Challenges combining quantum mechanics and molecular mechanics using CP2K and GROMACS
After Dmitry’s lecture, I found future cp2k/gromacs combination very promising for my group projects. I will be happy to contribute with test systems and compare it with ultrafast Thomas Kubar Gromacs/DFTb .
Hi Andrey, yes the testing help would be definitely appreciated. Once I finish with suitable version, I will post a link to the Gitlab branch here, so you could check it yourself. However, experience in compiling CP2K itself would be required.
Also, adjusting parameters for CP2K is not fully finished yet. So if you have some experience with them (especially for biological systems), your suggestions will be useful.
Dmitry, Hi.
Compiling cp2k on Lomonosov2 as I remember is many ways easier than CPMD and GROMACS/DFTb. Probably linking cpk2k and gromacs binaries could be an issue.
What kind of parameters from QM part you are thinking about? theory, basis, system size, kind of braking bond? Most of them in my experience are heavily dependent on the reaction or just obvious. For example, we use metadynamics and timestep 1fs is too big, we used to 0.2-0.6 fs, and again it very depends on the methods and system itself.
Kind regards, Andrey
Well if you manage to compile and test CP2K itself, then linking against Gromacs would be quiet an easy procedure.
Yes, that kind of things like DFT functional, basis set, multi-grid setup and scf parameters. For now what we observe is that MOLOPT-DZVP basis set / GTH pseudopotentials in combination with PBE functional and QM box approximately twice as big as size of QM system itself, represents the best compromise between accuracy and speed.
What kind of that parameters you are usually using? Also, B3LYP is too slow in CP2K, because HF exchange are not very well working with GPW methodology.
Timestep is something Gromacs user should be aware of, I think it is not directly connected to QMMM.
Dmitry, sorry
I missed your reply, during workshop I checked forum and returned to it.
Parameters: We focused on fast methods, DFTb (D3), we used to sampling for 0.1 -1 ns in system. In most of cases we have quite reasonable output and convergence.
Some time we do really need to improve theory level and here cp2k is just in time for our gromacs ready files