DLVO Potential and Diverging Forces

[huangtu577]

Hello everybody,
I'm quite new to using hoomd-blue and am trying to simulate a colloidal suspension of spheres. It is essential that the particles can stick together, so I'm using the DLVO potential. However, the potential is not defined for dist < radius_1 + radius_2, and diverges for r \to radius_1 + radius_2. But then that results in a diverging force that does not allow the particles to stick together. Is there a smart way to avoid this?
Thanks in advance and best regards from germany :)

[mphoward]

I think that pair.DLVO might require you to add your own repulsive core to it. Have you considered using an integrated Lennard-Jones potential for the van der Waals attraction in DLVO? It includes both the repulsive core and the attractive tail, and we have implemented this in azplugins as pair.Colloid:

https://azplugins.readthedocs.io/en/stable/module-azplugins-pair.html#hoomd.azplugins.pair.Colloid

Here, the meaning of the radii of the particles is where the Lennard-Jones medium is, so the minimum of the potential is a bit larger than the nominal contact distance due to the finite size of the LJ beads that comprise it.

You will want to use pair.Yukawa with it to get the electrostatic interaction in DLVO theory.

[huangtu577]

Thanks a lot! I think, that's what I'm looking for. I built hoomd from source to install the plugin, which worked fine, but on any GPU older than A100 (like P100, V100), I get an error that I need a higher compute capability:

The device Tesla V100-PCIE-32GB with compute capability 7.0 does not support HOOMD-blue.

What is the minimal CC version supported by hoomd? In the changelog I found that it supports CC >= 6.0, but this is back from the switch to v3.0. From than I did not find any information about supported CCs.
Thanks a lot in advance

[joaander]

The CUDA_ARCH_LIST variable controls which architectures HOOMD-blue builds for. To silence the many compile warnings issued when CUDA 12.8 builds for architectures that NVIDIA has deprecated, this defaults to 80 when you build with CUDA 12.8 or later

hoomd-blue/CMake/hoomd/HOOMDCUDASetup.cmake

Lines 5 to 6 in 2e39291

	if (CMAKE_CUDA_COMPILER_VERSION VERSION_GREATER_EQUAL 12.8)
	set(CUDA_ARCH_LIST 80 CACHE STRING "List of target sm_ architectures to compile CUDA code for. Separate with semicolons.")

You can configure CMake to build with older architectures:

cmake . -DCUDA_ARCH_LIST="60;70;80"

and then rebuild:

ninja

This should work with CUDA 12.8. A future CUDA release is likely to remove support for the deprecated architectures.

[huangtu577]

Thanks a lot for the quick help, that worked very well.