File size: 11,743 Bytes
d975267 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 | #!/usr/bin/env python
"""Compare QE, HPRO, and ML (DeepH-E3) band structures for diamond.
Processes both the pristine unit cell (UC) and pristine 2×2×2 supercell (SC).
For each cell, runs DeepH-E3 inference and plots QE / HPRO / ML bands.
Outputs: band_compare_uc_ml.png, band_compare_sc_ml.png
Usage: python compare_bands.py [params.json]
"""
import glob
import json
import os
import subprocess
import sys
import numpy as np
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
from scipy.linalg import eigh
SCRIPT_DIR = os.path.dirname(os.path.abspath(__file__))
DATA_DIR = os.path.abspath(os.path.join(SCRIPT_DIR, '..', '..', '1_data_prepare', 'data'))
PARAMS_DEFAULT = os.path.abspath(os.path.join(SCRIPT_DIR, '..', '..', '1_data_prepare', 'params.json'))
RESULTS_DIR = os.path.join(SCRIPT_DIR, 'results')
def load_params(path=None):
with open(path or PARAMS_DEFAULT) as f:
return json.load(f)
def find_latest_model():
"""Return path to the latest trained model directory in results/."""
dirs = sorted(glob.glob(os.path.join(RESULTS_DIR, '*')))
dirs = [d for d in dirs if os.path.isdir(d)
and os.path.exists(os.path.join(d, 'best_model.pkl'))]
return dirs[-1] if dirs else None
def infer_paths(cell_label):
"""Return (dataset_dir, graph_dir, output_dir, ini_path) for cell_label."""
base = os.path.join(SCRIPT_DIR, f'infer_{cell_label}')
return (
os.path.join(base, 'dataset'),
os.path.join(base, 'graph'),
os.path.join(base, 'output'),
os.path.join(base, 'eval.ini'),
)
def build_infer_dataset(aodir, dataset_dir):
"""Create dataset_dir/00/ with per-file symlinks to aodir."""
dest = os.path.join(dataset_dir, '00')
os.makedirs(dest, exist_ok=True)
for fname in os.listdir(aodir):
link = os.path.join(dest, fname)
if not os.path.exists(link):
os.symlink(os.path.join(aodir, fname), link)
print(f' Inference dataset ready: {dest}')
def write_eval_ini(ini_path, model_dir, dataset_dir, graph_dir, output_dir,
dataset_name, params):
"""Write eval.ini for DeepH-E3 inference."""
t = params.get('hamiltonian', {})
ini = f"""; DeepH-E3 eval config — generated by compare_bands.py
[basic]
device = {t.get('device', 'cuda')}
dtype = float
trained_model_dir = {model_dir}
output_dir = {output_dir}
target = hamiltonian
inference = False
test_only = False
[data]
graph_dir =
DFT_data_dir =
processed_data_dir = {dataset_dir}
save_graph_dir = {graph_dir}
target_data = hamiltonian
dataset_name = {dataset_name}
get_overlap = False
"""
with open(ini_path, 'w') as f:
f.write(ini)
print(f' Written {ini_path}')
def run_inference(ini_path, params):
"""Launch DeepH-E3 inference in the deeph conda environment."""
t = params.get('hamiltonian', {})
conda_env = t.get('conda_env', 'deeph')
conda_base = params['paths']['conda_base']
deeph_e3_dir = t.get('deeph_e3_dir', '/home/apolyukhin/Development/DeepH-E3')
launcher = os.path.join(SCRIPT_DIR, '_eval_launcher.py')
with open(launcher, 'w') as f:
f.write(f"""import sys, torch
torch.serialization.add_safe_globals([slice])
try:
from torch_geometric.data.data import DataEdgeAttr, DataTensorAttr
from torch_geometric.data.storage import GlobalStorage
torch.serialization.add_safe_globals([DataEdgeAttr, DataTensorAttr, GlobalStorage])
except ImportError:
pass
sys.path.insert(0, '{deeph_e3_dir}')
from deephe3 import DeepHE3Kernel
kernel = DeepHE3Kernel()
kernel.eval('{ini_path}')
""")
activate = (f'source {conda_base}/etc/profile.d/conda.sh'
f' && conda activate {conda_env}')
print(' Running DeepH-E3 inference...')
subprocess.run(['bash', '-c', f'{activate} && python {launcher}'], check=True)
def load_kpath():
with open(os.path.join(DATA_DIR, 'bands', 'kpath.json')) as f:
return json.load(f)
def parse_bands_gnu(gnu_path):
"""Parse QE bands.dat.gnu → (nk, nbnd) array in eV."""
bands, block = [], []
with open(gnu_path) as f:
for line in f:
line = line.strip()
if line:
block.append(float(line.split()[1]))
else:
if block:
bands.append(block)
block = []
if block:
bands.append(block)
if not bands:
raise FileNotFoundError(f'No data in {gnu_path}')
return np.array(bands).T # (nk, nbnd)
def compute_bands(aodir, h5_name, kpts_all, nbnd):
"""Diagonalize H(R) from h5_name in aodir along kpts_all.
