Open research questions in Machine Learning in Materials Science

What the literature leaves open

• In this work, we develop a ML potential to explore the Ag-Cu compositional space and identify stable AgxCu1−xGaSe2 con- figurations. The ML potential reproduces structural and elastic properties in good agreement with available ab initio and ex- perimental data and enables the calculation of thermodynamic observables and the phase diagram, consistent with prior com- putational studies (Refs. [15, 16]). Atomistic Monte Carlo simulations further show that co- herency constraints couple chemistry with mechanical proper- ties: a mismatch between local chemical composition and co- herent lattice constants generates a substantial elastic energy that can dominate the mixing thermodynamics. Accounting for this coherency-strain contribution leads to complete Ag-Cu miscibility under coherent boundary conditions. These findings explain the apparently conflicting experimen- tal reports on the Ag-Cu miscibility: the observed behavior depends on processing-induced mechanical boundary condi- tions. Co-evaporated thin films can sustain a largely coherent lattice over relevant length scales, favoring solid-solution for- mation during thermal processing, whereas ingot-based routes may more readily relax mismatch (e.g., via defects and local lattice accommodation), promoting demixing. Generally, this physical mechanism should be relevant to chalcopyrite alloys with small chemical driving forces for mix- ing. Moreover, alkali substitutions (Na, K) on Ag-Cu sites ex- hibit mixing enthalpies comparable to those in Figure 5(a) [68], suggesting that phase stability predictions can change qualita- tively when elastic contributions are neglected. 10 CRediT authorship contribution statement Vasilios Karanikolas: Writing – review and editing, Writ- ing – original draft, Resources, Visualization, Validation, Soft- ware, Methodology, Investigation, Formal analysis, Conceptu- alization. Delwin Perera: Writing – review and editing, Visualization, Validation, Software, Methodology, Investigation, Formal anal- ysis, Conceptualization. Linus Erhard: Visualization, Software, Methodology. Jochen Rohrer: Writing – review and editing, Validation, Software, Methodology, Investigation, Formal analysis. Karsten Albe: Writing – review and editing, Validation, Supervision, Resources, Project administration, Methodology, Funding acquisition, Formal analysis, Data curation, Concep- tualization.

  Chemo-mechanical coupling stabilizes mixed $\mathrm{Ag}_{x}\mathrm{Cu}_{1-x}\mathrm{GaSe}_{2}$ solar-cell absorbers: Insights from Monte-Carlo simulations assisted by ab initio informed machine-learning potentials · 2026

• While first-principles thermodynamic tools, including surface phase 411 and Pourbaix diagrams, 412 can provide helpful equilibrium baselines, their predictions are limited by the surface terminations, reconstructions, coverages, and metastable states included in the candidate sets.

  Atomistic Modeling of Chemical Disorder in Materials: Bridging Classical Methods and AI-Assisted Approaches · 2026

• Developing a scalable heat-treatment optimization framework that can handle variations in component size, geometry, and thermal mass. Capturing the complex interactions among thermal dynamics, phase transformations, and mechanical behavior. Ensuring that the optimization process remains within physically meaningful and industrially relevant operating conditions.

  Data-Driven Optimization of Size-Aware T6 Heat Treatment Parameters for A356 Aluminum Alloy · 2026 · DOI

• Further research is needed to explore the application of the framework to other alloys, - Investigation of the effects of other factors on the heat treatment response of A356 aluminum alloy

  Data-Driven Optimization of Size-Aware T6 Heat Treatment Parameters for A356 Aluminum Alloy · 2026 · DOI

• Follow-up studies could investigate molecular dynamics-based simulations. Further research could explore the application of MTPs to other materials and systems. The study suggests examining the impact of various aspects of the MTP training procedure.

  Using Machine-Learned Force Fields for Describing Heat-Transport-Related Quantities in AlGaN and Derived Materials · 2026 · DOI

• The high computational cost of first-principles methods like density functional theory. The need for accurate and efficient simulation of thermal transport in AlGaN and derived materials. The limitations of empirical force-field potentials in capturing the complex physics of these materials.

