Implement Lees-Edwards boundary conditions for shear flow

[wkdarko]

Description

Simulations of bulk shear flow are currently not supported in HOOMD-Blue. Even though box deformations are supported, the necessary periodic boundary conditions to simulate the bulk behavior of shear rheology are not available. We would like to implement Lees-Edwards boundary conditions to support these types of simulations.

Proposed solution

We will implement the Lagrangian-Rhomboid version of Lees-Edwards boundary conditions. The following steps will be taken:

• modify the simulation box object to include deformation of the lattice vectors
• modify the wrap, minImage and shift methods of BoxDim to account for velocity wrapping alongside positions, as particles cross periodic boundaries
• modify integration methods or equations of motion that use PBCs, to include particle velocities
• define a box deformation updater that will trigger at special points in the run loop to set the deformation rate on the box; this object will also have methods for resetting the tilt factors if necessary

Additional context

No response

[mphoward]

We are working on this now and will have a PR implementing the BoxDim modifications ready soon. We have confirmed these changes do not significantly affect performance in benchmarks.

[joaander]

@SchoeniPhlippsn has also worked on code for Lees-Edwards boundary conditions.

[mphoward]

Great! We have completed the first 3 points above and should have a PR open today or Monday. We implemented a full deformation matrix a la LAMMPS fix deform, and I think we propagated the changes everywhere needed in the code but a careful review would be helpful.

@SchoeniPhlippsn do you have a branch you can share with any relevant code? It could be very helpful for finishing the last point to get the deformations set on the box.

@astatt also has a hacked implementation of Lees–Edwards BCs as well as a SLLOD integration method. We are planning to work with her to get the integrator moved into HOOMD once the boundary conditions are taken care of.

[joaander]

How do your changes to wrap work with HPMC? Velocity is unused in most HPMC code paths.

[mphoward]

We overloaded versions of wrap, minImage, and shift that take a velocity, then we reviewed all the code by grep for places they were called. We made no changes in HPMC because we assumed velocity would not be used, so it has all the old methods without velocity still. In MD and MPCD, we checked whether the calls we found needed to also wrap velocity one-by-one. For example, we use the new minImage with velocity for DPD potential, but we use the old minImage without velocity in the other pair potentials.

Before we implemented the methods, we also added the deformation rate members to BoxDim and ran the benchmarks to check if there were any performance regressions from the extra memory requirement for that object. We didn't see anything negative.