1: \begin{abstract}

2: The subject of this article is the introduction of a new concept of well-posedness of Bayesian inverse problems. 

3: The conventional concept of (Lipschitz, Hellinger) well-posedness in [Stuart 2010, Acta Numerica 19, pp. 451-559] is difficult to verify in practice and may be inappropriate in some contexts.

4: Our concept simply replaces the Lipschitz continuity of the posterior measure in the Hellinger distance by continuity in an appropriate distance between probability measures.

5: Aside from the Hellinger distance, we investigate well-posedness with respect to  weak convergence, the total variation distance, the Wasserstein distance, and also the Kullback--Leibler divergence.

6: We demonstrate that the weakening to continuity is tolerable and that the generalisation to other distances is important.

7: The main results of this article are proofs of well-posedness with respect to some of the aforementioned distances for large classes of Bayesian inverse problems. Here, little or no information about the underlying model is necessary; making these results particularly interesting for practitioners using black-box models.

8: We illustrate our findings with numerical examples motivated from machine learning and image processing.

9: %Finally, here-discussed well-posedness results are reproduced and verified numerically in low- and high-dimensional Bayesian inverse problems with up to $1\mathrm{E}4$ parameter dimensions.

10: \end{abstract}

11: