File size: 15,592 Bytes
e9e349d | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 | #!/usr/bin/env julia
#=
EPC with ML/HPRO AO brackets for diamond + comparison plots.
Loads precomputed AO brackets from ml_epc.py and runs EPC using
ElectronPhonon.jl with ML/HPRO eigenvectors instead of DFT wavefunctions.
Compares results against DFT EPC from diamond.jl (out_dft/).
Prerequisites:
1. diamond.jl (create + run + prepare + calc_ep) → displacements/out_dft/
2. ml_epc.py (all steps) → displacements/scf_0/ao_brackets_{hpro,e3}.npz
Usage:
cd example/diamond/3_epc
julia diamond_ml.jl
=#
using ElectronPhonon, PythonCall, NPZ, Printf, Statistics
# ============================================================
# Paths (same model as diamond.jl)
# ============================================================
SCRIPT_DIR = @__DIR__
path_to_calc = SCRIPT_DIR * "/"
path_to_qe = "/home/apolyukhin/Development/q-e_tmp/"
mpi_ranks = 8
a = 3.567
sc_size = [1, 1, 1]
k_mesh = [6, 6, 6]
natoms = 2
Ndisplace = 6 * natoms
pseudo_dir = SCRIPT_DIR * "/../pseudos/"
unitcell = Dict(
:symbols => pylist(["C", "C"]),
:cell => pylist([
[0.0, a/2, a/2],
[a/2, 0.0, a/2],
[a/2, a/2, 0.0]
]),
:scaled_positions => pylist([
(0.0, 0.0, 0.0),
(0.25, 0.25, 0.25)
]),
:masses => pylist([12.011, 12.011]),
)
scf_parameters = Dict(
:format => "espresso-in",
:kpts => pytuple((k_mesh[1], k_mesh[2], k_mesh[3])),
:calculation => "scf",
:prefix => "scf",
:outdir => "./tmp/",
:pseudo_dir => pseudo_dir,
:ecutwfc => 60,
:conv_thr => 1.0e-13,
:pseudopotentials => Dict("C" => "C.upf"),
:diagonalization => "david",
:mixing_mode => "plain",
:mixing_beta => 0.7,
:crystal_coordinates => true,
:verbosity => "high",
:tstress => false,
:ibrav => 0,
:tprnfor => true,
:nbnd => 8,
:electron_maxstep => 1000,
:nosym => true,
:noinv => true,
)
model = create_model(
path_to_calc = path_to_calc,
abs_disp = 1e-3,
path_to_qe = path_to_qe,
mpi_ranks = mpi_ranks,
sc_size = sc_size,
k_mesh = k_mesh,
Ndispalce = Ndisplace,
unitcell = unitcell,
scf_parameters = scf_parameters,
use_symm = false,
)
# Load DFT electrons/phonons (for k_list and phonons reference)
println("Loading DFT electrons and phonons...")
electrons_dft = load_electrons(model)
phonons = load_phonons(model)
disp_dir = path_to_calc * "displacements/"
scf0_dir = disp_dir * "scf_0/"
# ============================================================
# Helper: load AO brackets from npz
# ============================================================
function load_ao_brackets(npz_file)
data = npzread(npz_file)
U_raw = data["U_list"] # (n_dirs, nk, nk, nbands, nbands) complex128
V_raw = data["V_list"]
ek_raw = data["ek_list"] # (nk, nbands)
ep_raw = data["ep_list"] # (n_dirs, nk, nbands)
epm_raw = data["epm_list"]
ndisp = size(U_raw, 1)
nk = size(U_raw, 2)
U_list = [U_raw[d, :, :, :, :] for d in 1:ndisp]
V_list = [V_raw[d, :, :, :, :] for d in 1:ndisp]
ek_list = [ek_raw[ik, :] for ik in 1:nk]
ep_list = [[ep_raw[d, ik, :] for ik in 1:nk] for d in 1:ndisp]
epm_list = [[epm_raw[d, ik, :] for ik in 1:nk] for d in 1:ndisp]
return U_list, V_list, ek_list, ep_list, epm_list
end
# ============================================================
# Helper: run EPC for one source and save to out_{source}_ao/
# ============================================================
function run_epc_ao(source::String)
npz_file = scf0_dir * "ao_brackets_$(source).npz"
if !isfile(npz_file)
println("WARNING: $(npz_file) not found — run ml_epc.py first")
return false
end
println("\n" * "="^60)
println("Running EPC with AO brackets from $(uppercase(source))")
println("="^60)
U_list, V_list, ek_list, ep_list, epm_list = load_ao_brackets(npz_file)
println(" Displacements: $(length(U_list)), k-points: $(length(ek_list))")
println(" Gamma eigenvalues (eV): ", round.(ek_list[1], digits=4))
nk_total = prod(k_mesh)
out_dir = disp_dir * "out/"
out_ao_dir = disp_dir * "out_$(source)_ao/"
out_backup = disp_dir * "out_backup/"
# Skip if output already complete
if isdir(out_ao_dir) && length(readdir(out_ao_dir)) >= nk_total
println(" Cached output found ($(length(readdir(out_ao_dir))) files) — skipping EPC run")
return true
end
electrons_ao = Electrons(
U_list, V_list,
ek_list, ep_list, epm_list,
electrons_dft.k_list,
)
if isdir(out_dir)
mv(out_dir, out_backup; force=true)
end
mkpath(out_dir)
println("Calculating EPC for $(nk_total) k-points...")
for ik in 1:nk_total
electron_phonon(model, ik, 1, electrons_ao, phonons; phonons_dfpt=false)
if ik % 50 == 0
println(" Done $(ik) / $(nk_total)")
end
end
if isdir(out_ao_dir)
rm(out_ao_dir; recursive=true)
end
mv(out_dir, out_ao_dir)
if isdir(out_backup)
mv(out_backup, out_dir)
end
println("Done! Files saved to: out_$(source)_ao/ ($(length(readdir(out_ao_dir))) files)")
return true
end
# Run EPC for both sources
run_epc_ao("hpro")
run_epc_ao("e3")
# ============================================================
# EPC comparison: parse comparison files
# ============================================================
function parse_epc_dir(dir::String, nk::Int=216)
g = Dict{Tuple{Int,Int,Int}, Float64}()
for ik in 1:nk
fn = joinpath(dir, "comparison_$(ik)_1.txt")
isfile(fn) || continue
open(fn) do f
for line in eachline(f)
cols = split(strip(line))
length(cols) >= 8 || continue
i = parse(Int, cols[1])
j = parse(Int, cols[2])
nu = parse(Int, cols[3])
gv = parse(Float64, cols[8])
g[(i, j, nu)] = gv
end
end
end
return g
end
println("\n" * "="^60)
println("EPC comparison: DFT vs HPRO_AO vs E3_AO")
println("="^60)
out_dft_dir = disp_dir * "out_dft/"
out_hpro_dir = disp_dir * "out_hpro_ao/"
out_e3_dir = disp_dir * "out_e3_ao/"
nk = prod(k_mesh)
nbands = 8
nmodes = 6
n_occ = 4 # occupied bands
if !isdir(out_dft_dir)
println("WARNING: out_dft/ not found — run diamond.jl calc_ep first")
else
g_dft = parse_epc_dir(out_dft_dir, nk)
g_hpro = isdir(out_hpro_dir) ? parse_epc_dir(out_hpro_dir, nk) : Dict()
g_e3 = isdir(out_e3_dir) ? parse_epc_dir(out_e3_dir, nk) : Dict()
println(" DFT: $(length(g_dft)) elements")
println(" HPRO_AO: $(length(g_hpro)) elements")
println(" E3_AO: $(length(g_e3)) elements")
# Per-mode MAE statistics
@printf("\nPer-mode MAE (optical modes ν≥4, |g_DFT|>0.1 meV):\n")
for nu in 4:nmodes
errs_hpro = Float64[]
errs_e3 = Float64[]
for (key, gd) in g_dft
key[3] == nu || continue
abs(gd) > 1e-4 || continue
if haskey(g_hpro, key)
push!(errs_hpro, abs(g_hpro[key] - gd))
end
if haskey(g_e3, key)
push!(errs_e3, abs(g_e3[key] - gd))
end
end
n = length(errs_hpro)
hpro_mae = n > 0 ? mean(errs_hpro) * 1000 : NaN
e3_mae = length(errs_e3) > 0 ? mean(errs_e3) * 1000 : NaN
@printf(" Mode %d: HPRO MAE=%.2f meV E3 MAE=%.2f meV (N=%d)\n",
nu, hpro_mae, e3_mae, n)
end
# Diagonal vs off-diagonal breakdown (optical modes)
for (label, src_g) in [("HPRO", g_hpro), ("E3", g_e3)]
isempty(src_g) && continue
diag_errs = Float64[]
offdiag_errs = Float64[]
vc_errs = Float64[]
for (key, gd) in g_dft
key[3] >= 4 || continue
abs(gd) > 1e-4 || continue
haskey(src_g, key) || continue
i, j = key[1], key[2]
err = abs(src_g[key] - gd)
if i == j
push!(diag_errs, err)
elseif (i <= n_occ) != (j <= n_occ)
push!(vc_errs, err)
else
push!(offdiag_errs, err)
end
end
@printf("\n %s_AO vs DFT (optical, |g_DFT|>0.1 meV):\n", label)
@printf(" occ-occ diagonal MAE = %.2f meV (N=%d)\n",
isempty(diag_errs) ? NaN : mean(diag_errs)*1000, length(diag_errs))
@printf(" occ-occ off-diag MAE = %.2f meV (N=%d)\n",
isempty(offdiag_errs) ? NaN : mean(offdiag_errs)*1000, length(offdiag_errs))
@printf(" occ-cond/cond-occ MAE = %.2f meV (N=%d)\n",
isempty(vc_errs) ? NaN : mean(vc_errs)*1000, length(vc_errs))
all_errs = vcat(diag_errs, offdiag_errs, vc_errs)
if !isempty(all_errs)
@printf(" Overall optical MAE = %.2f meV (N=%d)\n",
mean(all_errs)*1000, length(all_errs))
end
end
# ============================================================
# Comparison plots — write Python script to tempfile and run
# ============================================================
println("\nGenerating comparison plots...")
plot_path1 = SCRIPT_DIR * "/epc_comparison_ml.png"
plot_path2 = SCRIPT_DIR * "/epc_per_kpoint_ml.png"
plot_script = tempname() * ".py"
write(plot_script, """
import sys, os
import numpy as np
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
def parse_epc_dir(dir_path, nk=216):
g = {}
for ik in range(1, nk+1):
fn = os.path.join(dir_path, f"comparison_{ik}_1.txt")
if not os.path.isfile(fn):
continue
with open(fn) as f:
for line in f:
cols = line.split()
if len(cols) < 8:
continue
i, j, nu = int(cols[0]), int(cols[1]), int(cols[2])
g[(ik, i, j, nu)] = float(cols[7])
return g
def make_epc_plot(out_dft_dir, ml_dirs, labels, n_occ=4, nk=216, out_path="epc_comparison.png"):
\"\"\"Scatter plot: 3 columns (occ-occ / cond-cond / occ-cond) per ML method row.\"\"\"
g_dft = parse_epc_dir(out_dft_dir, nk)
optical_keys = [k for k in g_dft if k[3] >= 4 and abs(g_dft[k]) > 1e-4]
g_dft_arr = np.array([g_dft[k] for k in optical_keys]) * 1000 # meV
is_vv = np.array([k[1] <= n_occ and k[2] <= n_occ for k in optical_keys])
is_cc = np.array([k[1] > n_occ and k[2] > n_occ for k in optical_keys])
is_vc = ~is_vv & ~is_cc
cats = [('occ-occ', is_vv, 'C0'), ('cond-cond', is_cc, 'C1'), ('occ-cond', is_vc, 'C2')]
nrows, ncols = len(ml_dirs), 3
fig, axes = plt.subplots(nrows, ncols, figsize=(5*ncols, 5*nrows), squeeze=False)
for row, (ml_dir, ml_label) in enumerate(zip(ml_dirs, labels)):
if not os.path.isdir(ml_dir):
continue
g_ml = parse_epc_dir(ml_dir, nk)
g_ml_arr = np.array([g_ml.get(k, 0.0) for k in optical_keys]) * 1000
for col, (cat_label, mask, color) in enumerate(cats):
ax = axes[row, col]
if mask.sum() == 0:
ax.set_visible(False); continue
gd = g_dft_arr[mask]; gm = g_ml_arr[mask]
mae = np.mean(np.abs(gm - gd))
lim = max(abs(gd).max(), abs(gm).max()) * 1.05
ax.scatter(gd, gm, s=4, alpha=0.3, c=color, rasterized=True)
ax.plot([-lim, lim], [-lim, lim], 'k--', lw=0.8, alpha=0.6)
ax.set_xlim(-lim, lim); ax.set_ylim(-lim, lim); ax.set_aspect('equal')
ax.set_xlabel('DFT g (meV)', fontsize=9)
ax.set_ylabel(f'{ml_label} g (meV)', fontsize=9)
ax.set_title(f'{ml_label} — {cat_label}\\nMAE={mae:.1f} meV N={mask.sum()}', fontsize=9)