  Using Machine-Learned Force Fields for Describing Heat-Transport-Related Quantities in AlGaN and Derived Materials · 2026 · DOI

• Exploring the application of the machine-learning framework to other materials and applications, - Further assessment of the optimized structures to ensure the choice of stable lowest-energy polymorphs for each composition

  Stoichiometry-based machine learning enables discovery of new salt hydration reactions for thermochemical heat storage · 2026 · DOI

• further evaluation of TXL Fusion on larger and more diverse datasets - exploration of the application of TXL Fusion to other material discovery tasks - investigation of methods to mitigate the increase in overconfidence with TXL Fusion

  TXL Fusion: A Hybrid Machine Learning Framework Integrating Chemical Heuristics and Large Language Models for Topological Materials Discovery · 2026 · DOI

• The high cost of first-principles calculations and experimental validation limits the discovery of topological materials. Symmetry-based approaches face inherent limitations, such as the inability to capture complex material properties. Prior machine learning approaches lack interpretability and scalability, limiting their applicability to large-scale material discovery.

  TXL Fusion: A Hybrid Machine Learning Framework Integrating Chemical Heuristics and Large Language Models for Topological Materials Discovery · 2026 · DOI

• Extending the method to more complex systems, - Improving the accuracy of the P/B quantification scheme, - Integrating AutoEMX with other characterization techniques

  Accurate SEM-EDS quantification, automation, and machine learning enable high-throughput compositional characterization of powders · 2026 · DOI

• Investigating the application of RCI to larger molecules and more complex systems - Exploring the potential of RCI in combination with other machine learning approaches - Analyzing the impact of different hyperparameter settings on the performance of RCI

  Learning to Rank for Selected Configuration Interaction · 2026 · DOI

• Existing ML integrations suffer from an objective-loss mismatch. Conventional regression or classification approaches do not account for the relative importance ranking of determinants.

  Learning to Rank for Selected Configuration Interaction · 2026 · DOI

• the HSS dataset was assembled from heterogeneous literature sources, - microstructural reporting completeness varied substantially across samples, - direct removal of incomplete records would have further reduced the already limited sample size, - the sample size was limited (n = 73)

  Descriptor completion and cascade transfer for strength prediction in high strength steels · 2026 · DOI

• further integration of physical metallurgy with data-driven modelling, - examination of the role of descriptor completeness in other alloy systems, - investigation of the potential benefits of cascade prediction in other materials informatics applications

  Descriptor completion and cascade transfer for strength prediction in high strength steels · 2026 · DOI

• The inability of traditional methods to model non-local geometric effects in molecules. The limited computational efficiency of existing machine learning methods for designing energy materials. The need for a new approach that can efficiently model geometric dependencies on arbitrary length scales.

  Recent Advances in Machine Learning‐Assisted Multiscale Design of Energy Materials · 2024 · DOI

• RGB cues are used to resolve ambiguities when grayscale contrast is insufficient.

  Qumus: Realization of An Embodied AI Quantum Material Experimentalist · 2026

• Autonomous laboratories that combine quantum mechanical calculations, large language models, and experimental validations

  Recent Advances in Machine Learning‐Assisted Multiscale Design of Energy Materials · 2024 · DOI

• These results establish that target-space accuracy alone is insufficient: a learned initializer must also be compatible with the nonlinear solver trajectory, and its utility must be measured in the loop.

  Evaluating Predicted Densities, Hamiltonians, and Density Matrices as Periodic SCF Initializers · 2026

• These results show that held-out prediction errors alone are insufficient for selecting interatomic potentials for finite, surface-dominated nanostructures.

  Benchmarking Machine-Learning Interatomic Potentials for Dynamical Stability in Inorganic Semiconductor Nanocrystals: A CdSe Case Study · 2026

• Determining molecular structures from spectroscopic data remains fundamentally challenging because the inverse problem is intrinsically underdetermined: individual spectra are sparse, low-dimensional, and encode only partial structural evidence relative to the vast space of possible molecules.

