2021
DOI: 10.1051/cocv/2021029
|View full text |Cite
|
Sign up to set email alerts
|

The back-and-forth method for Wasserstein gradient flows

Abstract: We present a method to efficiently compute Wasserstein gradient flows.  Our approach is based on a generalization of the back-and-forth method (BFM) introduced by Jacobs and Leger to solve optimal transport problems.  We evolve the gradient flow by solving the dual problem to the JKO scheme. In general, the dual problem is much better behaved than the primal problem.  This allows us to efficiently run large scale gradient flows simulations for a large class of internal energies including singular and non-conve… Show more

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
1
1
1

Citation Types

0
3
0

Year Published

2022
2022
2023
2023

Publication Types

Select...
4
2
1

Relationship

3
4

Authors

Journals

citations
Cited by 9 publications
(3 citation statements)
references
References 21 publications
0
3
0
Order By: Relevance
“…As we mentioned above, we obtain this result via a variational approximation scheme, which is a simplified version of the more general scheme introduced in [JKT21]. Although there are many different ways to establish the well-posedness of the system (such as the degenerate diffusion approach considered in [PQV14] and [GKM22]), we emphasize our variational scheme, as it has a very efficient numerical implementation via the Back-and-Forth method [JL20,JLL21]. For instance, the images displayed in Figure 1 are computed on a high-resolution 1024 × 1024 grid.…”
Section: Summary Of Main Resultsmentioning
confidence: 99%
“…As we mentioned above, we obtain this result via a variational approximation scheme, which is a simplified version of the more general scheme introduced in [JKT21]. Although there are many different ways to establish the well-posedness of the system (such as the degenerate diffusion approach considered in [PQV14] and [GKM22]), we emphasize our variational scheme, as it has a very efficient numerical implementation via the Back-and-Forth method [JL20,JLL21]. For instance, the images displayed in Figure 1 are computed on a high-resolution 1024 × 1024 grid.…”
Section: Summary Of Main Resultsmentioning
confidence: 99%
“…As we mentioned above, we obtain this result via a variational approximation scheme, which is a simplified version of the more general scheme introduced in [JKT21]. Although there are many different ways to establish the well-posedness of the system (such as the degenerate diffusion approach considered in [PQV14] and [GKM22]), we emphasize our variational scheme, as it has a very efficient numerical implementation via the Back-and-Forth method [JL20,JLL21]. For instance, the images displayed in Figure 1 are computed on a high-resolution 1024 × 1024 grid.…”
Section: 𝜓(𝑛 − 𝑏)𝜌 𝑑𝑥 𝑑𝑡;mentioning
confidence: 96%
“…The advantage of this perspective is that our scheme can approximate any flow of the form (P ). Furthermore, the dual problem associated to our scheme has a very efficient numerical implementation using the recently introduced backand-forth method [JL19,JWL]. In particular, the numerical implementation via the back-andforth method does not require introducing an additional time dimension, which allows for a faster computation time than schemes based around the Benamou-Brenier formula.…”
Section: Introductionmentioning
confidence: 99%