NaiveDynamics Tutorial

Bounding Volume Hierarchy Neighbor List

In order to generate a neighbor list using BVH traversal, we must

• aquire positional data
• initialize BVH data
• generate a BVH
• traverse the BVH

Here in Julia the full process is as follows:

using NaiveDynamics
using JLD2

f = jldopen("assets/positions/positions.jld2", "r")
myposition = deepcopy(read(f, "pos5000"))
close(f)

spec = SpheresBVHSpecs(; neighbor_distance=0.2, 
                            atom_count=length(myposition),
                            floattype=Float32,
                            atomsperleaf=4 
)


treeData = TreeData(myposition, spec)

my_neighbor_list =  neighbor_traverse(treeData.tree, treeData.position, spec)

In the first three lines, we used JLD2.jl to open a file of pregenerated positions stored in the data folder of NaiveDynamics. We selected pos5000 from the file, indicating a vector of 5000 mutable vectors, which are each 3 Float32's long.

The neighbor_distance is the maximum distance at which two points are considered neighbors. The atomsperleaf determines the number of morton-encoded and sorted atoms that will be assigned to each leaf. The atomsperleaf' can be any whole number that divides evenly into the number of points, or atoms. Increasingatomsperleafwill reduce the height of the bvh, while increasing the amount of distance checks between points. The height of the bvh can be balanced with theatom_count` specification to achieve optimal performance.

With a specification and an array of positions, we construct a bvh, along with related reusable data.

Finally, neighbor traverse will test the overlap of each element of the myposition vector –which has been transformed into the position field of the TreeData structure– in a time efficient way, using the BVH constructed before, and return a list of neighbors. This list is a tuple with 3 components: index to first element of myposition, index to second element, and the Euclidean distance between them.