Pull Request №681 RosettaCommons/rosetta/main ← everyday847/rosetta/differentiable-parametric
Merge: cf532f89b6bbb7e67fd75031f3cc2aa95fb5ce19←f84a8d234b4b9854783474b432ca98d0cba1aa82
Add differentiable parametric backbone minimization
----------------
Merge commit message:
Add differentiable parametric backbone minimization
Enables MinMover to co-optimize parametric DOFs (r0, omega0,
delta_omega0, etc.) alongside standard atom-tree DOFs (chi angles,
backbone torsions for non-parametric residues, jumps).
Components:
- Analytical Crick equation derivatives (dXYZ/dParam) in
numeric/crick_equations/BundleParams_derivatives for r0, omega0,
delta_omega0, delta_omega1, and delta_t
- MoveMap::set_parametric(true) flag to enable parametric DOFs
- ParametricAtomTreeMultifunc: extends the DOF vector with parametric
variables, computes gradients via Jacobian chain rule
(dE/dParam = sum_atoms dE/dXYZ . dXYZ/dParam)
- AtomTreeMinimizer dispatch: detects parametric poses and
automatically excludes backbone torsions of parametric residues
to prevent redundant DOF control
- Integration test: 3-helix ELLKAIA bundle minimized with and
without chi angle co-optimization