DMRG / SA Benchmark Pipeline

This Repository implements the Benchmark test used in the paper.

Run everything from inside src/ (paths are relative).

cd src

1. Check Julia

julia --version

Tested with Julia 1.12.6 (ITensors/ITensorMPS need ≥1.9).

2. Create environment + install packages

julia setup_env.jl

→ creates ../.env, installs all important libs

3. Build sysimage

Requires precompile_script.jl in the same folder.

julia --project=../.env build_image.jl

→ produces itensors_image.so.

4. R setup

Rscript r_setup.R

Installs ggplot2, dplyr, tidyr, stringr, ggh4x, readr, patchwork, tikzDevice. Checks for pdflatex (needed by tikzDevice) — install TeX Live / MacTeX if missing.

5. Run pipelines

mkdir -p ../data/DMRG_QUBO ../data/DMRG_PUBO ../data/SA_QUBO ../data/SA_PUBO

For DMRG:

julia -t <number of thread> --project=../.env -Jitensors_image.so DMRG_pipeline.jl --mode <qubo,pubo> --maxfiles <1-90> --total_runs <N> 

For Simulated Annealing:

julia -t <number of thread> --project=../.env -Jitensors_image.so SA_pipeline.jl --mode <qubo,pubo> --maxfiles <1-90> --total_runs <N>

Flag	Meaning
-t 256	thread count — set to your actual core count
-Jitensors_image.so	load precompiled sysimage
--mode qubo/pubo	encoding
--maxfiles N	limit instances processed
--total_runs N	repeats per instance

For our paper we used for maxfiles = 90 and total_runs = 100

To see how the PUBO/QUBO instances are created see:

https://github.com/lfd/QSW25-SAT-Strikes-Back

Checklist

• .env/ created
• itensors_image.so built
• R packages + pdflatex OK
• output dirs exist
• -t matches CPU cores