Movatterモバイル変換

[0]ホーム
URL:

Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
Thehttps:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

NIH NLM Logo
Log inShow account info
Access keysNCBI HomepageMyNCBI HomepageMain ContentMain Navigation
pubmed logo
Advanced Clipboard
User Guide

Full text links

Free PMC article
Full text links

Actions

.2022 Aug 23;34(16):7323-7336.
doi: 10.1021/acs.chemmater.2c01293. Epub 2022 Aug 5.

Machine-Learning Rationalization and Prediction of Solid-State Synthesis Conditions

Affiliations

Machine-Learning Rationalization and Prediction of Solid-State Synthesis Conditions

Haoyan Huo et al. Chem Mater..

Abstract

There currently exist no quantitative methods to determine the appropriate conditions for solid-state synthesis. This not only hinders the experimental realization of novel materials but also complicates the interpretation and understanding of solid-state reaction mechanisms. Here, we demonstrate a machine-learning approach that predicts synthesis conditions using large solid-state synthesis data sets text-mined from scientific journal articles. Using feature importance ranking analysis, we discovered that optimal heating temperatures have strong correlations with the stability of precursor materials quantified using melting points and formation energies (ΔGf , ΔHf ). In contrast, features derived from the thermodynamics of synthesis-related reactions did not directly correlate to the chosen heating temperatures. This correlation between optimal solid-state heating temperature and precursor stability extends Tamman's rule from intermetallics to oxide systems, suggesting the importance of reaction kinetics in determining synthesis conditions. Heating times are shown to be strongly correlated with the chosen experimental procedures and instrument setups, which may be indicative of human bias in the data set. Using these predictive features, we constructed machine-learning models with good performance and general applicability to predict the conditions required to synthesize diverse chemical systems.

© 2022 The Authors. Published by American Chemical Society.

PubMed Disclaimer

Conflict of interest statement

The authors declare no competing financial interest.

Figures

Figure 1
Figure 1
Schematic ofthe ML methods developed in this work for predictingsolid-state synthesis conditions.
Figure 2
Figure 2
DI values andrankings of the top 15 synthesis features for heatingtemperature models (a and b) and heating time models (c and d). Thedata set is split into carbonate reactions (reactions with at leastone carbonate precursor) (a and c) and noncarbonate reactions (reactionswith no carbonate precursors) (b and d). Interactional DI (IADI):decrease of modelR2 when a feature isremoved from the whole model that uses all features. Individual DI(IDI):R2 of models trained using onlyone feature. Average partial DI (APDI): averageR2 increase when a feature is added to a submodel. Featuresare ordered according to the sum of all three DI values.
Figure 3
Figure 3
Regression result of linear models. The scatter plotsshow reportedconditions vs predicted conditions for temperature prediction (a)and time prediction (b). Opacity of the markers indicates the weightsof data points. Histograms of prediction errors are also shown.
Figure 4
Figure 4
Average effect of eachchemical element to predicted heating temperatures(a) and times (b) in trained linear models. The values are coefficientsof the corresponding features in the linear models, quantifying howmuch the predicted value changes relatively if a new chemical elementis added to (or removed from) the synthesis.
Figure 5
Figure 5
Modelperformance versus number of training features for both linearand nonlinear (gradient boosting tree regressor) models. Thex-axis shows the number of features used. The features areadded in the order of DI value rankings. The first row shows performancesof temperature prediction models trained on carbonate reactions (a)and noncarbonate reactions (b). The second row shows performancesof time prediction models trained on reactions with (c) and without(d) carbonate precursors.
Figure 6
Figure 6
LOOCV predictionerror distributions of synthesis temperature andtime. Plotted are prediction error median values (shown with whitedots), interquartile ranges (IQR, or the spread of errors between25% and 75% percentiles, shown with thick lines), and 1.5× IQR(shown with thin lines). Shaded areas are probabilistic density estimationsof the errors. Our models are expected to make prediction errors withinthe IQR approximately half of the time and within 1.5× IQR mostof the time.
Figure 7
Figure 7
Performance of the modelsversus the number of features evaluatedon the PCD data set.X-axes show the number of featuresused in each model. Features are added in the order of DI value rankingsas in Figure 2. Theleft panels (a) and (c) show models trained on carbonate reactions,and the right panels (b) and (d) show models trained on noncarbonatereactions. The top panels (a) and (b) show the performance of modelstrained and evaluated on the PCD data set, which represent the upperbounds of OOS scores (c) and (d), which show performance of the modelstrained on the TMR data set. A higher OOS score indicates better modelgeneralizability.
Figure 8
Figure 8
Curves arethe estimated distribution of heating temperatures foreach group of reactions in the training data set. The dashed/dottedlines show temperature distributions for the reaction TiO2 + BaCO3 → BaTiO3 + CO2 (reddashed line for single-heating reactions and blue dotted line formultiple-heating reactions). Green solid line shows the temperaturedistribution for the entire data set.
Figure 9
Figure 9
Fitting resultof Tamman’s rule, i.e., synthesis temperatureis proportional to the average precursor melting point. (a) Scatterplot of the reported vs predicted synthesis temperatures and histogramof prediction error. Opacity indicates data point weights. (b) Regressionparameters and F-test for model significance. A very smallp-value indicates that it is extremely unlikely the resultis due to random noise.
See this image and copyright information in PMC

References

    1. Kohlmann H. Looking into the Black Box of Solid-State Synthesis. Eur. J. Inorg. Chem. 2019, 2019, 4174–4180. 10.1002/ejic.201900733. - DOI
    1. Chamorro J. R.; McQueen T. M. Progress toward solid state synthesis by design. Accounts of chemical research 2018, 51, 2918–2925. 10.1021/acs.accounts.8b00382. - DOI - PubMed
    1. Shoemaker D. P.; Hu Y.-J.; Chung D. Y.; Halder G. J.; Chupas P. J.; Soderholm L.; Mitchell J.; Kanatzidis M. G. In situ studies of a platform for metastable inorganic crystal growth and materials discovery. Proc. Natl. Acad. Sci. U. S. A. 2014, 111, 10922–10927. 10.1073/pnas.1406211111. - DOI - PMC - PubMed
    1. McClain R.; Malliakas C. D.; Shen J.; He J.; Wolverton C.; González G. B.; Kanatzidis M. G. Mechanistic insight of KBiQ2 (Q = S,Se) using panoramic synthesis towards synthesis-by-design. Chemical Science 2021, 12, 1378–1391. 10.1039/D0SC04562D. - DOI - PMC - PubMed
    1. Ito H.; Shitara K.; Wang Y.; Fujii K.; Yashima M.; Goto Y.; Moriyoshi C.; Rosero-Navarro N. C.; Miura A.; Tadanaga K. Kinetically Stabilized Cation Arrangement in Li3YCl6 Superionic Conductor during Solid-State Reaction. Advanced Science 2021, 8, 2101413. 10.1002/advs.202101413. - DOI - PMC - PubMed

LinkOut - more resources

Full text links
Free PMC article
Cite
Send To

NCBI Literature Resources

MeSHPMCBookshelfDisclaimer

The PubMed wordmark and PubMed logo are registered trademarks of the U.S. Department of Health and Human Services (HHS). Unauthorized use of these marks is strictly prohibited.

[8]ページ先頭
©2009-2026 Movatter.jp