Back to search
2505.04815

Causal Discovery in Symmetric Dynamic Systems with Convergent Cross Mapping

Yiting Duan, Yi Guo, Jack Yang, Ming Yin

correctmedium confidence
Category
Not specified
Journal tier
Strong Field
Processed
Sep 28, 2025, 12:56 AM

Audit review

The paper proves that for C2 symmetry Rxn(π):(x1,…,xn)→(−x1,…,−x_{n−1},xn), the nth-coordinate observation h(x)=xn is even, so all time-derivatives retain even parity, giving F_{xn,n}(R·x)=F_{xn,n}(x) and thus a generically two-to-one, noninjective reconstruction; in contrast, xi (i<n) is odd, giving inversion symmetry in its reconstruction. This produces a noninjective cross-map Π_{xixn}=F_{xn,n}∘f′∘(F_{xi,n})^{-1}, so CCM falsely indicates unidirectional causality even when coupling is bidirectional. These steps are stated and sketched in Proposition 1 with equations (4.6)–(4.10) and (4.11) in the paper, and the general limit of single-observable differential reconstructions to at most Z2 symmetry is stated as Theorem 3 (Cross). The candidate solution reproduces the same parity argument, the two-to-one/noninjectivity conclusion, and the Π construction, matching the paper’s logic and conclusions.

Referee report (LaTeX)

\textbf{Recommendation:} minor revisions

\textbf{Journal Tier:} strong field

\textbf{Justification:}

The manuscript gives a coherent theoretical account of why CCM can misidentify causality in the presence of two-fold rotational symmetry and complements this with a practical segmentation remedy. The core arguments—parity propagation through derivatives, noninjectivity of the differential reconstruction from an invariant coordinate, and the induced noninjective cross-map—are correct and clearly linked to CCM’s projection/homeomorphism framework. Minor revisions are warranted to explicitly state embedding assumptions, clarify the scope of two-to-one mapping (excluding fixed sets), and provide a precise source or proof sketch for the Z2 symmetry bound.