools of real algebraic geometry. We deduce the optimal martingal transport structure in the case of the power costs of the Euclidean norm, which makes it possible to solve a conjecture that dates from 2015. Finally, we compared the existingnumerical methods and proposed a new method which proves more efficient and allows to treat an intrinsic problem of the martingale constraint which is the defect of convex order. Techniques are also provided to manage digital problems in practice.
Document type :
Complete list of metadatas

Cited literature [80 references]  Display  Hide  Download
Contributor : Abes Star <>
Submitted on : Tuesday, January 8, 2019 - 11:25:06 AM
Last modification on : Tuesday, April 16, 2019 - 5:48:49 AM
Long-term archiving on : Tuesday, April 9, 2019 - 2:58:13 PM


Version validated by the jury (STAR)


  • HAL Id : tel-01973279, version 1


Hadrien de March. Multidimensional martingale optimal transport.. Probability [math.PR]. Université Paris-Saclay, 2018. English. ⟨NNT : 2018SACLX042⟩. ⟨tel-01973279⟩



Record views


Files downloads