With
couple_lambda0 = q
couple_lambda1 = none
you would definitely want to use soft-core for coulomb. But unless I’m completely mistaken sc-coul will not make any difference in this case. Just make sure that you have set the sc-alpha and the other soft-core parameters.
I might be wrong, but I think it’s only when using a setup like
couple_lambda0 = vdwq
couple_lambda1 = none
(or when using modifications in the topology) that sc-alpha makes a difference, i.e., if the soft-core parameters should be applied to both the vdw and q transformations.
I would recommend doing two sets of sequences of runs:
couple_lambda0 = vdwq
couple_lambda1 = vdw
and from the lambda 1 state output in that set of runs you run:
couple_lambda0 = vdw
couple_lambda1 = none
In theory, it is most efficient to run the respective lambda state simulations in each set in sequence, since you will then start each lambda state simulation from a state that is similar. But if you want to run very many simulations in parallel (better hardware utilisation) you will need to discard more time at the beginning of each simulation since it will take longer to reach equilibrium.
In the first of these sets you can set sc-alpha = 0 since no soft-core modifications are needed when there are still vdw interactions. In the second you should set the proper sc parameters. sc-coul should not be necessary in either of these cases.