Reproducing the Paper's Figures

Reproducing the Paper’s Figures¶

Every figure in the paper is produced by one function call. This page is deliberately thin — that it can be thin is the point. The model code lives in enclose, not scattered across notebook cells, so reproducing a figure takes a line rather than a transcription.

Getting set up¶

git clone https://github.com/open-enclose/open-enclose.github.io
cd open-enclose.github.io
pip install -e ".[dev]"

To regenerate everything at once:

python scripts/make_figures.py --outdir Figures

That single command writes all seven paper figures plus the four explanatory figures used elsewhere on this site. It runs in CI on every push, immediately before the site build, so the figures published here cannot drift from the code that generates them.

The paper’s figures¶

Each returns a matplotlib (fig, ax); nothing is written to disk unless you save it.

All seven, in the order the paper prints them:

from enclose import figures

fig, ax = figures.social_optimum(cond_opt=False)     # Fig. 1  social_optimum.png
fig, ax = figures.nash_equilibrium(full_diag=False)  # Fig. 2  nash_eq.png
fig, ax = figures.nash_equilibrium(full_diag=True)   # Fig. 3  nash_so_comp.png
fig, ax = figures.social_optimum(cond_opt=True)      # Fig. 4  social_opt_cond.png
fig, ax = figures.comparison()                       # Fig. 5  comparison.png
fig, ax = figures.combined_4x4()                     # Fig. 6  new_comp_fig4x4.png
fig, ax = figures.monopoly()                         # Fig. 7  monopoly.png

The \label is carried alongside the number because numbers renumber when a figure is added or moved, and labels do not — if the two ever disagree, trust the label.

Verified against main.tex by scripts/check_figure_map.py, which reads the figure environments in source order — which is what determines the printed number — and compares them with the table above.

Explanatory figures¶

Four more are used on this site but appear nowhere in the paper. Two fill placeholders the Mathematical Appendix numbers but never had generators for; two support the manufacturing extension.

fig, ax = figures.labor_reaction(te=0.5, th=1.6, alp=0.5, mu=0.5)
fig, ax = figures.labor_misallocation(te=0.5, alp=0.5, th=1.5, mu=0.0)
fig, ax = figures.manufacturing_equilibrium(te_values=(0.0, 1.0), th=1.0, alp=0.4, b=0.7)
fig, ax = figures.structural_transformation(alp=0.5, mu=0.0, b=0.5)

Changing the parameters¶

Everything is keyword-driven, so exploring is a matter of passing different values. All the paper’s diagrams are drawn at $\alpha = 2/3$ and $c/A = 1$; the loci carry those as defaults.

from enclose import loci
import numpy as np

theta = np.linspace(1.1, 2.1, 200)
loci.ln_l01(theta)                 # first-best no-enclosure locus, eq. (6)
loci.ln_ld0(theta, tau=0.5)        # decentralized, with partial compensation
loci.ln_ld0(theta, mu=1.0, tau=1.0)  # both wedges closed -- equals the planner's locus
loci.ln_gg(theta, tau=1.0)         # global-games locus under full compensation

How the figures are checked¶

Two mechanisms, because figures are easy to get quietly wrong:

• The numeric layer is checked against symbolic derivations. enclose/symbolic.py derives every locus from its objective — differentiating, integrating or comparing endpoints and solving for $\bar l$ — and the test suite asserts the closed forms in loci.py match. A mistranscribed formula fails the build.

• The figures are checked against the paper’s own PNGs. The set generated here was compared pixel-by-pixel against the figures the published paper compiles, and agrees to within the axis-label wording that was deliberately standardised.

pytest tests/