Machine Learning-Guided Discovery of Temperature-Induced Solid-Solid Phase Transitions in Inorganic Materials
Cibrán López, Joshua Ojih, Ming Hu, Josep-Lluís Tamarit, Edgardo Saucedo, Claudio Cazorla · 2025
Read Machine Learning-Guided Discovery of Temperature-Induced Solid-Solid Phase Transitions in Inorganic Materials by Cibrán López, Joshua Ojih, Ming Hu, Josep-Lluís Tamarit, Edgardo Saucedo, Claudio Cazorla (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-01449.txt.
Introduction
Cibrán López and five co-authors send a machine learning framework hunting through fifty thousand crystals for ones that switch structure as they warm, in this 2025 paper. Solid-to-solid phase transitions power cooling, memory, and thermal management technologies, but predicting them needs costly finite-temperature physics.
Their pipeline pairs first-principles calculations with graph neural networks trained to estimate vibrational free energies, with uncertainty tracked along the way. Screening the curated compound set turned up over two thousand candidate transitions between 300 and 600 kelvin. Some showed entropy changes above three hundred joules per kilogram-kelvin near room temperature, promising for solid-state cooling. Twenty-one compounds changed their heat conduction by twenty to seventy percent across the switch, marking them as thermal switches. Checks against experiments and heavier calculations supported the predictions. The authors call it a scalable route to functional phase-change materials.
Catalog record
| CREATOR | Cibrán López, Joshua Ojih, Ming Hu, Josep-Lluís Tamarit, Edgardo Saucedo, Claudio Cazorla |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 64 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2506-01449
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Machine Learning-Guided Discovery of Temperature-Induced Solid-Solid Phase Transitions in Inorganic Materials Cibrán López, Joshua Ojih, Ming Hu, Josep-Lluís Tamarit, Edgardo Saucedo, Claudio Cazorla Abstract Abstract. Predicting solid-solid phase transitions remains a long-standing challenge in materials science. Solid-solid transformations underpin a wide range of functional properties critical to energy conversion, information storage, and thermal management technologies. However, their prediction is computationally intensive due to the need to account for finite-temperature effects. Here, we present an uncertainty-aware machine-learning-guided framework for high-throughput prediction of temperature-induced polymorphic phase transitions in inorganic crystals. By combining density functional theory calculations with graph-based neural networks trained to estimate vibrational free energies, we screened a curated dataset of ∼ 50,000 \sim 50,000 inorganic compounds and identified over 2,000 2,000 potential solid-solid transitions within the technologically relevant temperature interval 300 300 – 600 600 K. Among our key findings, we uncover numerous phase transitions exhibiting large entropy changes ( > 300 >300 J K -1 kg -1 ), many of which occur near room temperature hence offering strong potential for solid-state cooling applications. We also identify 21 21 compounds that exhibit substantial relative changes in lattice thermal conductivity ( 20 20 – 70 70 %) across a phase transition, highlighting them as promising thermal switching materials. Validation against experimental observations and first-principles calculations supports the robustness and predictive power of our approach. Overall, this work establishes a scalable route to discover functional phase-change materials under realistic thermal conditions, and lays the foundation for future high-throughput studies leveraging generative models and expanding open-access materials databases. Keywords: solid-solid phase transitions, machine learning, first-principles calculations, high-throughput screening, solid-state cooling, thermal switching, inorganic crystals I Introduction Solid-solid phase transitions play a crucial role in governing a wide range of physical phenomena and underpin numerous technological applications. For example, polymorphic transitions in crystalline materials influence mechanical properties in metallurgy, as seen in the martensitic transformation of steel, which enhances its strength and toughness. In functional materials, phase transitions enable key applications, such as the ferroelectric-paraelectric transition in BaTiO 3 , which is essential for capacitors, or the shape-memory effect in NiTi alloys used in medical devices. The ability to control and predict these phase transitions is fundamental for optimizing material performance in areas ranging from structural engineering to electronic and energy technologies. During a phase transition a material undergoes a transformation between different states of matter (e.g., solid to liquid) or structural phases (e.g., martensitic transformation) due to changes in external conditions such as temperature, pressure, or composition pt1 ; pt2 ; pt3 ; pt4 . These phase transitions are broadly classified into first-order and second-order types: first-order transitions involve latent heat and discontinuous changes in order parameters, as seen in melting, while second-order transitions occur continuously without latent heat, as in ferromagnetic to paramagnetic transitions. The nature of a phase transition is governed by underlying thermodynamic and atomistic mechanisms, which can be described using concepts from quantum physics, statistical mechanics and symmetry breaking. Understanding