Thursday, January 10, 2013

Enhanced N-body Merger Tree Handling

In addition to generating its own merger trees, Galacticus can read trees from file and process them to populate each tree with galaxies. Typically, these merger trees have been extracted from an N-body simulation.

The original implementation of processing such trees in Galacticus had some limitations - which were necessary to make the trees fit with certain assumptions made in the way in which Galacticus processes trees as it creates and evolves galaxies. This processing modified the structure of trees and so could affect the resulting physical properties of galaxies.

We've just finished some improvements to v0.9.2 of Galacticus which remove these assumptions and allow N-body trees to be handled without the need to modify them in any way1.

The original, simplified trees always looked something like this:

Subhalos (orange circles) were supported, but once a halo became a subhalo, it had to stay a subhalo forever. Also, subhalos could never move between branches of the tree (or between different trees). Furthermore, there was no way for a subhalo to exist unless it had, at some point, a non-subhalo (i.e. isolated halo; green circles) progenitor.

This last point is now fixed, allowing tree configurations such as this:

in which a node begins its life as a subhalo, having never had a progenitor2.

The second improvement was to allow "branch jumps", in which a subhalo jumps to another branch as shown below:
This can happen if the subhalo becomes unbound from its host (or, perhaps, was never bound), leaves the host and travels to a new host.

The final improvement was to allow subhalos to become isolated halos once again, as in the following tree:

Obviously, these are highly simplified trees - trees extracted from typical N-body simulations often have ~100,000 nodes and much more complicated structure.

It's worth stating briefly Galacticus' philosophy regarding merger trees read from file. Galacticus assumes that the trees are "correct" - that is, whatever the nodes in the tree do, it should be treated as real, physical behavior3. This means that if you put garbage in, you'll get garbage out! Therefore, the input merger trees should be as realistic as possible - typically this means processing them to remove numerical artifacts as much as is possible. This approach seems to be the only reasonable one - Galacticus doesn't know enough about the simulation, halo finding, or merger tree construction algorithms used to create a particular set of trees to intelligently figure out if the trees are physically reasonable. That's left as a task for whoever builds the trees.

Technically, these improvements involved modifying Galacticus to allow arbitrary numbers of "events" to be attached to each node of the merger tree. An event object simply specifies the time at which the event occurs, a paired node involved in the event, and a function to handle the event when it occurs. This allows the evolution of nodes in merger trees to be "locked" until any dependent events have occurred - even if those events occur in other merger trees. Currently, two types of event are handled. The first type are "branch jumps", in which a subhalo jumps from one host halo to another. The second event type is "subhalo promotion" in which a subhalo stops being a subhalo and becomes an isolated halo.

1 The new code has been tested by running on the entire set of trees from the Millennium Simulation which contain a total of 800 million halos. These contain a lot of edge cases and halos doing things you wouldn't expect - Galacticus now handles all of these cases robustly. If you try running Galacticus on trees from some other simulation and find problems - let us know!

2 Such an "initial subhalo" won't do anything interesting at present - it has no way to accrete gas and so can't form a galaxy. That could change in future though, and, in any case, the presence of the subhalo is important as it may interact gravitationally with other subhalos and galaxies.

3 Some modification of the tree is possible within Galacticus. For example, you can tell Galacticus to prune branches of the tree below a mass threshold (if you think that the trees are not reliable below that mass scale for example).

1 comment: