[gmx-users] Implicit solvent setup for large protein

[Mark Abraham]

On 07/03/2012 11:01 PM, Justin A. Lemkul wrote:
>
>
> On 7/3/12 8:57 AM, Ramon Crehuet Simon wrote:
>> Dear all,
>> I know several questions about implict solvent have already been 
>> asked in this list. I think and hope the question I have has not been 
>> raised. Forgive me if I am wrong.
>> I have read that all-to-all kernels are the best option when doing 
>> implicit solvent. Otherwise one should use large cutoffs.
>> I wanto to simulate a tetramer which has more than 1200 residues. I'm 
>> afraid all-to-all interactions are much too expensive in this case. 
>> So I should use largecutoffs. My question is, how large? Or, to be 
>> more precise, what should I monitor to check the the cutoffs are 
>> large enough?
>> As an extra questions, if I don't use all-to-all kernels, is it still 
>> worth using the GPU to accelerate the calculation?
>
> In my experience, using long cutoffs leads to unstable trajectories 
> and very poor energy conservation.  The all-vs-all kernels are the 
> only ones I use for implicit calculations.

... and you should check out DOI 10.1002/jcc.21552 and 
10.1021/ct700053u, as well as the literature for any other implicit 
solvation protocol you might like to implement in GROMACS. Making stuff 
up is worth the effort you put into it...

Mark