Uses the standard generalized eigenproblem eigh(Hk, Sk) matching the
reference approach (add_qe_bands.py): hermitianize both Hk and Sk in
k-space, then solve directly without Löwdin truncation.
Returns (nk, nbnd) eigenvalues in eV.
"""
from HPRO.deephio import load_deeph_HS
from HPRO.constants import hartree2ev
matH = load_deeph_HS(aodir, h5_name, energy_unit=True)
matS = load_deeph_HS(aodir, 'overlaps.h5', energy_unit=False)
matS.hermitianize()
nk = len(kpts_all)
eigs_all = np.empty((nk, nbnd))
print(f' Diagonalizing {h5_name} at {nk} k-points...')
for ik, kpt in enumerate(kpts_all):
if ik % 50 == 0:
print(f' k-point {ik}/{nk}')
Hk = matH.r2k(kpt).toarray()
Sk = matS.r2k(kpt).toarray()
Hk = 0.5 * (Hk + Hk.conj().T)
Sk = 0.5 * (Sk + Sk.conj().T) # hermitianize Sk in k-space (matches reference)
eigs_k = np.sort(np.real(eigh(Hk, Sk, eigvals_only=True)))
n = min(nbnd, len(eigs_k))
eigs_all[ik, :n] = eigs_k[:n] * hartree2ev
if n < nbnd:
eigs_all[ik, n:] = np.nan
return eigs_all
def plot_comparison(x, eigs_qe, eigs_hpro, eigs_ml, x_hs, labels, n_occ,
title, outpath, n_cond=10, ewin=20.0):
"""Plot QE (blue), HPRO (red dashed), ML (green dotted), all VBM-aligned.
MAE is reported separately for occupied bands and the n_cond lowest
conduction bands. Only bands passing through [-ewin, +ewin] eV relative
to VBM are plotted (avoids unphysical high-energy AO-basis artifacts).
"""
vbm_qe = np.max(eigs_qe[:, :n_occ])
vbm_hpro = np.max(eigs_hpro[:, :n_occ])
vbm_ml = np.max(eigs_ml[:, :n_occ])
nbnd_all = min(eigs_qe.shape[1], eigs_hpro.shape[1], eigs_ml.shape[1])
eq = eigs_qe[:, :nbnd_all] - vbm_qe
eh = eigs_hpro[:, :nbnd_all] - vbm_hpro
em = eigs_ml[:, :nbnd_all] - vbm_ml
# MAE only over occupied + n_cond lowest conduction bands
n_cmp = min(n_occ + n_cond, nbnd_all)
def _mae(a, b):
d = np.abs(a - b)
return np.nanmean(d)
mae_hpro_occ = _mae(eq[:, :n_occ], eh[:, :n_occ])
mae_ml_occ = _mae(eq[:, :n_occ], em[:, :n_occ])
mae_hpro_cond = _mae(eq[:, n_occ:n_cmp], eh[:, n_occ:n_cmp]) if n_cmp > n_occ else 0.0
mae_ml_cond = _mae(eq[:, n_occ:n_cmp], em[:, n_occ:n_cmp]) if n_cmp > n_occ else 0.0
print(f' MAE (occ) HPRO: {mae_hpro_occ*1000:.1f} meV ML: {mae_ml_occ*1000:.1f} meV')
print(f' MAE (cond) HPRO: {mae_hpro_cond*1000:.1f} meV ML: {mae_ml_cond*1000:.1f} meV')
fig, ax = plt.subplots(figsize=(7, 5))
for ib in range(nbnd_all):
bq, bh, bm = eq[:, ib], eh[:, ib], em[:, ib]
in_win = np.any((bq > -ewin) & (bq < ewin))
if not in_win:
continue
ax.plot(x, bq, 'b-', lw=1.0, alpha=0.7, label='QE' if ib == 0 else '')
ax.plot(x, bh, 'r--', lw=0.9, alpha=0.7, label='HPRO' if ib == 0 else '')
ax.plot(x, bm, 'g:', lw=1.0, alpha=0.8, label='ML' if ib == 0 else '')
for xv in x_hs:
ax.axvline(xv, color='k', lw=0.8, ls='--')
ax.axhline(0, color='k', lw=0.5, ls=':')
ax.set_xticks(x_hs)
ax.set_xticklabels(labels, fontsize=11)
ax.set_ylabel('Energy (eV)', fontsize=11)
ax.set_xlim(x[0], x[-1])
ax.set_ylim(-ewin, ewin)
ax.set_title(
title + f'\nMAE occ: HPRO={mae_hpro_occ*1000:.1f} meV ML={mae_ml_occ*1000:.1f} meV'
+ f' | cond: HPRO={mae_hpro_cond*1000:.1f} meV ML={mae_ml_cond*1000:.1f} meV',
fontsize=9)
ax.legend(fontsize=9)
fig.tight_layout()
fig.savefig(outpath, dpi=200)
plt.close(fig)
print(f' Saved: {outpath}')
def process_cell(cell_label, model_dir, kpath, params):
"""Run inference + plot for one cell (uc or sc)."""
