Making Merger-trees

treefarm can compute merger-trees either for all halos, starting at the beginning of the simulation, or for specific halos, starting at the final output and moving backward. These two use-cases are covered separately. Halo catalogs must be in the form created by the Gadget FoF halo finder or Subfind substructure finder.

Computing a Full Merger-tree

treefarm accepts a yt time-series object over which the merger-tree will be computed.

import yt
from treefarm import TreeFarm

ts = yt.DatasetSeries("data/groups_*/*.0.hdf5")
my_tree = TreeFarm(ts)
my_tree.trace_descendents("Group", filename="all_halos/")

The first argument to trace_descendents specifies the type of halo object to use. This will typically be either “Group” for FoF groups or Subhalo for Subfind groups. This process will create a new halo catalogs with the additional field representing the descendent ID for each halo. These can be loaded using yt like any other catalogs. Once complete, the final merger-tree can be loaded into ytree.

Computing a Targeted Merger-tree

Computing a full merger-tree can be extremely expensive when the simulation is large. Instead, merger-trees can be created for specific halos in the final dataset, then working backward. Below is an example of computing the merger-tree for only the most massive halo.

import yt
from treefarm import TreeFarm

ds = yt.load("fof_subfind/groups_025/fof_subhalo_tab_025.0.hdf5")
i_max = np.argmax(ds.r["Group", "particle_mass"])
my_id = ds.r["particle_identifier"][i_max]

ts = yt.DatasetSeries("data/groups_*/*.0.hdf5")
my_tree = TreeFarm(ts)
my_tree.trace_ancestors("Group", my_id, filename="my_halo/")

Just as above, the resulting catalogs can then be loaded into a TreeFarm Arbor.

Optimizing Merger-tree Creation

Computing merger-trees can often be an expensive task. Below are some tips for speeding up the process.

Running in Parallel

ytree uses the parallel capabilities of yt to divide up the halo ancestor/descendent search over multiple processors. In order to do this, yt must be set up to run in parallel. See here for instructions. Once this is done, a call to yt.enable_parallelism() must be added to your script.

import yt
yt.enable_parallelism()
from treefarm import TreeFarm

ts = yt.DatasetSeries("data/groups_*/*.0.hdf5")
my_tree = TreeFarm(ts)
my_tree.trace_descendents("Group", filename="all_halos/")

That script must then be run with mpirun.

mpirun -np 4 python my_script.py

Optimizing Halo Candidate Selection

Halo ancestors and descendents are typically found by comparing particle IDs between two halos. The method of selecting which halos should be compared can greatly affect performance. By default, treefarm will compare a halo against all halos in the next dataset. This is both the most robust and slowest method of matching ancestors and descendents. A smarter method is to select candidate matches from only a region around the target halo. For example, treefarm can be configured to select halos from a sphere centered on the current halo.

my_tree = TreeFarm(ts)
my_tree.set_selector("sphere", "virial_radius", factor=5)
my_tree.trace_descendents("Group", filename="all_halos/")

In the above example, candidate halos will be selected from a sphere that is five times the value of the “virial_radius” field. While this will speed up the calculation, a match will not be found if the ancestor/descendent is outside of this region. Some experimentation is recommended to find the optimal balance between speed and robustness.

Currently, the “sphere” selector is the only other selection method implemented, although others can be created easily. For an example, see sphere_selector.

Searching for Fewer Ancestors

When computing a merger-tree for specific halos (Computing a Targeted Merger-tree), you only be interested in the most massive or the few most massive progenitors. If this is the case, treefarm can be configured to end the ancestor search when these have been found, rather than searching for all possible progenitors.

The set_ancestry_filter function places a filter on which ancestors of any given halo will be returned and followed in successive rounds of the merger-tree process. The “most_massive” filter instructs the treefarm to only keep the most massive ancestor. This will greatly reduce the number of halos included in the merger-tree and, therefore, speed up the calculation considerably. For an example of how to create a new filter, see most_massive.

The filtering will only occur after all candidates have been checked for ancestry. An additional operation an be added to end the ancestry search after certain criteria have been met. In the call to set_ancestry_short below, the ancestry search will end as soon as an ancestor with at least 50% of the mass of the target halo has been found. For an example of how to create a new function of this type, see most_massive.

ts = yt.DatasetSeries("data/groups_*/*.0.hdf5")
my_tree = TreeFarm(ts)
my_tree.trace_ancestors("Group", my_id, filename="my_halo/")
my_tree.set_ancestry_filter("most_massive")
my_tree.set_ancestry_short("above_mass_fraction", 0.5)