Questions about the features - Learning Mesh-Based Simulation with Graph Networks

Hi, I'm really interested in you guys work, "Learning Mesh-Based Simulation with Graph Networks", especially regarding to cloth domain.

I believe the key point of this work was putting "Mesh-space" information as features.

1. May I ask what exactly is this "Mesh-space" ? Is it like a UV-coordinate?
2. (Follow-up) What can be the substitution if the mesh is 3D garment, instead of a cloth, which is not easy to spread-out into 2D space? Do you think the UV coordinate is still valid for the 3D garment?
3. Lastly, is there a reason for splitting and processing separately the "Mesh-space" and the "World-space" edge features? Would the performance vary if the features were just concatenated?(u_ij, |u_ij|, x_ij, |x_ij|)
4. Why are x_ij, |x_ij| included in the inputs for Mesh-edge-features?

Really appreciate if you can help me..!

No files, tests, or entry points are mentioned. The issue asks for conceptual clarification about mesh-space and world-space features in cloth simulation, so there is no defined code change or completion criterion to pursue.

Written by the indexing model from the issue text.