Neighbor list with type dependent cutoff radius

[LangLuke]

Hi all,

I am running MD simulations with two particle types (A,B) such that B is much larger than A. This results in the pairwise interactions involving B having a much larger cutoff radius than the A-A interactions.

It seems excessive to store all A-A separations with the cutoff radius necessary for the B interactions. I was wondering if it was possible to use a neighbor list (or combination of neighbor lists) such that cutoff radius for A-B and B-B separations is large but the cutoff radius for A-A separations is small.

The documentation for the neighbor list structures state the following:
"The actual cutoff distance is always the maximum $r_{cut, i,j}$ of the base cutoff and associated pair potentials."
"Neighbor lists nominally include all particles within the chosen cutoff distances."

These two statements seem to imply what I am asking to do is impossible, but am curious if there is a workaround.

Thanks,
Luke

[mphoward]

The documentation is slightly unclear there, but the maximum is done per pair of types (i, j). So, the cutoff for the A-A interactions can indeed be much smaller than A-B or B-B, and HOOMD will handle this correctly out of the box.

I rewrote the neighbor list code several years ago because of exactly the issue and application you are describing, and I recommend using md.nlist.Tree for your neighbor list method. It was optimized for size asymmetric systems.

[joaander]

To clarify, the sentence "The actual cutoff distance is always the maximum $r_{cut, i,j}$ of the base cutoff and associated pair potentials" is trying to say that the r_cut_ij_overall = maximum(r_cut_ij_base, r_cut_ij_potential0, r_cut_ij, potential1, ...). In other words, the maximum is over potentials, not over i and j - the full matrix is kept:

hoomd-blue/hoomd/md/NeighborList.cc

Lines 463 to 485 in 4bc6e20

	// first: loop over the consumer r_cut matrices and take their max in h_r_cut
	for (unsigned int matrix = 0; matrix < m_consumer_r_cut.size(); matrix++)
	{
	ArrayHandle<Scalar> h_consumer_r_cut(*m_consumer_r_cut[matrix],
	access_location::host,
	access_mode::read);
	if (m_consumer_r_cut[matrix]->getNumElements() != m_r_cut.getNumElements())
	{
	throw std::invalid_argument("given r_cut_matrix is not the right size");
	}
	// take the maximum
	for (unsigned int i = 0; i < m_pdata->getNTypes(); ++i)
	{
	for (unsigned int j = 0; j < m_pdata->getNTypes(); ++j)
	{
	h_r_cut.data[m_typpair_idx(i, j)]
	= std::max(h_r_cut.data[m_typpair_idx(i, j)],
	h_consumer_r_cut.data[m_typpair_idx(i, j)]);
	}
	}
	}

The "nominally includes" line is in the context of exclusions, where the documentation states how specifically excluded pairs will be removed from the neighbor list even when they are within a given r_cut distance.

[LangLuke]

Thanks!