variationaLLL

Reference implementation of the deep-insertion LLL selectors studied in Variational and Majorization Principles in Lattice Reduction (Blanco-Romero & Almenares Mendoza). The four selectors share one greedy loop and differ only in the score they maximise:

Selector	Score
Deep-Var	$\Delta V = -\Delta\sum p_i^2$
SS-GG	$\Delta\sum r_i$
Thermal-Adaptive	$\Delta\sum r_i^{\alpha}$, $\alpha$ from initial CV
G-DLLL	$\Delta V/(k-j)$

The C++ benchmark in src/ is the canonical implementation built on fplll; the Python scripts in scripts/ drive the runs and produce the figures and tables referenced in the paper.

Build

cd src
bash install_fplll.sh
export PKG_CONFIG_PATH=$PWD/deps/lib/pkgconfig:${PKG_CONFIG_PATH:-}
export LD_LIBRARY_PATH=$PWD/deps/lib:${LD_LIBRARY_PATH:-}
mkdir -p build && cd build
cmake .. && make

Reproduce the paper

bash scripts/run_gdlll_final.sh         # writes results/gdlll_final.json
python scripts/generate_gdlll_plots.py  # writes figures/*.pdf
python scripts/generate_paper_tables.py # writes figures/tables/*.tex

A single run of the full suite takes a few hours on 12 threads. For a quick smoke test, invoke src/build/gdlll_benchmark --dims 40 --nlat 4 --families gaussian -o /tmp/q.json directly.

Layout

src/        C++ canonical implementation (gdlll_benchmark)
scripts/    Python: lattice generators, plot/table builders, run drivers
results/    JSON benchmark output (created on first run)
figures/    Plots and LaTeX table snippets (created on first run)

License

MIT.