New Material Modelling Technique Predicts Microstructure Formation Without Complex Calculations

Spinodal decomposition, a fundamental process governing microstructure formation in materials, frequently hinders accurate predictive modelling due to the difficulty of establishing parameter-free methods that effectively represent local energy landscapes. Simon Divilov, Hagen Eckert, and Nico Hotz, all from the Materials Science department at Duke University, alongside Xiomara Campilongo and Stefano Curtarolo, also of Duke University, present a novel approach to predict spinodal behaviour by implementing a disorder viscosity correction to bulk free energies calculated from small, representative cells. This research is significant as it approximates the energy penalty associated with transitioning to a disordered state, stabilising regions of concave bulk free energy crucial for interface formation, whilst simultaneously suppressing long-range concentration fluctuations. By avoiding complex ab initio parameterisation of interfacial properties, the team delivers a scalable route suitable for high-throughput computational frameworks, ultimately offering improved prediction of spinodal regions in complex, high-entropy materials.

This work addresses a longstanding challenge in materials modelling, the need for predictive capability without relying on experimentally determined parameters.

The research introduces a ‘disorder viscosity correction’ applied to standard free energy calculations, enabling the prediction of when and how materials will separate into distinct compositional phases. By accounting for the energetic cost of disorder, the approach stabilises the initial stages of interface formation, a key aspect of spinodal decomposition, while avoiding unphysical long-range fluctuations.

This innovative technique circumvents the complexities of directly calculating interfacial properties from first principles, offering a pathway towards high-throughput materials design and integration with machine-learning frameworks. The ability to accurately predict spinodal regions opens doors to designing materials with tailored microstructures for improved hardness, magnetism, and other critical performance characteristics.

Current approaches frequently assume thermodynamic equilibrium, an unrealistic condition for many real materials containing defects or exhibiting sluggish kinetics. This new method, however, acknowledges that imperfections and kinetic constraints can stabilise local concavity in the bulk free energy, initiating spinodal decomposition. By correcting free energy calculations from small, representative cells, the researchers effectively mimic the influence of these stabilising factors.

The disorder viscosity correction not only predicts the temperature at which spinodal decomposition begins, but also the characteristic wavelength of the resulting microstructure. This dual prediction capability is essential for controlling the final material properties, as both temperature and wavelength influence performance. The researchers demonstrate the scalability of their approach, paving the way for rapid screening of high-entropy materials, alloys containing multiple principal elements, to identify promising candidates with desirable microstructural features and enhanced functionality.

Spinodal decomposition temperatures for AuPt alloys reach 1300 K at a composition of 0.5, as determined by direct comparison with experimental data. These values align closely with experimental observations, demonstrating the predictive power of the computational approach. For CuNi alloys, the calculated spinodal temperature peaks at approximately 1000 K, again exhibiting strong agreement with published experimental results.

Discrepancies observed in CuNi at compositions far from the centre of the spinodal curve are likely due to overestimation of the transition temperature in the Clapp, Moss model used to derive experimental values. Analysis of the maximum spinodal wavelength for Au/Pt alloys reveals values around 200 Å at an undercooling temperature of 200 K. Similarly, Cu/Ni alloys exhibit a maximum spinodal wavelength of approximately 150 Å under the same temperature conditions.

While the disorder viscosity correction to the free energy does not significantly influence the predicted wavelengths, the calculated values provide a reasonable estimate, despite inherent limitations of the regular solid-solution approximation. Extending the analysis to ternary systems, the spinodal temperature for (Ti,Zr)C reaches 3000 K at a composition of 0.6, closely matching experimental findings.

For (Ga,In)N, the calculated spinodal temperature peaks at 2000 K, again demonstrating good agreement with experimental data. Similarly, (Ca,Sr)O and (K,Na)Cl exhibit spinodal temperatures of 1500 K and 1000 K, respectively, aligning well with available experimental measurements. The consistent validation across binary and ternary systems underscores the robustness and scalability of the methodology.

A disorder viscosity correction (DVC) approach, based on finite cells, forms the core of this work to predict spinodal behaviour in materials. The methodology circumvents the need for detailed ab initio parameterisation of interfacial properties, a significant challenge in modelling spinodal decomposition. Initial calculations involve generating periodic order cells (POCC), representative cells defining the compositional space, and determining their ab initio energies using density functional theory (DFT) within the AFLOW standard, a computational framework for materials science.

These energies are then used to compute statistical moments and cumulants, quantifying the energy distribution within the disordered system at each concentration grid point. A key innovation lies in approximating the self-interacting energy of the disordered system, a computationally intensive quantity. Rather than a full calculation, this energy is initially estimated as the ensemble average energy of the disordered system, simplifying the process while retaining essential physics.

Subsequently, multivariate polynomials are fitted to the calculated cumulants and self-interacting energies, enabling the reconstruction of the bulk free energy as a function of concentration and temperature. This reconstructed free energy is then corrected by a disorder viscosity term, effectively introducing a penalty for transitioning into a disordered state.

The magnitude of this disorder viscosity is carefully chosen to preserve local concavity within the bulk free energy, a crucial condition for stabilising interfaces and initiating spinodal kinetics. By preventing long-range compositional fluctuations, the DVC method avoids unphysical, infinite phase separation. This approach allows for the efficient determination of the spinodal temperature and wavelength, facilitating high-throughput screening of compositionally complex and high-entropy materials for desirable microstructural features.

The workflow culminates in storing the calculated spinodal temperature and wavelength for each composition, providing a predictive map of spinodal regions within the miscibility gap. Scientists have long struggled to accurately model spinodal decomposition, a process fundamental to creating materials with tailored microstructures. The difficulty lies in predicting how materials break down into distinct phases without relying on empirically-derived parameters, which limits the ability to design new compositions.

This new work offers a significant step forward by introducing a ‘disorder viscosity’ correction to existing free energy calculations. Rather than attempting a full, computationally expensive description of the interfaces forming during decomposition, the approach cleverly approximates the energy cost of disorder, effectively stabilising the initial stages of phase separation.

By focusing on the energy landscape itself, and acknowledging the inherent resistance to disorder, researchers have created a more scalable method for predicting spinodal behaviour in complex materials, including high-entropy alloys where compositional space is vast. The implications extend beyond fundamental materials science, potentially accelerating the discovery of new alloys with specific properties for applications ranging from high-temperature coatings to advanced energy storage.

However, the approximation inherent in the ‘disorder viscosity’ does introduce a degree of uncertainty. While the method captures the essential physics, validating its accuracy across a wider range of material systems will be crucial. Furthermore, spinodal decomposition is rarely an isolated event; it often occurs alongside other phenomena like grain growth or diffusion. Future work must integrate this predictive capability with broader modelling frameworks to account for these complexities.