Introduction
This article serves as a formal erratum to the 2018 study "Equivalence of weak and strong modes of measures on topological vector spaces," published in the journal Inverse Problems. While primarily technical, these corrections are vital for the mathematical consistency of theories regarding "modes"—the most likely outcomes—within complex, high-dimensional Bayesian inference frameworks.
Problem Statement: Modes in Infinite Dimensions
In standard statistics, identifying the mode of a distribution is a matter of finding the peak of a density function. However, in the field of inverse problems—such as medical imaging reconstruction or seismic analysis—researchers often operate in infinite-dimensional topological vector spaces. In these settings, the definition of a "mode" is not unique.
Mathematical definitions vary based on the underlying topology or "neighborhood" considered, leading to the concepts of weak modes and strong modes. The original 2018 paper sought to establish the conditions under which these definitions are equivalent. This erratum addresses specific errors in the abstract and a core proof on page 14 that were necessary to solidify the paper's conclusions regarding Maximum A Posteriori (MAP) estimators.
Technical Corrections and Methodology
The erratum focuses on refining the measure-theoretic machinery used to bridge weak and strong modes. The authors utilize Radon-Nikodym derivatives ($\frac{d\mu_v}{d\mu}$) to characterize how a measure $\mu$ behaves when shifted by a vector $v$.
A critical correction involves the representation of lower semicontinuous functions. The authors clarify that such functions can be represented as the limit of an increasing sequence of continuous functions. To complete the proof that an $E$-weak mode is necessarily an $E$-strong mode, the erratum introduces a specific sequence of functions, $g_m(x)$:
$$g_m (x) := \max \left\{ \frac{d\mu_{v_n}}{d\mu} (x) \mid n \in \mathbb{N}, n \leqslant m \right\}, \forall x \in X$$
This construction ensures the continuity of the supremum of measure ratios, a necessary step for the equivalence proof that was previously underspecified.
Significance for Bayesian Inference
The most impactful conceptual refinement appears in the corrected abstract. The authors explicitly link their measure-theoretic results to Maximum A Posteriori (MAP) estimation in non-parametric Bayesian inference. By ensuring the equivalence of weak and strong modes, the authors provide a more robust theoretical foundation for researchers using optimization techniques to find "best-fit" models in high-dimensional spaces.
Conclusion
The erratum restores the integrity of the original paper’s primary result: the equivalence of weak and strong modes in topological vector spaces. By addressing gaps in the sequence construction and clarifying the relationship to MAP estimators, the authors ensure that this theoretical framework remains a reliable tool for the inverse problems community.
Summary updated