plt.tight_layout()
plt.savefig(out_path, dpi=150, bbox_inches='tight')
plt.close(fig)
print(f' Saved: {out_path}')
def make_per_kpoint_plot(out_dft_dir, ml_dirs, labels, n_occ=4, nk=216,
out_path="epc_per_kpoint.png"):
\"\"\"Stacked bar chart: per-k-point MAE for occ-occ / cond-cond / occ-cond.\"\"\"
g_dft = parse_epc_dir(out_dft_dir, nk)
optical_keys = [k for k in g_dft if k[3] >= 4 and abs(g_dft[k]) > 1e-4]
g_dft_arr = np.array([g_dft[k] for k in optical_keys]) * 1000
is_vv = np.array([k[1] <= n_occ and k[2] <= n_occ for k in optical_keys])
is_cc = np.array([k[1] > n_occ and k[2] > n_occ for k in optical_keys])
is_vc = ~is_vv & ~is_cc
ik_arr = np.array([k[0] for k in optical_keys])
k_idx = np.arange(1, nk + 1)
fig, axes = plt.subplots(len(ml_dirs), 1,
figsize=(14, 4.5 * len(ml_dirs)), squeeze=False)
for row, (ml_dir, ml_label) in enumerate(zip(ml_dirs, labels)):
if not os.path.isdir(ml_dir):
continue
g_ml = parse_epc_dir(ml_dir, nk)
g_ml_arr = np.array([g_ml.get(k, 0.0) for k in optical_keys]) * 1000
def per_k_mae(mask):
out = np.zeros(nk)
for ik in k_idx:
sel = (ik_arr == ik) & mask
if sel.any():
out[ik - 1] = np.mean(np.abs(g_ml_arr[sel] - g_dft_arr[sel]))
return out
vv_mae = per_k_mae(is_vv)
cc_mae = per_k_mae(is_cc)
vc_mae = per_k_mae(is_vc)
avg_vv = vv_mae[vv_mae > 0].mean() if vv_mae.any() else 0.0
avg_cc = cc_mae[cc_mae > 0].mean() if cc_mae.any() else 0.0
avg_vc = vc_mae[vc_mae > 0].mean() if vc_mae.any() else 0.0
ax = axes[row, 0]
ax.bar(k_idx, vv_mae, color='C0', label=f'val-val (avg={avg_vv:.1f})', width=1.0)
ax.bar(k_idx, cc_mae, bottom=vv_mae, color='C1',
label=f'cond-cond (avg={avg_cc:.1f})', width=1.0)
ax.bar(k_idx, vc_mae, bottom=vv_mae + cc_mae, color='C2',
label=f'val-cond (avg={avg_vc:.1f})', width=1.0)
ax.set_xlabel('k-point index', fontsize=10)
ax.set_ylabel('MAE (meV)', fontsize=10)
ax.set_title(f'{ml_label}: Per k-point EPC MAE', fontsize=11)
ax.legend(fontsize=9, loc='upper right')
plt.tight_layout()
plt.savefig(out_path, dpi=150, bbox_inches='tight')
plt.close(fig)
print(f' Saved: {out_path}')
if __name__ == "__main__":
out_dft_dir, out_hpro_dir, out_e3_dir, path1, path2 = sys.argv[1:6]
make_epc_plot(out_dft_dir, [out_hpro_dir, out_e3_dir], ['HPRO_AO', 'E3_AO'], out_path=path1)
make_per_kpoint_plot(out_dft_dir, [out_hpro_dir, out_e3_dir], ['HPRO_AO', 'E3_AO'], out_path=path2)
""")
try
# Use epc_ml conda env python; fall back to Sys.which (honours PATH from conda run)
python_exe = let p = expanduser("~/anaconda3/envs/epc_ml/bin/python")
isfile(p) ? p : something(Sys.which("python"), "python")
end
run(`$python_exe $plot_script $out_dft_dir $out_hpro_dir $out_e3_dir $plot_path1 $plot_path2`)
finally
rm(plot_script; force=true)
end
println("\ndiamond_ml.jl done.")
println(" epc_comparison_ml.png — scatter: DFT vs HPRO_AO vs E3_AO")
println(" epc_error_dist_ml.png — error histograms per band category")
end
|