A publication-style plot

This example combines concise data loading, deterministic series styles, an inset axis, labels, legends, and square panel geometry. It uses the public paper and output defaults without rebuilding the profile in application code. The dense three-panel layout uses an explicit 8.3-by-2.85-inch design canvas, producing an approximately 5000-pixel-wide PNG at the standard 600 DPI while keeping its labels and inset within the canvas. Ordinary multi-panel calls retain the 170 mm automatic default.

"""Build the concise publication figure used by the documentation."""

from pathlib import Path

import matplotlib.pyplot as plt
import numpy as np

import gsplot as gs

ROOT = Path(__file__).resolve().parent
DATA = ROOT / "data"
NAMES = ("A1g", "A2u", "B1g", "B1u", "Eg1i", "Eg10", "Eu10", "Eg11", "Eu11")
EVEN = ("A1g", "B1g", "Eg1i", "Eg10", "Eg11")
LABELS = (
    r"$A_{1g}$",
    r"$A_{2u}$",
    r"$B_{1g}$",
    r"$B_{1u}$",
    r"$E_g(1,i)$",
    r"$E_g(1,0)$",
    r"$E_u(1,0)$",
    r"$E_g(1,1)$",
    r"$E_u(1,1)$",
)
AXES = (
    (r"$T/T_c$", r"$\Delta_0(T)/k_BT_c$", (0, 1.2), (0, 3)),
    (r"$T/T_c$", r"$C_s/C_n$", (0, 1.2), (0, 3)),
    (r"$T/T_c$", r"$Y(T)$", (0, 1), (0, 1)),
)


def read(group: str, prefix: str, names: tuple[str, ...]) -> list[np.ndarray]:
    return [
        gs.read(DATA / group / f"{prefix}{name}.dat", skip_header=1, delimiter="\t")
        for name in names
    ]


def main() -> None:
    """Build, save, display, and close the publication figure."""
    gap = read("gap", "Gapeq_", NAMES)
    hs = read("heat_capacity", "C_", NAMES)
    ys = read("yosida", "Y(T)_", EVEN)
    fig, ax = gs.subplots("ABC", size=(8.3, 2.85), xspace=0.08)
    sq = gs.inset(
        ax["B"],
        (0.22, 0.64, 0.36, 0.30),
        label=(r"$(T/T_c)^2$", r"$C_s/C_n$", (0, 0.25), (0, 1)),
    )
    for i, (g, h, name) in enumerate(zip(gap, hs, LABELS)):
        t = np.append(h[0], [1, 1.5])
        cap = np.append(h[1], [1, 1])
        kw = {"series": i, "label": name, "ms": 0, "lw": 2}
        gs.line(ax["A"], *g, **kw)
        gs.line(ax["B"], t, cap, **kw)
        gs.line(sq, t**2, cap, **kw)
    for v, name in zip(ys, EVEN):
        i = NAMES.index(name)
        gs.line(ax["C"], *v, series=i, label=LABELS[i], ms=0, lw=2)
    p = {"fontsize": 7, "columnspacing": 0.4, "handlelength": 1.4}
    gs.legend(ax["A"], loc="lower left", props={**p, "ncols": 2})
    gs.legend(ax["C"], loc="upper left", props=p)
    sq.tick_params(labelsize=7, pad=5)
    plt.setp((sq.xaxis.label, sq.yaxis.label), fontsize=7)
    gs.label(ax, AXES, square=True, index="out")
    try:
        gs.save(fig, ROOT / "SC_cal")
    finally:
        plt.close(fig)


if __name__ == "__main__":
    main()

The example data is included in the repository and is also available in the publication data directory.

Running the example writes SC_cal.png and SC_cal.pdf transactionally from the same Figure. gs.save uses 600 DPI for the PNG, Type 42 PDF fonts, and a tight crop with 0.1-inch padding. Its default show=True displays the Figure after successful output; the script then closes its explicitly owned Figure. Sphinx regenerates both ignored build products in a fresh headless subprocess and rejects missing or stale outputs.

The complete example is 73 physical lines, 61 executable lines, 2007 executable characters, and 11 gsplot calls. It retains the reviewed scientific content while staying within the tracked source budgets.

Download the generated vector PDF.

The publication guide explains when to retain the design canvas with crop=False, how these defaults relate to journal dimensions, and which submission requirements still need independent verification.