Exporting DeePTB Models to External Tools (PythTB)#
DeePTB not only provides its own powerful training and analysis tools but also allows you to export your trained tight-binding models to popular external formats. This opens up compatibility with a vast ecosystem of post-processing tools.
In this tutorial, we will demonstrate:
DeePTB Native Analysis: Calculating band structure using DeePTB’s built-in tools.
Safety Check: Verifying that the model is compatible with external tools (no overlap).
Export: Converting the DeePTB model to a PythTB model object.
PythTB Analysis: Calculating band structure with PythTB and comparing.
Prerequisites#
dptbpythtbasematplotlib
import os
from dptb.nn import build_model
from dptb.data import AtomicData
from ase.io import read
from dptb.postprocess.interfaces import ToWannier90, ToPythTB
import matplotlib.pyplot as plt
import numpy as np
from dptb.postprocess.bandstructure.band import Band
---------------------------------------------------------------------------
ModuleNotFoundError Traceback (most recent call last)
Cell In[1], line 2
1 import os
----> 2 from dptb.nn import build_model
3 from dptb.data import AtomicData
4 from ase.io import read
ModuleNotFoundError: No module named 'dptb'
1. Load Pre-trained Model#
# Define paths to example data
root_dir = os.path.abspath("path_to_DeePTB/examples/ToPythTB")
model_path = os.path.join(root_dir, "models", "nnsk.ep20.pth")
struct_path = os.path.join(root_dir, "silicon.vasp")
print(f"Loading model from {model_path}...")
model = build_model(model_path)
model.eval()
print("Model loaded.")
The model option atomic_radius in nnsk is not defined in input model_options, set to v1.
Loading model from ...DeePTB/examples/ToPythTB/models/nnsk.ep20.pth...
Model loaded.
2. DeePTB Native Band Structure#
First, we calculate the band structure using DeePTB’s Band class. This serves as our reference.
# Configuration for Band Calculation
jdata={
"task_options": {
"task": "band",
"kline_type":"abacus",
"kpath":[[0.0000000000, 0.0000000000, 0.0000000000, 50],
[0.5000000000, 0.0000000000, 0.5000000000, 50],
[0.6250000000, 0.2500000000, 0.6250000000, 1],
[0.3750000000, 0.3750000000, 0.7500000000, 50],
[0.0000000000, 0.0000000000, 0.0000000000, 50],
[0.5000000000, 0.5000000000, 0.5000000000, 50],
[0.5000000000, 0.2500000000, 0.7500000000, 50],
[0.5000000000, 0.0000000000, 0.5000000000, 1 ]
],
"klabels":["G","X","X/U","K","G","L","W","X"],
"nel_atom":{"Si":4},
"E_fermi":-4.722,
"emin":-18,
"emax":10,
"ref_band": None
}
}
kpath_kwargs = jdata["task_options"]
# Calculate Bands
bcal = Band(model=model, use_gui=False, device=model.device) # use_gui=False for notebook
eigens = bcal.get_bands(data=struct_path, kpath_kwargs=kpath_kwargs)
# Plot DeePTB Bands
print("Plotting DeePTB Band Structure...")
bcal.band_plot(ref_band = None,
E_fermi = kpath_kwargs["E_fermi"],
emin = kpath_kwargs["emin"],
emax = kpath_kwargs["emax"])
eig_solver is not set, using default 'torch'.
Plotting DeePTB Band Structure...
3. Export to PythTB and Verification#
Now we export the valid orthogonal model to PythTB and verify that the band structure matches.
# Initialize Exporter
exporter = ToPythTB(model)
# Export to PythTB model object
print("Exporting to PythTB...")
tb_model = exporter.get_model(struct_path)
print("Successfully exported!")
print(tb_model)
Exporting to PythTB...
Successfully exported!
<pythtb.tb_model object at 0x368c1f400>
Calculate and Plot PythTB Bands#
We use PythTB’s native methods to calculate the bands along the same k-path (approximate mapping) to verify consistency.
