2107.04346
Generalization of the Change of Variables Formula with Applications to Residual Flows
Niklas Koenen, Marvin N. Wright, Peter Maaß, Jens Behrmann
correctmedium confidence
- Category
- Not specified
- Journal tier
- Specialist/Solid
- Processed
- Sep 28, 2025, 12:56 AM
- arXiv Links
- Abstract ↗PDF ↗
Audit review
The paper’s Theorem 4 states and (essentially) proves the CVF for L-diffeomorphisms, yielding pf(x) = pZ(f^{-1}(x)) |det J f^{-1}(x)| almost everywhere, by removing null sets so the restriction is a diffeomorphism and then applying the standard CVF; see the definition of L-diffeomorphism and Theorem 4 with proof in the main text and Appendix A (e.g., Definition 2 and Theorem 4; proof lines leading to eqs. (8)–(11)) . The candidate solution proves the same result by a patchwise change-of-variables argument on countably many local inverse charts and careful disjointization, avoiding any set-theoretic slippage. The only issue we found in the paper is the asserted set equality f^{-1}(B) \ NZ = f^{-1}(B \ NX) (eq. (9)), which need not hold absent an assumption like f(NZ) ⊆ NX; however, this can be fixed with a subset relation plus a measure-zero correction, leaving the main theorem intact . Hence both are correct; the model’s proof is different (more granular) and fully resolves the technical point.
Referee report (LaTeX)
\textbf{Recommendation:} minor revisions
\textbf{Journal Tier:} specialist/solid
\textbf{Justification:}
The paper clearly formulates L-diffeomorphisms and extends the CVF to this setting, which is mathematically clean and practically relevant for flows with non-smooth activations. The core argument is standard and correct once negligible sets are removed. A minor set-theoretic equality in Appendix A should be weakened to an inclusion with an accompanying null-set argument. With this correction, the result is sound and the exposition is strong.