Cluster Search

Install the package, then run a 38-point Lennard-Jones search with the measured configuration.

pip install anneal

import numpy as np
from anneal import Config, cluster_search

def lj(x):
    p = x.reshape(-1, 3)
    d = p[:, None] - p
    r2 = (d * d).sum(-1)
    iu = np.triu_indices(len(p), 1)
    inv6 = (1.0 / r2[iu]) ** 3
    return float(4.0 * np.sum(inv6 * inv6 - inv6))

def lj_g(x):
    p = x.reshape(-1, 3)
    d = p[:, None] - p
    r2 = (d * d).sum(-1)
    np.fill_diagonal(r2, np.inf)
    inv2 = 1.0 / r2
    inv6 = inv2 ** 3
    c = 24.0 * inv2 * (2.0 * inv6 * inv6 - inv6)
    return np.einsum("ij,ijk->ik", c, d).ravel()

cfg = Config.recommended(38)
out = cluster_search(lj, lj_g, cfg.n_points, 4000, seed=0, recommended=True)
print(out["hops"], round(out["best_energy"], 3), len(out["best"]))

Expected output (seed 0)::

PLACEHOLDER

Direct quantum-chemistry profiles

The molecular-cluster and slab examples use the same persistent ProfileEngine. The adapter loads a conforming potential library once, turns each energy-and-force result into the optimizer’s value-and-gradient pair, and owns the session for the full hop loop. Selecting nwchemc calls libnwchemc in process; no RPC server or result cache participates. Molecular requests omit a simulation cell and slab requests carry their periodic cell through the same request type.

POTENTIAL_CONFIG=/path/to/PotentialConfig.bin \
POTENTIAL_LIBRARY=/path/to/libnwchemc.so \
cargo run --locked --release --features rgpot-ex \
  --example molecular_cluster -- 6 1200 8 nwchemc

The shared adapter is examples/common/profile_engine.rs. The molecular and periodic consumers are examples/molecular_cluster.rs and examples/slab_adsorption.rs.