phase transitions is essential for predicting crystal properties and designing advanced functional materials. Figure 1: Machine learning-guided high-throughput approach for prediction of solid-solid phase transitions in inorganic crystals. a. Sketch of the employed computational workflow relying on DFT and vibrational free-energy calculations. b. Diagram of a graph encoding a crystal structure. The graph is characterized by nodes (dots), edges (lines), and a cutoff distance ( R , dashed line). c. Determination of temperature-induced phase transitions, both stable and metastable, based on free-energy curve calculations. Phase transitions can occur under a broad range of thermodynamic conditions, including variations in temperature, pressure, and external fields, making their precise characterization inherently complex. Crucially, the occurrence and nature of these transitions cannot always be predicted solely from a material’s chemical composition or atomic structure, as they often arise from intricate many-body interactions and thermal effects. This lack of predictability poses a significant challenge for the rational design of materials as, for example, unexpected phase transitions may hinder or alter their intended functionality. Thus, the difficulty in foreseeing phase behavior limits the discovery of novel physical phenomena and constrains the development of advanced technological applications. Despite the challenges in predicting phase transitions, several well-established theoretical and computational approaches have been developed to accurately model and anticipate solid-solid phase transformations met1 ; met2 ; met3 ; met4 . Methods such as free-energy calculations, Monte Carlo and molecular dynamics simulations enable the quantitative evaluation of phase stability by capturing the interplay between enthalpic and entropic contributions. However, achieving reliable predictions requires extremely high numerical accuracy, as small errors in energy estimations can lead to significant deviations in transition temperatures and stability trends. Consequently, these techniques demand rigorous computational protocols to ensure the precise determination of phase behavior in materials. Furthermore, the computational cost associated with free-energy calculations and molecular dynamics simulations is substantial, as these methods require extensive sampling and highly accurate energy and forces evaluations. Consequently, their application is often limited to specific case studies rather than large-scale, high-throughput screenings of extensive material datasets. This technical constraint significantly hinders the systematic exploration of phase behavior across diverse material families, limiting the discovery of novel physical phenomena and emergent functionalities. As a result, promising materials with transformative technological potential may remain unexplored due to the prohibitive computational demands of current predictive approaches. In this study, we have developed a computational framework that combines first-principles calculations with machine-learning techniques to overcome the limitations of conventional phase transition predictions. Specifically, we integrate density functional theory (DFT) energy calculations with uncertainty-aware machine-learning-assisted quasi-harmonic free-energy evaluations, enabling the systematic and high-throughput prediction of temperature-induced solid-solid phase transitions. By applying this approach to a dataset of approximately 50,000 inorganic materials, we have identified around 2,000 polymorphic phase transitions, some of which present potential technological relevance. Using these predictions, we propose novel materials for two distinct thermal management applications: solid-state cooling and thermal conductivity switching. Our findings not only identify promising phase transitions for energy-related technologies but also establish a scalable and efficient computational strategy for materials discovery under realistic finite-temperature conditions. II Computational Approach We present a novel high-throughput methodology for predicting temperature-induced solid-solid phase transitions in crystals. By integrating first-principles DFT calculations with uncertainty-aware, machine-learning-assisted quasi-harmonic free-energy evaluations, this approach enables a robust and reliable prediction of vibrational free energies. II.1 Quasi-harmonic free-energy formalism In the quasi-harmonic (QH) approximation qh1 ; qh2 ; qh4 , the Helmholtz free energy of a nonmetallic and nonmagnetic crystal, F , is expressed as a function of volume ( V ) and temperature ( T ) like: F(V,T)=E_{0}(V)+F_{\rm v}(V,T)~, (1) where E_{0} is the static lattice energy at zero temperature, and F_{\rm v} is the vibrational free energy (Fig. 1 a). The static lattice energy can be directly computed using first-principles methods such as density functional theory (DFT), E_{0}=E_{\rm DFT} . Similarly, the vibrational free energy can be obtained from the crystal’s phonon frequencies, \omega_{\bm{q}s}(V) , as: F_{\rm v}(V,T)=k_{B}T\sum_{\bm{q}s}\ln\left[2\sinh\left(\frac{\hbar\omega_{\bm{q}s}}{2k_{\rm B}T}\right)\right]~, (2) where the summation extends over the reciprocal-space vectors in the Brillouin zone, \bm{q} , and phonon branches, s . It is important to note that the QH approach is applicable only to vibrationally stable systems with strictly positive phonon frequencies, as imaginary (or negative) frequencies render F_{\rm v} ill-defined. Once F is known,