1: \begin{abstract}

2: Entropy regularized optimal transport and its multi-marginal generalization have 

3: attracted increasing attention in various applications, in particular due to efficient Sinkhorn-like algorithms for computing optimal transport plans. % along their dual formulations.

4: However, it is often desirable that the marginals of the optimal transport plan do not match the given measures exactly, which led to the introduction of the so-called unbalanced optimal transport.

5: Since unbalanced methods were not examined for the multi-marginal setting so far, we address this topic in the present paper.

6: More precisely, we introduce the unbalanced multi-marginal optimal transport problem and its dual, and show that a unique optimal transport plan exists under mild assumptions.

7: Further, we generalize the Sinkhorn algorithm for regularized unbalanced optimal transport to the multi-marginal setting and prove its convergence.

8: If the cost function decouples according to a tree, the iterates can be computed efficiently. 

9: At the end, we discuss three applications of our framework, namely two barycenter problems and a transfer operator approach, where we establish a relation between the barycenter problem and the multi-marginal optimal transport with an appropriate tree-structured cost function.

10: \end{abstract}

11: