Predicting Coherent B2 Stability in Ru-Containing Refractory Alloys Through Thermodynamic Elastic Design Maps

arXiv:2601.01326 · cond-mat.mtrl-sci · Submitted 2026-01-04 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Condensed Matter. The day in condensed matter: materials, phases and the measurements under them.

Kai: We're looking at the paper: "Predicting Coherent B2 Stability in Ru-Containing Refractory Alloys Through Thermodynamic Elastic Design Maps".

Mira: Here's the gist of the paper. The development of next-generation refractory superalloys is fundamentally constrained by the competing requirements of high thermodynamic stability and low interfacial energy Through a physics-guided machine learning framework, this paper resolves the "binary paradox" observed in recent experimental surveys where stoichiometric compounds like RuHf fail to achieve their theoretical solvus temperatures despite high melting points The resulting framework establishes a rigorous, constraint-based protocol for alloy design, shifting the paradigm from trial-and-error composition searches to the precise engineering of zero-misfit, high-stability microstructures Computational Methodology The study employs a multi-stage approach integrating high-throughput Density Functional Theory (DFT) calculations with Random Forest screening and Symbolic Regression DFT calculations were performed using the Vienna Ab initio Simulation Package (VASP) within the framework of density functional theory The overall DFT workflow consisted of three stages: (i) systematic generation of BCC and B2 structural models, (ii) full relaxation of cell shape and atomic positions, and (iii) static single-point total energy calculations on the relaxed geometries Descriptor Construction and Rule-of-Mixtures Features Each alloy configuration was represented using a set of physically motivated descriptors derived directly from density functional theory Thermodynamic stability was characterized using DFT formation energies, defined as the formation-energy difference between BCC and B2 phases, denoted as ∆EB2−BCC = E B2 form −E BCC form (1) which quantifies the energetic preference for partitioning into an ordered B2 phase rather than remaining as a disordered BCC solid solution at the same overall composition Elastic coherency between B2 precipitates and BCC matrices was quantified using a lattice-parameter-based misfit descriptor defined by δ = 2(aB2 −aBCC)/ (aB2 +aBCC) (2) which captures both the magnitude and sign of the misfit, which are relevant for elastic strain energy, coherency loss, and precipitate morphology Machine-Learning Optimization of the BCC–B2 Design Space The random forest (RF) model was selected as the primary model because it provides the most stable balance between predictive accuracy, robustness to noisy descriptors, and scalability to larger datasets The RF model maps multidimensional relationships between composition, lattice coherency, and thermodynamic stability across all candidate matrix–B2 combinations The resulting design map reveals a compact region of favorable solutions centered near zero misfit, with compositions falling inside the ±1% interval forming the most coherent class of candidates and those within ±2% remaining viable for controlled precipitation at elevated temperatures Symbolic Regression Formulation of Coherent Stability Symbolic regression (SR) was applied to the RF-screened dataset to derive an interpretable solvus relation, constrained to operate on descriptors with direct thermodynamic or elastic meaning, namely the enthalpy–entropy ratio ∆H/∆Smix and the coherent lattice misfit δ The resulting expression is Tsolvus = 0.1121(∆H/∆Smix) − 20000δ (6) which expresses the solvus temperature as a direct competition between a thermodynamic ordering term and an elastic coherency penalty This relation shows that the thermodynamic ratio forms the backbone of the solvus temperature, and the misfit terms act as subtractive corrections whose influence increases with lattice strain The study establishes that controlling coherency through targeted alloying additions is essential for translating theoretical thermodynamic stability into experimentally realized B2 precipitates The framework developed here offers a transferable methodology for integrating high-throughput calculations, machine learning, and experimental validation to guide alloy discovery The work establishes that misfit is the primary practical bottleneck for realizing the high enthalpic stability of Ru-containing B2 phases Acknowledgments This work was supported by the National Science Foundation through ACCESS supercomputing allocations MAT250103 and MAT240094 Additional computational resources were provided by Argonne National Laboratory under the Director’s Discretionary allocation for the project AIAlloyLW Data Availability All datasets generated and analyzed during this study, including the high-throughput DFT formation energies and lattice parameters, are openly available in the project repository The repository also contains the trained Random Forest regressor, the Symbolic Regression pipeline scripts (based on PySR), and the complete optimization results for the candidate alloy systems References

1: C.

Kai: Jin, K.

Mira: Xiong, W.

Kai: Li, H.

Title and authors: Kai: Let's talk about who wrote this, Avik Mahata is from Merrimack College. He's tackling this complex materials design problem by combining different computational tools.

Mira: The title itself tells you a lot: "Predicting Coherent B2 Stability in Ru-Containing Refractory Alloys Through Thermodynamic Elastic Design Maps." It immediately points to the two main drivers they are balancing—thermodynamics and elastic strain.

Lev: It sounds like they are trying to create a map that shows exactly where the stable solutions live, not just a list of possibilities. That’s much more useful for real alloy development than just running simulations one by one.

Kai: Right, because they're using this framework to resolve that "binary paradox" we mentioned earlier—where compounds like RuHf look promising but don't actually reach the theoretical solvus temperatures they should have.

Mira: They are addressing that gap directly by building a model constrained by classical theory of coherent elastic misfit and precipitate shape transitions, following the ideas of Fratzl, Jog, and co-workers. That grounds their machine learning in established physics.

Lev: If they can build a map that links specific chemical substitutions to changes in B2 stability and misfit, that’s the kind of data we need to actually guide alloy selection for things like turbine blades or high-temperature components.

The paper's summary: Kai: So, what does the actual process look like? They start with generating BCC and B2 structural models using Density Functional Theory calculations via VASP. Then they do full relaxation of the cell shape and atomic positions.

Mira: After that, they move to static single-point total energy calculations on those relaxed geometries to get their initial data points for stability. They then construct two key descriptors: formation energy, which is the difference between BCC and B2 phases, called Delta E B2 minus BCC.

Lev: That formation energy tells you how much energetic preference there is for the ordered B2 phase compared to staying as a disordered BCC solid solution at the same overall composition. That's a fundamental thermodynamic measure.

Kai: And they also have this elastic coherency descriptor, which is based on lattice parameters, defined by Delta delta, which captures both the magnitude and sign of the misfit between those two structures.

Mira: So they take these descriptors—the formation energy and the misfit—and feed them into a Random Forest model to map out the relationship between composition, lattice coherency, and thermodynamic stability across all candidate matrix to B2 combinations.

Lev: That Random Forest is where the screening happens, right? It’s trying to find which combinations are viable for forming coherent precipitates in a disordered BCC matrix.

The paper's improvements: Kai: One of the big improvements they propose is using a dual-surrogate architecture instead of just one black box model. They train two distinct Random Forest ensembles, one for energy and one for misfit, to see how they work independently.

Mira: That decoupling allows them to explicitly identify "High Stability / High Misfit" candidates as structurally unviable and "Zero Misfit / Low Stability" candidates that are thermodynamically unstable. It’s much clearer what the constraints are.

Lev: From an engineering standpoint, that separation helps us know exactly which parts of our search space we can immediately discard without wasting time running expensive DFT calculations on them.

Kai: They also apply Symbolic Regression to this screened dataset to derive an interpretable solvus relation. This relation is constrained by descriptors that have direct thermodynamic or elastic meaning, like the enthalpy entropy ratio, Delta H over Delta S mix and that lattice misfit we talked about earlier.

Mira: The resulting expression for the solvus temperature is Tsolvus equals zero point one one two one times Delta H over Delta S mix minus twenty thousand times the absolute value of that lattice misfit. It shows a direct competition between a thermodynamic ordering term and an elastic coherency penalty.

Lev: That final equation, showing that the misfit terms act as subtractive corrections whose influence increases with lattice strain, is really useful because it tells us exactly how much we need to tune our composition to control the temperature at which precipitation occurs.

Conclusion: Kai: So, wrapping up this whole process, the paper shows that controlling coherency through targeted alloying additions is essential for translating theoretical thermodynamic stability into experimentally realized B2 precipitates.

Mira: The framework they developed offers a transferable methodology—integrating high-throughput calculations with machine learning and experimental validation—to guide alloy discovery. They establish that misfit is the primary practical bottleneck for realizing the high enthalpic stability of Ru-containing B2 phases.

Lev: It confirms that we need to focus on engineering low misfit, not just aiming for high formation energy, because the elastic penalty is what actually stops those precipitates from forming or staying stable in practice.

Kai: They’ve mapped out a compact region of favorable solutions centered near zero misfit, with compositions falling inside a plus minus one percent interval forming the most coherent class of candidates and those within plus minus two percent remaining viable for controlled precipitation at elevated temperatures.

Mira: It moves alloy design away from just guessing compositions toward precisely engineering zero-misfit microstructures using this physics-informed approach. It’s a strong step forward in how we use AI to help materials science.

Lev: I think the ability to quickly screen thousands of candidates against both stability and coherency requirements simultaneously is what makes this design map really powerful for real hardware development.

Kai: That's the summary of "Predicting Coherent B2 Stability in Ru-Containing Refractory Alloys Through Thermodynamic Elastic Design Maps." It’s a framework that uses physics to guide the AI search for stable, coherent B2 phases.

Department of Mechanical and Electrical Engineering, Merrimack College

cond-mat.mtrl-sci

Submitted: 2026-01-04

Updated: 2026-11-11

Code: https://github.com/mahata-lab/Ru-B2-BCC-Optimization

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 87/100

The gist: The development of next-generation refractory superalloys is fundamentally constrained by the competing requirements of high thermodynamic stability and low interfacial energy Through a…

Key concepts

Thermodynamic Stability ($\Delta E_{B2-BCC}$)
This measures how energetically favorable it is for an alloy to form the ordered B2 phase compared to a disordered BCC phase. A positive value indicates a preference for the ordered structure, which is essential for high-temperature stability in superalloys.
Lattice Misfit ($\delta$)
Misfit quantifies the difference in lattice parameters between the B2 precipitate and its surrounding BCC matrix. This descriptor captures both how large the mismatch is and whether it causes tensile or compressive strain, directly influencing elastic strain energy and coherency.
Symbolic Regression (SR) Formulation
SR was used to derive a mathematical equation relating solvus temperature to thermodynamic driving forces ($\Delta H/\Delta S$) and lattice misfit. This provides an interpretable, closed-form physical law ($T_{solvus} = 0.1121(\Delta H/\Delta Smix) - 20000|\delta|$) for predicting stability.
Random Forest (RF) Model
The RF model was used to map complex relationships between composition, lattice coherency, and thermodynamic stability. It was chosen for its robustness and ability to handle noisy data, allowing the framework to efficiently screen vast design spaces for favorable alloy combinations.

Terminology

Summary

The development of next-generation refractory superalloys is fundamentally constrained by the competing requirements of high thermodynamic stability and low interfacial energy Through a physics-guided machine learning framework, this paper resolves the binary paradox observed in recent experimental surveys where stoichiometric compounds like RuHf fail to achieve their theoretical solvus temperatures despite high melting points The resulting framework establishes a rigorous, constraint-based protocol for alloy design, shifting the paradigm from trial-and-error composition searches to the precise engineering of zero-misfit, high-stability microstructures

Computational Methodology

The study employs a multi-stage approach integrating high-throughput Density Functional Theory (DFT) calculations with Random Forest screening and Symbolic Regression DFT calculations were performed using the Vienna Ab initio Simulation Package (VASP) within the framework of density functional theory The overall DFT workflow consisted of three stages: (i) systematic generation of BCC and B2 structural models, (ii) full relaxation of cell shape and atomic positions, and (iii) static single-point total energy calculations on the relaxed geometries

Descriptor Construction and Rule-of-Mixtures Features

Each alloy configuration was represented using a set of physically motivated descriptors derived directly from density functional theory Thermodynamic stability was characterized using DFT formation energies, defined as the formation-energy difference between BCC and B2 phases, denoted as ∆EB2−BCC = E B2 form −E BCC form (1) which quantifies the energetic preference for partitioning into an ordered B2 phase rather than remaining as a disordered BCC solid solution at the same overall composition Elastic coherency between B2 precipitates and BCC matrices was quantified using a lattice-parameter-based misfit descriptor defined by δ = 2(aB2 −aBCC)/ (aB2 +aBCC) (2) which captures both the magnitude and sign of the misfit, which are relevant for elastic strain energy, coherency loss, and precipitate morphology

Machine-Learning Optimization of the BCC–B2 Design Space

The random forest (RF) model was selected as the primary model because it provides the most stable balance between predictive accuracy, robustness to noisy descriptors, and scalability to larger datasets The RF model maps multidimensional relationships between composition, lattice coherency, and thermodynamic stability across all candidate matrix–B2 combinations The resulting design map reveals a compact region of favorable solutions centered near zero misfit, with compositions falling inside the ±1% interval forming the most coherent class of candidates and those within ±2% remaining viable for controlled precipitation at elevated temperatures

Symbolic Regression Formulation of Coherent Stability

Symbolic regression (SR) was applied to the RF-screened dataset to derive an interpretable solvus relation, constrained to operate on descriptors with direct thermodynamic or elastic meaning, namely the enthalpy–entropy ratio ∆H/∆Smix and the coherent lattice misfit δ The resulting expression is Tsolvus = 0.1121(∆H/∆Smix) − 20000δ (6) which expresses the solvus temperature as a direct competition between a thermodynamic ordering term and an elastic coherency penalty This relation shows that the thermodynamic ratio forms the backbone of the solvus temperature, and the misfit terms act as subtractive corrections whose influence increases with lattice strain

Conclusion

The study establishes that controlling coherency through targeted alloying additions is essential for translating theoretical thermodynamic stability into experimentally realized B2 precipitates The framework developed here offers a transferable methodology for integrating high-throughput calculations, machine learning, and experimental validation to guide alloy discovery The work establishes that misfit is the primary practical bottleneck for realizing the high enthalpic stability of Ru-containing B2 phases

Acknowledgments

This work was supported by the National Science Foundation through ACCESS supercomputing allocations MAT250103 and MAT240094 Additional computational resources were provided by Argonne National Laboratory under the Director’s Discretionary allocation for the project AIAlloyLW

Data Availability

All datasets generated and analyzed during this study, including the high-throughput DFT formation energies and lattice parameters, are openly available in the project repository The repository also contains the trained Random Forest regressor, the Symbolic Regression pipeline scripts (based on PySR), and the complete optimization results for the candidate alloy systems

References

[1] C. Jin, K. Xiong, W. Li, H. Wu, H. Dai, F. Guo, R. Luo, Y. Mao, Y Wang Effects of v on the microstructure and mechanical properties of hfnbtativx refractory multi-principle element alloys: A combined experimental and computational study Materials Science and Engineering: A 898 (2024) 146401 doi:10. p. 0

Where the paper says it

Keywords: Refractory alloy, Ruthenium, machine learning, First principle calculation 1. Introduction Refractory multi-principal element alloys that are based on group IV and group V elements such as Nb, Mo, Ta, Hf, Zr, and Ti are promising for use above about 1200 ◦C, where Ni-based superalloys begin to approach intrinsic limits [1–5].

1016/j.msea.2024.146401

[5] C Frey, R Silverstein, T M Pollock A high stability b2-containing refractory multiprincipal element alloy Acta Materialia 229 (2022) 117767 doi:10.1016/j.actamat. 2022.117767

[8] E P George D Raabe R Ritchie High-entropy alloys Nature Reviews Materials 4 (8) (2019) 515–534

[9] D B Miracle O N Senkov A critical review of high entropy alloys and related concepts Acta Materialia 122 (2017) 448–511

[33] J M Rickman T Lookman S R Kalidindi Materials informatics: From the atomic-level to the continuum Acta Materialia 168 (2019) 95–110

Supplementary Material: Predicting Coherent B2 Stability in Ru-Containing Refractory Alloys Through Thermodynamic–Elastic Design Maps

S1. Density Functional Theory (DFT) Methodology

S1.1. Computational Parameters and Settings

All first-principles calculations were performed using the Vienna Ab initio Simulation Package (VASP) with the Projector Augmented Wave (PAW) method Exchange-correlation effects were treated using the Generalized Gradient Approximation (GGA) parameterized by Perdew, Burke, and Ernzerhof (PBE) The specific convergence criteria and lattice relaxation protocols used for the high-throughput screening are summarized in Table S1

S2. Random Forest Model Development

S2.3. Dual-Surrogate Architecture

To independently resolve the competing drivers of stability, the machine learning framework utilizes a dual-regressor architecture rather than a single black-box model As evidenced by the serialized models rf energy.joblib and rf misfit. p. 0

Where the paper says it

11(∆H/∆S) − 20000|δ|, we quantify the subtractive penalty imposed by lattice strain. Experimental validation confirms both the predicted stability hierarchy and the strong, linear sensitivity of the solvus temperature to coherent misfit, providing direct support for the symbolic-regression-based design rule.

joblib, two distinct Random Forest ensembles were trained in parallel This decoupling allows for the explicit identification of High Stability / High Misfit candidates (structurally unviable) versus Zero Misfit / Low Stability candidates (thermodynamically unstable)

S3. Symbolic Regression Arithmetic Verification

To demonstrate the derivation of the solvus temperatures presented in the main text, we provide explicit calculations for three representative candidates using the derived law: Tsolvus(◦C) ≈ 0.1121∆Hform∆Smix − 20000 ·δ −273. p. 0

Where the paper says it

By distilling the complex decision boundaries of the ensemble model into a closed-form physical law, Tsolvus ≈ 0. 11(∆H/∆S) − 20000|δ|, we quantify the subtractive penalty imposed by lattice strain.