# Define K-path in PythTB format (Si High Symmetry Points)
# Gamma=(0,0,0), X=(0,0.5,0.5), L=(0.5,0.5,0.5), W=(0.25, 0.5, 0.75), K=(0.375, 0.375, 0.75)
# Note: DeePTB and PythTB might define paths slightly differently, but critical points match.
k_path_nodes = [
[0.0, 0.0, 0.0], # Gamma
[0.0, 0.5, 0.5], # X
[0.25, 0.5, 0.75], # W
[0.0, 0.0, 0.0], # Gamma
[0.5, 0.5, 0.5], # L
]
k_path_labels = [r'$\Gamma$', 'X', 'W', r'$\Gamma$', 'L']
# Create k-path
(k_vec, k_dist, k_node) = tb_model.k_path(k_path_nodes, 100)
# Solve for eigenvalues
print("Solving with PythTB...")
evals = tb_model.solve_all(k_vec)
# Plotting
fig, ax = plt.subplots(figsize=(8, 6))
# PythTB returns shape (num_orbitals, num_kpoints)
for i in range(evals.shape[0]):
ax.plot(k_dist, evals[i,:], 'r-', lw=1.5, alpha=0.6, label='PythTB' if i==0 else "")
# Decorate
for n in k_node:
ax.axvline(n, color='k', lw=0.5)
ax.set_xticks(k_node)
ax.set_xticklabels(k_path_labels)
ax.set_ylabel("Energy (eV)")
ax.set_title("PythTB Calculated Band Structure")
ax.grid(True, alpha=0.3)
plt.tight_layout()
plt.show()
----- k_path report begin ----------
real-space lattice vectors
[[0. 2.715 2.715]
[2.715 0. 2.715]
[2.715 2.715 0. ]]
k-space metric tensor
[[ 0.10175 -0.03392 -0.03392]
[-0.03392 0.10175 -0.03392]
[-0.03392 -0.03392 0.10175]]
internal coordinates of nodes
[[0. 0. 0. ]
[0. 0.5 0.5 ]
[0.25 0.5 0.75]
[0. 0. 0. ]
[0.5 0.5 0.5 ]]
reciprocal-space lattice vectors
[[-0.18416 0.18416 0.18416]
[ 0.18416 -0.18416 0.18416]
[ 0.18416 0.18416 -0.18416]]
cartesian coordinates of nodes
[[0.00000e+00 0.00000e+00 0.00000e+00]
[1.84162e-01 0.00000e+00 0.00000e+00]
[1.84162e-01 9.20810e-02 6.93889e-18]
[0.00000e+00 0.00000e+00 0.00000e+00]
[9.20810e-02 9.20810e-02 9.20810e-02]]
list of segments:
length = 0.18416 from [0. 0. 0.] to [0. 0.5 0.5]
length = 0.09208 from [0. 0.5 0.5] to [0.25 0.5 0.75]
length = 0.2059 from [0.25 0.5 0.75] to [0. 0. 0.]
length = 0.15949 from [0. 0. 0.] to [0.5 0.5 0.5]
node distance list: [0. 0.18416 0.27624 0.48214 0.64163]
node index list: [ 0 28 43 74 99]
----- k_path report end ------------
Solving with PythTB...
# Configuration for Band Calculation
jdata={
"task_options": {
"task": "band",
"kline_type":"abacus",
"kpath":[[0.0000000000, 0.0000000000, 0.0000000000, 50],
[0.5000000000, 0.0000000000, 0.5000000000, 50],
[0.6250000000, 0.2500000000, 0.6250000000, 1],
[0.3750000000, 0.3750000000, 0.7500000000, 50],
[0.0000000000, 0.0000000000, 0.0000000000, 50],
[0.5000000000, 0.5000000000, 0.5000000000, 50],
[0.5000000000, 0.2500000000, 0.7500000000, 50],
[0.5000000000, 0.0000000000, 0.5000000000, 1 ]
],
"klabels":["G","X","X/U","K","G","L","W","X"],
"nel_atom":{"Si":4},
"E_fermi":-4.722,
"emin":-18,
"emax":10,
"ref_band": None
}
}
kpath_kwargs = jdata["task_options"]
# Calculate Bands
bcal = Band(model=model, use_gui=False, device=model.device) # use_gui=False for notebook
eigens = bcal.get_bands(data=struct_path, kpath_kwargs=kpath_kwargs)
# Plot DeePTB Bands
print("Plotting DeePTB Band Structure...")
bcal.band_plot(ref_band = None,
E_fermi = kpath_kwargs["E_fermi"],
emin = kpath_kwargs["emin"],
emax = kpath_kwargs["emax"])
eig_solver is not set, using default 'torch'.
Plotting DeePTB Band Structure...
PythTB计算能带，使用DeePTB内置的 k点#
evals = tb_model.solve_all(eigens['klist'])
refeig = evals[:,:].T
plt.plot(eigens['xlist'],refeig - refeig.min() ,'r')
plt.plot(eigens['xlist'],eigens['eigenvalues']-eigens['eigenvalues'].min(),'b--')
plt.show()
现在DeePTB完成了与 PythTB 的对接，你可以基于PythTB进行一些性质计算
要深入了解 PythTB 的使用，请参考其官方文档，其中包含详细的安装指南、API 参考和教程：官方文档