This package is designed to model solid-state syntheses in the synthesis text-mining project. It has the following objectives:
- Compute thermodynamic quantities for arbitrary compounds by interpolation using DFT data (from the Materials Project, MP).
- Decompose solid-state reactions into pairwise intermediate reactions by optimizing grand potential.
- Calculate synthesis features for machine-learning the prediction of solid-state synthesis conditions.
- Train machine-learning models by properly performing feature engineering, feature selection, and model validation methods.
Please refer to this documentation for a list of package details, the algorithms, and API references.
In this package, the thermodynamic quantity can be calculated using either experimental databases and DFT data from MP. Because the lack of experimental data, not every compound's experimental thermodynamic data can be found.
On the other hand, DFT databases contain enough entries covering much of the chemical space. We also perform interpolation if the compound has no direct match in the DFT database. Therefore, thermodynamic quantities of most compounds can be calculated in this way.
We used the FREED database to compute thermodynamic quantities using experimentally determined data. FREED database is an electronic compilation of the U.S. Bureau of Mines (USBM) experimental thermodynamic data. We don't perform any interpolation to ensure the experimental thermodynamic data are accurate.
As an example, the following code computes the Gibbs formation energy and formation enthalpy for BaCO3.
from s4.thermo.exp.freed import database
print(database.dhf('BaCO3', 300, unit='ev/atom'))
# Prints -2.509443985939046
print(database.dgf('BaCO3', 300, unit='ev/atom'))
# Prints -2.345862145732135Since not every compound has direct match in MP, we perform interpolation of compounds, which is developed by Chris Bartel. Please read the documentation for the details of this interpolation algorithm.
As an example, the following code computes the Gibbs formation energy and formation enthalpy for BaCO3.
from s4.thermo.calc.mp import database
print(database.dhf('BaCO3', 300, unit='ev/atom'))
# Prints -2.5211004719999996
print(database.dgf('BaCO3', 300, unit='ev/atom'))
# Prints -2.361179038215705For BaCO3, the experimental values and the DFT-derived values are very close. This is because we performed additional corrections to the thermodynamic data, please see the documentation for details.
The basic assumption of this cascade construction is the maximum reaction driving force hypothesis, which states that the pairwise reaction happening on reactant interfaces are the ones with the maximum reaction driving force (grand potential). This is demonstrated in the paper by Bianchini et al. and Miura et al.. The details of this algorithm can be found in the documentation.
The following demonstrates an example from Bianchini et al. on the phase evolution of Na2O2 + CoO = NaxCoO2. It also demonstrates the usage of using MP entries to interpolate thermodynamic quantities.
from s4.tmr import (
ReactionEnergies,
MaterialWithEnergy,
MPInterpolatedMaterial,
MPUniverseInterpolation)
from s4.cascade.analysis import compute_cascade
from s4.thermo.calc.mp import query_system
from pymatgen.core import Composition as C
interp = MPUniverseInterpolation()
def dft_thermo(comp_str):
interpolated = interp.interpolate(comp_str)
compositions, info = zip(*interpolated.items())
return MPInterpolatedMaterial(
compositions=compositions,
amounts=[x['amt'] for x in info],
mp_entries=[query_system(x.formula)[0] for x in compositions])
reaction = ReactionEnergies(
target=C('Na2(CoO2)3'),
vars_sub={},
species=[
MaterialWithEnergy(
thermo=dft_thermo('Na2(CoO2)3'),
composition=C('Na2(CoO2)3'),
is_target=True, side='product', amt=1./3),
MaterialWithEnergy(
thermo=dft_thermo('CoO'),
composition=C('CoO'),
