Dear Axel, Steve, and other developers
I’d like to simulate a liquid system in NPT ensemble consists of a set of TIP4P water enclosed by some organic molecules.
However, I cannot find GPU version for pair style TIP4P. Is there such an option or going to be developed later? If not, how should I manage such a system in my GPU system?
Has anybody ever encountered such case?
Many Thanks in advance.
Dear Axel, Steve, and other developers
I’d like to simulate a liquid system in NPT ensemble consists of a set of TIP4P water enclosed by some organic molecules.
However, I cannot find GPU version for pair style TIP4P.
I’m not sure there is much demand for tip4p
Is there such an option or going to be developed later? If not, how should I manage such a system in my GPU system?
You can simulate a tip4p molecule using a normal Lennard Jones pair style (with long range electrostatics), using fix rigid to keep the 4 particles in each tip4p molecule from moving relative to each other. There was a long discussion of this a month or two ago for using tip5p, which is similar
https://sourceforge.net/p/lammps/mailman/message/36376988/
I’m not sure how whether using fix rigid will slow down a simulation using gpu or kokkos force field styles, but we were able to get away with using large timesteps for this system (1-2fs). It was tricky to run simulations at constant pressure. With Yaliehanda’s permission, I can send you the tip5p model we have and you can easily modify it to createl tip4p. (We were going to post it online anyway.)
Cheers
Andrew
Sorry, no one has yet done the coding work to implement the TIP4P pair
or PPPM solver for GPU. So you’d have to run that model
in LAMMPS on a CPU.
Steve
Sorry, no one has yet done the coding work to implement the TIP4P pair
or PPPM solver for GPU. So you’d have to run that model
in LAMMPS on a CPU.
If you are using the GPU package you can try using pair_style lj/cut/coul/long/gpu
…along with a combination of fix rigid (to keep the tip4p molecules rigid) along with fix npt (or fix nvh+fix langevin, to integrate the equations of.motion for the flexible molecules).
I have no idea whether this would be faster than running the simulations using a CPU.
In my system water (TIP4P) molecules are sandwiched by two set of organics and I’d like calculating interfacial tension for such system. So, I have to establish NPT with setting pressure normal to the interfaces, say z axis. Do you think combing rigid/NPT/small and conventional NPT would work correctly?
In my system water (TIP4P) molecules are sandwiched by two set of organics and I’d like calculating interfacial tension for such system. So, I have to establish NPT with setting pressure normal to the interfaces, say z axis. Do you think combing rigid/NPT/small and conventional NPT would work correctly?
If you plan to run a system with liquid at the boundary of an interface, AND if you plan to completely immobilize the organic layer, then this adds a little more difficulty to the simulation. (See the post on the mailing list today titled “pressure on liquid enclosed by solid”).
Otherwise, if the atoms in the organic molecules are integrated in the normal way (ie fix npt or fix nph+langevin), then this detail does not matter. In that case look at the protocol I mentioned in a link included with my original reply to your post for tip5p
Either way, when you run the npt simulations (with fix not or fix nph+langevin) be sure to use an anisotropic pressure. (Don’t use the “iso” keyword.)
Cheers
Andrew