SubFind in Practice: Tips, Tricks, and Best Uses

SubFind in Practice: Tips, Tricks, and Best UsesSubFind is a widely used algorithm and software tool for identifying gravitationally bound substructures (subhaloes) within dark matter haloes in cosmological N-body simulations. This practical guide focuses on getting the most out of SubFind: when to use it, how to configure and run it efficiently, how to interpret and validate results, and common pitfalls to avoid. The target audience includes simulation users (students to advanced researchers) working with cosmological outputs from codes such as GADGET, as well as people integrating SubFind into analysis pipelines.

1. What SubFind does (concise overview)

SubFind locates locally overdense regions inside Friends-of-Friends (FoF) groups and determines which of those regions are gravitationally self-bound. The general steps are:

• Identify candidate density peaks by computing a smoothed density field from particle positions.
• Grow substructure candidates around peaks by following density contours.
• For each candidate, iteratively remove unbound particles based on total energy until a bound set remains.
• Report subhalo properties: mass, center, velocity, particle lists, and optionally internal structure measures.

Key takeaway: SubFind finds gravitationally bound substructures inside pre-identified host haloes (FoF groups), not isolated halo finding outside FoF.

2. When to use SubFind vs. alternatives

• Use SubFind if you need physically bound subhaloes inside larger haloes and are working with FoF groups (e.g., GADGET outputs).
• Alternatives/complimentary methods:
  • Rockstar / AHF / VELOCIraptor: often faster or better at temporal tracking and merger trees.
  • 6D phase-space finders (e.g., Rockstar, VELOCIraptor) can separate overlapping substructures better in dense environments.
  • SUBFIND remains reliable for density-based, bound subhalo identification and integrates naturally with many GADGET-based pipelines.

3. Preparing input: best practices

• Start with high-quality snapshots: consistent units, no missing particle types, and appropriate softening lengths recorded.
• Run a Friends-of-Friends (FoF) grouping first with a standard linking length (commonly 0.2 times mean interparticle spacing).
• Ensure sufficient particle resolution in regions of interest: subhalo detection scales with particle number—subhaloes with fewer than ~20–50 particles are highly uncertain.
• If you plan on tracking subhalo evolution, store snapshots at suitably fine time intervals to enable reliable merger trees.

4. Configuration tips

• Smoothing kernel and neighbor numbers: choose neighbor counts that balance noise and resolution. Too few neighbors produce noisy density estimates; too many smear small peaks.
• Unbinding criteria: verify the energy calculation includes all relevant contributions (kinetic + gravitational potential). For baryonic runs include gas/star potentials if using those particle types.
• Minimum particle threshold: set a sensible minimum particle count (e.g., 20–50) for reporting subhaloes to reduce false positives.
• Parallel settings: tune the domain decomposition and MPI processes to balance memory and communication overheads for your cluster.

5. Performance optimizations

• Run FoF and SubFind in the same pipeline to reuse domain decomposition and minimize I/O.
• Use appropriate parallel resources: SubFind’s unbinding step can be expensive; allocate more CPU per snapshot if possible.
• Reduce I/O by producing trimmed snapshots containing only particles inside FoF groups when analyzing a subset of the volume.
• Profile memory usage: ensure each MPI rank has enough memory for local particle buffers and potential arrays.
• Consider using compiled, optimized builds and enabling compiler optimizations (e.g., -O3, OpenMP where supported).

6. Common pitfalls and how to avoid them

• Overinterpreting low-particle subhaloes: treat subhaloes with <50 particles cautiously; they’re often numerical artifacts.
• Boundary effects: subhaloes near FoF boundaries can be split or lost—check linking length and FoF choices.
• Missing particle types: if SubFind isn’t configured for gas/star particles, baryonic components may be ignored, biasing boundness checks.
• Inconsistent units or coordinate systems between FoF and SubFind runs can produce wrong potentials and energies—verify headers and conversion factors.

7. Output interpretation and useful diagnostics

• Particle lists: examine particle membership for suspicious subhaloes (e.g., spatially diffuse sets).
• Mass functions: compare subhalo mass functions to literature expectations as a sanity check.
• Radial distributions: check radial distribution of subhaloes inside host haloes; deviations may indicate finder issues.
• Track conservation: for merger trees, ensure mass/particle continuity across snapshots to detect spurious splitting/merging.
• Visual checks: render density maps and overlay subhalo centers to validate localization.

8. Using SubFind with baryons

• Include all relevant particle types when computing potentials for unbinding if you want physically consistent subhaloes in hydrodynamic runs.
• Gas can be hot and diffuse—decide whether to include unbound gas particles in subhalo lists or filter by temperature/density.
• Stars can tightly trace subhalo centers; using star particles for center-of-mass estimates often improves robustness for luminous subhaloes.

9. Integration into analysis pipelines

• Store particle IDs and group/subhalo IDs to rebuild custom properties later (density profiles, shape tensors).
• Precompute and store properties often reused (e.g., center, velocity, virial mass) to avoid rerunning SubFind unnecessarily.
• For merger trees, use consistent subhalo identifiers or adopt a tracking tool (e.g., consistent-trees, HBT+) that uses SubFind outputs.

10. Advanced usage and research tips

• Substructure within substructure: for high-resolution zooms, run SubFind iteratively at different smoothing scales to probe nested substructure.
• Custom unbinding: modify the unbinding routine to include tidal fields or host potential approximations when studying extreme environments.
• Cross-compare: run multiple finders on the same snapshots to quantify finder-dependent systematics for science results.
• Public data and reproducibility: when publishing, provide SubFind configuration and minimum-particle thresholds to aid reproducibility.

11. Example sanity-check workflow (concise)

1. Run FoF with b=0.2 on snapshot.
2. Run SubFind on FoF groups with min_particles = 32 and neighbor_count tuned to resolution.
3. Inspect subhalo mass function and radial distribution for a sample of host haloes.
4. Visually inspect a few subhaloes with <100 particles; discard if diffuse.
5. Build merger trees using a consistent ID-tracking method.

12. Further reading and resources

• SubFind original papers and method descriptions for mathematical details.
• Comparison studies between halo finders (Rockstar, AHF, VELOCIraptor) for systematic differences.
• Code repositories and user guides for GADGET and SubFind implementations.

If you want, I can: provide a sample SubFind parameter file for GADGET outputs, create scripts to parse SubFind outputs into HDF5/CSV, or compare SubFind vs. Rockstar results for a specific snapshot—tell me which and I’ll generate it.