Network Robustness: Early Warning via Successive KL Divergence

Reproducible experiments and paper for:

Structural Early Warning of Connectivity Collapse in Heavy-Tailed Networks

One-command reproduction

bash run_docker.sh

Requires Docker. Builds the environment, runs all experiments, and compiles paper.pdf.

Outputs:

• paper.pdf — compiled manuscript
• paper/tables/ — LaTeX table fragments
• paper/figures/ — PDF/PNG figures
• paper/data/ — raw CSVs

CAIDA empirical network data

The CAIDA AS-relationship dataset is required for the empirical network results (Step 9). It is not bundled in this repository due to size. To include it:

1. Register at https://www.caida.org/data/request_user_info_forms/as-relationships.xml
2. Download the 20260101.as-rel2.txt.bz2 snapshot
3. Extract and convert to an undirected edge list:

bunzip2 20260101.as-rel2.txt.bz2
awk '!/^#/ {print $1, $2}' 20260101.as-rel2.txt > caida_as_edges.txt

4. Place caida_as_edges.txt in the repo root before running bash run_docker.sh.

If caida_as_edges.txt is absent, the pipeline will skip Step 9 and produce paper.pdf without the CAIDA figure (all synthetic results remain intact).

Running without Docker

python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
bash remake.sh

LaTeX (pdflatex, bibtex) must be installed separately.

Repository structure

core/           — graph models, failure models, metrics, experiment runner
runner/         — sweep orchestration, export, figure generation
experiments/    — control experiments (null control, low-damage control)
scripts/        — plotting scripts
paper/          — generated artifacts (tables, figures, data)
paper.tex       — manuscript source
references.bib  — bibliography
remake.sh       — full pipeline script
run_docker.sh   — one-command Docker entry point

License

Apache 2.0 — see paper.tex for full license notice.