Revised velocity verlet scheme to prevent unphysical sticking in DEM simulations.

[dhairyaiitb]

Summary

The half-step lag in the standard velocity-Verlet can lead to errors in DEM simulations. These errors accumulate when we calculate the integral history term using velocity components which have been calculated using velocities from half-step and positions from full step. The revised velocity-Verlet approach resolves this issue.

Related Issue(s)

Author(s)

Dhairya R. Vyas, Postdoctoral researcher, Department of Mechanical Engineering, Northwestern University

Licensing

By submitting this pull request, I agree, that my contribution will be included in LAMMPS and redistributed under either the GNU General Public License version 2 (GPL v2) or the GNU Lesser General Public License version 2.1 (LGPL v2.1).

Backward Compatibility

Implementation Notes

Post Submission Checklist

• The feature or features in this pull request is complete
• Licensing information is complete
• Corresponding author information is complete
• The source code follows the LAMMPS formatting guidelines
• Suitable new documentation files and/or updates to the existing docs are included
• The added/updated documentation is integrated and tested with the documentation build system
• The feature has been verified to work with the conventional build system
• The feature has been verified to work with the CMake based build system
• Suitable tests have been added to the unittest tree.
• A package specific README file has been included or updated
• One or more example input decks are included

Further Information, Files, and Links

[akohlmey]

@dhairyaiitb please go to the src folder, type make fix-whitespace and commit the resulting changes.

[jtclemm]

As discussed, a few changes are needed:

1. an option to turn on/off the new integration scheme and preserve the default behavior (with documentation)
2. rescalings ideally should be moved into general methods to simplify future additions of new contact models
3. if there's a minimal input script that demonstrates the impact of these changes, it would be very helpful to include as an example

[dhairyaiitb]

I have cleaned up the comments and added a test example. I will add documentation and possibly another example in the coming days.

[dhairyaiitb]

Wishing you a very Happy New Year, Joel!

I have updated the documentation with examples illustrating how to use this modification. Additionally, I have some good news: we have received reviews from the journal Computer Physics Communications. The reviewers suggested a few minor changes, which are related to presentation rather than the algorithm. I have also cited this paper into the documentation. Please note that this version of documentation is not for publishing online but specifically for you to review the documentation. I will update the details once the paper is accepted and published online, which will hopefully happen soon.

[akohlmey]

@dhairyaiitb your pull request has a conflict with upstream. You need to check out the current develop branch and merge it into your feature branch and resolve the merge conflict in the process and push it back.

Please also note that your branch has whitespace issues again(!).

[jtclemm]

Thanks for the update, glad to hear you got good reviews. I'll try and find some time to review soon, hopefully next week.

[jtclemm]

Thanks, those changes cover all my major comments. There are a few minor stylistic changes (e.g. spacing and variable names) I want to make. Then, if those look good with you, I'll get Dan to review.

[dhairyaiitb]

Thanks Joel.

[jtclemm]

Hey Dhairya, can you check the changes I just made? Primarily, I swapped the example so synchronization is the default (and added a wall so particles don't get lost) and renaming nx->nx_unrotated and nxuse->nx so the default name is more intuitive. Thanks!

I'll ping Dan to review.

[jtclemm]

Blocking comment until Dan can review

[dhairyaiitb]

Hey Dhairya, can you check the changes I just made? Primarily, I swapped the example so synchronization is the default (and added a wall so particles don't get lost) and renaming nx->nx_unrotated and nxuse->nx so the default name is more intuitive. Thanks!

I'll ping Dan to review.

Looks great. Thanks Joel.

[dsbolin]

Dhairya,
Sorry for the delay. I had a look through the code and the paper. While I wish there was a way to implement it without having modifications in so many places, I can't think of a way to do it, and this seems to work. More importantly, I think this will be a nice way to bring attention to issues with velocity Verlet for history/velocity-dependent pairwise interactions. Looks good to me!

[dhairyaiitb]

Hi Dan,

Thank you for reviewing the code and paper. Glad it looks good to you. We were able to implement it thanks to your and Joel’s help. I really appreciate both of your guidance and support!