Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Activity
    Mohammad Soroush Barhaghi
    @msoroush
    Mie units is in K and you must define the repulsion exponent, here is an example
    Dr. Anand Bharti
    @gandhiforbes
    Thanks @msoroush for the suggestions. I will go through the mentioned examples.
    Zeke A. Piskulich
    @piskuliche
    Hello, I hope this is the right place to ask this. I've been running NPT GEMC simulations of CO2+Acetonitrile to calculate the phase equilibria; however, when I take the liquid composition calculated from GOMC and run an NPT lammps simulation, the density does not agree (and is less than the GEMC value by ~.030 g/cc, which is outside of uncertainty determined as a 95% CI) I am seeing this over a range of pressures. I'm wondering, is there anything I should know about how GOMC calculates its forcefield (I'm using a MIE 12-6 Lennard-Jones force field, with the switch option) that could explain this?
    Thanks!
    I've attached above the input file for my GEMC simulation
    Zeke A. Piskulich
    @piskuliche
    And here is my force field file for GEMC
    Zeke A. Piskulich
    @piskuliche
    (Also I just noticed there is a separate issues page, please let me know if I should post there instead) @msoroush
    Mohammad Soroush Barhaghi
    @msoroush
    @piskuliche how much is the uncertainty in density and molfraction from gomc simulation? I would suggest to add regrowth move to make sure we sample internal configuration of molecule.
    For sanity check, you can run NPT simulation in gomc to see what density you are getting. If you are getting different density, it might means that your gemc simulation hasn’t been equilibrated yet
    One another test that you might be interested to do is to not use switch function, use standard VDW without truncation to see if your results from gomc matches the lammps.
    Zeke A. Piskulich
    @piskuliche
    @msoroush Thanks for the quick reply! I didn't have regrowth moves because they are rigid 3 site molecules (for both MeCN and CO2). I didn't think they were necessary, since there really isn't a ton of internal configurations. Let me know if I am thinking about this incorrectly.
    The uncertainty in the density and the molfraction is fairly small
    for instance, at 10 bar density is 734.277 kg/cc +/- 0.384 kg/cc
    (the uncertainty represents a 95% CI as found from Student's T-distribution)
    and the molfraction of CO2 is 0.1173 +/- 0.00127
    (Again, with the uncertainty being a 95% CI)
    Zeke A. Piskulich
    @piskuliche
    I ran a GOMC test simulation at 10 bar in the NPT ensemble, and verified that it agreed with the GEMC result.
    733 kg/cc +/- 2.088 kg/cc
    So I am fairly sure that my simulations are equilibrated, since the two agree.
    I currently have the last test that you suggested running and I will see if that solves my problem, but I wanted to double check a couple of things:
    1) Is there anything I should know about input pressure? I know in GOMC pressures are in bar and in Lammps they are in atm, which I have taken into account. Is there any sort of correction that occurs with the force field due to the pressure that might be different between the two?
    Zeke A. Piskulich
    @piskuliche
    2) Are there any Long Range Corrections that occur by default? I don't have them enabled, but I am just wondering if there is an additional one that might happen in the background that I am missing
    In the meantime, I am running that last simulation that you suggested and I will update you once that finishes
    Thank you again for your help!
    Mohammad Soroush Barhaghi
    @msoroush
    @piskuliche as long as the initial molecular geometry (bonds and angles) are correct, there would be no need for regrowth move. When truncated functions such as switch and shift is activated in GOMC, it would turn off the VDW LRC automatically.
    The only difference might be how we implemented the switch function. Switch function smoothly shift the energy to zero after switch distance. You can refer to our manual for exact mathematical equation. Using your last test case, we can verify if the problem is the switch function or not.
    Zeke A. Piskulich
    @piskuliche
    @msoroush Sorry for the delay getting back to you - I've been continuing working on troubleshooting this. I have run the last test case that we discussed. When I run NVT GEMC using GOMC with a constant cutoff for pure acetonitrile, I find a density of 730.4 +/- 1.2 kg/cc. When I run an NPT MC simulation using GOMC or with Cassandra, I find a similar density (Cassandra value 731.1 +/- 1.1 kg/cc), but when I run NPT MD with lammps it comes to 715.1 kg/cc, so it seems like the MC programs are consistent with each other.
    Zeke A. Piskulich
    @piskuliche
    I'm currently running a test for a system with no electrostatics to see how they compare between the two programs
    Mohammad Soroush Barhaghi
    @msoroush
    @piskuliche both GOMC and Cassandra are using standard Ewald Summation method to calculate electrostatic interactions. What is your Potential function in GOMC config file (VDW, Shift, Switch)? Are you using Long Range Correction (LRC)? Most MD codes (e.g. GROMACS), if they support LRC, only use attraction part of LJ potential to calculate the LRC. While MC codes considers both repulsion and attraction part of LJ potential. We should also consider the possible error in Lammps's barostat.
