Initial commit. Stage one work done.

tests/libxc_reference.py

@@ -0,0 +1,67 @@
+"""Loaders for libxc's vendored testsuite data (golden reference).
+
+Input files (libxc/testsuite/input/<system>): first line npoints, then
+9 columns: rhoa rhob sigmaaa sigmaab sigmabb lapla laplb taua taub.
+
+Regression files (libxc/testsuite/regression/<family>/<name>.<system>.<spin>.<order>.bz2):
+first line "func_id npoints order", then a column-header line, then values.
+"""
+
+import bz2
+from pathlib import Path
+
+import numpy as np
+
+LIBXC_ROOT = Path(__file__).resolve().parent.parent / "libxc"
+INPUT_DIR = LIBXC_ROOT / "testsuite" / "input"
+REGRESSION_DIR = LIBXC_ROOT / "testsuite" / "regression"
+
+SYSTEMS = ("BrOH", "BrOH+", "H", "Li")
+
+
+def load_input(system: str) -> dict[str, np.ndarray]:
+    """Return the 9-column grid data for one test system."""
+    raw = np.loadtxt(INPUT_DIR / system, skiprows=1)
+    return {
+        "rho": raw[:, 0:2],  # (N, 2): rhoa, rhob
+        "sigma": raw[:, 2:5],  # (N, 3): sigmaaa, sigmaab, sigmabb
+        "lapl": raw[:, 5:7],
+        "tau": raw[:, 7:9],
+    }
+
+
+def unpolarize(inp: dict[str, np.ndarray]) -> dict[str, np.ndarray]:
+    """Collapse spin channels the way xc-regression.c does for nspin=1."""
+    return {
+        "rho": inp["rho"].sum(axis=1),
+        "sigma": inp["sigma"][:, 0] + 2.0 * inp["sigma"][:, 1] + inp["sigma"][:, 2],
+        "lapl": inp["lapl"].sum(axis=1),
+        "tau": inp["tau"].sum(axis=1),
+    }
+
+
+def load_regression(
+    family_dir: str, functional: str, system: str, spin: str, order: int
+) -> dict[str, np.ndarray]:
+    """Return reference outputs keyed by column name (zk, vrho(a), ...)."""
+    path = (
+        REGRESSION_DIR
+        / family_dir
+        / f"{functional}.{system}.{spin}.{order}.bz2"
+    )
+    with bz2.open(path, "rt") as fh:
+        header = fh.readline().split()
+        func_id, npoints, forder = int(header[0]), int(header[1]), int(header[2])
+        assert forder == order
+        names = fh.readline().split()
+        data = np.loadtxt(fh)
+    data = np.asarray(data).reshape(npoints, len(names))
+    out = {name: data[:, i] for i, name in enumerate(names)}
+    out["_func_id"] = func_id
+    return out
+
+
+def has_regression(family_dir: str, functional: str) -> bool:
+    return any(
+        (REGRESSION_DIR / family_dir).glob(f"{functional}.*.bz2")
+    )