15 (S1) Values for ∆H, ∆S, and misfit δ are taken directly from the dataset (Table S8)

Table S8: Extended list of B2-BCC candidates

Tsolv is the SR-predicted solvus temperature.

BCC Matrix B2 Precipitate Misfit δ ∆E

B2 f ∆E BCC f Tsolv (◦C) Nb75-V25 Ru75Cu25 Hf 0.0037-0.97 0.05 1897 Ta50-V50 Ru75 Al25 HfAl 0.0166-1.1 (Note: The table shows a slight inconsistency in the provided text vs the description, but this is what is presented) p. 8

Where the paper says it

Representative examples from the BCC and B2 dataset are listed in Table 1. The complete dataset derived from the CSV file is provided in the Supplementary Information.

Table S8: Extended list of B2-BCC candidates

Tsolv is the SR-predicted solvus temperature.

BCC Matrix B2 Precipitate Misfit δ ∆E BCC f Tsolv (◦C) Nb75-V25 Ru75Cu25 Hf 0. p. 4

Where the paper says it

The coherent lattice misfit was defined using the symmetric form δ = 2(aB2 −aBCC) aB2 +aBCC , (2) which is widely used in experimental and computational studies of coherent BCC–B2 systems. Equation (2) treats the BCC matrix as the elastic parent phase and captures both the magnitude and sign of the misfit, which are relevant for elastic strain energy, coherency loss, and precipitate morphology.

0037-0.97 0.05 1897 Ta25-V75 Ru Ti75 V25 0.0166-1.1 (Note: The table shows a slight inconsistency in the provided text vs the description, but this is what is presented)

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Incorporate a physics-guided symbolic regression model into materials design workflows to derive closed-form physical laws for solvus temperature based on thermodynamic driving force and lattice misfit. This allows AI systems to move beyond trial-and-error composition searches by predicting stability using Tsolvus ≈ 0.11(∆H/∆S) − 20000δ.

  2. Develop a dual-surrogate machine learning architecture consisting of a thermodynamic surrogate and a structural surrogate to independently predict formation enthalpy and lattice misfit. This decoupling enables the explicit identification of High Stability / High Misfit candidates versus Zero Misfit / Low Stability candidates, providing clearer design constraints.

  3. Implement feature importance analysis using SHAP values to verify that the machine learning model captures physical causality rather than spurious correlations in misfit prediction. This ensures that the AI system relies on physically meaningful descriptors like atomic size mismatch and lattice parameter difference for structural predictions.

  4. Integrate a constrained design protocol where composition is primarily tuned to zero out the elastic penalty, treating thermodynamic driving force as a secondary resource to be preserved. This provides a quantitative design rule: optimization of Ru-refractory alloys requires a multi-component approach where composition is tuned primarily to zero out the elastic penalty.

  5. Utilize the resulting design map from machine learning optimization to identify compact, physically meaningful regions of the BCC–B2 design space, such as the ±1% interval for coherent candidates and ±2% for near-coherent candidates. This allows AI systems to rapidly screen thousands of candidate compositions against both stability and coherency requirements simultaneously.

Abstract

The development of next-generation refractory superalloys is fundamentally constrained by competing requirements of high thermodynamic stability and low interfacial energy. Ru-based B2 intermetallics possess strong intrinsic ordering, but their use as coherent reinforcing phases requires balancing formation enthalpy against elastic strain. Here, we present a physics-guided machine-learning framework that integrates high-throughput density functional theory (DFT), Random Forest screening, and symbolic regression to map this design space. The approach addresses the "binary paradox" observed in recent experimental studies, where strongly ordered Ru-refractory compounds do not necessarily achieve the highest solvus temperatures. The resulting closed-form relation, T solvus 0.11(Delta H/Delta S) - 20000delta, quantifies the competing effects of thermodynamic stabilization and lattice misfit. Because the electronic-structure calculations are static 0 K calculations, the relation is treated as an effective calibrated design equation rather than a directly calculated finite-temperature free-energy boundary. The model is benchmarked against published B2 solvus measurements, while calculated formation enthalpies and lattice parameters are independently compared with calorimetric and diffraction data. Within the fitted relation, a 1% lattice misfit corresponds to an effective solvus penalty of approximately 200 degrees C. Al/Si substitution further shows that improved lattice matching is accompanied by reduced thermodynamic stabilization, with zero misfit near 11 at.% Al and 7 at.% Si for Nb75V25.

Related papers