rec = params['reconstruction']
if cell_label == 'uc':
nbnd = rec['nbnd']
n_occ = 4 # 4 valence electrons in the UC
else:
nbnd = rec.get('nbnd_sc', rec['nbnd'])
n_occ = 4 * 8 # 4 valence e⁻ × 8 UC per 2×2×2 SC
bands_dir = os.path.join(DATA_DIR, 'bands', cell_label)
aodir = os.path.join(bands_dir, 'reconstruction', 'aohamiltonian')
gnu = os.path.join(bands_dir, 'scf', 'bands.dat.gnu')
if not os.path.exists(os.path.join(aodir, 'hamiltonians.h5')):
print(f'[{cell_label}] hamiltonians.h5 not found, skipping')
return
if not os.path.exists(gnu):
print(f'[{cell_label}] bands.dat.gnu not found, skipping')
return
dataset_dir, graph_dir, output_dir, ini_path = infer_paths(cell_label)
pred_h5 = os.path.join(output_dir, '00', 'hamiltonians_pred.h5')
pred_link = os.path.join(dataset_dir, '00', 'hamiltonians_pred.h5')
if not os.path.exists(pred_h5):
print(f'[{cell_label}] Step 1: Building inference dataset...')
build_infer_dataset(aodir, dataset_dir)
os.makedirs(graph_dir, exist_ok=True)
os.makedirs(output_dir, exist_ok=True)
print(f'[{cell_label}] Step 2: Writing eval.ini...')
write_eval_ini(ini_path, model_dir, dataset_dir, graph_dir, output_dir,
dataset_name=f'diamond_qe_e3_pristine_{cell_label}',
params=params)
print(f'[{cell_label}] Step 3: Running DeepH-E3 inference...')
run_inference(ini_path, params)
else:
print(f'[{cell_label}] Using cached prediction: {pred_h5}')
if not os.path.exists(pred_link):
os.symlink(pred_h5, pred_link)
kpts_all = np.array(kpath['kpts_all'])
x_ref = np.array(kpath['x'])
x_hs = kpath['x_hs']
labels = kpath['labels']
print(f'\n[{cell_label}] Loading QE bands...')
eigs_qe = parse_bands_gnu(gnu)
print(f'[{cell_label}] Computing HPRO bands...')
eigs_hpro = compute_bands(aodir, 'hamiltonians.h5', kpts_all, nbnd)
print(f'[{cell_label}] Computing ML bands...')
eigs_ml = compute_bands(dataset_dir + '/00', 'hamiltonians_pred.h5', kpts_all, nbnd)
outpath = os.path.join(SCRIPT_DIR, f'band_compare_{cell_label}_ml.png')
cell_title = 'UC' if cell_label == 'uc' else 'SC (2×2×2)'
print(f'[{cell_label}] Plotting...')
plot_comparison(x_ref, eigs_qe, eigs_hpro, eigs_ml, x_hs, labels, n_occ,
title=f'Diamond {cell_title}: QE vs HPRO vs ML (DeepH-E3)',
outpath=outpath)
def main():
params_path = next((a for a in sys.argv[1:] if not a.startswith('--')), None)
params = load_params(params_path)
model_dir = find_latest_model()
if model_dir is None:
print('ERROR: No trained model found in results/. Run train_ham.py first.')
sys.exit(1)
print(f'Using model: {os.path.basename(model_dir)}\n')
kpath = load_kpath()
for cell_label in ('uc', 'sc'):
print(f'{"="*50}')
print(f'Processing {cell_label.upper()}')
print(f'{"="*50}')
process_cell(cell_label, model_dir, kpath, params)
print('\ncompare_bands.py done.')
if __name__ == '__main__':
main()
|