Umbrella issue for multidimensional lambda design
Umbrella issue for multidimensional lambda functionality. Some previous discussions on having different lambdas for different parts of the system are here: - Different parts of system have different lambdas #1654 (closed) #1653 (closed) #1652 (closed)
We have already identified a large amount of potential use cases (in particular in combination with #4431 (closed) and #4432 (closed)). They are outlined below together with the basic algorithmic ideas that we want to support. However, further discussion is needed to clarify the design of the actual data structure.
The resulting n-dimensional lambda space can be sampled in three ways, which are described below together with some potential applications:
- Follow a 1D path in the n-dimensional lambda space.
- FEP/REST2
- One vector changing the softening of some parts of the potential, one vector controlling soft-core FEP
- FEP/REST2
- Follow a 1D path, but make the trajectory deviate from that path.
- Together with lambda vectors fully supporting pull code CVs and AWH, we can provide path metadynamics-like simulations that way.
- Sample the full n-dimensional surface.
- Constant pH with multiple residues.
- Combinatorial exploration of ligands
- Different vectors for separate atoms lists (like separate R-groups or residues)
- Combinatorial testing of parameters.
- Replica exchange with multiple pull code CVs.
- Mimic PT-MetaD, bias exchange and PB-MetaD with the help of AWH.
Note that the multi-dimensional lambda vectors can also be used for non-equilibrium simulations via the slow growth interface (init-lambda
and delta-lambda
). Together with a gmxapi-based replica exchange API enabling users to implement custom rules for configuration exchanges/updates and Monte Carlo criteria, we gain a versatile setup for non-equilibrium candidate Monte Carlo (NCMC) simulations with GROMACS as a side effect. If we go the final step and implement a meta-lambda vector controlling which dimension of the multi-dimensional lambda vectors is currently active, we can readily implement replica exchange, expanded ensemble and AWH simulations with NCMC as integrator as well as expanded ensemble versions of replica exchange with multiple pull code CVs as well as of our PT-MetaD, bias exchange and PB-MetaD mimics.