Artifact with including Kspace in capillary flow

Hi everybody

I’m going to simulate spontaneous uptake of NaCl solution within a mineral slit. But, I encountered difficulty while running my model.
As depicted in the attached file, the aqueous solution is placed below the mineral with an invisible wall separating them. Upon equilibration at 300K via NVT and then removing the wall(Figure 1), the solution enters the channel but its movement does not make sense! A hole (Figure 2) appears at the pore entrance and solution creeps up over the mineral surface (Figure 3). I performed this simulation for a long period, 10 ns, but the expected meniscus interface does not form.
After many trials, finally, I figured out ignoring Kspace results in the desired capillary flow, however, with too slow progress, i.e., solution front rises too slowly, due to eliminating the long-range electrostatic interactions between water and mineral. Judging upon open literature, I believe Kspace must be considered for such systems.

I’m really frustrated with my project and could not work around the weird effect of kspace (pppm in my case).

One more thing the system is periodic in all directions and removing PBC in the flow axis would not change the result (in this case, Kspace modify slab 3).

Would you please let me know if any trick or solution going with MD simulation of capillarity?

The discussed image and associating input file are attached.

Any hint is welcome.

Sincerely
Badizad

in.System (3.78 KB)

2.png

3.png

1.png

Hi everybody

I’m going to simulate spontaneous uptake of NaCl solution within a mineral slit. But, I encountered difficulty while running my model.
As depicted in the attached file, the aqueous solution is placed below the mineral with an invisible wall separating them. Upon equilibration at 300K via NVT and then removing the wall(Figure 1), the solution enters the channel but its movement does not make sense! A hole (Figure 2) appears at the pore entrance and solution creeps up over the mineral surface (Figure 3). I performed this simulation for a long period, 10 ns, but the expected meniscus interface does not form.

you expect a meniscus based on what information/evidence?

the behavior shown in the pictures does not look so unreasonable to me.

After many trials, finally, I figured out ignoring Kspace results in the desired capillary flow, however, with too slow progress, i.e., solution front rises too slowly, due to eliminating the long-range electrostatic interactions between water and mineral. Judging upon open literature, I believe Kspace must be considered for such systems.

I’m really frustrated with my project and could not work around the weird effect of kspace (pppm in my case).

One more thing the system is periodic in all directions and removing PBC in the flow axis would not change the result (in this case, Kspace modify slab 3).

Would you please let me know if any trick or solution going with MD simulation of capillarity?

please keep in mind, that this is science and not computer animation, so if there is something that needs changing, then it is either your setup or your model or your understanding of the physics. tricking a simulation into looking like you expect it, is not science.

one thing that looks strange in your input is your low convergence for pppm. please let us also know how you compiled the GPU package and what version of LAMMPS you are using. if you worry about getting accurate results, you should have a kspace convergence of 1.0-6.

axel.

Dear Axel
Thanks for your prompt response.

Through my question sounds physically, it concerns the malfunction of the LAMMPS.

Before proceeding to simulation, I thoroughly skimmed over papers, All of them notified a smooth meniscus front for the flow of water in mineral channels like below.

image.png

Kspace is necessary for realistic simulation but it’s not working at all for my case!

As notified before, my system nicely reproduces the expected flow front by ignoring long-range forces, however, it’s progress rate does not comply with what reported in the literature.

I think something is wrong with the settings of the kspace but I don’t know how to deal with.
It may be a bug or numerical issue with the current stable release of LAMMPS.

The GPU package complied with last released CUDA under instruction in the manual.

Dear Axel
Thanks for your prompt response.

Through my question sounds physically, it concerns the malfunction of the LAMMPS.

as i said before, there are many possible reasons for things not working as expected. claiming there is a malfunction in LAMMPS is making a strong statement and conflicting with many other calculations that work as expected. if you are convinced that LAMMPS is at fault, you need to provide much more convincing evidence than the simulation not behaving as you expect. like real numerical differences.

Before proceeding to simulation, I thoroughly skimmed over papers, All of them notified a smooth meniscus front for the flow of water in mineral channels like below.

image.png

what i see here is water creeping up the walls just in your system. however, there seems to be a much larger reservoir than in your, a vacuum region in z-direction, and non periodicity in x. at least that is what the pictures you show here suggests. the fact that the water does not follow could be attributed to the too small water reservoir. water has to come from somewhere to fill the slit.

Kspace is necessary for realistic simulation but it’s not working at all for my case!

As notified before, my system nicely reproduces the expected flow front by ignoring long-range forces, however, it’s progress rate does not comply with what reported in the literature.

I think something is wrong with the settings of the kspace but I don’t know how to deal with.
It may be a bug or numerical issue with the current stable release of LAMMPS.

i strongly disagree. there is no evidence here confirming this. your setup does not look identical to the images you quote.

The GPU package complied with last released CUDA under instruction in the manual.

that doesn’t tell me the settings you choose to compile and since “last stable” and “last released” version is a term depending on the date when somebody looked it up, it is more scientific and accurate to state the exact version information. and while you are at it, also provide information on the platform you are running on.

looking back at the history of people claiming that LAMMPS has a bug because simulations don’t behave as expected, your probability of being correct is extremely low. almost all of the cases where people had similar claims that yours could be traced back to bad input parameters or bad construction of the simulation setup. so i caution you to not make any claims that cannot be sustained. the scientific approach is to carefully investigate the situation and provide convincing evidence, that numbers produced by LAMMPS are off. e.g. forces for non-kspace and kspace coulomb interactions have to converge to the same values for growing cutoffs of the non-kspace coulomb compared to kspace with sufficiently well converged settings. also, you have to check that GPU and CPU are producing converging results when compiling with since, mixed and then double precision.

axel.