is_target=False, side='reactant', amt=1.),
MaterialWithEnergy(
thermo=dft_thermo('Na2O2'),
composition=C('Na2O2'),
is_target=False, side='reactant', amt=1./3),
]
)
compute_cascade(reaction, [500]*10, only_icsd=False)
# Prints
# [{'driving_force': -0.4195086425659669,
# 'temperature': 500,
# 'previous_vessel': {1.0 CoO, 0.3333 Na2O2},
# 'current_vessel': {0.3333 CoO, 0.6667 Na1Co1O2},
# 'reason': 'cascade: determined by minimizing dG/m.a'},
# {'driving_force': -0.037765379636589554,
# 'temperature': 500,
# 'previous_vessel': {0.3333 CoO, 0.6667 Na1Co1O2},
# 'current_vessel': {0.5 Na1Co1O2, 0.1667 Na1Co3O6},
# 'reason': 'cascade: determined by minimizing dG/m.a'},
# {'driving_force': -0.021780617177087437,
# 'temperature': 500,
# 'previous_vessel': {0.5 Na1Co1O2, 0.1667 Na1Co3O6},
# 'current_vessel': {0.2222 Na1Co1O2, 0.1111 Na4Co7O14},
# 'reason': 'cascade: determined by minimizing dG/m.a'},
# {'driving_force': -0.0015646056138710655,
# 'temperature': 500,
# 'previous_vessel': {0.2222 Na1Co1O2, 0.1111 Na4Co7O14},
# 'current_vessel': {0.0667 Na4Co7O14, 0.1333 Na3Co4O8},
# 'reason': 'cascade: determined by minimizing dG/m.a'},
# {'driving_force': 0.0018870757762980397,
# 'temperature': 500,
# 'previous_vessel': {0.0667 Na4Co7O14, 0.1333 Na3Co4O8},
# 'current_vessel': {0.1111 Na3Co4O8, 0.1111 Na3Co5O10},
# 'reason': 'cascade: determined by minimizing dG/m.a'},
# {'driving_force': 0.08069345991540304,
# 'temperature': 500,
# 'previous_vessel': {0.1111 Na3Co4O8, 0.1111 Na3Co5O10},
# 'current_vessel': {0.3333 Na2Co3O6},
# 'reason': 'cascade: determined by minimizing dG/m.a'}]In this package, we compute four types of synthesis features (133 features in total).
Precursor compound properties. The first type of features (12 in total) are the average/ minimum/ maximum/ difference of melting points, standard enthalpy of formation, standard Gibbs free energy of formation of precursors. The melting points are retrieved from the NIST Chemistry WebBook + Wikipedia, while the experimental thermodynamic properties were retrieved from the FREED database.
Target compound compositional features. The second type of features are 74 indicator variables representing the presence of different chemical elements in target compounds.
Reaction thermodynamics features. We used 32 thermodynamic features, including the total reaction driving force, first and last pairwise reaction driving force, and the ratio between first/last pairwise reaction driving force and the total reaction driving force, evaluated at different temperatures T=800, 900, 1000, 1100, 1200, and 1300 degrees Celsius. We also calculated the slope of total/first driving forces by assuming they are linear with respect to temperature and used the slopes as additional features.
Experiment-adjacent features. The fourth type of features are 15 experiment-adjacent features, i.e., indicator variables representing whether certain devices, experiment procedures, and aiding materials were used in the synthesis.
To understand how they are calculated, please refer to features.py. The following code calculates the features for the reaction above as a dictionary.
from s4.ml.features import Featurizer
featurizer = Featurizer()
features = featurizer.featurize(reaction, exp_t=800, exp_time=6)If you find this package useful, please consider citing the following paper:
@article{
doi:10.1021/acs.chemmater.2c01293,
author = {Huo, Haoyan and Bartel, Christopher J. and He, Tanjin and Trewartha, Amalie and Dunn, Alexander and Ouyang, Bin and Jain, Anubhav and Ceder, Gerbrand},
title = {Machine-Learning Rationalization and Prediction of Solid-State Synthesis Conditions},
journal = {Chemistry of Materials},
volume = {34},
number = {16},
pages = {7323-7336},
year = {2022},
doi = {10.1021/acs.chemmater.2c01293},
URL = {https://doi.org/10.1021/acs.chemmater.2c01293},
eprint = {https://doi.org/10.1021/acs.chemmater.2c01293}
}