ProdNet Generator

Ashwin Bhattathiripad·Vipin P. Veetil v1

Generate a weighted, primitive production network whose sector-to-sector flows reproduce the empirical input–output table, and in which each firm's purchases and sales add up to its real-world size. Reconstructing Large Scale Production Networks — arXiv ↗

Inputs

Economyprebuilt bundle, or a custom IO upload

Data folderserver-side path holding firm_size_by_sector.csv + sector_io_table.csv

Pick a folder (or use the bundled US set) and click validate.

A Scale

Number of firmsBernoulli subsample of the census

Mean degreetarget average links per firm (sets z)

Number of networksgenerate an ensemble (each a fresh draw)

Backbone drawthe binary/backbone draw is the only GPU stage; all others run on CPU

B Constraint bounds

δ^firm — size RMS, all firmsRMS of (inflow − size)/size over every firm

δfirmtail — size RMS, worst 10%the same RMS over the worst-10% mismatched firms

δ^sec — sector RMSRMS of each sector's aggregate inflow vs its total size

δsectail — sector RMS, worst 10%over the worst-10% of sectors

Firm caps are hard; sector caps are a driven target the solver reports. Tail fraction (10%), per-edge floor (1e-7) and degree floor (2) are locked.

C Tolerance & solver

Per-firm toleranceRMS consensus gap per firm (N-invariant)

Max iterationshard cap

· Run

Run labelnames the output folder

Fit λ-familyneeded once per economy / IO table

λ-fit solveronly used when fitting a new economy

Auto fits the λ-family the first time it sees an IO table, then reuses the cached fit.

Solver: MUMPS is license-free. MA97 (HSL) needs your own license installed & linked; if neither is available, the fit can't run in the browser and you'll be told so.

ProdNet Generator

Inputs

A Scale

B Constraint bounds

C Tolerance & solver

· Run

Progressidle

Figures

Tables