[Paper Review] Predictions and correlation analyses of Ellingham diagrams in binary oxides

Shun‐Li Shang, Shuang Lin+3 more|arXiv (Cornell University)|Aug 10, 2023

Advanced Materials Characterization TechniquesEngineering3 citations

TL;DR

This study presents a comprehensive thermodynamic analysis of 137 binary oxides using newly predicted Ellingham diagrams, enabling qualitative prediction of oxide stability in multicomponent systems. Machine learning-based correlation analysis identifies electronic structures (e.g., d- and s-valence electrons, Mendeleev numbers) and thermochemical properties (e.g., melting point, standard entropy) as key predictors of oxide-forming ability, validated by CALPHAD calculations and experimental data for Fe-20Cr-20Ni and AlCoCrFeNi high-entropy alloys.

ABSTRACT

Knowing oxide-forming ability is vital to gain desired or avoid deleterious oxides formation through tuning oxidizing environment and materials chemistry. Here, we have conducted a comprehensive thermodynamic analysis of 137 binary oxides using the presently predicted Ellingham diagrams. It is found that the active elements to form oxides easily are the f-block elements (lanthanides and actinides), elements in the groups II, III, and IV (alkaline earth, Sc, Y, Ti, Zr, and Hf), and Al and Li; while the noble elements with their oxides nonstable and easily reduced are coinage metals (Cu, Ag, and especially Au), Pt-group elements, and Hg and Se. Machine learning based sequential feature selection indicates that oxide-forming ability can be represented by electronic structures of pure elements, for example, their d- and s-valence electrons, Mendeleev numbers, and the groups, making the periodic table a useful tool to tailor oxide-forming ability. The other key elemental features to correlate oxide-forming ability are thermochemical properties such as melting points and standard entropy at 298 K of pure elements. It further shows that the present Ellingham diagrams enable qualitatively understanding and even predicting oxides formed in multicomponent materials, such as the Fe-20Cr-20Ni alloy (in wt.%) and the equimolar high entropy alloy of AlCoCrFeNi, which are in accordance with thermodynamic calculations using the CALPHAD approach and experimental observations in the literature.

Motivation & Objective

• To develop a comprehensive set of Ellingham diagrams for 137 binary oxides based on thermodynamic predictions.
• To identify fundamental elemental properties that correlate with oxide-forming ability using machine learning.
• To enable predictive understanding of oxide stability in complex multicomponent systems like Fe-20Cr-20Ni and AlCoCrFeNi high-entropy alloys.
• To validate predictions using CALPHAD calculations and experimental observations from the literature.

Proposed method

• Constructed Ellingham diagrams by calculating standard Gibbs energy changes (ΔG°) for oxide formation reactions at varying temperatures.
• Applied two reaction scenarios: (1) metal to MxOy oxide, and (2) one oxide to another (MaOb to MxOy).
• Used the CALPHAD approach for thermodynamic consistency and validation in multicomponent systems.
• Performed machine learning-based sequential feature selection with Gaussian process regression (kernel: matern52) to identify key elemental descriptors.
• Evaluated 42 elemental features (e.g., valence electrons, melting point, electronegativity) for correlation with logPO2 at 1100 K.
• Ranked features by R² goodness-of-fit to determine predictive significance for oxide stability.

Experimental results

Research questions

• RQ1 Which elemental properties most strongly correlate with the thermodynamic stability of binary oxides as indicated by Ellingham diagrams?
• RQ2 Can electronic structure descriptors such as d- and s-valence electrons predict oxide-forming ability across the periodic table?
• RQ3 How do thermochemical properties like melting point and standard entropy at 298 K influence oxide stability?
• RQ4 To what extent can the predicted Ellingham diagrams qualitatively predict oxide formation in complex multicomponent alloys like Fe-20Cr-20Ni and AlCoCrFeNi?
• RQ5 Can machine learning models trained on elemental features accurately predict equilibrium oxygen partial pressures (PO2) for oxide formation?

Key findings

• The f-block elements (lanthanides, actinides), Group II, III, and IV elements (e.g., Sc, Y, Ti, Zr, Hf), and Al and Li are identified as 'active' elements with high oxide-forming ability.
• Coinage metals (Cu, Ag, Au), Pt-group elements, Hg, and Se are classified as 'noble' elements due to thermodynamically unstable oxides easily reducible under low PO2.
• The top predictive features for oxide-forming ability are NdVal (d-valence electrons, R² = 0.727), M_Num2 (Mendeleev number, R² = 0.408), and EleNeg_Pauling (Pauling electronegativity, R² = 0.593).
• Thermochemical properties such as melting point (R² = 0.004) and standard entropy at 298 K (R² = 0.073) show weaker but measurable correlations with oxide stability.
• Predicted Ellingham diagrams successfully explain oxide formation in Fe-20Cr-20Ni and equimolar AlCoCrFeNi high-entropy alloys, consistent with CALPHAD calculations and experimental observations.
• The sequential feature selection identified 15 key descriptors from 42 elemental features, with electronic structure and electronegativity emerging as dominant predictors of oxide-forming tendency.