  Hypothesis‐and‐Refinement Learning of Organic Structures From Multimodal Spectroscopic Data · 2026 · DOI

• However, conventional experiments are costly and time-consuming, resulting in limited data and challenges for accurate modeling under small-sample conditions.

  Enhancing bending load prediction of CaCO₃-filled polypropylene composites via data augmentation and ensemble machine learning · 2026 · DOI

• Accordingly, it remains unclear how and to what extent AI can accelerate innovation.

  Can artificial intelligence accelerate technological progress? Researchers' perspectives on AI in manufacturing and materials science · 2026 · DOI

• Current challenges, including data quality and coverage, DFT accuracy limitations, model interpretability, out of distribution generalisation, and the reproducibility of several landmark large-scale demonstrations, are addressed, and future directions toward autonomous materials discovery platforms integrating artificial intelligence, robotics, and quantum computing are outlined together with their present-day limitations and motivating open problems.

  Integration of density functional theory and machine learning for materials discovery in energy applications · 2026 · DOI

• The lack of long-range electrostatics is a key limitation of modern machine learning interatomic potentials (MLIPs), hindering reliable applications to interfaces, charge-transfer reactions, polar and ionic materials, and biomolecules.

  Long-range electrostatics for machine learning interatomic potentials is easier than we thought · 2026 · DOI

• Ion dynamics in carbonate electrolytes are fundamentally governed by the solvation power of cyclic solvents, a property whose quantification remains elusive because of the intricate competition between electronic and steric effects.

  Quantifying and Interpreting Solvation Power of Cyclic Carbonate by Chemical Calculation and Machine Learning · 2026 · DOI

Most-cited papers in Machine Learning in Materials Science

• Software update: The ORCA program system—Version 5.0 · Wiley Interdisciplinary Reviews Computational Molecular Science · 2022 · 4,894 citations
• Machine learning for molecular and materials science · Nature · 2018 · 4,557 citations
• Small data machine learning in materials science · npj Computational Materials · 2023 · 800 citations
• Learning local equivariant representations for large-scale atomistic dynamics · Nature Communications · 2023 · 662 citations
• Structured information extraction from scientific text with large language models · Nature Communications · 2024 · 594 citations
• Augmenting large language models with chemistry tools · Nature Machine Intelligence · 2024 · 565 citations
• A generative model for inorganic materials design · Nature · 2025 · 493 citations
• Interpretable machine learning for knowledge generation in heterogeneous catalysis · Nature Catalysis · 2022 · 417 citations
• A foundation model for atomistic materials chemistry · The Journal of Chemical Physics · 2025 · 361 citations
• Epik: p K a and Protonation State Prediction through Machine Learning · Journal of Chemical Theory and Computation · 2023 · 356 citations

Most recent work

• A Dual-Engine Artificial Intelligence Framework Accelerates Sustainable Aviation Fuel Component Synthesis · Journal of the American Chemical Society · 2026
• AI-driven biomaterial design: an intelligent closed loop from reverse design to biological response · Frontiers in Cell and Developmental Biology · 2026
• In situ Studies of Electrochemical Energy Conversion and Storage Technologies: From Materials, Intermediates, and Products to Surroundings · Nano-Micro Letters · 2026
• Artificial Intelligence-Driven Materials Design for Next-Generation Sustainable Energy Technologies · ACS Sustainable Chemistry & Engineering · 2026
• Artificial Intelligence Empowered New Materials: Discovery, Synthesis, Prediction to Validation · Nano-Micro Letters · 2026
• NEP89: universal neuroevolution potential for inorganic and organic materials across 89 elements · Nature Computational Science · 2026
• Advancing vapor pressure prediction: A machine learning approach with directed message passing neural networks · Journal of the Taiwan Institute of Chemical Engineers · 2026
• metatensor and metatomic : Foundational libraries for interoperable atomistic machine learning · The Journal of Chemical Physics · 2026
• Artificial Intelligence in Analytical Chemistry: Towards Yet Undiscovered Opportunities · Analytical Chemistry · 2026
• Machine Learning Interatomic Potentials for Energy Materials: Architectures, Training Strategies, and Applications · Advanced Energy Materials · 2026

Related topics in Materials Science