    Zeke A. Piskulich
    @piskuliche
    Hi - I've done a few further tests. In my MD simulations, PPPM vs Ewald result in the same density (that still disagrees with the MC results). In my GOMC config file, I am using VDW as the potential function. I've run tests with and without LRC in both my MD and MC simulations. With LRC, MD gives 715.1 +/- 0.5 kg/cc and GOMC gives 730.39 +/- 1.178 kg/cc (and as I mentioned earlier Cassandra agrees with the GOMC result) Without LRC, MD gives 694.8 +/- 0.7 kg/cc and MC gives 722.83 +/- 0.862 . Thus, even without LRC the two codes disagree.
    I'm running some tests of different barostat parameters currently, to see if that solves the issue. I really appreciate all of your help with this.
    Zeke A. Piskulich
    @piskuliche
    Thanks so much @msoroush, I'll keep you in the loop.
    Mohammad Soroush Barhaghi
    @msoroush
    @piskuliche no problem. Let me know if you had any other question.
    Zeke A. Piskulich
    @piskuliche
    @msoroush I solved the issue. You were correct in that it was not a GOMC related issue, but was instead an issue with the way the LAMMPS barostat was coupled to the system. Now that I have fixed that, everything agrees.
    Mohammad Soroush Barhaghi
    @msoroush
    @piskuliche I am glad that everything is sorted out. Let me know if you have any other problem/question.
    Ambarish Kulkarni
    @arkulkarni245_gitlab
    Hi - I'm trying to compare the performance of Raspa vs GOMC for simple adsorption calculations. I looked through the GOMC_Examples, but I could not find one which was run using a GPU. It may be useful to have examples (on github) that compare the expected CPU vs GPU performance. Specifically, is the GPU executable called differently, compared to the CPU one?
    Jeffrey Potoff
    @jpotoff
    @arkulkarni245_gitlab I assume you've seen Figure 6 of the GOMC paper where we compare the performance of GOMC to Raspa and other Monte Carlo codes? https://www.sciencedirect.com/science/article/pii/S2352711018301171#fig6
    Note that in this figure, we take into account sampling efficiency, not just the time to produce X number of Monte Carlo steps.
    All of the related files for the benchmarks given in the GOMC paper can be found on GitHub: https://github.com/GOMC-WSU/Benchmarks
    Also, if you're interested in doing adsorption, you might want to take a look at the python scripts we've built to setup high throughput screening calculations: https://github.com/GOMC-WSU/Tools/tree/master/HTS
    We also have some tutorials we put together on adsorption that can be found here: https://github.com/GOMC-WSU/Workshop/tree/AIChE2019
    Jeffrey Potoff
    @jpotoff
    As for the CPU vs. GPU question, yes, they are called differently. If you take a look at the user's guide, you will see that when you run metamake.sh, separate CPU and GPU executables are created. For most adsorption calculations, there's generally not enough work for the GPU to have a significant performance advantage over the CPU. That said, we've made substantial improvements to the GPU performance recently, and we're working on some other ways to leverage to GPU so we can run multiple small simulations (like adsorption) on a single card.
    Also, both the CPU and GPU versions of the code use OpenMP for parallelization, so you can also run them on multicore CPUs and get a pretty good performance boost as long as you have enough work to do.
    Finally, if you have any feature requests or other feedback for us, please let us know!
    Matias
    @matiasf86
    Hello GOMC community, I wanted to ask how difficult it would be to add a new potential to GOMC, specifically I am interested in the Stillinger Weber potential which has tree body interactions. I have previously implemented this kind of potential for monte carlo moves, in a homemade program, but i am not familiarize with GOMC code structure, maybe pointing out in which file is the energy calculation taking place for the lennar jones potential would be a good start... thanks for any suggestions !
    Jeffrey Potoff
    @jpotoff
    @matiasf86 The best place to start is to take a look at CalculateEnergy.h and CalculateEnergy.cpp and FFparticle.h . Those are all built around the idea of a pair-wise additive force field, so you may need to build equivalent classes for the Stillinger Weber potential.
    Matias
    @matiasf86
    @jpotoff thanks for the hints, I will take a look !
    RajshreeBhardwaj
    @RajshreeBhardwaj
    Respected GOMC Community. I have one query. I have been running NPT GEMC simulation of CO2+DES (choline chloride: ethylene glycol); however output is being displayed for few steps but after that is is showing segmentation fault (core dumped).