Alchemical transformation — on hold¶
On hold since 2026-10-13
MD-tools is a conformational-sampling toolkit for now. Alchemical transformation is not part of the current release: its tutorials are not taught from the main tutorial list, and it is not re-validated or tested for a release. Everything below was last run against 0.6.4. See On hold for what else is waiting.
What exists¶
| part | what it is | last run against |
|---|---|---|
| Alchemical topology | md-openmm combine-topology: two endpoint Systems over one index space, and the protocol: alchemical ladder build-md generates from it |
0.6.4 |
| What the runtime computes | the evidence: a nine-leg hydration campaign checked against experiment and against a second alchemical route | 0.6.4 |
| Ethanol: hydration free energy | the smallest alchemical free energy there is, with an experimental number to check it against | 0.6.4 |
| Ethane → chloroethane: relative hydration free energy | one atom apart: the relative-free-energy pattern a protein–ligand campaign uses | 0.6.4 |
What is implemented, and what is not¶
Implemented: a plan with four modes (single, hybrid, dual, decoupling), the five λ
components named as OpenFE names them, an Amber18 softcore, and a generated ladder that md-run
dispatches. Not implemented, and refused by name: lambda_path: staged, mode: separated, a
decoupling of a charged ligand, and atom mapping across a ring. The detail is on the
method page.
What would take it off hold¶
- A decision that alchemical work is back in scope.
- The three failing CUDA tests in
test_alchemy_hamiltonian_cuda.pyexplained: they fail the same way onmain, deterministically and without MPS, by 0.0063 kJ/mol against a tolerance of 0.0021 on one of them. - The two tutorials re-run on the then-current release, because every number on them was produced by 0.6.4.
- The development plans — FEP–REST2 and relatives and water sampling — are separate and come after.